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 | |
d068b506 | 14 | * through a control-dependency in io_get_cqe (smp_store_release to |
1e84b97b SB |
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> |
ef98eb04 | 81 | #include <linux/tracehook.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
4ce8ad95 | 93 | #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8 |
65e19f54 | 94 | |
187f08c1 | 95 | /* only define max */ |
042b0d85 | 96 | #define IORING_MAX_FIXED_FILES (1U << 15) |
21b55dbc SG |
97 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
98 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 99 | |
187f08c1 | 100 | #define IO_RSRC_TAG_TABLE_SHIFT (PAGE_SHIFT - 3) |
2d091d62 PB |
101 | #define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT) |
102 | #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1) | |
103 | ||
489809e2 PB |
104 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
105 | ||
b16fed66 PB |
106 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
107 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
108 | IOSQE_BUFFER_SELECT) | |
c854357b PB |
109 | #define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \ |
110 | REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS) | |
b16fed66 | 111 | |
09899b19 PB |
112 | #define IO_TCTX_REFS_CACHE_NR (1U << 10) |
113 | ||
2b188cc1 JA |
114 | struct io_uring { |
115 | u32 head ____cacheline_aligned_in_smp; | |
116 | u32 tail ____cacheline_aligned_in_smp; | |
117 | }; | |
118 | ||
1e84b97b | 119 | /* |
75b28aff HV |
120 | * This data is shared with the application through the mmap at offsets |
121 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
122 | * |
123 | * The offsets to the member fields are published through struct | |
124 | * io_sqring_offsets when calling io_uring_setup. | |
125 | */ | |
75b28aff | 126 | struct io_rings { |
1e84b97b SB |
127 | /* |
128 | * Head and tail offsets into the ring; the offsets need to be | |
129 | * masked to get valid indices. | |
130 | * | |
75b28aff HV |
131 | * The kernel controls head of the sq ring and the tail of the cq ring, |
132 | * and the application controls tail of the sq ring and the head of the | |
133 | * cq ring. | |
1e84b97b | 134 | */ |
75b28aff | 135 | struct io_uring sq, cq; |
1e84b97b | 136 | /* |
75b28aff | 137 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
138 | * ring_entries - 1) |
139 | */ | |
75b28aff HV |
140 | u32 sq_ring_mask, cq_ring_mask; |
141 | /* Ring sizes (constant, power of 2) */ | |
142 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
143 | /* |
144 | * Number of invalid entries dropped by the kernel due to | |
145 | * invalid index stored in array | |
146 | * | |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application (i.e. get number of "new events" by comparing to | |
149 | * cached value). | |
150 | * | |
151 | * After a new SQ head value was read by the application this | |
152 | * counter includes all submissions that were dropped reaching | |
153 | * the new SQ head (and possibly more). | |
154 | */ | |
75b28aff | 155 | u32 sq_dropped; |
1e84b97b | 156 | /* |
0d9b5b3a | 157 | * Runtime SQ flags |
1e84b97b SB |
158 | * |
159 | * Written by the kernel, shouldn't be modified by the | |
160 | * application. | |
161 | * | |
162 | * The application needs a full memory barrier before checking | |
163 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
164 | */ | |
75b28aff | 165 | u32 sq_flags; |
0d9b5b3a SG |
166 | /* |
167 | * Runtime CQ flags | |
168 | * | |
169 | * Written by the application, shouldn't be modified by the | |
170 | * kernel. | |
171 | */ | |
fe7e3257 | 172 | u32 cq_flags; |
1e84b97b SB |
173 | /* |
174 | * Number of completion events lost because the queue was full; | |
175 | * this should be avoided by the application by making sure | |
0b4295b5 | 176 | * there are not more requests pending than there is space in |
1e84b97b SB |
177 | * the completion queue. |
178 | * | |
179 | * Written by the kernel, shouldn't be modified by the | |
180 | * application (i.e. get number of "new events" by comparing to | |
181 | * cached value). | |
182 | * | |
183 | * As completion events come in out of order this counter is not | |
184 | * ordered with any other data. | |
185 | */ | |
75b28aff | 186 | u32 cq_overflow; |
1e84b97b SB |
187 | /* |
188 | * Ring buffer of completion events. | |
189 | * | |
190 | * The kernel writes completion events fresh every time they are | |
191 | * produced, so the application is allowed to modify pending | |
192 | * entries. | |
193 | */ | |
75b28aff | 194 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
195 | }; |
196 | ||
45d189c6 PB |
197 | enum io_uring_cmd_flags { |
198 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 199 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
200 | }; |
201 | ||
edafccee JA |
202 | struct io_mapped_ubuf { |
203 | u64 ubuf; | |
4751f53d | 204 | u64 ubuf_end; |
edafccee | 205 | unsigned int nr_bvecs; |
de293938 | 206 | unsigned long acct_pages; |
41edf1a5 | 207 | struct bio_vec bvec[]; |
edafccee JA |
208 | }; |
209 | ||
50238531 BM |
210 | struct io_ring_ctx; |
211 | ||
6c2450ae PB |
212 | struct io_overflow_cqe { |
213 | struct io_uring_cqe cqe; | |
214 | struct list_head list; | |
215 | }; | |
216 | ||
a04b0ac0 PB |
217 | struct io_fixed_file { |
218 | /* file * with additional FFS_* flags */ | |
219 | unsigned long file_ptr; | |
220 | }; | |
221 | ||
269bbe5f BM |
222 | struct io_rsrc_put { |
223 | struct list_head list; | |
b60c8dce | 224 | u64 tag; |
50238531 BM |
225 | union { |
226 | void *rsrc; | |
227 | struct file *file; | |
bd54b6fe | 228 | struct io_mapped_ubuf *buf; |
50238531 | 229 | }; |
269bbe5f BM |
230 | }; |
231 | ||
aeca241b | 232 | struct io_file_table { |
042b0d85 | 233 | struct io_fixed_file *files; |
31b51510 JA |
234 | }; |
235 | ||
b895c9a6 | 236 | struct io_rsrc_node { |
05589553 XW |
237 | struct percpu_ref refs; |
238 | struct list_head node; | |
269bbe5f | 239 | struct list_head rsrc_list; |
b895c9a6 | 240 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 241 | struct llist_node llist; |
e297822b | 242 | bool done; |
05589553 XW |
243 | }; |
244 | ||
40ae0ff7 PB |
245 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
246 | ||
b895c9a6 | 247 | struct io_rsrc_data { |
05f3fb3c JA |
248 | struct io_ring_ctx *ctx; |
249 | ||
2d091d62 PB |
250 | u64 **tags; |
251 | unsigned int nr; | |
40ae0ff7 | 252 | rsrc_put_fn *do_put; |
3e942498 | 253 | atomic_t refs; |
05f3fb3c | 254 | struct completion done; |
8bad28d8 | 255 | bool quiesce; |
05f3fb3c JA |
256 | }; |
257 | ||
5a2e745d JA |
258 | struct io_buffer { |
259 | struct list_head list; | |
260 | __u64 addr; | |
d1f82808 | 261 | __u32 len; |
5a2e745d JA |
262 | __u16 bid; |
263 | }; | |
264 | ||
21b55dbc SG |
265 | struct io_restriction { |
266 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
267 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
268 | u8 sqe_flags_allowed; | |
269 | u8 sqe_flags_required; | |
7e84e1c7 | 270 | bool registered; |
21b55dbc SG |
271 | }; |
272 | ||
37d1e2e3 JA |
273 | enum { |
274 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
275 | IO_SQ_THREAD_SHOULD_PARK, | |
276 | }; | |
277 | ||
534ca6d6 JA |
278 | struct io_sq_data { |
279 | refcount_t refs; | |
9e138a48 | 280 | atomic_t park_pending; |
09a6f4ef | 281 | struct mutex lock; |
69fb2131 JA |
282 | |
283 | /* ctx's that are using this sqd */ | |
284 | struct list_head ctx_list; | |
69fb2131 | 285 | |
534ca6d6 JA |
286 | struct task_struct *thread; |
287 | struct wait_queue_head wait; | |
08369246 XW |
288 | |
289 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
290 | int sq_cpu; |
291 | pid_t task_pid; | |
5c2469e0 | 292 | pid_t task_tgid; |
37d1e2e3 JA |
293 | |
294 | unsigned long state; | |
37d1e2e3 | 295 | struct completion exited; |
534ca6d6 JA |
296 | }; |
297 | ||
6dd0be1e | 298 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 299 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 300 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 | 301 | |
a1ab7b35 PB |
302 | struct io_submit_link { |
303 | struct io_kiocb *head; | |
304 | struct io_kiocb *last; | |
305 | }; | |
306 | ||
258b29a9 PB |
307 | struct io_submit_state { |
308 | struct blk_plug plug; | |
a1ab7b35 | 309 | struct io_submit_link link; |
258b29a9 PB |
310 | |
311 | /* | |
312 | * io_kiocb alloc cache | |
313 | */ | |
bf019da7 | 314 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
315 | unsigned int free_reqs; |
316 | ||
317 | bool plug_started; | |
318 | ||
319 | /* | |
320 | * Batch completion logic | |
321 | */ | |
cd0ca2e0 PB |
322 | struct io_kiocb *compl_reqs[IO_COMPL_BATCH]; |
323 | unsigned int compl_nr; | |
324 | /* inline/task_work completion list, under ->uring_lock */ | |
325 | struct list_head free_list; | |
258b29a9 | 326 | |
258b29a9 PB |
327 | unsigned int ios_left; |
328 | }; | |
329 | ||
2b188cc1 | 330 | struct io_ring_ctx { |
b52ecf8c | 331 | /* const or read-mostly hot data */ |
2b188cc1 JA |
332 | struct { |
333 | struct percpu_ref refs; | |
2b188cc1 | 334 | |
b52ecf8c | 335 | struct io_rings *rings; |
2b188cc1 | 336 | unsigned int flags; |
e1d85334 | 337 | unsigned int compat: 1; |
e1d85334 RD |
338 | unsigned int drain_next: 1; |
339 | unsigned int eventfd_async: 1; | |
21b55dbc | 340 | unsigned int restricted: 1; |
f18ee4cf | 341 | unsigned int off_timeout_used: 1; |
10c66904 | 342 | unsigned int drain_active: 1; |
b52ecf8c | 343 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 344 | |
7f1129d2 | 345 | /* submission data */ |
b52ecf8c | 346 | struct { |
0499e582 PB |
347 | struct mutex uring_lock; |
348 | ||
75b28aff HV |
349 | /* |
350 | * Ring buffer of indices into array of io_uring_sqe, which is | |
351 | * mmapped by the application using the IORING_OFF_SQES offset. | |
352 | * | |
353 | * This indirection could e.g. be used to assign fixed | |
354 | * io_uring_sqe entries to operations and only submit them to | |
355 | * the queue when needed. | |
356 | * | |
357 | * The kernel modifies neither the indices array nor the entries | |
358 | * array. | |
359 | */ | |
360 | u32 *sq_array; | |
c7af47cf | 361 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
362 | unsigned cached_sq_head; |
363 | unsigned sq_entries; | |
de0617e4 | 364 | struct list_head defer_list; |
7f1129d2 PB |
365 | |
366 | /* | |
367 | * Fixed resources fast path, should be accessed only under | |
368 | * uring_lock, and updated through io_uring_register(2) | |
369 | */ | |
370 | struct io_rsrc_node *rsrc_node; | |
371 | struct io_file_table file_table; | |
372 | unsigned nr_user_files; | |
373 | unsigned nr_user_bufs; | |
374 | struct io_mapped_ubuf **user_bufs; | |
375 | ||
376 | struct io_submit_state submit_state; | |
5262f567 | 377 | struct list_head timeout_list; |
ef9dd637 | 378 | struct list_head ltimeout_list; |
1d7bb1d5 | 379 | struct list_head cq_overflow_list; |
7f1129d2 PB |
380 | struct xarray io_buffers; |
381 | struct xarray personalities; | |
382 | u32 pers_next; | |
383 | unsigned sq_thread_idle; | |
2b188cc1 JA |
384 | } ____cacheline_aligned_in_smp; |
385 | ||
d0acdee2 PB |
386 | /* IRQ completion list, under ->completion_lock */ |
387 | struct list_head locked_free_list; | |
388 | unsigned int locked_free_nr; | |
3c1a2ead | 389 | |
7c30f36a | 390 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
391 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
392 | ||
90554200 | 393 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 394 | struct list_head sqd_list; |
75b28aff | 395 | |
5ed7a37d PB |
396 | unsigned long check_cq_overflow; |
397 | ||
206aefde JA |
398 | struct { |
399 | unsigned cached_cq_tail; | |
400 | unsigned cq_entries; | |
0499e582 | 401 | struct eventfd_ctx *cq_ev_fd; |
311997b3 | 402 | struct wait_queue_head poll_wait; |
0499e582 PB |
403 | struct wait_queue_head cq_wait; |
404 | unsigned cq_extra; | |
405 | atomic_t cq_timeouts; | |
206aefde | 406 | struct fasync_struct *cq_fasync; |
0499e582 | 407 | unsigned cq_last_tm_flush; |
206aefde | 408 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 409 | |
2b188cc1 JA |
410 | struct { |
411 | spinlock_t completion_lock; | |
e94f141b | 412 | |
89850fce JA |
413 | spinlock_t timeout_lock; |
414 | ||
def596e9 | 415 | /* |
540e32a0 | 416 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
417 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
418 | * For SQPOLL, only the single threaded io_sq_thread() will | |
419 | * manipulate the list, hence no extra locking is needed there. | |
420 | */ | |
540e32a0 | 421 | struct list_head iopoll_list; |
78076bb6 JA |
422 | struct hlist_head *cancel_hash; |
423 | unsigned cancel_hash_bits; | |
915b3dde | 424 | bool poll_multi_queue; |
2b188cc1 | 425 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 426 | |
21b55dbc | 427 | struct io_restriction restrictions; |
3c1a2ead | 428 | |
b13a8918 PB |
429 | /* slow path rsrc auxilary data, used by update/register */ |
430 | struct { | |
431 | struct io_rsrc_node *rsrc_backup_node; | |
432 | struct io_mapped_ubuf *dummy_ubuf; | |
433 | struct io_rsrc_data *file_data; | |
434 | struct io_rsrc_data *buf_data; | |
435 | ||
436 | struct delayed_work rsrc_put_work; | |
437 | struct llist_head rsrc_put_llist; | |
438 | struct list_head rsrc_ref_list; | |
439 | spinlock_t rsrc_ref_lock; | |
440 | }; | |
441 | ||
3c1a2ead | 442 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
443 | struct { |
444 | #if defined(CONFIG_UNIX) | |
445 | struct socket *ring_sock; | |
446 | #endif | |
447 | /* hashed buffered write serialization */ | |
448 | struct io_wq_hash *hash_map; | |
449 | ||
450 | /* Only used for accounting purposes */ | |
451 | struct user_struct *user; | |
452 | struct mm_struct *mm_account; | |
453 | ||
454 | /* ctx exit and cancelation */ | |
9011bf9a PB |
455 | struct llist_head fallback_llist; |
456 | struct delayed_work fallback_work; | |
b986af7e PB |
457 | struct work_struct exit_work; |
458 | struct list_head tctx_list; | |
459 | struct completion ref_comp; | |
460 | }; | |
2b188cc1 JA |
461 | }; |
462 | ||
53e043b2 SM |
463 | struct io_uring_task { |
464 | /* submission side */ | |
09899b19 | 465 | int cached_refs; |
53e043b2 SM |
466 | struct xarray xa; |
467 | struct wait_queue_head wait; | |
ee53fb2b SM |
468 | const struct io_ring_ctx *last; |
469 | struct io_wq *io_wq; | |
53e043b2 | 470 | struct percpu_counter inflight; |
b303fe2e | 471 | atomic_t inflight_tracked; |
53e043b2 | 472 | atomic_t in_idle; |
53e043b2 SM |
473 | |
474 | spinlock_t task_lock; | |
475 | struct io_wq_work_list task_list; | |
53e043b2 | 476 | struct callback_head task_work; |
6294f368 | 477 | bool task_running; |
53e043b2 SM |
478 | }; |
479 | ||
09bb8394 JA |
480 | /* |
481 | * First field must be the file pointer in all the | |
482 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
483 | */ | |
221c5eb2 JA |
484 | struct io_poll_iocb { |
485 | struct file *file; | |
018043be | 486 | struct wait_queue_head *head; |
221c5eb2 | 487 | __poll_t events; |
8c838788 | 488 | bool done; |
221c5eb2 | 489 | bool canceled; |
392edb45 | 490 | struct wait_queue_entry wait; |
221c5eb2 JA |
491 | }; |
492 | ||
9d805892 | 493 | struct io_poll_update { |
018043be | 494 | struct file *file; |
9d805892 PB |
495 | u64 old_user_data; |
496 | u64 new_user_data; | |
497 | __poll_t events; | |
b69de288 JA |
498 | bool update_events; |
499 | bool update_user_data; | |
018043be PB |
500 | }; |
501 | ||
b5dba59e JA |
502 | struct io_close { |
503 | struct file *file; | |
b5dba59e JA |
504 | int fd; |
505 | }; | |
506 | ||
ad8a48ac JA |
507 | struct io_timeout_data { |
508 | struct io_kiocb *req; | |
509 | struct hrtimer timer; | |
510 | struct timespec64 ts; | |
511 | enum hrtimer_mode mode; | |
50c1df2b | 512 | u32 flags; |
ad8a48ac JA |
513 | }; |
514 | ||
8ed8d3c3 JA |
515 | struct io_accept { |
516 | struct file *file; | |
517 | struct sockaddr __user *addr; | |
518 | int __user *addr_len; | |
519 | int flags; | |
aaa4db12 | 520 | u32 file_slot; |
09952e3e | 521 | unsigned long nofile; |
8ed8d3c3 JA |
522 | }; |
523 | ||
524 | struct io_sync { | |
525 | struct file *file; | |
526 | loff_t len; | |
527 | loff_t off; | |
528 | int flags; | |
d63d1b5e | 529 | int mode; |
8ed8d3c3 JA |
530 | }; |
531 | ||
fbf23849 JA |
532 | struct io_cancel { |
533 | struct file *file; | |
534 | u64 addr; | |
535 | }; | |
536 | ||
b29472ee JA |
537 | struct io_timeout { |
538 | struct file *file; | |
bfe68a22 PB |
539 | u32 off; |
540 | u32 target_seq; | |
135fcde8 | 541 | struct list_head list; |
90cd7e42 PB |
542 | /* head of the link, used by linked timeouts only */ |
543 | struct io_kiocb *head; | |
89b263f6 JA |
544 | /* for linked completions */ |
545 | struct io_kiocb *prev; | |
b29472ee JA |
546 | }; |
547 | ||
0bdf7a2d PB |
548 | struct io_timeout_rem { |
549 | struct file *file; | |
550 | u64 addr; | |
9c8e11b3 PB |
551 | |
552 | /* timeout update */ | |
553 | struct timespec64 ts; | |
554 | u32 flags; | |
f1042b6c | 555 | bool ltimeout; |
0bdf7a2d PB |
556 | }; |
557 | ||
9adbd45d JA |
558 | struct io_rw { |
559 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
560 | struct kiocb kiocb; | |
561 | u64 addr; | |
562 | u64 len; | |
563 | }; | |
564 | ||
3fbb51c1 JA |
565 | struct io_connect { |
566 | struct file *file; | |
567 | struct sockaddr __user *addr; | |
568 | int addr_len; | |
569 | }; | |
570 | ||
e47293fd JA |
571 | struct io_sr_msg { |
572 | struct file *file; | |
fddaface | 573 | union { |
4af3417a PB |
574 | struct compat_msghdr __user *umsg_compat; |
575 | struct user_msghdr __user *umsg; | |
576 | void __user *buf; | |
fddaface | 577 | }; |
e47293fd | 578 | int msg_flags; |
bcda7baa | 579 | int bgid; |
fddaface | 580 | size_t len; |
bcda7baa | 581 | struct io_buffer *kbuf; |
e47293fd JA |
582 | }; |
583 | ||
15b71abe JA |
584 | struct io_open { |
585 | struct file *file; | |
586 | int dfd; | |
b9445598 | 587 | u32 file_slot; |
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; | |
38134ada | 634 | __u32 len; |
ddf0322d JA |
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 | ||
e34a02dc DK |
670 | struct io_mkdir { |
671 | struct file *file; | |
672 | int dfd; | |
673 | umode_t mode; | |
674 | struct filename *filename; | |
675 | }; | |
676 | ||
7a8721f8 DK |
677 | struct io_symlink { |
678 | struct file *file; | |
679 | int new_dfd; | |
680 | struct filename *oldpath; | |
681 | struct filename *newpath; | |
682 | }; | |
683 | ||
cf30da90 DK |
684 | struct io_hardlink { |
685 | struct file *file; | |
686 | int old_dfd; | |
687 | int new_dfd; | |
688 | struct filename *oldpath; | |
689 | struct filename *newpath; | |
690 | int flags; | |
691 | }; | |
692 | ||
3ca405eb PB |
693 | struct io_completion { |
694 | struct file *file; | |
8c3f9cd1 | 695 | u32 cflags; |
3ca405eb PB |
696 | }; |
697 | ||
f499a021 JA |
698 | struct io_async_connect { |
699 | struct sockaddr_storage address; | |
700 | }; | |
701 | ||
03b1230c JA |
702 | struct io_async_msghdr { |
703 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
704 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
705 | struct iovec *free_iov; | |
03b1230c JA |
706 | struct sockaddr __user *uaddr; |
707 | struct msghdr msg; | |
b537916c | 708 | struct sockaddr_storage addr; |
03b1230c JA |
709 | }; |
710 | ||
f67676d1 JA |
711 | struct io_async_rw { |
712 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
713 | const struct iovec *free_iovec; |
714 | struct iov_iter iter; | |
227c0c96 | 715 | size_t bytes_done; |
bcf5a063 | 716 | struct wait_page_queue wpq; |
f67676d1 JA |
717 | }; |
718 | ||
6b47ee6e PB |
719 | enum { |
720 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
721 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
722 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
723 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
724 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 725 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 726 | |
dddca226 | 727 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 728 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
729 | REQ_F_INFLIGHT_BIT, |
730 | REQ_F_CUR_POS_BIT, | |
731 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 732 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 733 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 734 | REQ_F_POLLED_BIT, |
bcda7baa | 735 | REQ_F_BUFFER_SELECTED_BIT, |
e342c807 | 736 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 737 | REQ_F_REISSUE_BIT, |
8c130827 | 738 | REQ_F_DONT_REISSUE_BIT, |
b8e64b53 | 739 | REQ_F_CREDS_BIT, |
20e60a38 | 740 | REQ_F_REFCOUNT_BIT, |
4d13d1a4 | 741 | REQ_F_ARM_LTIMEOUT_BIT, |
7b29f92d | 742 | /* keep async read/write and isreg together and in order */ |
b191e2df PB |
743 | REQ_F_NOWAIT_READ_BIT, |
744 | REQ_F_NOWAIT_WRITE_BIT, | |
7b29f92d | 745 | REQ_F_ISREG_BIT, |
84557871 JA |
746 | |
747 | /* not a real bit, just to check we're not overflowing the space */ | |
748 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
749 | }; |
750 | ||
751 | enum { | |
752 | /* ctx owns file */ | |
753 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
754 | /* drain existing IO first */ | |
755 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
756 | /* linked sqes */ | |
757 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
758 | /* doesn't sever on completion < 0 */ | |
759 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
760 | /* IOSQE_ASYNC */ | |
761 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
762 | /* IOSQE_BUFFER_SELECT */ |
763 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 764 | |
6b47ee6e | 765 | /* fail rest of links */ |
93d2bcd2 | 766 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 767 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
768 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
769 | /* read/write uses file position */ | |
770 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
771 | /* must not punt to workers */ | |
772 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 773 | /* has or had linked timeout */ |
6b47ee6e | 774 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
775 | /* needs cleanup */ |
776 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
777 | /* already went through poll handler */ |
778 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
779 | /* buffer already selected */ |
780 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
e342c807 PB |
781 | /* completion is deferred through io_comp_state */ |
782 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
783 | /* caller should reissue async */ |
784 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
785 | /* don't attempt request reissue, see io_rw_reissue() */ |
786 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d | 787 | /* supports async reads */ |
b191e2df | 788 | REQ_F_NOWAIT_READ = BIT(REQ_F_NOWAIT_READ_BIT), |
7b29f92d | 789 | /* supports async writes */ |
b191e2df | 790 | REQ_F_NOWAIT_WRITE = BIT(REQ_F_NOWAIT_WRITE_BIT), |
7b29f92d JA |
791 | /* regular file */ |
792 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
793 | /* has creds assigned */ |
794 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
20e60a38 PB |
795 | /* skip refcounting if not set */ |
796 | REQ_F_REFCOUNT = BIT(REQ_F_REFCOUNT_BIT), | |
4d13d1a4 PB |
797 | /* there is a linked timeout that has to be armed */ |
798 | REQ_F_ARM_LTIMEOUT = BIT(REQ_F_ARM_LTIMEOUT_BIT), | |
d7718a9d JA |
799 | }; |
800 | ||
801 | struct async_poll { | |
802 | struct io_poll_iocb poll; | |
807abcb0 | 803 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
804 | }; |
805 | ||
f237c30a | 806 | typedef void (*io_req_tw_func_t)(struct io_kiocb *req, bool *locked); |
5b0a6acc | 807 | |
7cbf1722 | 808 | struct io_task_work { |
5b0a6acc PB |
809 | union { |
810 | struct io_wq_work_node node; | |
811 | struct llist_node fallback_node; | |
812 | }; | |
813 | io_req_tw_func_t func; | |
7cbf1722 JA |
814 | }; |
815 | ||
992da01a PB |
816 | enum { |
817 | IORING_RSRC_FILE = 0, | |
818 | IORING_RSRC_BUFFER = 1, | |
819 | }; | |
820 | ||
09bb8394 JA |
821 | /* |
822 | * NOTE! Each of the iocb union members has the file pointer | |
823 | * as the first entry in their struct definition. So you can | |
824 | * access the file pointer through any of the sub-structs, | |
825 | * or directly as just 'ki_filp' in this struct. | |
826 | */ | |
2b188cc1 | 827 | struct io_kiocb { |
221c5eb2 | 828 | union { |
09bb8394 | 829 | struct file *file; |
9adbd45d | 830 | struct io_rw rw; |
221c5eb2 | 831 | struct io_poll_iocb poll; |
9d805892 | 832 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
833 | struct io_accept accept; |
834 | struct io_sync sync; | |
fbf23849 | 835 | struct io_cancel cancel; |
b29472ee | 836 | struct io_timeout timeout; |
0bdf7a2d | 837 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 838 | struct io_connect connect; |
e47293fd | 839 | struct io_sr_msg sr_msg; |
15b71abe | 840 | struct io_open open; |
b5dba59e | 841 | struct io_close close; |
269bbe5f | 842 | struct io_rsrc_update rsrc_update; |
4840e418 | 843 | struct io_fadvise fadvise; |
c1ca757b | 844 | struct io_madvise madvise; |
3e4827b0 | 845 | struct io_epoll epoll; |
7d67af2c | 846 | struct io_splice splice; |
ddf0322d | 847 | struct io_provide_buf pbuf; |
1d9e1288 | 848 | struct io_statx statx; |
36f4fa68 | 849 | struct io_shutdown shutdown; |
80a261fd | 850 | struct io_rename rename; |
14a1143b | 851 | struct io_unlink unlink; |
e34a02dc | 852 | struct io_mkdir mkdir; |
7a8721f8 | 853 | struct io_symlink symlink; |
cf30da90 | 854 | struct io_hardlink hardlink; |
3ca405eb PB |
855 | /* use only after cleaning per-op data, see io_clean_op() */ |
856 | struct io_completion compl; | |
221c5eb2 | 857 | }; |
2b188cc1 | 858 | |
e8c2bc1f JA |
859 | /* opcode allocated if it needs to store data for async defer */ |
860 | void *async_data; | |
d625c6ee | 861 | u8 opcode; |
65a6543d XW |
862 | /* polled IO has completed */ |
863 | u8 iopoll_completed; | |
2b188cc1 | 864 | |
4f4eeba8 | 865 | u16 buf_index; |
9cf7c104 | 866 | u32 result; |
4f4eeba8 | 867 | |
010e8e6b PB |
868 | struct io_ring_ctx *ctx; |
869 | unsigned int flags; | |
abc54d63 | 870 | atomic_t refs; |
010e8e6b PB |
871 | struct task_struct *task; |
872 | u64 user_data; | |
d7718a9d | 873 | |
f2f87370 | 874 | struct io_kiocb *link; |
269bbe5f | 875 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 876 | |
b303fe2e | 877 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 878 | struct list_head inflight_entry; |
5b0a6acc | 879 | struct io_task_work io_task_work; |
010e8e6b PB |
880 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
881 | struct hlist_node hash_node; | |
882 | struct async_poll *apoll; | |
883 | struct io_wq_work work; | |
fe7e3257 | 884 | const struct cred *creds; |
c10d1f98 | 885 | |
eae071c9 PB |
886 | /* store used ubuf, so we can prevent reloading */ |
887 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 888 | }; |
05589553 | 889 | |
13bf43f5 PB |
890 | struct io_tctx_node { |
891 | struct list_head ctx_node; | |
892 | struct task_struct *task; | |
13bf43f5 PB |
893 | struct io_ring_ctx *ctx; |
894 | }; | |
895 | ||
27dc8338 PB |
896 | struct io_defer_entry { |
897 | struct list_head list; | |
898 | struct io_kiocb *req; | |
9cf7c104 | 899 | u32 seq; |
2b188cc1 JA |
900 | }; |
901 | ||
d3656344 | 902 | struct io_op_def { |
d3656344 JA |
903 | /* needs req->file assigned */ |
904 | unsigned needs_file : 1; | |
d3656344 JA |
905 | /* hash wq insertion if file is a regular file */ |
906 | unsigned hash_reg_file : 1; | |
907 | /* unbound wq insertion if file is a non-regular file */ | |
908 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
909 | /* opcode is not supported by this kernel */ |
910 | unsigned not_supported : 1; | |
8a72758c JA |
911 | /* set if opcode supports polled "wait" */ |
912 | unsigned pollin : 1; | |
913 | unsigned pollout : 1; | |
bcda7baa JA |
914 | /* op supports buffer selection */ |
915 | unsigned buffer_select : 1; | |
26f0505a PB |
916 | /* do prep async if is going to be punted */ |
917 | unsigned needs_async_setup : 1; | |
27926b68 JA |
918 | /* should block plug */ |
919 | unsigned plug : 1; | |
e8c2bc1f JA |
920 | /* size of async data needed, if any */ |
921 | unsigned short async_size; | |
d3656344 JA |
922 | }; |
923 | ||
0918682b | 924 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
925 | [IORING_OP_NOP] = {}, |
926 | [IORING_OP_READV] = { | |
d3656344 JA |
927 | .needs_file = 1, |
928 | .unbound_nonreg_file = 1, | |
8a72758c | 929 | .pollin = 1, |
4d954c25 | 930 | .buffer_select = 1, |
26f0505a | 931 | .needs_async_setup = 1, |
27926b68 | 932 | .plug = 1, |
e8c2bc1f | 933 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 934 | }, |
0463b6c5 | 935 | [IORING_OP_WRITEV] = { |
d3656344 JA |
936 | .needs_file = 1, |
937 | .hash_reg_file = 1, | |
938 | .unbound_nonreg_file = 1, | |
8a72758c | 939 | .pollout = 1, |
26f0505a | 940 | .needs_async_setup = 1, |
27926b68 | 941 | .plug = 1, |
e8c2bc1f | 942 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 943 | }, |
0463b6c5 | 944 | [IORING_OP_FSYNC] = { |
d3656344 JA |
945 | .needs_file = 1, |
946 | }, | |
0463b6c5 | 947 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
948 | .needs_file = 1, |
949 | .unbound_nonreg_file = 1, | |
8a72758c | 950 | .pollin = 1, |
27926b68 | 951 | .plug = 1, |
e8c2bc1f | 952 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 953 | }, |
0463b6c5 | 954 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
955 | .needs_file = 1, |
956 | .hash_reg_file = 1, | |
957 | .unbound_nonreg_file = 1, | |
8a72758c | 958 | .pollout = 1, |
27926b68 | 959 | .plug = 1, |
e8c2bc1f | 960 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 961 | }, |
0463b6c5 | 962 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
963 | .needs_file = 1, |
964 | .unbound_nonreg_file = 1, | |
965 | }, | |
0463b6c5 PB |
966 | [IORING_OP_POLL_REMOVE] = {}, |
967 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
968 | .needs_file = 1, |
969 | }, | |
0463b6c5 | 970 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
971 | .needs_file = 1, |
972 | .unbound_nonreg_file = 1, | |
8a72758c | 973 | .pollout = 1, |
26f0505a | 974 | .needs_async_setup = 1, |
e8c2bc1f | 975 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 976 | }, |
0463b6c5 | 977 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
978 | .needs_file = 1, |
979 | .unbound_nonreg_file = 1, | |
8a72758c | 980 | .pollin = 1, |
52de1fe1 | 981 | .buffer_select = 1, |
26f0505a | 982 | .needs_async_setup = 1, |
e8c2bc1f | 983 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 984 | }, |
0463b6c5 | 985 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 986 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 987 | }, |
9c8e11b3 PB |
988 | [IORING_OP_TIMEOUT_REMOVE] = { |
989 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 990 | }, |
0463b6c5 | 991 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
992 | .needs_file = 1, |
993 | .unbound_nonreg_file = 1, | |
8a72758c | 994 | .pollin = 1, |
d3656344 | 995 | }, |
0463b6c5 PB |
996 | [IORING_OP_ASYNC_CANCEL] = {}, |
997 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 998 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 999 | }, |
0463b6c5 | 1000 | [IORING_OP_CONNECT] = { |
d3656344 JA |
1001 | .needs_file = 1, |
1002 | .unbound_nonreg_file = 1, | |
8a72758c | 1003 | .pollout = 1, |
26f0505a | 1004 | .needs_async_setup = 1, |
e8c2bc1f | 1005 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 1006 | }, |
0463b6c5 | 1007 | [IORING_OP_FALLOCATE] = { |
d3656344 | 1008 | .needs_file = 1, |
d3656344 | 1009 | }, |
44526bed JA |
1010 | [IORING_OP_OPENAT] = {}, |
1011 | [IORING_OP_CLOSE] = {}, | |
1012 | [IORING_OP_FILES_UPDATE] = {}, | |
1013 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 1014 | [IORING_OP_READ] = { |
3a6820f2 JA |
1015 | .needs_file = 1, |
1016 | .unbound_nonreg_file = 1, | |
8a72758c | 1017 | .pollin = 1, |
bcda7baa | 1018 | .buffer_select = 1, |
27926b68 | 1019 | .plug = 1, |
e8c2bc1f | 1020 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1021 | }, |
0463b6c5 | 1022 | [IORING_OP_WRITE] = { |
3a6820f2 | 1023 | .needs_file = 1, |
7b3188e7 | 1024 | .hash_reg_file = 1, |
3a6820f2 | 1025 | .unbound_nonreg_file = 1, |
8a72758c | 1026 | .pollout = 1, |
27926b68 | 1027 | .plug = 1, |
e8c2bc1f | 1028 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1029 | }, |
0463b6c5 | 1030 | [IORING_OP_FADVISE] = { |
4840e418 | 1031 | .needs_file = 1, |
c1ca757b | 1032 | }, |
44526bed | 1033 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1034 | [IORING_OP_SEND] = { |
fddaface JA |
1035 | .needs_file = 1, |
1036 | .unbound_nonreg_file = 1, | |
8a72758c | 1037 | .pollout = 1, |
fddaface | 1038 | }, |
0463b6c5 | 1039 | [IORING_OP_RECV] = { |
fddaface JA |
1040 | .needs_file = 1, |
1041 | .unbound_nonreg_file = 1, | |
8a72758c | 1042 | .pollin = 1, |
bcda7baa | 1043 | .buffer_select = 1, |
fddaface | 1044 | }, |
0463b6c5 | 1045 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1046 | }, |
3e4827b0 JA |
1047 | [IORING_OP_EPOLL_CTL] = { |
1048 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1049 | }, |
7d67af2c PB |
1050 | [IORING_OP_SPLICE] = { |
1051 | .needs_file = 1, | |
1052 | .hash_reg_file = 1, | |
1053 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1054 | }, |
1055 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1056 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1057 | [IORING_OP_TEE] = { |
1058 | .needs_file = 1, | |
1059 | .hash_reg_file = 1, | |
1060 | .unbound_nonreg_file = 1, | |
1061 | }, | |
36f4fa68 JA |
1062 | [IORING_OP_SHUTDOWN] = { |
1063 | .needs_file = 1, | |
1064 | }, | |
44526bed JA |
1065 | [IORING_OP_RENAMEAT] = {}, |
1066 | [IORING_OP_UNLINKAT] = {}, | |
e34a02dc | 1067 | [IORING_OP_MKDIRAT] = {}, |
7a8721f8 | 1068 | [IORING_OP_SYMLINKAT] = {}, |
cf30da90 | 1069 | [IORING_OP_LINKAT] = {}, |
d3656344 JA |
1070 | }; |
1071 | ||
0756a869 PB |
1072 | /* requests with any of those set should undergo io_disarm_next() */ |
1073 | #define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL) | |
1074 | ||
7a612350 | 1075 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1076 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1077 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1078 | struct task_struct *task, | |
3dd0c97a | 1079 | bool cancel_all); |
78cc687b | 1080 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
1ffc5422 | 1081 | |
d4d19c19 PB |
1082 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1083 | long res, unsigned int cflags); | |
ec9c02ad | 1084 | static void io_put_req(struct io_kiocb *req); |
91c2f697 | 1085 | static void io_put_req_deferred(struct io_kiocb *req); |
c7dae4ba | 1086 | static void io_dismantle_req(struct io_kiocb *req); |
94ae5e77 | 1087 | static void io_queue_linked_timeout(struct io_kiocb *req); |
fdecb662 | 1088 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1089 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1090 | unsigned nr_args); |
68fb8979 | 1091 | static void io_clean_op(struct io_kiocb *req); |
ac177053 | 1092 | static struct file *io_file_get(struct io_ring_ctx *ctx, |
8371adf5 | 1093 | struct io_kiocb *req, int fd, bool fixed); |
c5eef2b9 | 1094 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1095 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1096 | |
907d1df3 | 1097 | static void io_req_task_queue(struct io_kiocb *req); |
2a2758f2 | 1098 | static void io_submit_flush_completions(struct io_ring_ctx *ctx); |
179ae0d1 | 1099 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1100 | |
b9445598 PB |
1101 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
1102 | unsigned int issue_flags, u32 slot_index); | |
f1042b6c | 1103 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); |
b9445598 | 1104 | |
2b188cc1 JA |
1105 | static struct kmem_cache *req_cachep; |
1106 | ||
0918682b | 1107 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1108 | |
1109 | struct sock *io_uring_get_socket(struct file *file) | |
1110 | { | |
1111 | #if defined(CONFIG_UNIX) | |
1112 | if (file->f_op == &io_uring_fops) { | |
1113 | struct io_ring_ctx *ctx = file->private_data; | |
1114 | ||
1115 | return ctx->ring_sock->sk; | |
1116 | } | |
1117 | #endif | |
1118 | return NULL; | |
1119 | } | |
1120 | EXPORT_SYMBOL(io_uring_get_socket); | |
1121 | ||
f237c30a PB |
1122 | static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) |
1123 | { | |
1124 | if (!*locked) { | |
1125 | mutex_lock(&ctx->uring_lock); | |
1126 | *locked = true; | |
1127 | } | |
1128 | } | |
1129 | ||
f2f87370 PB |
1130 | #define io_for_each_link(pos, head) \ |
1131 | for (pos = (head); pos; pos = pos->link) | |
1132 | ||
21c843d5 PB |
1133 | /* |
1134 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1135 | * see commit f958d7b528b1 for details. | |
1136 | */ | |
1137 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1138 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1139 | ||
1140 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) | |
1141 | { | |
20e60a38 | 1142 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1143 | return atomic_inc_not_zero(&req->refs); |
1144 | } | |
1145 | ||
21c843d5 PB |
1146 | static inline bool req_ref_put_and_test(struct io_kiocb *req) |
1147 | { | |
20e60a38 PB |
1148 | if (likely(!(req->flags & REQ_F_REFCOUNT))) |
1149 | return true; | |
1150 | ||
21c843d5 PB |
1151 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1152 | return atomic_dec_and_test(&req->refs); | |
1153 | } | |
1154 | ||
1155 | static inline void req_ref_put(struct io_kiocb *req) | |
1156 | { | |
20e60a38 | 1157 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1158 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
1159 | } | |
1160 | ||
1161 | static inline void req_ref_get(struct io_kiocb *req) | |
1162 | { | |
20e60a38 | 1163 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1164 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1165 | atomic_inc(&req->refs); | |
1166 | } | |
1167 | ||
48dcd38d | 1168 | static inline void __io_req_set_refcount(struct io_kiocb *req, int nr) |
20e60a38 PB |
1169 | { |
1170 | if (!(req->flags & REQ_F_REFCOUNT)) { | |
1171 | req->flags |= REQ_F_REFCOUNT; | |
48dcd38d | 1172 | atomic_set(&req->refs, nr); |
20e60a38 PB |
1173 | } |
1174 | } | |
1175 | ||
48dcd38d PB |
1176 | static inline void io_req_set_refcount(struct io_kiocb *req) |
1177 | { | |
1178 | __io_req_set_refcount(req, 1); | |
1179 | } | |
1180 | ||
b895c9a6 | 1181 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1182 | { |
1183 | struct io_ring_ctx *ctx = req->ctx; | |
1184 | ||
269bbe5f | 1185 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1186 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1187 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1188 | } |
1189 | } | |
1190 | ||
f70865db PB |
1191 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1192 | { | |
1193 | bool got = percpu_ref_tryget(ref); | |
1194 | ||
1195 | /* already at zero, wait for ->release() */ | |
1196 | if (!got) | |
1197 | wait_for_completion(compl); | |
1198 | percpu_ref_resurrect(ref); | |
1199 | if (got) | |
1200 | percpu_ref_put(ref); | |
1201 | } | |
1202 | ||
3dd0c97a PB |
1203 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1204 | bool cancel_all) | |
08d23634 PB |
1205 | { |
1206 | struct io_kiocb *req; | |
1207 | ||
68207680 | 1208 | if (task && head->task != task) |
08d23634 | 1209 | return false; |
3dd0c97a | 1210 | if (cancel_all) |
08d23634 PB |
1211 | return true; |
1212 | ||
1213 | io_for_each_link(req, head) { | |
b05a1bcd | 1214 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1215 | return true; |
08d23634 PB |
1216 | } |
1217 | return false; | |
1218 | } | |
1219 | ||
93d2bcd2 | 1220 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1221 | { |
93d2bcd2 | 1222 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1223 | } |
4a38aed2 | 1224 | |
a8295b98 HX |
1225 | static inline void req_fail_link_node(struct io_kiocb *req, int res) |
1226 | { | |
1227 | req_set_fail(req); | |
1228 | req->result = res; | |
1229 | } | |
1230 | ||
2b188cc1 JA |
1231 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1232 | { | |
1233 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1234 | ||
0f158b4c | 1235 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1236 | } |
1237 | ||
8eb7e2d0 PB |
1238 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1239 | { | |
1240 | return !req->timeout.off; | |
1241 | } | |
1242 | ||
f56165e6 PB |
1243 | static void io_fallback_req_func(struct work_struct *work) |
1244 | { | |
1245 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
1246 | fallback_work.work); | |
1247 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
1248 | struct io_kiocb *req, *tmp; | |
f237c30a | 1249 | bool locked = false; |
f56165e6 PB |
1250 | |
1251 | percpu_ref_get(&ctx->refs); | |
1252 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) | |
f237c30a | 1253 | req->io_task_work.func(req, &locked); |
5636c00d | 1254 | |
f237c30a PB |
1255 | if (locked) { |
1256 | if (ctx->submit_state.compl_nr) | |
1257 | io_submit_flush_completions(ctx); | |
1258 | mutex_unlock(&ctx->uring_lock); | |
1259 | } | |
f56165e6 | 1260 | percpu_ref_put(&ctx->refs); |
f237c30a | 1261 | |
f56165e6 PB |
1262 | } |
1263 | ||
2b188cc1 JA |
1264 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1265 | { | |
1266 | struct io_ring_ctx *ctx; | |
78076bb6 | 1267 | int hash_bits; |
2b188cc1 JA |
1268 | |
1269 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1270 | if (!ctx) | |
1271 | return NULL; | |
1272 | ||
78076bb6 JA |
1273 | /* |
1274 | * Use 5 bits less than the max cq entries, that should give us around | |
1275 | * 32 entries per hash list if totally full and uniformly spread. | |
1276 | */ | |
1277 | hash_bits = ilog2(p->cq_entries); | |
1278 | hash_bits -= 5; | |
1279 | if (hash_bits <= 0) | |
1280 | hash_bits = 1; | |
1281 | ctx->cancel_hash_bits = hash_bits; | |
1282 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1283 | GFP_KERNEL); | |
1284 | if (!ctx->cancel_hash) | |
1285 | goto err; | |
1286 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1287 | ||
6224843d PB |
1288 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1289 | if (!ctx->dummy_ubuf) | |
1290 | goto err; | |
1291 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1292 | ctx->dummy_ubuf->ubuf = -1UL; | |
1293 | ||
21482896 | 1294 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1295 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1296 | goto err; | |
2b188cc1 JA |
1297 | |
1298 | ctx->flags = p->flags; | |
90554200 | 1299 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1300 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1301 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1302 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1303 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1304 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1305 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1306 | mutex_init(&ctx->uring_lock); |
311997b3 | 1307 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1308 | spin_lock_init(&ctx->completion_lock); |
89850fce | 1309 | spin_lock_init(&ctx->timeout_lock); |
540e32a0 | 1310 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1311 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1312 | INIT_LIST_HEAD(&ctx->timeout_list); |
ef9dd637 | 1313 | INIT_LIST_HEAD(&ctx->ltimeout_list); |
d67d2263 BM |
1314 | spin_lock_init(&ctx->rsrc_ref_lock); |
1315 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1316 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1317 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1318 | INIT_LIST_HEAD(&ctx->tctx_list); |
cd0ca2e0 | 1319 | INIT_LIST_HEAD(&ctx->submit_state.free_list); |
d0acdee2 | 1320 | INIT_LIST_HEAD(&ctx->locked_free_list); |
9011bf9a | 1321 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
2b188cc1 | 1322 | return ctx; |
206aefde | 1323 | err: |
6224843d | 1324 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1325 | kfree(ctx->cancel_hash); |
206aefde JA |
1326 | kfree(ctx); |
1327 | return NULL; | |
2b188cc1 JA |
1328 | } |
1329 | ||
8f6ed49a PB |
1330 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1331 | { | |
1332 | struct io_rings *r = ctx->rings; | |
1333 | ||
1334 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1335 | ctx->cq_extra--; | |
1336 | } | |
1337 | ||
9cf7c104 | 1338 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1339 | { |
2bc9930e JA |
1340 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1341 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1342 | |
8f6ed49a | 1343 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1344 | } |
de0617e4 | 1345 | |
9d858b21 | 1346 | return false; |
de0617e4 JA |
1347 | } |
1348 | ||
c97d8a0f PB |
1349 | #define FFS_ASYNC_READ 0x1UL |
1350 | #define FFS_ASYNC_WRITE 0x2UL | |
1351 | #ifdef CONFIG_64BIT | |
1352 | #define FFS_ISREG 0x4UL | |
1353 | #else | |
1354 | #define FFS_ISREG 0x0UL | |
1355 | #endif | |
1356 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
1357 | ||
1358 | static inline bool io_req_ffs_set(struct io_kiocb *req) | |
1359 | { | |
1360 | return IS_ENABLED(CONFIG_64BIT) && (req->flags & REQ_F_FIXED_FILE); | |
1361 | } | |
1362 | ||
ce3d5aae PB |
1363 | static void io_req_track_inflight(struct io_kiocb *req) |
1364 | { | |
ce3d5aae | 1365 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1366 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1367 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1368 | } |
1369 | } | |
1370 | ||
906c6caa PB |
1371 | static inline void io_unprep_linked_timeout(struct io_kiocb *req) |
1372 | { | |
1373 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
1374 | } | |
1375 | ||
fd08e530 PB |
1376 | static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req) |
1377 | { | |
906c6caa PB |
1378 | if (WARN_ON_ONCE(!req->link)) |
1379 | return NULL; | |
1380 | ||
4d13d1a4 PB |
1381 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
1382 | req->flags |= REQ_F_LINK_TIMEOUT; | |
fd08e530 PB |
1383 | |
1384 | /* linked timeouts should have two refs once prep'ed */ | |
48dcd38d | 1385 | io_req_set_refcount(req); |
4d13d1a4 PB |
1386 | __io_req_set_refcount(req->link, 2); |
1387 | return req->link; | |
fd08e530 PB |
1388 | } |
1389 | ||
1390 | static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) | |
1391 | { | |
4d13d1a4 | 1392 | if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) |
fd08e530 PB |
1393 | return NULL; |
1394 | return __io_prep_linked_timeout(req); | |
1395 | } | |
1396 | ||
1e6fa521 JA |
1397 | static void io_prep_async_work(struct io_kiocb *req) |
1398 | { | |
1399 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1400 | struct io_ring_ctx *ctx = req->ctx; |
1401 | ||
b8e64b53 PB |
1402 | if (!(req->flags & REQ_F_CREDS)) { |
1403 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1404 | req->creds = get_current_cred(); |
b8e64b53 | 1405 | } |
003e8dcc | 1406 | |
e1d675df PB |
1407 | req->work.list.next = NULL; |
1408 | req->work.flags = 0; | |
feaadc4f PB |
1409 | if (req->flags & REQ_F_FORCE_ASYNC) |
1410 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1411 | ||
1e6fa521 JA |
1412 | if (req->flags & REQ_F_ISREG) { |
1413 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1414 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1415 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1416 | if (def->unbound_nonreg_file) |
1417 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1418 | } | |
e1d675df PB |
1419 | |
1420 | switch (req->opcode) { | |
1421 | case IORING_OP_SPLICE: | |
1422 | case IORING_OP_TEE: | |
e1d675df PB |
1423 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1424 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1425 | break; | |
1426 | } | |
561fb04a | 1427 | } |
cccf0ee8 | 1428 | |
cbdcb435 | 1429 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1430 | { |
cbdcb435 | 1431 | struct io_kiocb *cur; |
54a91f3b | 1432 | |
44eff40a PB |
1433 | if (req->flags & REQ_F_LINK_TIMEOUT) { |
1434 | struct io_ring_ctx *ctx = req->ctx; | |
1435 | ||
79ebeaee | 1436 | spin_lock(&ctx->completion_lock); |
44eff40a PB |
1437 | io_for_each_link(cur, req) |
1438 | io_prep_async_work(cur); | |
79ebeaee | 1439 | spin_unlock(&ctx->completion_lock); |
44eff40a PB |
1440 | } else { |
1441 | io_for_each_link(cur, req) | |
1442 | io_prep_async_work(cur); | |
1443 | } | |
561fb04a JA |
1444 | } |
1445 | ||
f237c30a | 1446 | static void io_queue_async_work(struct io_kiocb *req, bool *locked) |
561fb04a | 1447 | { |
a197f664 | 1448 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1449 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1450 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1451 | |
f237c30a PB |
1452 | /* must not take the lock, NULL it as a precaution */ |
1453 | locked = NULL; | |
1454 | ||
3bfe6106 JA |
1455 | BUG_ON(!tctx); |
1456 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1457 | |
cbdcb435 PB |
1458 | /* init ->work of the whole link before punting */ |
1459 | io_prep_async_link(req); | |
991468dc JA |
1460 | |
1461 | /* | |
1462 | * Not expected to happen, but if we do have a bug where this _can_ | |
1463 | * happen, catch it here and ensure the request is marked as | |
1464 | * canceled. That will make io-wq go through the usual work cancel | |
1465 | * procedure rather than attempt to run this request (or create a new | |
1466 | * worker for it). | |
1467 | */ | |
1468 | if (WARN_ON_ONCE(!same_thread_group(req->task, current))) | |
1469 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1470 | ||
d07f1e8a PB |
1471 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1472 | &req->work, req->flags); | |
ebf93667 | 1473 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1474 | if (link) |
1475 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1476 | } |
1477 | ||
1ee4160c | 1478 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1479 | __must_hold(&req->ctx->completion_lock) |
89850fce | 1480 | __must_hold(&req->ctx->timeout_lock) |
5262f567 | 1481 | { |
e8c2bc1f | 1482 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1483 | |
fd9c7bc5 | 1484 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1485 | atomic_set(&req->ctx->cq_timeouts, |
1486 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1487 | list_del_init(&req->timeout.list); |
d4d19c19 | 1488 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
91c2f697 | 1489 | io_put_req_deferred(req); |
5262f567 JA |
1490 | } |
1491 | } | |
1492 | ||
441b8a78 | 1493 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1494 | { |
441b8a78 | 1495 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1496 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1497 | struct io_defer_entry, list); | |
de0617e4 | 1498 | |
9cf7c104 | 1499 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1500 | break; |
27dc8338 | 1501 | list_del_init(&de->list); |
907d1df3 | 1502 | io_req_task_queue(de->req); |
27dc8338 | 1503 | kfree(de); |
441b8a78 | 1504 | } |
04518945 PB |
1505 | } |
1506 | ||
360428f8 | 1507 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
89850fce | 1508 | __must_hold(&ctx->completion_lock) |
de0617e4 | 1509 | { |
441b8a78 | 1510 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1511 | |
79ebeaee | 1512 | spin_lock_irq(&ctx->timeout_lock); |
f18ee4cf | 1513 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1514 | u32 events_needed, events_got; |
360428f8 | 1515 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1516 | struct io_kiocb, timeout.list); |
de0617e4 | 1517 | |
8eb7e2d0 | 1518 | if (io_is_timeout_noseq(req)) |
360428f8 | 1519 | break; |
f010505b MDG |
1520 | |
1521 | /* | |
1522 | * Since seq can easily wrap around over time, subtract | |
1523 | * the last seq at which timeouts were flushed before comparing. | |
1524 | * Assuming not more than 2^31-1 events have happened since, | |
1525 | * these subtractions won't have wrapped, so we can check if | |
1526 | * target is in [last_seq, current_seq] by comparing the two. | |
1527 | */ | |
1528 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1529 | events_got = seq - ctx->cq_last_tm_flush; | |
1530 | if (events_got < events_needed) | |
360428f8 | 1531 | break; |
bfe68a22 | 1532 | |
135fcde8 | 1533 | list_del_init(&req->timeout.list); |
1ee4160c | 1534 | io_kill_timeout(req, 0); |
f18ee4cf | 1535 | } |
f010505b | 1536 | ctx->cq_last_tm_flush = seq; |
79ebeaee | 1537 | spin_unlock_irq(&ctx->timeout_lock); |
360428f8 | 1538 | } |
5262f567 | 1539 | |
2335f6f5 | 1540 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1541 | { |
2335f6f5 PB |
1542 | if (ctx->off_timeout_used) |
1543 | io_flush_timeouts(ctx); | |
1544 | if (ctx->drain_active) | |
1545 | io_queue_deferred(ctx); | |
1546 | } | |
1547 | ||
1548 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) | |
1549 | { | |
1550 | if (unlikely(ctx->off_timeout_used || ctx->drain_active)) | |
1551 | __io_commit_cqring_flush(ctx); | |
ec30e04b PB |
1552 | /* order cqe stores with ring update */ |
1553 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1554 | } |
1555 | ||
90554200 JA |
1556 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1557 | { | |
1558 | struct io_rings *r = ctx->rings; | |
1559 | ||
a566c556 | 1560 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1561 | } |
1562 | ||
888aae2e PB |
1563 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1564 | { | |
1565 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1566 | } | |
1567 | ||
d068b506 | 1568 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1569 | { |
75b28aff | 1570 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1571 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1572 | |
115e12e5 SB |
1573 | /* |
1574 | * writes to the cq entry need to come after reading head; the | |
1575 | * control dependency is enough as we're using WRITE_ONCE to | |
1576 | * fill the cq entry | |
1577 | */ | |
a566c556 | 1578 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1579 | return NULL; |
1580 | ||
888aae2e | 1581 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1582 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1583 | } |
1584 | ||
f2842ab5 JA |
1585 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1586 | { | |
44c769de | 1587 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1588 | return false; |
7e55a19c SG |
1589 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1590 | return false; | |
44c769de | 1591 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1592 | } |
1593 | ||
2c5d763c JA |
1594 | /* |
1595 | * This should only get called when at least one event has been posted. | |
1596 | * Some applications rely on the eventfd notification count only changing | |
1597 | * IFF a new CQE has been added to the CQ ring. There's no depedency on | |
1598 | * 1:1 relationship between how many times this function is called (and | |
1599 | * hence the eventfd count) and number of CQEs posted to the CQ ring. | |
1600 | */ | |
b41e9852 | 1601 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1602 | { |
5fd46178 JA |
1603 | /* |
1604 | * wake_up_all() may seem excessive, but io_wake_function() and | |
1605 | * io_should_wake() handle the termination of the loop and only | |
1606 | * wake as many waiters as we need to. | |
1607 | */ | |
1608 | if (wq_has_sleeper(&ctx->cq_wait)) | |
1609 | wake_up_all(&ctx->cq_wait); | |
534ca6d6 JA |
1610 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1611 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1612 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1613 | eventfd_signal(ctx->cq_ev_fd, 1); |
311997b3 PB |
1614 | if (waitqueue_active(&ctx->poll_wait)) { |
1615 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1616 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1617 | } | |
1d7bb1d5 JA |
1618 | } |
1619 | ||
80c18e4a PB |
1620 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1621 | { | |
c57a91fb PB |
1622 | /* see waitqueue_active() comment */ |
1623 | smp_mb(); | |
1624 | ||
80c18e4a | 1625 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c57a91fb | 1626 | if (waitqueue_active(&ctx->cq_wait)) |
5fd46178 | 1627 | wake_up_all(&ctx->cq_wait); |
80c18e4a PB |
1628 | } |
1629 | if (io_should_trigger_evfd(ctx)) | |
1630 | eventfd_signal(ctx->cq_ev_fd, 1); | |
311997b3 PB |
1631 | if (waitqueue_active(&ctx->poll_wait)) { |
1632 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1633 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1634 | } | |
80c18e4a PB |
1635 | } |
1636 | ||
c4a2ed72 | 1637 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1638 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1639 | { |
b18032bb | 1640 | bool all_flushed, posted; |
1d7bb1d5 | 1641 | |
a566c556 | 1642 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1643 | return false; |
1d7bb1d5 | 1644 | |
b18032bb | 1645 | posted = false; |
79ebeaee | 1646 | spin_lock(&ctx->completion_lock); |
6c2450ae | 1647 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1648 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1649 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1650 | |
1d7bb1d5 JA |
1651 | if (!cqe && !force) |
1652 | break; | |
6c2450ae PB |
1653 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1654 | struct io_overflow_cqe, list); | |
1655 | if (cqe) | |
1656 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1657 | else | |
8f6ed49a PB |
1658 | io_account_cq_overflow(ctx); |
1659 | ||
b18032bb | 1660 | posted = true; |
6c2450ae PB |
1661 | list_del(&ocqe->list); |
1662 | kfree(ocqe); | |
1d7bb1d5 JA |
1663 | } |
1664 | ||
09e88404 PB |
1665 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1666 | if (all_flushed) { | |
5ed7a37d | 1667 | clear_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1668 | WRITE_ONCE(ctx->rings->sq_flags, |
1669 | ctx->rings->sq_flags & ~IORING_SQ_CQ_OVERFLOW); | |
09e88404 | 1670 | } |
46930143 | 1671 | |
b18032bb JA |
1672 | if (posted) |
1673 | io_commit_cqring(ctx); | |
79ebeaee | 1674 | spin_unlock(&ctx->completion_lock); |
b18032bb JA |
1675 | if (posted) |
1676 | io_cqring_ev_posted(ctx); | |
09e88404 | 1677 | return all_flushed; |
1d7bb1d5 JA |
1678 | } |
1679 | ||
90f67366 | 1680 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx) |
6c503150 | 1681 | { |
ca0a2651 JA |
1682 | bool ret = true; |
1683 | ||
5ed7a37d | 1684 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1685 | /* iopoll syncs against uring_lock, not completion_lock */ |
1686 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1687 | mutex_lock(&ctx->uring_lock); | |
90f67366 | 1688 | ret = __io_cqring_overflow_flush(ctx, false); |
6c503150 PB |
1689 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1690 | mutex_unlock(&ctx->uring_lock); | |
1691 | } | |
ca0a2651 JA |
1692 | |
1693 | return ret; | |
6c503150 PB |
1694 | } |
1695 | ||
6a290a14 PB |
1696 | /* must to be called somewhat shortly after putting a request */ |
1697 | static inline void io_put_task(struct task_struct *task, int nr) | |
1698 | { | |
1699 | struct io_uring_task *tctx = task->io_uring; | |
1700 | ||
e98e49b2 PB |
1701 | if (likely(task == current)) { |
1702 | tctx->cached_refs += nr; | |
1703 | } else { | |
1704 | percpu_counter_sub(&tctx->inflight, nr); | |
1705 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1706 | wake_up(&tctx->wait); | |
1707 | put_task_struct_many(task, nr); | |
1708 | } | |
6a290a14 PB |
1709 | } |
1710 | ||
9a10867a PB |
1711 | static void io_task_refs_refill(struct io_uring_task *tctx) |
1712 | { | |
1713 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
1714 | ||
1715 | percpu_counter_add(&tctx->inflight, refill); | |
1716 | refcount_add(refill, ¤t->usage); | |
1717 | tctx->cached_refs += refill; | |
1718 | } | |
1719 | ||
1720 | static inline void io_get_task_refs(int nr) | |
1721 | { | |
1722 | struct io_uring_task *tctx = current->io_uring; | |
1723 | ||
1724 | tctx->cached_refs -= nr; | |
1725 | if (unlikely(tctx->cached_refs < 0)) | |
1726 | io_task_refs_refill(tctx); | |
1727 | } | |
1728 | ||
d4d19c19 PB |
1729 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1730 | long res, unsigned int cflags) | |
2b188cc1 | 1731 | { |
cce4b8b0 | 1732 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1733 | |
cce4b8b0 PB |
1734 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1735 | if (!ocqe) { | |
1736 | /* | |
1737 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1738 | * or cannot allocate an overflow entry, then we need to drop it | |
1739 | * on the floor. | |
1740 | */ | |
8f6ed49a | 1741 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1742 | return false; |
2b188cc1 | 1743 | } |
cce4b8b0 | 1744 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1745 | set_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1746 | WRITE_ONCE(ctx->rings->sq_flags, |
1747 | ctx->rings->sq_flags | IORING_SQ_CQ_OVERFLOW); | |
1748 | ||
cce4b8b0 | 1749 | } |
d4d19c19 | 1750 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1751 | ocqe->cqe.res = res; |
1752 | ocqe->cqe.flags = cflags; | |
1753 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1754 | return true; | |
2b188cc1 JA |
1755 | } |
1756 | ||
d4d19c19 PB |
1757 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1758 | long res, unsigned int cflags) | |
2b188cc1 JA |
1759 | { |
1760 | struct io_uring_cqe *cqe; | |
1761 | ||
d4d19c19 | 1762 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1763 | |
2b188cc1 JA |
1764 | /* |
1765 | * If we can't get a cq entry, userspace overflowed the | |
1766 | * submission (by quite a lot). Increment the overflow count in | |
1767 | * the ring. | |
1768 | */ | |
d068b506 | 1769 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1770 | if (likely(cqe)) { |
d4d19c19 | 1771 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1772 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1773 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1774 | return true; |
2b188cc1 | 1775 | } |
d4d19c19 | 1776 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1777 | } |
1778 | ||
8d13326e | 1779 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1780 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1781 | long res, unsigned int cflags) | |
bcda7baa | 1782 | { |
d4d19c19 | 1783 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1784 | } |
1785 | ||
7a612350 PB |
1786 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1787 | unsigned int cflags) | |
2b188cc1 | 1788 | { |
78e19bbe | 1789 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 | 1790 | |
79ebeaee | 1791 | spin_lock(&ctx->completion_lock); |
d4d19c19 | 1792 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1793 | /* |
1794 | * If we're the last reference to this request, add to our locked | |
1795 | * free_list cache. | |
1796 | */ | |
de9b4cca | 1797 | if (req_ref_put_and_test(req)) { |
7a612350 | 1798 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
0756a869 | 1799 | if (req->flags & IO_DISARM_MASK) |
7a612350 PB |
1800 | io_disarm_next(req); |
1801 | if (req->link) { | |
1802 | io_req_task_queue(req->link); | |
1803 | req->link = NULL; | |
1804 | } | |
1805 | } | |
c7dae4ba JA |
1806 | io_dismantle_req(req); |
1807 | io_put_task(req->task, 1); | |
bb943b82 | 1808 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
d0acdee2 | 1809 | ctx->locked_free_nr++; |
180f829f PB |
1810 | } else { |
1811 | if (!percpu_ref_tryget(&ctx->refs)) | |
1812 | req = NULL; | |
1813 | } | |
7a612350 | 1814 | io_commit_cqring(ctx); |
79ebeaee | 1815 | spin_unlock(&ctx->completion_lock); |
7a612350 | 1816 | |
180f829f PB |
1817 | if (req) { |
1818 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1819 | percpu_ref_put(&ctx->refs); |
180f829f | 1820 | } |
229a7b63 JA |
1821 | } |
1822 | ||
4e3d9ff9 JA |
1823 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1824 | { | |
c854357b | 1825 | return req->flags & IO_REQ_CLEAN_FLAGS; |
4e3d9ff9 JA |
1826 | } |
1827 | ||
a38d68db | 1828 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1829 | unsigned int cflags) |
229a7b63 | 1830 | { |
4e3d9ff9 | 1831 | if (io_req_needs_clean(req)) |
68fb8979 | 1832 | io_clean_op(req); |
a38d68db PB |
1833 | req->result = res; |
1834 | req->compl.cflags = cflags; | |
e342c807 | 1835 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1836 | } |
1837 | ||
889fca73 PB |
1838 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1839 | long res, unsigned cflags) | |
bcda7baa | 1840 | { |
889fca73 PB |
1841 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1842 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1843 | else |
c7dae4ba | 1844 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1845 | } |
1846 | ||
a38d68db | 1847 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1848 | { |
889fca73 | 1849 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1850 | } |
1851 | ||
f41db273 PB |
1852 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1853 | { | |
93d2bcd2 | 1854 | req_set_fail(req); |
f41db273 PB |
1855 | io_req_complete_post(req, res, 0); |
1856 | } | |
1857 | ||
c6d3d9cb PB |
1858 | static void io_req_complete_fail_submit(struct io_kiocb *req) |
1859 | { | |
1860 | /* | |
1861 | * We don't submit, fail them all, for that replace hardlinks with | |
1862 | * normal links. Extra REQ_F_LINK is tolerated. | |
1863 | */ | |
1864 | req->flags &= ~REQ_F_HARDLINK; | |
1865 | req->flags |= REQ_F_LINK; | |
1866 | io_req_complete_failed(req, req->result); | |
1867 | } | |
1868 | ||
864ea921 PB |
1869 | /* |
1870 | * Don't initialise the fields below on every allocation, but do that in | |
1871 | * advance and keep them valid across allocations. | |
1872 | */ | |
1873 | static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
1874 | { | |
1875 | req->ctx = ctx; | |
1876 | req->link = NULL; | |
1877 | req->async_data = NULL; | |
1878 | /* not necessary, but safer to zero */ | |
1879 | req->result = 0; | |
1880 | } | |
1881 | ||
dac7a098 | 1882 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
cd0ca2e0 | 1883 | struct io_submit_state *state) |
dac7a098 | 1884 | { |
79ebeaee | 1885 | spin_lock(&ctx->completion_lock); |
cd0ca2e0 | 1886 | list_splice_init(&ctx->locked_free_list, &state->free_list); |
d0acdee2 | 1887 | ctx->locked_free_nr = 0; |
79ebeaee | 1888 | spin_unlock(&ctx->completion_lock); |
dac7a098 PB |
1889 | } |
1890 | ||
dd78f492 | 1891 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1892 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1893 | { |
c7dae4ba | 1894 | struct io_submit_state *state = &ctx->submit_state; |
dd78f492 | 1895 | int nr; |
0ddf92e8 | 1896 | |
c7dae4ba JA |
1897 | /* |
1898 | * If we have more than a batch's worth of requests in our IRQ side | |
1899 | * locked cache, grab the lock and move them over to our submission | |
1900 | * side cache. | |
1901 | */ | |
d0acdee2 | 1902 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
cd0ca2e0 | 1903 | io_flush_cached_locked_reqs(ctx, state); |
0ddf92e8 | 1904 | |
dd78f492 | 1905 | nr = state->free_reqs; |
cd0ca2e0 PB |
1906 | while (!list_empty(&state->free_list)) { |
1907 | struct io_kiocb *req = list_first_entry(&state->free_list, | |
bb943b82 | 1908 | struct io_kiocb, inflight_entry); |
dd78f492 | 1909 | |
bb943b82 | 1910 | list_del(&req->inflight_entry); |
dd78f492 PB |
1911 | state->reqs[nr++] = req; |
1912 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1913 | break; |
1b4c351f JA |
1914 | } |
1915 | ||
dd78f492 PB |
1916 | state->free_reqs = nr; |
1917 | return nr != 0; | |
0ddf92e8 JA |
1918 | } |
1919 | ||
5d5901a3 PB |
1920 | /* |
1921 | * A request might get retired back into the request caches even before opcode | |
1922 | * handlers and io_issue_sqe() are done with it, e.g. inline completion path. | |
1923 | * Because of that, io_alloc_req() should be called only under ->uring_lock | |
1924 | * and with extra caution to not get a request that is still worked on. | |
1925 | */ | |
e5d1bc0a | 1926 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
5d5901a3 | 1927 | __must_hold(&ctx->uring_lock) |
2b188cc1 | 1928 | { |
e5d1bc0a | 1929 | struct io_submit_state *state = &ctx->submit_state; |
864ea921 PB |
1930 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
1931 | int ret, i; | |
e5d1bc0a | 1932 | |
fe7e3257 | 1933 | BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); |
e5d1bc0a | 1934 | |
864ea921 PB |
1935 | if (likely(state->free_reqs || io_flush_cached_reqs(ctx))) |
1936 | goto got_req; | |
e5d1bc0a | 1937 | |
864ea921 PB |
1938 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1939 | state->reqs); | |
fd6fab2c | 1940 | |
864ea921 PB |
1941 | /* |
1942 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1943 | * retry single alloc to be on the safe side. | |
1944 | */ | |
1945 | if (unlikely(ret <= 0)) { | |
1946 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1947 | if (!state->reqs[0]) | |
1948 | return NULL; | |
1949 | ret = 1; | |
2b188cc1 | 1950 | } |
864ea921 PB |
1951 | |
1952 | for (i = 0; i < ret; i++) | |
1953 | io_preinit_req(state->reqs[i], ctx); | |
1954 | state->free_reqs = ret; | |
e5d1bc0a | 1955 | got_req: |
291b2821 PB |
1956 | state->free_reqs--; |
1957 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1958 | } |
1959 | ||
e1d767f0 | 1960 | static inline void io_put_file(struct file *file) |
8da11c19 | 1961 | { |
e1d767f0 | 1962 | if (file) |
8da11c19 PB |
1963 | fput(file); |
1964 | } | |
1965 | ||
4edf20f9 | 1966 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1967 | { |
094bae49 | 1968 | unsigned int flags = req->flags; |
929a3af9 | 1969 | |
3a0a6902 PB |
1970 | if (io_req_needs_clean(req)) |
1971 | io_clean_op(req); | |
e1d767f0 PB |
1972 | if (!(flags & REQ_F_FIXED_FILE)) |
1973 | io_put_file(req->file); | |
269bbe5f BM |
1974 | if (req->fixed_rsrc_refs) |
1975 | percpu_ref_put(req->fixed_rsrc_refs); | |
99ebe4ef | 1976 | if (req->async_data) { |
094bae49 | 1977 | kfree(req->async_data); |
99ebe4ef PB |
1978 | req->async_data = NULL; |
1979 | } | |
e65ef56d JA |
1980 | } |
1981 | ||
216578e5 | 1982 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1983 | { |
51a4cc11 | 1984 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1985 | |
216578e5 | 1986 | io_dismantle_req(req); |
7c660731 | 1987 | io_put_task(req->task, 1); |
c6ca97b3 | 1988 | |
79ebeaee | 1989 | spin_lock(&ctx->completion_lock); |
bb943b82 | 1990 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
c34b025f | 1991 | ctx->locked_free_nr++; |
79ebeaee | 1992 | spin_unlock(&ctx->completion_lock); |
c34b025f | 1993 | |
ecfc5177 | 1994 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1995 | } |
1996 | ||
f2f87370 PB |
1997 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1998 | { | |
1999 | struct io_kiocb *nxt = req->link; | |
2000 | ||
2001 | req->link = nxt->link; | |
2002 | nxt->link = NULL; | |
2003 | } | |
2004 | ||
33cc89a9 PB |
2005 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
2006 | __must_hold(&req->ctx->completion_lock) | |
89b263f6 | 2007 | __must_hold(&req->ctx->timeout_lock) |
2665abfd | 2008 | { |
33cc89a9 | 2009 | struct io_kiocb *link = req->link; |
f2f87370 | 2010 | |
b97e736a | 2011 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
c9abd7ad | 2012 | struct io_timeout_data *io = link->async_data; |
7c86ffee | 2013 | |
f2f87370 | 2014 | io_remove_next_linked(req); |
90cd7e42 | 2015 | link->timeout.head = NULL; |
fd9c7bc5 | 2016 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ef9dd637 | 2017 | list_del(&link->timeout.list); |
d4d19c19 PB |
2018 | io_cqring_fill_event(link->ctx, link->user_data, |
2019 | -ECANCELED, 0); | |
91c2f697 | 2020 | io_put_req_deferred(link); |
d4729fbd | 2021 | return true; |
c9abd7ad PB |
2022 | } |
2023 | } | |
d4729fbd | 2024 | return false; |
7c86ffee PB |
2025 | } |
2026 | ||
d148ca4b | 2027 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 2028 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 2029 | { |
33cc89a9 | 2030 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 2031 | |
f2f87370 | 2032 | req->link = NULL; |
f2f87370 | 2033 | while (link) { |
a8295b98 HX |
2034 | long res = -ECANCELED; |
2035 | ||
2036 | if (link->flags & REQ_F_FAIL) | |
2037 | res = link->result; | |
2038 | ||
f2f87370 PB |
2039 | nxt = link->link; |
2040 | link->link = NULL; | |
2665abfd | 2041 | |
f2f87370 | 2042 | trace_io_uring_fail_link(req, link); |
a8295b98 | 2043 | io_cqring_fill_event(link->ctx, link->user_data, res, 0); |
91c2f697 | 2044 | io_put_req_deferred(link); |
f2f87370 | 2045 | link = nxt; |
9e645e11 | 2046 | } |
33cc89a9 | 2047 | } |
9e645e11 | 2048 | |
33cc89a9 PB |
2049 | static bool io_disarm_next(struct io_kiocb *req) |
2050 | __must_hold(&req->ctx->completion_lock) | |
2051 | { | |
2052 | bool posted = false; | |
2053 | ||
0756a869 PB |
2054 | if (req->flags & REQ_F_ARM_LTIMEOUT) { |
2055 | struct io_kiocb *link = req->link; | |
2056 | ||
906c6caa | 2057 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
0756a869 PB |
2058 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
2059 | io_remove_next_linked(req); | |
2060 | io_cqring_fill_event(link->ctx, link->user_data, | |
2061 | -ECANCELED, 0); | |
2062 | io_put_req_deferred(link); | |
2063 | posted = true; | |
2064 | } | |
2065 | } else if (req->flags & REQ_F_LINK_TIMEOUT) { | |
89b263f6 JA |
2066 | struct io_ring_ctx *ctx = req->ctx; |
2067 | ||
2068 | spin_lock_irq(&ctx->timeout_lock); | |
33cc89a9 | 2069 | posted = io_kill_linked_timeout(req); |
89b263f6 JA |
2070 | spin_unlock_irq(&ctx->timeout_lock); |
2071 | } | |
93d2bcd2 | 2072 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 2073 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
2074 | posted |= (req->link != NULL); |
2075 | io_fail_links(req); | |
2076 | } | |
2077 | return posted; | |
9e645e11 JA |
2078 | } |
2079 | ||
3fa5e0f3 | 2080 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2081 | { |
33cc89a9 | 2082 | struct io_kiocb *nxt; |
944e58bf | 2083 | |
9e645e11 JA |
2084 | /* |
2085 | * If LINK is set, we have dependent requests in this chain. If we | |
2086 | * didn't fail this request, queue the first one up, moving any other | |
2087 | * dependencies to the next request. In case of failure, fail the rest | |
2088 | * of the chain. | |
2089 | */ | |
0756a869 | 2090 | if (req->flags & IO_DISARM_MASK) { |
33cc89a9 | 2091 | struct io_ring_ctx *ctx = req->ctx; |
33cc89a9 PB |
2092 | bool posted; |
2093 | ||
79ebeaee | 2094 | spin_lock(&ctx->completion_lock); |
33cc89a9 PB |
2095 | posted = io_disarm_next(req); |
2096 | if (posted) | |
2097 | io_commit_cqring(req->ctx); | |
79ebeaee | 2098 | spin_unlock(&ctx->completion_lock); |
33cc89a9 PB |
2099 | if (posted) |
2100 | io_cqring_ev_posted(ctx); | |
f2f87370 | 2101 | } |
33cc89a9 PB |
2102 | nxt = req->link; |
2103 | req->link = NULL; | |
2104 | return nxt; | |
4d7dd462 | 2105 | } |
9e645e11 | 2106 | |
f2f87370 | 2107 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2108 | { |
cdbff982 | 2109 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2110 | return NULL; |
2111 | return __io_req_find_next(req); | |
2112 | } | |
2113 | ||
f237c30a | 2114 | static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked) |
2c32395d PB |
2115 | { |
2116 | if (!ctx) | |
2117 | return; | |
f237c30a | 2118 | if (*locked) { |
99c8bc52 HX |
2119 | if (ctx->submit_state.compl_nr) |
2120 | io_submit_flush_completions(ctx); | |
2c32395d | 2121 | mutex_unlock(&ctx->uring_lock); |
f237c30a | 2122 | *locked = false; |
2c32395d PB |
2123 | } |
2124 | percpu_ref_put(&ctx->refs); | |
2125 | } | |
2126 | ||
7cbf1722 | 2127 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2128 | { |
f237c30a | 2129 | bool locked = false; |
ebd0df2e | 2130 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
2131 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
2132 | task_work); | |
c40f6379 | 2133 | |
16f72070 | 2134 | while (1) { |
3f18407d PB |
2135 | struct io_wq_work_node *node; |
2136 | ||
8d4ad41e PB |
2137 | if (!tctx->task_list.first && locked && ctx->submit_state.compl_nr) |
2138 | io_submit_flush_completions(ctx); | |
2139 | ||
3f18407d | 2140 | spin_lock_irq(&tctx->task_lock); |
c6538be9 | 2141 | node = tctx->task_list.first; |
3f18407d | 2142 | INIT_WQ_LIST(&tctx->task_list); |
6294f368 PB |
2143 | if (!node) |
2144 | tctx->task_running = false; | |
3f18407d | 2145 | spin_unlock_irq(&tctx->task_lock); |
6294f368 PB |
2146 | if (!node) |
2147 | break; | |
3f18407d | 2148 | |
6294f368 | 2149 | do { |
3f18407d PB |
2150 | struct io_wq_work_node *next = node->next; |
2151 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2152 | io_task_work.node); | |
2153 | ||
2154 | if (req->ctx != ctx) { | |
f237c30a | 2155 | ctx_flush_and_put(ctx, &locked); |
3f18407d | 2156 | ctx = req->ctx; |
126180b9 PB |
2157 | /* if not contended, grab and improve batching */ |
2158 | locked = mutex_trylock(&ctx->uring_lock); | |
3f18407d PB |
2159 | percpu_ref_get(&ctx->refs); |
2160 | } | |
f237c30a | 2161 | req->io_task_work.func(req, &locked); |
3f18407d | 2162 | node = next; |
6294f368 PB |
2163 | } while (node); |
2164 | ||
7cbf1722 | 2165 | cond_resched(); |
3f18407d | 2166 | } |
ebd0df2e | 2167 | |
f237c30a | 2168 | ctx_flush_and_put(ctx, &locked); |
7cbf1722 JA |
2169 | } |
2170 | ||
e09ee510 | 2171 | static void io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 2172 | { |
c15b79de | 2173 | struct task_struct *tsk = req->task; |
7cbf1722 | 2174 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 2175 | enum task_work_notify_mode notify; |
e09ee510 | 2176 | struct io_wq_work_node *node; |
0b81e80c | 2177 | unsigned long flags; |
6294f368 | 2178 | bool running; |
7cbf1722 JA |
2179 | |
2180 | WARN_ON_ONCE(!tctx); | |
2181 | ||
0b81e80c | 2182 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2183 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
6294f368 PB |
2184 | running = tctx->task_running; |
2185 | if (!running) | |
2186 | tctx->task_running = true; | |
0b81e80c | 2187 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2188 | |
2189 | /* task_work already pending, we're done */ | |
6294f368 | 2190 | if (running) |
e09ee510 | 2191 | return; |
7cbf1722 | 2192 | |
c15b79de PB |
2193 | /* |
2194 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
2195 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2196 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2197 | * will do the job. | |
2198 | */ | |
2199 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
c15b79de PB |
2200 | if (!task_work_add(tsk, &tctx->task_work, notify)) { |
2201 | wake_up_process(tsk); | |
e09ee510 | 2202 | return; |
c15b79de | 2203 | } |
2215bed9 | 2204 | |
0b81e80c | 2205 | spin_lock_irqsave(&tctx->task_lock, flags); |
6294f368 | 2206 | tctx->task_running = false; |
e09ee510 PB |
2207 | node = tctx->task_list.first; |
2208 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2209 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 2210 | |
e09ee510 PB |
2211 | while (node) { |
2212 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
2213 | node = node->next; | |
2214 | if (llist_add(&req->io_task_work.fallback_node, | |
2215 | &req->ctx->fallback_llist)) | |
2216 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
2217 | } | |
eab30c4d PB |
2218 | } |
2219 | ||
f237c30a | 2220 | static void io_req_task_cancel(struct io_kiocb *req, bool *locked) |
c40f6379 | 2221 | { |
87ceb6a6 | 2222 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2223 | |
b18a1a45 | 2224 | /* not needed for normal modes, but SQPOLL depends on it */ |
f237c30a | 2225 | io_tw_lock(ctx, locked); |
2593553a | 2226 | io_req_complete_failed(req, req->result); |
c40f6379 JA |
2227 | } |
2228 | ||
f237c30a | 2229 | static void io_req_task_submit(struct io_kiocb *req, bool *locked) |
c40f6379 JA |
2230 | { |
2231 | struct io_ring_ctx *ctx = req->ctx; | |
2232 | ||
f237c30a | 2233 | io_tw_lock(ctx, locked); |
316319e8 | 2234 | /* req->task == current here, checking PF_EXITING is safe */ |
af066f31 | 2235 | if (likely(!(req->task->flags & PF_EXITING))) |
c5eef2b9 | 2236 | __io_queue_sqe(req); |
81b6d05c | 2237 | else |
2593553a | 2238 | io_req_complete_failed(req, -EFAULT); |
c40f6379 JA |
2239 | } |
2240 | ||
2c4b8eb6 | 2241 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2242 | { |
2c4b8eb6 | 2243 | req->result = ret; |
5b0a6acc | 2244 | req->io_task_work.func = io_req_task_cancel; |
e09ee510 | 2245 | io_req_task_work_add(req); |
c40f6379 JA |
2246 | } |
2247 | ||
2c4b8eb6 | 2248 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2249 | { |
5b0a6acc | 2250 | req->io_task_work.func = io_req_task_submit; |
e09ee510 | 2251 | io_req_task_work_add(req); |
a3df7698 PB |
2252 | } |
2253 | ||
773af691 JA |
2254 | static void io_req_task_queue_reissue(struct io_kiocb *req) |
2255 | { | |
2256 | req->io_task_work.func = io_queue_async_work; | |
2257 | io_req_task_work_add(req); | |
2258 | } | |
2259 | ||
f2f87370 | 2260 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2261 | { |
9b5f7bd9 | 2262 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2263 | |
2264 | if (nxt) | |
906a8c3f | 2265 | io_req_task_queue(nxt); |
c69f8dbe JL |
2266 | } |
2267 | ||
c3524383 | 2268 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2269 | { |
c3524383 PB |
2270 | io_queue_next(req); |
2271 | __io_free_req(req); | |
2272 | } | |
8766dd51 | 2273 | |
f237c30a PB |
2274 | static void io_free_req_work(struct io_kiocb *req, bool *locked) |
2275 | { | |
2276 | io_free_req(req); | |
2277 | } | |
2278 | ||
2d6500d4 | 2279 | struct req_batch { |
5af1d13e PB |
2280 | struct task_struct *task; |
2281 | int task_refs; | |
1b4c351f | 2282 | int ctx_refs; |
2d6500d4 PB |
2283 | }; |
2284 | ||
5af1d13e PB |
2285 | static inline void io_init_req_batch(struct req_batch *rb) |
2286 | { | |
5af1d13e | 2287 | rb->task_refs = 0; |
9ae72463 | 2288 | rb->ctx_refs = 0; |
5af1d13e PB |
2289 | rb->task = NULL; |
2290 | } | |
2291 | ||
2d6500d4 PB |
2292 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2293 | struct req_batch *rb) | |
2294 | { | |
9ae72463 PB |
2295 | if (rb->ctx_refs) |
2296 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
e98e49b2 | 2297 | if (rb->task) |
e9dbe221 | 2298 | io_put_task(rb->task, rb->task_refs); |
2d6500d4 PB |
2299 | } |
2300 | ||
6ff119a6 PB |
2301 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2302 | struct io_submit_state *state) | |
2d6500d4 | 2303 | { |
f2f87370 | 2304 | io_queue_next(req); |
96670657 | 2305 | io_dismantle_req(req); |
2d6500d4 | 2306 | |
e3bc8e9d | 2307 | if (req->task != rb->task) { |
7c660731 PB |
2308 | if (rb->task) |
2309 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2310 | rb->task = req->task; |
2311 | rb->task_refs = 0; | |
5af1d13e | 2312 | } |
e3bc8e9d | 2313 | rb->task_refs++; |
9ae72463 | 2314 | rb->ctx_refs++; |
5af1d13e | 2315 | |
bd759045 | 2316 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2317 | state->reqs[state->free_reqs++] = req; |
bd759045 | 2318 | else |
cd0ca2e0 | 2319 | list_add(&req->inflight_entry, &state->free_list); |
7a743e22 PB |
2320 | } |
2321 | ||
2a2758f2 | 2322 | static void io_submit_flush_completions(struct io_ring_ctx *ctx) |
a141dd89 | 2323 | __must_hold(&ctx->uring_lock) |
905c172f | 2324 | { |
cd0ca2e0 PB |
2325 | struct io_submit_state *state = &ctx->submit_state; |
2326 | int i, nr = state->compl_nr; | |
905c172f PB |
2327 | struct req_batch rb; |
2328 | ||
79ebeaee | 2329 | spin_lock(&ctx->completion_lock); |
905c172f | 2330 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2331 | struct io_kiocb *req = state->compl_reqs[i]; |
5182ed2e | 2332 | |
d4d19c19 PB |
2333 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2334 | req->compl.cflags); | |
905c172f PB |
2335 | } |
2336 | io_commit_cqring(ctx); | |
79ebeaee | 2337 | spin_unlock(&ctx->completion_lock); |
905c172f | 2338 | io_cqring_ev_posted(ctx); |
5182ed2e PB |
2339 | |
2340 | io_init_req_batch(&rb); | |
905c172f | 2341 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2342 | struct io_kiocb *req = state->compl_reqs[i]; |
905c172f | 2343 | |
91c2f697 | 2344 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2345 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2346 | } |
2347 | ||
2348 | io_req_free_batch_finish(ctx, &rb); | |
cd0ca2e0 | 2349 | state->compl_nr = 0; |
7a743e22 PB |
2350 | } |
2351 | ||
ba816ad6 JA |
2352 | /* |
2353 | * Drop reference to request, return next in chain (if there is one) if this | |
2354 | * was the last reference to this request. | |
2355 | */ | |
0d85035a | 2356 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2357 | { |
9b5f7bd9 PB |
2358 | struct io_kiocb *nxt = NULL; |
2359 | ||
de9b4cca | 2360 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2361 | nxt = io_req_find_next(req); |
4d7dd462 | 2362 | __io_free_req(req); |
2a44f467 | 2363 | } |
9b5f7bd9 | 2364 | return nxt; |
2b188cc1 JA |
2365 | } |
2366 | ||
0d85035a | 2367 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2368 | { |
de9b4cca | 2369 | if (req_ref_put_and_test(req)) |
e65ef56d | 2370 | io_free_req(req); |
2b188cc1 JA |
2371 | } |
2372 | ||
91c2f697 | 2373 | static inline void io_put_req_deferred(struct io_kiocb *req) |
216578e5 | 2374 | { |
91c2f697 | 2375 | if (req_ref_put_and_test(req)) { |
f237c30a | 2376 | req->io_task_work.func = io_free_req_work; |
543af3a1 PB |
2377 | io_req_task_work_add(req); |
2378 | } | |
216578e5 PB |
2379 | } |
2380 | ||
6c503150 | 2381 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2382 | { |
2383 | /* See comment at the top of this file */ | |
2384 | smp_rmb(); | |
e23de15f | 2385 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2386 | } |
2387 | ||
fb5ccc98 PB |
2388 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2389 | { | |
2390 | struct io_rings *rings = ctx->rings; | |
2391 | ||
2392 | /* make sure SQ entry isn't read before tail */ | |
2393 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2394 | } | |
2395 | ||
8ff069bf | 2396 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2397 | { |
8ff069bf | 2398 | unsigned int cflags; |
e94f141b | 2399 | |
bcda7baa JA |
2400 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2401 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2402 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2403 | kfree(kbuf); |
2404 | return cflags; | |
e94f141b JA |
2405 | } |
2406 | ||
8ff069bf | 2407 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2408 | { |
4d954c25 | 2409 | struct io_buffer *kbuf; |
bcda7baa | 2410 | |
ae421d93 PB |
2411 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) |
2412 | return 0; | |
4d954c25 | 2413 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2414 | return io_put_kbuf(req, kbuf); |
2415 | } | |
2416 | ||
4c6e277c JA |
2417 | static inline bool io_run_task_work(void) |
2418 | { | |
ef98eb04 | 2419 | if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) { |
4c6e277c | 2420 | __set_current_state(TASK_RUNNING); |
ef98eb04 | 2421 | tracehook_notify_signal(); |
4c6e277c JA |
2422 | return true; |
2423 | } | |
2424 | ||
2425 | return false; | |
bcda7baa JA |
2426 | } |
2427 | ||
def596e9 JA |
2428 | /* |
2429 | * Find and free completed poll iocbs | |
2430 | */ | |
2431 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
a8576af9 | 2432 | struct list_head *done) |
def596e9 | 2433 | { |
8237e045 | 2434 | struct req_batch rb; |
def596e9 | 2435 | struct io_kiocb *req; |
bbde017a XW |
2436 | |
2437 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2438 | smp_rmb(); | |
def596e9 | 2439 | |
5af1d13e | 2440 | io_init_req_batch(&rb); |
def596e9 | 2441 | while (!list_empty(done)) { |
d21ffe7e | 2442 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2443 | list_del(&req->inflight_entry); |
2444 | ||
a8576af9 | 2445 | if (READ_ONCE(req->result) == -EAGAIN && |
8c130827 | 2446 | !(req->flags & REQ_F_DONT_REISSUE)) { |
bbde017a | 2447 | req->iopoll_completed = 0; |
773af691 | 2448 | io_req_task_queue_reissue(req); |
8c130827 | 2449 | continue; |
bbde017a | 2450 | } |
def596e9 | 2451 | |
ae421d93 PB |
2452 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2453 | io_put_rw_kbuf(req)); | |
def596e9 JA |
2454 | (*nr_events)++; |
2455 | ||
de9b4cca | 2456 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2457 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2458 | } |
def596e9 | 2459 | |
09bb8394 | 2460 | io_commit_cqring(ctx); |
80c18e4a | 2461 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2462 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2463 | } |
2464 | ||
def596e9 | 2465 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
a8576af9 | 2466 | long min) |
def596e9 JA |
2467 | { |
2468 | struct io_kiocb *req, *tmp; | |
2469 | LIST_HEAD(done); | |
2470 | bool spin; | |
def596e9 JA |
2471 | |
2472 | /* | |
2473 | * Only spin for completions if we don't have multiple devices hanging | |
2474 | * off our complete list, and we're under the requested amount. | |
2475 | */ | |
915b3dde | 2476 | spin = !ctx->poll_multi_queue && *nr_events < min; |
def596e9 | 2477 | |
d21ffe7e | 2478 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2479 | struct kiocb *kiocb = &req->rw.kiocb; |
a2416e1e | 2480 | int ret; |
def596e9 JA |
2481 | |
2482 | /* | |
581f9810 BM |
2483 | * Move completed and retryable entries to our local lists. |
2484 | * If we find a request that requires polling, break out | |
2485 | * and complete those lists first, if we have entries there. | |
def596e9 | 2486 | */ |
65a6543d | 2487 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2488 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2489 | continue; |
2490 | } | |
2491 | if (!list_empty(&done)) | |
2492 | break; | |
2493 | ||
2494 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
a2416e1e PB |
2495 | if (unlikely(ret < 0)) |
2496 | return ret; | |
2497 | else if (ret) | |
2498 | spin = false; | |
def596e9 | 2499 | |
3aadc23e PB |
2500 | /* iopoll may have completed current req */ |
2501 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2502 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2503 | } |
2504 | ||
2505 | if (!list_empty(&done)) | |
a8576af9 | 2506 | io_iopoll_complete(ctx, nr_events, &done); |
def596e9 | 2507 | |
a2416e1e | 2508 | return 0; |
def596e9 JA |
2509 | } |
2510 | ||
def596e9 JA |
2511 | /* |
2512 | * We can't just wait for polled events to come to us, we have to actively | |
2513 | * find and complete them. | |
2514 | */ | |
b2edc0a7 | 2515 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2516 | { |
2517 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2518 | return; | |
2519 | ||
2520 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2521 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2522 | unsigned int nr_events = 0; |
2523 | ||
a8576af9 | 2524 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2525 | |
b2edc0a7 PB |
2526 | /* let it sleep and repeat later if can't complete a request */ |
2527 | if (nr_events == 0) | |
2528 | break; | |
08f5439f JA |
2529 | /* |
2530 | * Ensure we allow local-to-the-cpu processing to take place, | |
2531 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2532 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2533 | */ |
3fcee5a6 PB |
2534 | if (need_resched()) { |
2535 | mutex_unlock(&ctx->uring_lock); | |
2536 | cond_resched(); | |
2537 | mutex_lock(&ctx->uring_lock); | |
2538 | } | |
def596e9 JA |
2539 | } |
2540 | mutex_unlock(&ctx->uring_lock); | |
2541 | } | |
2542 | ||
7668b92a | 2543 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2544 | { |
7668b92a | 2545 | unsigned int nr_events = 0; |
e9979b36 | 2546 | int ret = 0; |
500f9fba | 2547 | |
c7849be9 XW |
2548 | /* |
2549 | * We disallow the app entering submit/complete with polling, but we | |
2550 | * still need to lock the ring to prevent racing with polled issue | |
2551 | * that got punted to a workqueue. | |
2552 | */ | |
2553 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2554 | /* |
2555 | * Don't enter poll loop if we already have events pending. | |
2556 | * If we do, we can potentially be spinning for commands that | |
2557 | * already triggered a CQE (eg in error). | |
2558 | */ | |
5ed7a37d | 2559 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2560 | __io_cqring_overflow_flush(ctx, false); |
2561 | if (io_cqring_events(ctx)) | |
2562 | goto out; | |
def596e9 | 2563 | do { |
500f9fba JA |
2564 | /* |
2565 | * If a submit got punted to a workqueue, we can have the | |
2566 | * application entering polling for a command before it gets | |
2567 | * issued. That app will hold the uring_lock for the duration | |
2568 | * of the poll right here, so we need to take a breather every | |
2569 | * now and then to ensure that the issue has a chance to add | |
2570 | * the poll to the issued list. Otherwise we can spin here | |
2571 | * forever, while the workqueue is stuck trying to acquire the | |
2572 | * very same mutex. | |
2573 | */ | |
e9979b36 | 2574 | if (list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
2575 | u32 tail = ctx->cached_cq_tail; |
2576 | ||
500f9fba | 2577 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2578 | io_run_task_work(); |
500f9fba | 2579 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2580 | |
8f487ef2 PB |
2581 | /* some requests don't go through iopoll_list */ |
2582 | if (tail != ctx->cached_cq_tail || | |
2583 | list_empty(&ctx->iopoll_list)) | |
e9979b36 | 2584 | break; |
500f9fba | 2585 | } |
a8576af9 | 2586 | ret = io_do_iopoll(ctx, &nr_events, min); |
f39c8a5b PB |
2587 | } while (!ret && nr_events < min && !need_resched()); |
2588 | out: | |
500f9fba | 2589 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2590 | return ret; |
2591 | } | |
2592 | ||
491381ce | 2593 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2594 | { |
491381ce JA |
2595 | /* |
2596 | * Tell lockdep we inherited freeze protection from submission | |
2597 | * thread. | |
2598 | */ | |
2599 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2600 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2601 | |
1c98679d PB |
2602 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2603 | sb_end_write(sb); | |
2b188cc1 JA |
2604 | } |
2605 | } | |
2606 | ||
b63534c4 | 2607 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2608 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2609 | { |
ab454438 | 2610 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2611 | |
ab454438 PB |
2612 | if (!rw) |
2613 | return !io_req_prep_async(req); | |
2614 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2615 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2616 | return true; | |
b63534c4 | 2617 | } |
b63534c4 | 2618 | |
3e6a0d3c | 2619 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2620 | { |
355afaeb | 2621 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2622 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2623 | |
355afaeb JA |
2624 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2625 | return false; | |
3e6a0d3c JA |
2626 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2627 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2628 | return false; |
7c977a58 JA |
2629 | /* |
2630 | * If ref is dying, we might be running poll reap from the exit work. | |
2631 | * Don't attempt to reissue from that path, just let it fail with | |
2632 | * -EAGAIN. | |
2633 | */ | |
3e6a0d3c JA |
2634 | if (percpu_ref_is_dying(&ctx->refs)) |
2635 | return false; | |
ef046888 JA |
2636 | /* |
2637 | * Play it safe and assume not safe to re-import and reissue if we're | |
2638 | * not in the original thread group (or in task context). | |
2639 | */ | |
2640 | if (!same_thread_group(req->task, current) || !in_task()) | |
2641 | return false; | |
3e6a0d3c JA |
2642 | return true; |
2643 | } | |
e82ad485 | 2644 | #else |
a1ff1e3f | 2645 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2646 | { |
2647 | return false; | |
2648 | } | |
e82ad485 | 2649 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2650 | { |
b63534c4 JA |
2651 | return false; |
2652 | } | |
3e6a0d3c | 2653 | #endif |
b63534c4 | 2654 | |
8ef12efe | 2655 | static bool __io_complete_rw_common(struct io_kiocb *req, long res) |
a1d7c393 | 2656 | { |
b65c128f PB |
2657 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2658 | kiocb_end_write(req); | |
9532b99b PB |
2659 | if (res != req->result) { |
2660 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2661 | io_rw_should_reissue(req)) { | |
2662 | req->flags |= REQ_F_REISSUE; | |
8ef12efe | 2663 | return true; |
9532b99b | 2664 | } |
93d2bcd2 | 2665 | req_set_fail(req); |
8ef12efe | 2666 | req->result = res; |
9532b99b | 2667 | } |
8ef12efe JA |
2668 | return false; |
2669 | } | |
2670 | ||
f237c30a | 2671 | static void io_req_task_complete(struct io_kiocb *req, bool *locked) |
8ef12efe | 2672 | { |
126180b9 PB |
2673 | unsigned int cflags = io_put_rw_kbuf(req); |
2674 | long res = req->result; | |
2675 | ||
2676 | if (*locked) { | |
2677 | struct io_ring_ctx *ctx = req->ctx; | |
2678 | struct io_submit_state *state = &ctx->submit_state; | |
2679 | ||
2680 | io_req_complete_state(req, res, cflags); | |
2681 | state->compl_reqs[state->compl_nr++] = req; | |
2682 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2683 | io_submit_flush_completions(ctx); | |
2684 | } else { | |
2685 | io_req_complete_post(req, res, cflags); | |
2686 | } | |
8ef12efe JA |
2687 | } |
2688 | ||
2689 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, | |
2690 | unsigned int issue_flags) | |
2691 | { | |
2692 | if (__io_complete_rw_common(req, res)) | |
2693 | return; | |
63637853 | 2694 | __io_req_complete(req, issue_flags, req->result, io_put_rw_kbuf(req)); |
ba816ad6 JA |
2695 | } |
2696 | ||
2697 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2698 | { | |
9adbd45d | 2699 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2700 | |
8ef12efe JA |
2701 | if (__io_complete_rw_common(req, res)) |
2702 | return; | |
2703 | req->result = res; | |
2704 | req->io_task_work.func = io_req_task_complete; | |
2705 | io_req_task_work_add(req); | |
2b188cc1 JA |
2706 | } |
2707 | ||
def596e9 JA |
2708 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2709 | { | |
9adbd45d | 2710 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2711 | |
491381ce JA |
2712 | if (kiocb->ki_flags & IOCB_WRITE) |
2713 | kiocb_end_write(req); | |
9532b99b | 2714 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2715 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2716 | io_resubmit_prep(req))) { | |
93d2bcd2 | 2717 | req_set_fail(req); |
9532b99b PB |
2718 | req->flags |= REQ_F_DONT_REISSUE; |
2719 | } | |
8c130827 | 2720 | } |
bbde017a XW |
2721 | |
2722 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2723 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2724 | smp_wmb(); |
2725 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2726 | } |
2727 | ||
2728 | /* | |
2729 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2730 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2731 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2732 | * accessing the kiocb cookie. |
2733 | */ | |
cb3d8972 | 2734 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2735 | { |
2736 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2737 | const bool in_async = io_wq_current_is_worker(); |
2738 | ||
2739 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2740 | if (unlikely(in_async)) | |
2741 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2742 | |
2743 | /* | |
2744 | * Track whether we have multiple files in our lists. This will impact | |
2745 | * how we do polling eventually, not spinning if we're on potentially | |
2746 | * different devices. | |
2747 | */ | |
540e32a0 | 2748 | if (list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
2749 | ctx->poll_multi_queue = false; |
2750 | } else if (!ctx->poll_multi_queue) { | |
def596e9 | 2751 | struct io_kiocb *list_req; |
915b3dde | 2752 | unsigned int queue_num0, queue_num1; |
def596e9 | 2753 | |
540e32a0 | 2754 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2755 | inflight_entry); |
915b3dde HX |
2756 | |
2757 | if (list_req->file != req->file) { | |
2758 | ctx->poll_multi_queue = true; | |
2759 | } else { | |
2760 | queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie); | |
2761 | queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie); | |
2762 | if (queue_num0 != queue_num1) | |
2763 | ctx->poll_multi_queue = true; | |
2764 | } | |
def596e9 JA |
2765 | } |
2766 | ||
2767 | /* | |
2768 | * For fast devices, IO may have already completed. If it has, add | |
2769 | * it to the front so we find it first. | |
2770 | */ | |
65a6543d | 2771 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2772 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2773 | else |
d21ffe7e | 2774 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2775 | |
cb3d8972 PB |
2776 | if (unlikely(in_async)) { |
2777 | /* | |
2778 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2779 | * in sq thread task context or in io worker task context. If | |
2780 | * current task context is sq thread, we don't need to check | |
2781 | * whether should wake up sq thread. | |
2782 | */ | |
2783 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2784 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2785 | wake_up(&ctx->sq_data->wait); | |
2786 | ||
2787 | mutex_unlock(&ctx->uring_lock); | |
2788 | } | |
def596e9 JA |
2789 | } |
2790 | ||
4503b767 JA |
2791 | static bool io_bdev_nowait(struct block_device *bdev) |
2792 | { | |
9ba0d0c8 | 2793 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2794 | } |
2795 | ||
2b188cc1 JA |
2796 | /* |
2797 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2798 | * any file. For now, just ensure that anything potentially problematic is done | |
2799 | * inline. | |
2800 | */ | |
b191e2df | 2801 | static bool __io_file_supports_nowait(struct file *file, int rw) |
2b188cc1 JA |
2802 | { |
2803 | umode_t mode = file_inode(file)->i_mode; | |
2804 | ||
4503b767 | 2805 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2806 | if (IS_ENABLED(CONFIG_BLOCK) && |
2807 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2808 | return true; |
2809 | return false; | |
2810 | } | |
976517f1 | 2811 | if (S_ISSOCK(mode)) |
2b188cc1 | 2812 | return true; |
4503b767 | 2813 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2814 | if (IS_ENABLED(CONFIG_BLOCK) && |
2815 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2816 | file->f_op != &io_uring_fops) |
2817 | return true; | |
2818 | return false; | |
2819 | } | |
2b188cc1 | 2820 | |
c5b85625 JA |
2821 | /* any ->read/write should understand O_NONBLOCK */ |
2822 | if (file->f_flags & O_NONBLOCK) | |
2823 | return true; | |
2824 | ||
af197f50 JA |
2825 | if (!(file->f_mode & FMODE_NOWAIT)) |
2826 | return false; | |
2827 | ||
2828 | if (rw == READ) | |
2829 | return file->f_op->read_iter != NULL; | |
2830 | ||
2831 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2832 | } |
2833 | ||
b191e2df | 2834 | static bool io_file_supports_nowait(struct io_kiocb *req, int rw) |
7b29f92d | 2835 | { |
b191e2df | 2836 | if (rw == READ && (req->flags & REQ_F_NOWAIT_READ)) |
7b29f92d | 2837 | return true; |
b191e2df | 2838 | else if (rw == WRITE && (req->flags & REQ_F_NOWAIT_WRITE)) |
7b29f92d JA |
2839 | return true; |
2840 | ||
b191e2df | 2841 | return __io_file_supports_nowait(req->file, rw); |
7b29f92d JA |
2842 | } |
2843 | ||
a88fc400 | 2844 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2845 | { |
def596e9 | 2846 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2847 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2848 | struct file *file = req->file; |
09bb8394 JA |
2849 | unsigned ioprio; |
2850 | int ret; | |
2b188cc1 | 2851 | |
c97d8a0f | 2852 | if (!io_req_ffs_set(req) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2853 | req->flags |= REQ_F_ISREG; |
2854 | ||
2b188cc1 | 2855 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2856 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2857 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2858 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2859 | } |
2b188cc1 | 2860 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2861 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2862 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2863 | if (unlikely(ret)) | |
2864 | return ret; | |
2b188cc1 | 2865 | |
75c668cd PB |
2866 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2867 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2868 | req->flags |= REQ_F_NOWAIT; | |
2869 | ||
2b188cc1 JA |
2870 | ioprio = READ_ONCE(sqe->ioprio); |
2871 | if (ioprio) { | |
2872 | ret = ioprio_check_cap(ioprio); | |
2873 | if (ret) | |
09bb8394 | 2874 | return ret; |
2b188cc1 JA |
2875 | |
2876 | kiocb->ki_ioprio = ioprio; | |
2877 | } else | |
2878 | kiocb->ki_ioprio = get_current_ioprio(); | |
2879 | ||
def596e9 | 2880 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2881 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2882 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2883 | return -EOPNOTSUPP; |
2b188cc1 | 2884 | |
394918eb | 2885 | kiocb->ki_flags |= IOCB_HIPRI | IOCB_ALLOC_CACHE; |
def596e9 | 2886 | kiocb->ki_complete = io_complete_rw_iopoll; |
65a6543d | 2887 | req->iopoll_completed = 0; |
def596e9 | 2888 | } else { |
09bb8394 JA |
2889 | if (kiocb->ki_flags & IOCB_HIPRI) |
2890 | return -EINVAL; | |
def596e9 JA |
2891 | kiocb->ki_complete = io_complete_rw; |
2892 | } | |
9adbd45d | 2893 | |
eae071c9 PB |
2894 | if (req->opcode == IORING_OP_READ_FIXED || |
2895 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2896 | req->imu = NULL; | |
2897 | io_req_set_rsrc_node(req); | |
2898 | } | |
2899 | ||
3529d8c2 JA |
2900 | req->rw.addr = READ_ONCE(sqe->addr); |
2901 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2902 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2903 | return 0; |
2b188cc1 JA |
2904 | } |
2905 | ||
2906 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2907 | { | |
2908 | switch (ret) { | |
2909 | case -EIOCBQUEUED: | |
2910 | break; | |
2911 | case -ERESTARTSYS: | |
2912 | case -ERESTARTNOINTR: | |
2913 | case -ERESTARTNOHAND: | |
2914 | case -ERESTART_RESTARTBLOCK: | |
2915 | /* | |
2916 | * We can't just restart the syscall, since previously | |
2917 | * submitted sqes may already be in progress. Just fail this | |
2918 | * IO with EINTR. | |
2919 | */ | |
2920 | ret = -EINTR; | |
df561f66 | 2921 | fallthrough; |
2b188cc1 JA |
2922 | default: |
2923 | kiocb->ki_complete(kiocb, ret, 0); | |
2924 | } | |
2925 | } | |
2926 | ||
a1d7c393 | 2927 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2928 | unsigned int issue_flags) |
ba816ad6 | 2929 | { |
ba04291e | 2930 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2931 | struct io_async_rw *io = req->async_data; |
97284637 | 2932 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2933 | |
227c0c96 | 2934 | /* add previously done IO, if any */ |
e8c2bc1f | 2935 | if (io && io->bytes_done > 0) { |
227c0c96 | 2936 | if (ret < 0) |
e8c2bc1f | 2937 | ret = io->bytes_done; |
227c0c96 | 2938 | else |
e8c2bc1f | 2939 | ret += io->bytes_done; |
227c0c96 JA |
2940 | } |
2941 | ||
ba04291e JA |
2942 | if (req->flags & REQ_F_CUR_POS) |
2943 | req->file->f_pos = kiocb->ki_pos; | |
e149bd74 | 2944 | if (ret >= 0 && check_reissue) |
889fca73 | 2945 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2946 | else |
2947 | io_rw_done(kiocb, ret); | |
97284637 | 2948 | |
fe7e3257 | 2949 | if (check_reissue && (req->flags & REQ_F_REISSUE)) { |
97284637 | 2950 | req->flags &= ~REQ_F_REISSUE; |
a7be7c23 | 2951 | if (io_resubmit_prep(req)) { |
773af691 | 2952 | io_req_task_queue_reissue(req); |
8c130827 | 2953 | } else { |
93d2bcd2 | 2954 | req_set_fail(req); |
ae421d93 PB |
2955 | __io_req_complete(req, issue_flags, ret, |
2956 | io_put_rw_kbuf(req)); | |
97284637 PB |
2957 | } |
2958 | } | |
ba816ad6 JA |
2959 | } |
2960 | ||
eae071c9 PB |
2961 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2962 | struct io_mapped_ubuf *imu) | |
edafccee | 2963 | { |
9adbd45d | 2964 | size_t len = req->rw.len; |
75769e3f | 2965 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2966 | size_t offset; |
edafccee | 2967 | |
75769e3f | 2968 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2969 | return -EFAULT; |
2970 | /* not inside the mapped region */ | |
4751f53d | 2971 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2972 | return -EFAULT; |
2973 | ||
2974 | /* | |
2975 | * May not be a start of buffer, set size appropriately | |
2976 | * and advance us to the beginning. | |
2977 | */ | |
2978 | offset = buf_addr - imu->ubuf; | |
2979 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2980 | |
2981 | if (offset) { | |
2982 | /* | |
2983 | * Don't use iov_iter_advance() here, as it's really slow for | |
2984 | * using the latter parts of a big fixed buffer - it iterates | |
2985 | * over each segment manually. We can cheat a bit here, because | |
2986 | * we know that: | |
2987 | * | |
2988 | * 1) it's a BVEC iter, we set it up | |
2989 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2990 | * first and last bvec | |
2991 | * | |
2992 | * So just find our index, and adjust the iterator afterwards. | |
2993 | * If the offset is within the first bvec (or the whole first | |
2994 | * bvec, just use iov_iter_advance(). This makes it easier | |
2995 | * since we can just skip the first segment, which may not | |
2996 | * be PAGE_SIZE aligned. | |
2997 | */ | |
2998 | const struct bio_vec *bvec = imu->bvec; | |
2999 | ||
3000 | if (offset <= bvec->bv_len) { | |
3001 | iov_iter_advance(iter, offset); | |
3002 | } else { | |
3003 | unsigned long seg_skip; | |
3004 | ||
3005 | /* skip first vec */ | |
3006 | offset -= bvec->bv_len; | |
3007 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3008 | ||
3009 | iter->bvec = bvec + seg_skip; | |
3010 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3011 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3012 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3013 | } |
3014 | } | |
3015 | ||
847595de | 3016 | return 0; |
edafccee JA |
3017 | } |
3018 | ||
eae071c9 PB |
3019 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
3020 | { | |
3021 | struct io_ring_ctx *ctx = req->ctx; | |
3022 | struct io_mapped_ubuf *imu = req->imu; | |
3023 | u16 index, buf_index = req->buf_index; | |
3024 | ||
3025 | if (likely(!imu)) { | |
3026 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
3027 | return -EFAULT; | |
3028 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
3029 | imu = READ_ONCE(ctx->user_bufs[index]); | |
3030 | req->imu = imu; | |
3031 | } | |
3032 | return __io_import_fixed(req, rw, iter, imu); | |
3033 | } | |
3034 | ||
bcda7baa JA |
3035 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3036 | { | |
3037 | if (needs_lock) | |
3038 | mutex_unlock(&ctx->uring_lock); | |
3039 | } | |
3040 | ||
3041 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3042 | { | |
3043 | /* | |
3044 | * "Normal" inline submissions always hold the uring_lock, since we | |
3045 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3046 | * The only exception is when we've detached the request and issue it | |
3047 | * from an async worker thread, grab the lock for that case. | |
3048 | */ | |
3049 | if (needs_lock) | |
3050 | mutex_lock(&ctx->uring_lock); | |
3051 | } | |
3052 | ||
3053 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3054 | int bgid, struct io_buffer *kbuf, | |
3055 | bool needs_lock) | |
3056 | { | |
3057 | struct io_buffer *head; | |
3058 | ||
3059 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3060 | return kbuf; | |
3061 | ||
3062 | io_ring_submit_lock(req->ctx, needs_lock); | |
3063 | ||
3064 | lockdep_assert_held(&req->ctx->uring_lock); | |
3065 | ||
9e15c3a0 | 3066 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
3067 | if (head) { |
3068 | if (!list_empty(&head->list)) { | |
3069 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3070 | list); | |
3071 | list_del(&kbuf->list); | |
3072 | } else { | |
3073 | kbuf = head; | |
9e15c3a0 | 3074 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
3075 | } |
3076 | if (*len > kbuf->len) | |
3077 | *len = kbuf->len; | |
3078 | } else { | |
3079 | kbuf = ERR_PTR(-ENOBUFS); | |
3080 | } | |
3081 | ||
3082 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3083 | ||
3084 | return kbuf; | |
3085 | } | |
3086 | ||
4d954c25 JA |
3087 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3088 | bool needs_lock) | |
3089 | { | |
3090 | struct io_buffer *kbuf; | |
4f4eeba8 | 3091 | u16 bgid; |
4d954c25 JA |
3092 | |
3093 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3094 | bgid = req->buf_index; |
4d954c25 JA |
3095 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3096 | if (IS_ERR(kbuf)) | |
3097 | return kbuf; | |
3098 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3099 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3100 | return u64_to_user_ptr(kbuf->addr); | |
3101 | } | |
3102 | ||
3103 | #ifdef CONFIG_COMPAT | |
3104 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3105 | bool needs_lock) | |
3106 | { | |
3107 | struct compat_iovec __user *uiov; | |
3108 | compat_ssize_t clen; | |
3109 | void __user *buf; | |
3110 | ssize_t len; | |
3111 | ||
3112 | uiov = u64_to_user_ptr(req->rw.addr); | |
3113 | if (!access_ok(uiov, sizeof(*uiov))) | |
3114 | return -EFAULT; | |
3115 | if (__get_user(clen, &uiov->iov_len)) | |
3116 | return -EFAULT; | |
3117 | if (clen < 0) | |
3118 | return -EINVAL; | |
3119 | ||
3120 | len = clen; | |
3121 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3122 | if (IS_ERR(buf)) | |
3123 | return PTR_ERR(buf); | |
3124 | iov[0].iov_base = buf; | |
3125 | iov[0].iov_len = (compat_size_t) len; | |
3126 | return 0; | |
3127 | } | |
3128 | #endif | |
3129 | ||
3130 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3131 | bool needs_lock) | |
3132 | { | |
3133 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3134 | void __user *buf; | |
3135 | ssize_t len; | |
3136 | ||
3137 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3138 | return -EFAULT; | |
3139 | ||
3140 | len = iov[0].iov_len; | |
3141 | if (len < 0) | |
3142 | return -EINVAL; | |
3143 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3144 | if (IS_ERR(buf)) | |
3145 | return PTR_ERR(buf); | |
3146 | iov[0].iov_base = buf; | |
3147 | iov[0].iov_len = len; | |
3148 | return 0; | |
3149 | } | |
3150 | ||
3151 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3152 | bool needs_lock) | |
3153 | { | |
dddb3e26 JA |
3154 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3155 | struct io_buffer *kbuf; | |
3156 | ||
3157 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3158 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3159 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3160 | return 0; |
dddb3e26 | 3161 | } |
dd201662 | 3162 | if (req->rw.len != 1) |
4d954c25 JA |
3163 | return -EINVAL; |
3164 | ||
3165 | #ifdef CONFIG_COMPAT | |
3166 | if (req->ctx->compat) | |
3167 | return io_compat_import(req, iov, needs_lock); | |
3168 | #endif | |
3169 | ||
3170 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3171 | } | |
3172 | ||
847595de PB |
3173 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3174 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3175 | { |
9adbd45d JA |
3176 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3177 | size_t sqe_len = req->rw.len; | |
847595de | 3178 | u8 opcode = req->opcode; |
4d954c25 | 3179 | ssize_t ret; |
edafccee | 3180 | |
7d009165 | 3181 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3182 | *iovec = NULL; |
9adbd45d | 3183 | return io_import_fixed(req, rw, iter); |
edafccee | 3184 | } |
2b188cc1 | 3185 | |
bcda7baa | 3186 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3187 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3188 | return -EINVAL; |
3189 | ||
3a6820f2 | 3190 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3191 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3192 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3193 | if (IS_ERR(buf)) |
4d954c25 | 3194 | return PTR_ERR(buf); |
3f9d6441 | 3195 | req->rw.len = sqe_len; |
bcda7baa JA |
3196 | } |
3197 | ||
3a6820f2 JA |
3198 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3199 | *iovec = NULL; | |
10fc72e4 | 3200 | return ret; |
3a6820f2 JA |
3201 | } |
3202 | ||
4d954c25 JA |
3203 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3204 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3205 | if (!ret) |
3206 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3207 | *iovec = NULL; |
3208 | return ret; | |
3209 | } | |
3210 | ||
89cd35c5 CH |
3211 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3212 | req->ctx->compat); | |
2b188cc1 JA |
3213 | } |
3214 | ||
0fef9483 JA |
3215 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3216 | { | |
5b09e37e | 3217 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3218 | } |
3219 | ||
31b51510 | 3220 | /* |
32960613 JA |
3221 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3222 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3223 | */ |
4017eb91 | 3224 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3225 | { |
4017eb91 JA |
3226 | struct kiocb *kiocb = &req->rw.kiocb; |
3227 | struct file *file = req->file; | |
32960613 JA |
3228 | ssize_t ret = 0; |
3229 | ||
3230 | /* | |
3231 | * Don't support polled IO through this interface, and we can't | |
3232 | * support non-blocking either. For the latter, this just causes | |
3233 | * the kiocb to be handled from an async context. | |
3234 | */ | |
3235 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3236 | return -EOPNOTSUPP; | |
3237 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3238 | return -EAGAIN; | |
3239 | ||
3240 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3241 | struct iovec iovec; |
32960613 JA |
3242 | ssize_t nr; |
3243 | ||
311ae9e1 PB |
3244 | if (!iov_iter_is_bvec(iter)) { |
3245 | iovec = iov_iter_iovec(iter); | |
3246 | } else { | |
4017eb91 JA |
3247 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3248 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3249 | } |
3250 | ||
32960613 JA |
3251 | if (rw == READ) { |
3252 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3253 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3254 | } else { |
3255 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3256 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3257 | } |
3258 | ||
3259 | if (nr < 0) { | |
3260 | if (!ret) | |
3261 | ret = nr; | |
3262 | break; | |
3263 | } | |
3264 | ret += nr; | |
3265 | if (nr != iovec.iov_len) | |
3266 | break; | |
4017eb91 JA |
3267 | req->rw.len -= nr; |
3268 | req->rw.addr += nr; | |
32960613 JA |
3269 | iov_iter_advance(iter, nr); |
3270 | } | |
3271 | ||
3272 | return ret; | |
3273 | } | |
3274 | ||
ff6165b2 JA |
3275 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3276 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3277 | { |
e8c2bc1f | 3278 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3279 | |
ff6165b2 | 3280 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3281 | rw->free_iovec = iovec; |
227c0c96 | 3282 | rw->bytes_done = 0; |
ff6165b2 | 3283 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3284 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3285 | return; |
b64e3444 | 3286 | if (!iovec) { |
ff6165b2 JA |
3287 | unsigned iov_off = 0; |
3288 | ||
3289 | rw->iter.iov = rw->fast_iov; | |
3290 | if (iter->iov != fast_iov) { | |
3291 | iov_off = iter->iov - fast_iov; | |
3292 | rw->iter.iov += iov_off; | |
3293 | } | |
3294 | if (rw->fast_iov != fast_iov) | |
3295 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3296 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3297 | } else { |
3298 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3299 | } |
3300 | } | |
3301 | ||
6cb78689 | 3302 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3303 | { |
e8c2bc1f JA |
3304 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3305 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3306 | return req->async_data == NULL; | |
3d9932a8 XW |
3307 | } |
3308 | ||
ff6165b2 JA |
3309 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3310 | const struct iovec *fast_iov, | |
227c0c96 | 3311 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3312 | { |
26f0505a | 3313 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3314 | return 0; |
e8c2bc1f | 3315 | if (!req->async_data) { |
6cb78689 | 3316 | if (io_alloc_async_data(req)) { |
6bf985dc | 3317 | kfree(iovec); |
5d204bcf | 3318 | return -ENOMEM; |
6bf985dc | 3319 | } |
b7bb4f7d | 3320 | |
ff6165b2 | 3321 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3322 | } |
b7bb4f7d | 3323 | return 0; |
f67676d1 JA |
3324 | } |
3325 | ||
73debe68 | 3326 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3327 | { |
e8c2bc1f | 3328 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3329 | struct iovec *iov = iorw->fast_iov; |
847595de | 3330 | int ret; |
c3e330a4 | 3331 | |
2846c481 | 3332 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3333 | if (unlikely(ret < 0)) |
3334 | return ret; | |
3335 | ||
ab0b196c PB |
3336 | iorw->bytes_done = 0; |
3337 | iorw->free_iovec = iov; | |
3338 | if (iov) | |
3339 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3340 | return 0; |
3341 | } | |
3342 | ||
73debe68 | 3343 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3344 | { |
3529d8c2 JA |
3345 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3346 | return -EBADF; | |
93642ef8 | 3347 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3348 | } |
3349 | ||
c1dd91d1 JA |
3350 | /* |
3351 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3352 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3353 | * This gets called when the page is unlocked, and we generally expect that to | |
3354 | * happen when the page IO is completed and the page is now uptodate. This will | |
3355 | * queue a task_work based retry of the operation, attempting to copy the data | |
3356 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3357 | * do a thread based blocking retry of the operation. That's the unexpected | |
3358 | * slow path. | |
3359 | */ | |
bcf5a063 JA |
3360 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3361 | int sync, void *arg) | |
3362 | { | |
3363 | struct wait_page_queue *wpq; | |
3364 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3365 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3366 | |
3367 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3368 | ||
cdc8fcb4 LT |
3369 | if (!wake_page_match(wpq, key)) |
3370 | return 0; | |
3371 | ||
c8d317aa | 3372 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 | 3373 | list_del_init(&wait->entry); |
921b9054 | 3374 | io_req_task_queue(req); |
bcf5a063 JA |
3375 | return 1; |
3376 | } | |
3377 | ||
c1dd91d1 JA |
3378 | /* |
3379 | * This controls whether a given IO request should be armed for async page | |
3380 | * based retry. If we return false here, the request is handed to the async | |
3381 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3382 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3383 | * will either succeed because the page is now uptodate and unlocked, or it | |
3384 | * will register a callback when the page is unlocked at IO completion. Through | |
3385 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3386 | * That retry will attempt the buffered read again. The retry will generally | |
3387 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3388 | * async worker threads for a blocking retry. | |
3389 | */ | |
227c0c96 | 3390 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3391 | { |
e8c2bc1f JA |
3392 | struct io_async_rw *rw = req->async_data; |
3393 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3394 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3395 | |
bcf5a063 JA |
3396 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3397 | if (req->flags & REQ_F_NOWAIT) | |
3398 | return false; | |
f67676d1 | 3399 | |
227c0c96 | 3400 | /* Only for buffered IO */ |
3b2a4439 | 3401 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3402 | return false; |
3b2a4439 | 3403 | |
bcf5a063 JA |
3404 | /* |
3405 | * just use poll if we can, and don't attempt if the fs doesn't | |
3406 | * support callback based unlocks | |
3407 | */ | |
3408 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3409 | return false; | |
f67676d1 | 3410 | |
3b2a4439 JA |
3411 | wait->wait.func = io_async_buf_func; |
3412 | wait->wait.private = req; | |
3413 | wait->wait.flags = 0; | |
3414 | INIT_LIST_HEAD(&wait->wait.entry); | |
3415 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3416 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3417 | kiocb->ki_waitq = wait; |
3b2a4439 | 3418 | return true; |
bcf5a063 JA |
3419 | } |
3420 | ||
aeab9506 | 3421 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3422 | { |
3423 | if (req->file->f_op->read_iter) | |
3424 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3425 | else if (req->file->f_op->read) |
4017eb91 | 3426 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3427 | else |
3428 | return -EINVAL; | |
f67676d1 JA |
3429 | } |
3430 | ||
7db30437 ML |
3431 | static bool need_read_all(struct io_kiocb *req) |
3432 | { | |
3433 | return req->flags & REQ_F_ISREG || | |
3434 | S_ISBLK(file_inode(req->file)->i_mode); | |
3435 | } | |
3436 | ||
889fca73 | 3437 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3438 | { |
3439 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3440 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3441 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3442 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3443 | ssize_t io_size, ret, ret2; |
45d189c6 | 3444 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3445 | |
2846c481 | 3446 | if (rw) { |
e8c2bc1f | 3447 | iter = &rw->iter; |
2846c481 PB |
3448 | iovec = NULL; |
3449 | } else { | |
3450 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3451 | if (ret < 0) | |
3452 | return ret; | |
3453 | } | |
632546c4 | 3454 | io_size = iov_iter_count(iter); |
fa15bafb | 3455 | req->result = io_size; |
2b188cc1 | 3456 | |
fd6c2e4c JA |
3457 | /* Ensure we clear previously set non-block flag */ |
3458 | if (!force_nonblock) | |
29de5f6a | 3459 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3460 | else |
3461 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3462 | ||
24c74678 | 3463 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3464 | if (force_nonblock && !io_file_supports_nowait(req, READ)) { |
6713e7a6 | 3465 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3466 | return ret ?: -EAGAIN; |
6713e7a6 | 3467 | } |
9e645e11 | 3468 | |
632546c4 | 3469 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3470 | if (unlikely(ret)) { |
3471 | kfree(iovec); | |
3472 | return ret; | |
3473 | } | |
2b188cc1 | 3474 | |
227c0c96 | 3475 | ret = io_iter_do_read(req, iter); |
32960613 | 3476 | |
230d50d4 | 3477 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3478 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3479 | /* IOPOLL retry should happen for io-wq threads */ |
3480 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3481 | goto done; |
75c668cd PB |
3482 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3483 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3484 | goto done; |
84216315 | 3485 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3486 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3487 | ret = 0; |
230d50d4 JA |
3488 | } else if (ret == -EIOCBQUEUED) { |
3489 | goto out_free; | |
7335e3bf | 3490 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
7db30437 | 3491 | (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { |
7335e3bf | 3492 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3493 | goto done; |
227c0c96 JA |
3494 | } |
3495 | ||
227c0c96 | 3496 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3497 | if (ret2) |
3498 | return ret2; | |
3499 | ||
fe1cdd55 | 3500 | iovec = NULL; |
e8c2bc1f | 3501 | rw = req->async_data; |
227c0c96 | 3502 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3503 | iter = &rw->iter; |
227c0c96 | 3504 | |
b23df91b PB |
3505 | do { |
3506 | io_size -= ret; | |
3507 | rw->bytes_done += ret; | |
3508 | /* if we can retry, do so with the callbacks armed */ | |
3509 | if (!io_rw_should_retry(req)) { | |
3510 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3511 | return -EAGAIN; | |
3512 | } | |
3513 | ||
3514 | /* | |
3515 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3516 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3517 | * desired page gets unlocked. We can also get a partial read | |
3518 | * here, and if we do, then just retry at the new offset. | |
3519 | */ | |
3520 | ret = io_iter_do_read(req, iter); | |
3521 | if (ret == -EIOCBQUEUED) | |
3522 | return 0; | |
227c0c96 | 3523 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3524 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3525 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3526 | done: |
889fca73 | 3527 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3528 | out_free: |
3529 | /* it's faster to check here then delegate to kfree */ | |
3530 | if (iovec) | |
3531 | kfree(iovec); | |
5ea5dd45 | 3532 | return 0; |
2b188cc1 JA |
3533 | } |
3534 | ||
73debe68 | 3535 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3536 | { |
3529d8c2 JA |
3537 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3538 | return -EBADF; | |
93642ef8 | 3539 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3540 | } |
3541 | ||
889fca73 | 3542 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3543 | { |
3544 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3545 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3546 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3547 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3548 | ssize_t ret, ret2, io_size; |
45d189c6 | 3549 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3550 | |
2846c481 | 3551 | if (rw) { |
e8c2bc1f | 3552 | iter = &rw->iter; |
2846c481 PB |
3553 | iovec = NULL; |
3554 | } else { | |
3555 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3556 | if (ret < 0) | |
3557 | return ret; | |
3558 | } | |
632546c4 | 3559 | io_size = iov_iter_count(iter); |
fa15bafb | 3560 | req->result = io_size; |
2b188cc1 | 3561 | |
fd6c2e4c JA |
3562 | /* Ensure we clear previously set non-block flag */ |
3563 | if (!force_nonblock) | |
a88fc400 PB |
3564 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3565 | else | |
3566 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3567 | |
24c74678 | 3568 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3569 | if (force_nonblock && !io_file_supports_nowait(req, WRITE)) |
f67676d1 | 3570 | goto copy_iov; |
31b51510 | 3571 | |
10d59345 JA |
3572 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3573 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3574 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3575 | goto copy_iov; |
31b51510 | 3576 | |
632546c4 | 3577 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3578 | if (unlikely(ret)) |
3579 | goto out_free; | |
4ed734b0 | 3580 | |
fa15bafb PB |
3581 | /* |
3582 | * Open-code file_start_write here to grab freeze protection, | |
3583 | * which will be released by another thread in | |
3584 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3585 | * released so that it doesn't complain about the held lock when | |
3586 | * we return to userspace. | |
3587 | */ | |
3588 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3589 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3590 | __sb_writers_release(file_inode(req->file)->i_sb, |
3591 | SB_FREEZE_WRITE); | |
3592 | } | |
3593 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3594 | |
fa15bafb | 3595 | if (req->file->f_op->write_iter) |
ff6165b2 | 3596 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3597 | else if (req->file->f_op->write) |
4017eb91 | 3598 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3599 | else |
3600 | ret2 = -EINVAL; | |
4ed734b0 | 3601 | |
6ad7f233 PB |
3602 | if (req->flags & REQ_F_REISSUE) { |
3603 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3604 | ret2 = -EAGAIN; |
6ad7f233 | 3605 | } |
230d50d4 | 3606 | |
fa15bafb PB |
3607 | /* |
3608 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3609 | * retry them without IOCB_NOWAIT. | |
3610 | */ | |
3611 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3612 | ret2 = -EAGAIN; | |
75c668cd PB |
3613 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3614 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3615 | goto done; |
fa15bafb | 3616 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3617 | /* IOPOLL retry should happen for io-wq threads */ |
3618 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3619 | goto copy_iov; | |
355afaeb | 3620 | done: |
889fca73 | 3621 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3622 | } else { |
f67676d1 | 3623 | copy_iov: |
84216315 | 3624 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3625 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3626 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3627 | return ret ?: -EAGAIN; |
2b188cc1 | 3628 | } |
31b51510 | 3629 | out_free: |
f261c168 | 3630 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3631 | if (iovec) |
6f2cc166 | 3632 | kfree(iovec); |
2b188cc1 JA |
3633 | return ret; |
3634 | } | |
3635 | ||
80a261fd JA |
3636 | static int io_renameat_prep(struct io_kiocb *req, |
3637 | const struct io_uring_sqe *sqe) | |
3638 | { | |
3639 | struct io_rename *ren = &req->rename; | |
3640 | const char __user *oldf, *newf; | |
3641 | ||
ed7eb259 JA |
3642 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3643 | return -EINVAL; | |
26578cda | 3644 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
ed7eb259 | 3645 | return -EINVAL; |
80a261fd JA |
3646 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3647 | return -EBADF; | |
3648 | ||
3649 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3650 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3651 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3652 | ren->new_dfd = READ_ONCE(sqe->len); | |
3653 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3654 | ||
3655 | ren->oldpath = getname(oldf); | |
3656 | if (IS_ERR(ren->oldpath)) | |
3657 | return PTR_ERR(ren->oldpath); | |
3658 | ||
3659 | ren->newpath = getname(newf); | |
3660 | if (IS_ERR(ren->newpath)) { | |
3661 | putname(ren->oldpath); | |
3662 | return PTR_ERR(ren->newpath); | |
3663 | } | |
3664 | ||
3665 | req->flags |= REQ_F_NEED_CLEANUP; | |
3666 | return 0; | |
3667 | } | |
3668 | ||
45d189c6 | 3669 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3670 | { |
3671 | struct io_rename *ren = &req->rename; | |
3672 | int ret; | |
3673 | ||
45d189c6 | 3674 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3675 | return -EAGAIN; |
3676 | ||
3677 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3678 | ren->newpath, ren->flags); | |
3679 | ||
3680 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3681 | if (ret < 0) | |
93d2bcd2 | 3682 | req_set_fail(req); |
80a261fd JA |
3683 | io_req_complete(req, ret); |
3684 | return 0; | |
3685 | } | |
3686 | ||
14a1143b JA |
3687 | static int io_unlinkat_prep(struct io_kiocb *req, |
3688 | const struct io_uring_sqe *sqe) | |
3689 | { | |
3690 | struct io_unlink *un = &req->unlink; | |
3691 | const char __user *fname; | |
3692 | ||
22634bc5 JA |
3693 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3694 | return -EINVAL; | |
26578cda PB |
3695 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || |
3696 | sqe->splice_fd_in) | |
22634bc5 | 3697 | return -EINVAL; |
14a1143b JA |
3698 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3699 | return -EBADF; | |
3700 | ||
3701 | un->dfd = READ_ONCE(sqe->fd); | |
3702 | ||
3703 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3704 | if (un->flags & ~AT_REMOVEDIR) | |
3705 | return -EINVAL; | |
3706 | ||
3707 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3708 | un->filename = getname(fname); | |
3709 | if (IS_ERR(un->filename)) | |
3710 | return PTR_ERR(un->filename); | |
3711 | ||
3712 | req->flags |= REQ_F_NEED_CLEANUP; | |
3713 | return 0; | |
3714 | } | |
3715 | ||
45d189c6 | 3716 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3717 | { |
3718 | struct io_unlink *un = &req->unlink; | |
3719 | int ret; | |
3720 | ||
45d189c6 | 3721 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3722 | return -EAGAIN; |
3723 | ||
3724 | if (un->flags & AT_REMOVEDIR) | |
3725 | ret = do_rmdir(un->dfd, un->filename); | |
3726 | else | |
3727 | ret = do_unlinkat(un->dfd, un->filename); | |
3728 | ||
3729 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3730 | if (ret < 0) | |
93d2bcd2 | 3731 | req_set_fail(req); |
14a1143b JA |
3732 | io_req_complete(req, ret); |
3733 | return 0; | |
3734 | } | |
3735 | ||
e34a02dc DK |
3736 | static int io_mkdirat_prep(struct io_kiocb *req, |
3737 | const struct io_uring_sqe *sqe) | |
3738 | { | |
3739 | struct io_mkdir *mkd = &req->mkdir; | |
3740 | const char __user *fname; | |
3741 | ||
3742 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3743 | return -EINVAL; | |
3744 | if (sqe->ioprio || sqe->off || sqe->rw_flags || sqe->buf_index || | |
3745 | sqe->splice_fd_in) | |
3746 | return -EINVAL; | |
3747 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3748 | return -EBADF; | |
3749 | ||
3750 | mkd->dfd = READ_ONCE(sqe->fd); | |
3751 | mkd->mode = READ_ONCE(sqe->len); | |
3752 | ||
3753 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3754 | mkd->filename = getname(fname); | |
3755 | if (IS_ERR(mkd->filename)) | |
3756 | return PTR_ERR(mkd->filename); | |
3757 | ||
3758 | req->flags |= REQ_F_NEED_CLEANUP; | |
3759 | return 0; | |
3760 | } | |
3761 | ||
3762 | static int io_mkdirat(struct io_kiocb *req, int issue_flags) | |
3763 | { | |
3764 | struct io_mkdir *mkd = &req->mkdir; | |
3765 | int ret; | |
3766 | ||
3767 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3768 | return -EAGAIN; | |
3769 | ||
3770 | ret = do_mkdirat(mkd->dfd, mkd->filename, mkd->mode); | |
3771 | ||
3772 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3773 | if (ret < 0) | |
3774 | req_set_fail(req); | |
3775 | io_req_complete(req, ret); | |
3776 | return 0; | |
3777 | } | |
3778 | ||
7a8721f8 DK |
3779 | static int io_symlinkat_prep(struct io_kiocb *req, |
3780 | const struct io_uring_sqe *sqe) | |
3781 | { | |
3782 | struct io_symlink *sl = &req->symlink; | |
3783 | const char __user *oldpath, *newpath; | |
3784 | ||
3785 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3786 | return -EINVAL; | |
3787 | if (sqe->ioprio || sqe->len || sqe->rw_flags || sqe->buf_index || | |
3788 | sqe->splice_fd_in) | |
3789 | return -EINVAL; | |
3790 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3791 | return -EBADF; | |
3792 | ||
3793 | sl->new_dfd = READ_ONCE(sqe->fd); | |
3794 | oldpath = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3795 | newpath = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3796 | ||
3797 | sl->oldpath = getname(oldpath); | |
3798 | if (IS_ERR(sl->oldpath)) | |
3799 | return PTR_ERR(sl->oldpath); | |
3800 | ||
3801 | sl->newpath = getname(newpath); | |
3802 | if (IS_ERR(sl->newpath)) { | |
3803 | putname(sl->oldpath); | |
3804 | return PTR_ERR(sl->newpath); | |
3805 | } | |
3806 | ||
3807 | req->flags |= REQ_F_NEED_CLEANUP; | |
3808 | return 0; | |
3809 | } | |
3810 | ||
3811 | static int io_symlinkat(struct io_kiocb *req, int issue_flags) | |
3812 | { | |
3813 | struct io_symlink *sl = &req->symlink; | |
3814 | int ret; | |
3815 | ||
3816 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3817 | return -EAGAIN; | |
3818 | ||
3819 | ret = do_symlinkat(sl->oldpath, sl->new_dfd, sl->newpath); | |
3820 | ||
3821 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3822 | if (ret < 0) | |
3823 | req_set_fail(req); | |
3824 | io_req_complete(req, ret); | |
3825 | return 0; | |
3826 | } | |
3827 | ||
cf30da90 DK |
3828 | static int io_linkat_prep(struct io_kiocb *req, |
3829 | const struct io_uring_sqe *sqe) | |
3830 | { | |
3831 | struct io_hardlink *lnk = &req->hardlink; | |
3832 | const char __user *oldf, *newf; | |
3833 | ||
3834 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3835 | return -EINVAL; | |
3836 | if (sqe->ioprio || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) | |
3837 | return -EINVAL; | |
3838 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3839 | return -EBADF; | |
3840 | ||
3841 | lnk->old_dfd = READ_ONCE(sqe->fd); | |
3842 | lnk->new_dfd = READ_ONCE(sqe->len); | |
3843 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3844 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3845 | lnk->flags = READ_ONCE(sqe->hardlink_flags); | |
3846 | ||
3847 | lnk->oldpath = getname(oldf); | |
3848 | if (IS_ERR(lnk->oldpath)) | |
3849 | return PTR_ERR(lnk->oldpath); | |
3850 | ||
3851 | lnk->newpath = getname(newf); | |
3852 | if (IS_ERR(lnk->newpath)) { | |
3853 | putname(lnk->oldpath); | |
3854 | return PTR_ERR(lnk->newpath); | |
3855 | } | |
3856 | ||
3857 | req->flags |= REQ_F_NEED_CLEANUP; | |
3858 | return 0; | |
3859 | } | |
3860 | ||
3861 | static int io_linkat(struct io_kiocb *req, int issue_flags) | |
3862 | { | |
3863 | struct io_hardlink *lnk = &req->hardlink; | |
3864 | int ret; | |
3865 | ||
3866 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3867 | return -EAGAIN; | |
3868 | ||
3869 | ret = do_linkat(lnk->old_dfd, lnk->oldpath, lnk->new_dfd, | |
3870 | lnk->newpath, lnk->flags); | |
3871 | ||
3872 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3873 | if (ret < 0) | |
3874 | req_set_fail(req); | |
3875 | io_req_complete(req, ret); | |
3876 | return 0; | |
3877 | } | |
3878 | ||
36f4fa68 JA |
3879 | static int io_shutdown_prep(struct io_kiocb *req, |
3880 | const struct io_uring_sqe *sqe) | |
3881 | { | |
3882 | #if defined(CONFIG_NET) | |
3883 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3884 | return -EINVAL; | |
26578cda PB |
3885 | if (unlikely(sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || |
3886 | sqe->buf_index || sqe->splice_fd_in)) | |
36f4fa68 JA |
3887 | return -EINVAL; |
3888 | ||
3889 | req->shutdown.how = READ_ONCE(sqe->len); | |
3890 | return 0; | |
3891 | #else | |
3892 | return -EOPNOTSUPP; | |
3893 | #endif | |
3894 | } | |
3895 | ||
45d189c6 | 3896 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3897 | { |
3898 | #if defined(CONFIG_NET) | |
3899 | struct socket *sock; | |
3900 | int ret; | |
3901 | ||
45d189c6 | 3902 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3903 | return -EAGAIN; |
3904 | ||
48aba79b | 3905 | sock = sock_from_file(req->file); |
36f4fa68 | 3906 | if (unlikely(!sock)) |
48aba79b | 3907 | return -ENOTSOCK; |
36f4fa68 JA |
3908 | |
3909 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3910 | if (ret < 0) |
93d2bcd2 | 3911 | req_set_fail(req); |
36f4fa68 JA |
3912 | io_req_complete(req, ret); |
3913 | return 0; | |
3914 | #else | |
3915 | return -EOPNOTSUPP; | |
3916 | #endif | |
3917 | } | |
3918 | ||
f2a8d5c7 PB |
3919 | static int __io_splice_prep(struct io_kiocb *req, |
3920 | const struct io_uring_sqe *sqe) | |
7d67af2c | 3921 | { |
fe7e3257 | 3922 | struct io_splice *sp = &req->splice; |
7d67af2c | 3923 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 3924 | |
3232dd02 PB |
3925 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3926 | return -EINVAL; | |
7d67af2c PB |
3927 | |
3928 | sp->file_in = NULL; | |
7d67af2c PB |
3929 | sp->len = READ_ONCE(sqe->len); |
3930 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3931 | ||
3932 | if (unlikely(sp->flags & ~valid_flags)) | |
3933 | return -EINVAL; | |
3934 | ||
62906e89 | 3935 | sp->file_in = io_file_get(req->ctx, req, READ_ONCE(sqe->splice_fd_in), |
8371adf5 PB |
3936 | (sp->flags & SPLICE_F_FD_IN_FIXED)); |
3937 | if (!sp->file_in) | |
3938 | return -EBADF; | |
7d67af2c | 3939 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3940 | return 0; |
3941 | } | |
3942 | ||
f2a8d5c7 PB |
3943 | static int io_tee_prep(struct io_kiocb *req, |
3944 | const struct io_uring_sqe *sqe) | |
3945 | { | |
3946 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3947 | return -EINVAL; | |
3948 | return __io_splice_prep(req, sqe); | |
3949 | } | |
3950 | ||
45d189c6 | 3951 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3952 | { |
3953 | struct io_splice *sp = &req->splice; | |
3954 | struct file *in = sp->file_in; | |
3955 | struct file *out = sp->file_out; | |
3956 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3957 | long ret = 0; | |
3958 | ||
45d189c6 | 3959 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3960 | return -EAGAIN; |
3961 | if (sp->len) | |
3962 | ret = do_tee(in, out, sp->len, flags); | |
3963 | ||
e1d767f0 PB |
3964 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3965 | io_put_file(in); | |
f2a8d5c7 PB |
3966 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3967 | ||
f2a8d5c7 | 3968 | if (ret != sp->len) |
93d2bcd2 | 3969 | req_set_fail(req); |
e1e16097 | 3970 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3971 | return 0; |
3972 | } | |
3973 | ||
3974 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3975 | { | |
fe7e3257 | 3976 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
3977 | |
3978 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3979 | sp->off_out = READ_ONCE(sqe->off); | |
3980 | return __io_splice_prep(req, sqe); | |
3981 | } | |
3982 | ||
45d189c6 | 3983 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3984 | { |
3985 | struct io_splice *sp = &req->splice; | |
3986 | struct file *in = sp->file_in; | |
3987 | struct file *out = sp->file_out; | |
3988 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3989 | loff_t *poff_in, *poff_out; | |
c9687426 | 3990 | long ret = 0; |
7d67af2c | 3991 | |
45d189c6 | 3992 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3993 | return -EAGAIN; |
7d67af2c PB |
3994 | |
3995 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3996 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3997 | |
948a7749 | 3998 | if (sp->len) |
c9687426 | 3999 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 4000 | |
e1d767f0 PB |
4001 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4002 | io_put_file(in); | |
7d67af2c PB |
4003 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4004 | ||
7d67af2c | 4005 | if (ret != sp->len) |
93d2bcd2 | 4006 | req_set_fail(req); |
e1e16097 | 4007 | io_req_complete(req, ret); |
7d67af2c PB |
4008 | return 0; |
4009 | } | |
4010 | ||
2b188cc1 JA |
4011 | /* |
4012 | * IORING_OP_NOP just posts a completion event, nothing else. | |
4013 | */ | |
889fca73 | 4014 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
4015 | { |
4016 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4017 | |
def596e9 JA |
4018 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4019 | return -EINVAL; | |
4020 | ||
889fca73 | 4021 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4022 | return 0; |
4023 | } | |
4024 | ||
1155c76a | 4025 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4026 | { |
6b06314c | 4027 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4028 | |
09bb8394 JA |
4029 | if (!req->file) |
4030 | return -EBADF; | |
c992fe29 | 4031 | |
6b06314c | 4032 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4033 | return -EINVAL; |
26578cda PB |
4034 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4035 | sqe->splice_fd_in)) | |
c992fe29 CH |
4036 | return -EINVAL; |
4037 | ||
8ed8d3c3 JA |
4038 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4039 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4040 | return -EINVAL; | |
4041 | ||
4042 | req->sync.off = READ_ONCE(sqe->off); | |
4043 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4044 | return 0; |
4045 | } | |
4046 | ||
45d189c6 | 4047 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4048 | { |
8ed8d3c3 | 4049 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4050 | int ret; |
4051 | ||
ac45abc0 | 4052 | /* fsync always requires a blocking context */ |
45d189c6 | 4053 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4054 | return -EAGAIN; |
4055 | ||
9adbd45d | 4056 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4057 | end > 0 ? end : LLONG_MAX, |
4058 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4059 | if (ret < 0) | |
93d2bcd2 | 4060 | req_set_fail(req); |
e1e16097 | 4061 | io_req_complete(req, ret); |
c992fe29 CH |
4062 | return 0; |
4063 | } | |
4064 | ||
d63d1b5e JA |
4065 | static int io_fallocate_prep(struct io_kiocb *req, |
4066 | const struct io_uring_sqe *sqe) | |
4067 | { | |
26578cda PB |
4068 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags || |
4069 | sqe->splice_fd_in) | |
d63d1b5e | 4070 | return -EINVAL; |
3232dd02 PB |
4071 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4072 | return -EINVAL; | |
d63d1b5e JA |
4073 | |
4074 | req->sync.off = READ_ONCE(sqe->off); | |
4075 | req->sync.len = READ_ONCE(sqe->addr); | |
4076 | req->sync.mode = READ_ONCE(sqe->len); | |
4077 | return 0; | |
4078 | } | |
4079 | ||
45d189c6 | 4080 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4081 | { |
ac45abc0 PB |
4082 | int ret; |
4083 | ||
d63d1b5e | 4084 | /* fallocate always requiring blocking context */ |
45d189c6 | 4085 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4086 | return -EAGAIN; |
ac45abc0 PB |
4087 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4088 | req->sync.len); | |
ac45abc0 | 4089 | if (ret < 0) |
93d2bcd2 | 4090 | req_set_fail(req); |
e1e16097 | 4091 | io_req_complete(req, ret); |
5d17b4a4 JA |
4092 | return 0; |
4093 | } | |
4094 | ||
ec65fea5 | 4095 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4096 | { |
f8748881 | 4097 | const char __user *fname; |
15b71abe | 4098 | int ret; |
b7bb4f7d | 4099 | |
d3fddf6d PB |
4100 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4101 | return -EINVAL; | |
b9445598 | 4102 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4103 | return -EINVAL; |
ec65fea5 | 4104 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4105 | return -EBADF; |
03b1230c | 4106 | |
ec65fea5 PB |
4107 | /* open.how should be already initialised */ |
4108 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4109 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4110 | |
25e72d10 PB |
4111 | req->open.dfd = READ_ONCE(sqe->fd); |
4112 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4113 | req->open.filename = getname(fname); |
15b71abe JA |
4114 | if (IS_ERR(req->open.filename)) { |
4115 | ret = PTR_ERR(req->open.filename); | |
4116 | req->open.filename = NULL; | |
4117 | return ret; | |
4118 | } | |
b9445598 PB |
4119 | |
4120 | req->open.file_slot = READ_ONCE(sqe->file_index); | |
4121 | if (req->open.file_slot && (req->open.how.flags & O_CLOEXEC)) | |
4122 | return -EINVAL; | |
4123 | ||
4022e7af | 4124 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4125 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4126 | return 0; |
03b1230c JA |
4127 | } |
4128 | ||
ec65fea5 PB |
4129 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4130 | { | |
d3fddf6d PB |
4131 | u64 mode = READ_ONCE(sqe->len); |
4132 | u64 flags = READ_ONCE(sqe->open_flags); | |
ec65fea5 | 4133 | |
ec65fea5 PB |
4134 | req->open.how = build_open_how(flags, mode); |
4135 | return __io_openat_prep(req, sqe); | |
4136 | } | |
4137 | ||
cebdb986 | 4138 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4139 | { |
cebdb986 | 4140 | struct open_how __user *how; |
cebdb986 | 4141 | size_t len; |
0fa03c62 JA |
4142 | int ret; |
4143 | ||
cebdb986 JA |
4144 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4145 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4146 | if (len < OPEN_HOW_SIZE_VER0) |
4147 | return -EINVAL; | |
3529d8c2 | 4148 | |
cebdb986 JA |
4149 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4150 | len); | |
4151 | if (ret) | |
4152 | return ret; | |
3529d8c2 | 4153 | |
ec65fea5 | 4154 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4155 | } |
4156 | ||
45d189c6 | 4157 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4158 | { |
4159 | struct open_flags op; | |
15b71abe | 4160 | struct file *file; |
b9445598 PB |
4161 | bool resolve_nonblock, nonblock_set; |
4162 | bool fixed = !!req->open.file_slot; | |
15b71abe JA |
4163 | int ret; |
4164 | ||
cebdb986 | 4165 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4166 | if (ret) |
4167 | goto err; | |
3a81fd02 JA |
4168 | nonblock_set = op.open_flag & O_NONBLOCK; |
4169 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4170 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4171 | /* |
4172 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4173 | * it'll always -EAGAIN | |
4174 | */ | |
4175 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4176 | return -EAGAIN; | |
4177 | op.lookup_flags |= LOOKUP_CACHED; | |
4178 | op.open_flag |= O_NONBLOCK; | |
4179 | } | |
15b71abe | 4180 | |
b9445598 PB |
4181 | if (!fixed) { |
4182 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); | |
4183 | if (ret < 0) | |
4184 | goto err; | |
4185 | } | |
15b71abe JA |
4186 | |
4187 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 4188 | if (IS_ERR(file)) { |
944d1444 | 4189 | /* |
12dcb58a PB |
4190 | * We could hang on to this 'fd' on retrying, but seems like |
4191 | * marginal gain for something that is now known to be a slower | |
4192 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 4193 | */ |
b9445598 PB |
4194 | if (!fixed) |
4195 | put_unused_fd(ret); | |
3a81fd02 | 4196 | |
15b71abe | 4197 | ret = PTR_ERR(file); |
12dcb58a PB |
4198 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
4199 | if (ret == -EAGAIN && | |
4200 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
4201 | return -EAGAIN; | |
4202 | goto err; | |
15b71abe | 4203 | } |
12dcb58a PB |
4204 | |
4205 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
4206 | file->f_flags &= ~O_NONBLOCK; | |
4207 | fsnotify_open(file); | |
b9445598 PB |
4208 | |
4209 | if (!fixed) | |
4210 | fd_install(ret, file); | |
4211 | else | |
4212 | ret = io_install_fixed_file(req, file, issue_flags, | |
4213 | req->open.file_slot - 1); | |
15b71abe JA |
4214 | err: |
4215 | putname(req->open.filename); | |
8fef80bf | 4216 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 4217 | if (ret < 0) |
93d2bcd2 | 4218 | req_set_fail(req); |
0bdf3398 | 4219 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
4220 | return 0; |
4221 | } | |
4222 | ||
45d189c6 | 4223 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4224 | { |
e45cff58 | 4225 | return io_openat2(req, issue_flags); |
cebdb986 JA |
4226 | } |
4227 | ||
067524e9 JA |
4228 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4229 | const struct io_uring_sqe *sqe) | |
4230 | { | |
4231 | struct io_provide_buf *p = &req->pbuf; | |
4232 | u64 tmp; | |
4233 | ||
26578cda PB |
4234 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off || |
4235 | sqe->splice_fd_in) | |
067524e9 JA |
4236 | return -EINVAL; |
4237 | ||
4238 | tmp = READ_ONCE(sqe->fd); | |
4239 | if (!tmp || tmp > USHRT_MAX) | |
4240 | return -EINVAL; | |
4241 | ||
4242 | memset(p, 0, sizeof(*p)); | |
4243 | p->nbufs = tmp; | |
4244 | p->bgid = READ_ONCE(sqe->buf_group); | |
4245 | return 0; | |
4246 | } | |
4247 | ||
4248 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4249 | int bgid, unsigned nbufs) | |
4250 | { | |
4251 | unsigned i = 0; | |
4252 | ||
4253 | /* shouldn't happen */ | |
4254 | if (!nbufs) | |
4255 | return 0; | |
4256 | ||
4257 | /* the head kbuf is the list itself */ | |
4258 | while (!list_empty(&buf->list)) { | |
4259 | struct io_buffer *nxt; | |
4260 | ||
4261 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4262 | list_del(&nxt->list); | |
4263 | kfree(nxt); | |
4264 | if (++i == nbufs) | |
4265 | return i; | |
4266 | } | |
4267 | i++; | |
4268 | kfree(buf); | |
9e15c3a0 | 4269 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
4270 | |
4271 | return i; | |
4272 | } | |
4273 | ||
889fca73 | 4274 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4275 | { |
4276 | struct io_provide_buf *p = &req->pbuf; | |
4277 | struct io_ring_ctx *ctx = req->ctx; | |
4278 | struct io_buffer *head; | |
4279 | int ret = 0; | |
45d189c6 | 4280 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4281 | |
4282 | io_ring_submit_lock(ctx, !force_nonblock); | |
4283 | ||
4284 | lockdep_assert_held(&ctx->uring_lock); | |
4285 | ||
4286 | ret = -ENOENT; | |
9e15c3a0 | 4287 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
4288 | if (head) |
4289 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 4290 | if (ret < 0) |
93d2bcd2 | 4291 | req_set_fail(req); |
067524e9 | 4292 | |
9fb8cb49 PB |
4293 | /* complete before unlock, IOPOLL may need the lock */ |
4294 | __io_req_complete(req, issue_flags, ret, 0); | |
4295 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
4296 | return 0; |
4297 | } | |
4298 | ||
ddf0322d JA |
4299 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4300 | const struct io_uring_sqe *sqe) | |
4301 | { | |
38134ada | 4302 | unsigned long size, tmp_check; |
ddf0322d JA |
4303 | struct io_provide_buf *p = &req->pbuf; |
4304 | u64 tmp; | |
4305 | ||
26578cda | 4306 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
ddf0322d JA |
4307 | return -EINVAL; |
4308 | ||
4309 | tmp = READ_ONCE(sqe->fd); | |
4310 | if (!tmp || tmp > USHRT_MAX) | |
4311 | return -E2BIG; | |
4312 | p->nbufs = tmp; | |
4313 | p->addr = READ_ONCE(sqe->addr); | |
4314 | p->len = READ_ONCE(sqe->len); | |
4315 | ||
38134ada PB |
4316 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
4317 | &size)) | |
4318 | return -EOVERFLOW; | |
4319 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
4320 | return -EOVERFLOW; | |
4321 | ||
d81269fe PB |
4322 | size = (unsigned long)p->len * p->nbufs; |
4323 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
4324 | return -EFAULT; |
4325 | ||
4326 | p->bgid = READ_ONCE(sqe->buf_group); | |
4327 | tmp = READ_ONCE(sqe->off); | |
4328 | if (tmp > USHRT_MAX) | |
4329 | return -E2BIG; | |
4330 | p->bid = tmp; | |
4331 | return 0; | |
4332 | } | |
4333 | ||
4334 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4335 | { | |
4336 | struct io_buffer *buf; | |
4337 | u64 addr = pbuf->addr; | |
4338 | int i, bid = pbuf->bid; | |
4339 | ||
4340 | for (i = 0; i < pbuf->nbufs; i++) { | |
4341 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4342 | if (!buf) | |
4343 | break; | |
4344 | ||
4345 | buf->addr = addr; | |
d1f82808 | 4346 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
4347 | buf->bid = bid; |
4348 | addr += pbuf->len; | |
4349 | bid++; | |
4350 | if (!*head) { | |
4351 | INIT_LIST_HEAD(&buf->list); | |
4352 | *head = buf; | |
4353 | } else { | |
4354 | list_add_tail(&buf->list, &(*head)->list); | |
4355 | } | |
4356 | } | |
4357 | ||
4358 | return i ? i : -ENOMEM; | |
4359 | } | |
4360 | ||
889fca73 | 4361 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4362 | { |
4363 | struct io_provide_buf *p = &req->pbuf; | |
4364 | struct io_ring_ctx *ctx = req->ctx; | |
4365 | struct io_buffer *head, *list; | |
4366 | int ret = 0; | |
45d189c6 | 4367 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4368 | |
4369 | io_ring_submit_lock(ctx, !force_nonblock); | |
4370 | ||
4371 | lockdep_assert_held(&ctx->uring_lock); | |
4372 | ||
9e15c3a0 | 4373 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4374 | |
4375 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4376 | if (ret >= 0 && !list) { |
4377 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4378 | if (ret < 0) | |
067524e9 | 4379 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4380 | } |
ddf0322d | 4381 | if (ret < 0) |
93d2bcd2 | 4382 | req_set_fail(req); |
9fb8cb49 PB |
4383 | /* complete before unlock, IOPOLL may need the lock */ |
4384 | __io_req_complete(req, issue_flags, ret, 0); | |
4385 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4386 | return 0; |
cebdb986 JA |
4387 | } |
4388 | ||
3e4827b0 JA |
4389 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4390 | const struct io_uring_sqe *sqe) | |
4391 | { | |
4392 | #if defined(CONFIG_EPOLL) | |
26578cda | 4393 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
3e4827b0 | 4394 | return -EINVAL; |
2d74d042 | 4395 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4396 | return -EINVAL; |
3e4827b0 JA |
4397 | |
4398 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4399 | req->epoll.op = READ_ONCE(sqe->len); | |
4400 | req->epoll.fd = READ_ONCE(sqe->off); | |
4401 | ||
4402 | if (ep_op_has_event(req->epoll.op)) { | |
4403 | struct epoll_event __user *ev; | |
4404 | ||
4405 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4406 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4407 | return -EFAULT; | |
4408 | } | |
4409 | ||
4410 | return 0; | |
4411 | #else | |
4412 | return -EOPNOTSUPP; | |
4413 | #endif | |
4414 | } | |
4415 | ||
889fca73 | 4416 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4417 | { |
4418 | #if defined(CONFIG_EPOLL) | |
4419 | struct io_epoll *ie = &req->epoll; | |
4420 | int ret; | |
45d189c6 | 4421 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4422 | |
4423 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4424 | if (force_nonblock && ret == -EAGAIN) | |
4425 | return -EAGAIN; | |
4426 | ||
4427 | if (ret < 0) | |
93d2bcd2 | 4428 | req_set_fail(req); |
889fca73 | 4429 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4430 | return 0; |
4431 | #else | |
4432 | return -EOPNOTSUPP; | |
4433 | #endif | |
4434 | } | |
4435 | ||
c1ca757b JA |
4436 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4437 | { | |
4438 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
26578cda | 4439 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->splice_fd_in) |
c1ca757b | 4440 | return -EINVAL; |
3232dd02 PB |
4441 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4442 | return -EINVAL; | |
c1ca757b JA |
4443 | |
4444 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4445 | req->madvise.len = READ_ONCE(sqe->len); | |
4446 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4447 | return 0; | |
4448 | #else | |
4449 | return -EOPNOTSUPP; | |
4450 | #endif | |
4451 | } | |
4452 | ||
45d189c6 | 4453 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4454 | { |
4455 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4456 | struct io_madvise *ma = &req->madvise; | |
4457 | int ret; | |
4458 | ||
45d189c6 | 4459 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4460 | return -EAGAIN; |
4461 | ||
0726b01e | 4462 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4463 | if (ret < 0) |
93d2bcd2 | 4464 | req_set_fail(req); |
e1e16097 | 4465 | io_req_complete(req, ret); |
c1ca757b JA |
4466 | return 0; |
4467 | #else | |
4468 | return -EOPNOTSUPP; | |
4469 | #endif | |
4470 | } | |
4471 | ||
4840e418 JA |
4472 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4473 | { | |
26578cda | 4474 | if (sqe->ioprio || sqe->buf_index || sqe->addr || sqe->splice_fd_in) |
4840e418 | 4475 | return -EINVAL; |
3232dd02 PB |
4476 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4477 | return -EINVAL; | |
4840e418 JA |
4478 | |
4479 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4480 | req->fadvise.len = READ_ONCE(sqe->len); | |
4481 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4482 | return 0; | |
4483 | } | |
4484 | ||
45d189c6 | 4485 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4486 | { |
4487 | struct io_fadvise *fa = &req->fadvise; | |
4488 | int ret; | |
4489 | ||
45d189c6 | 4490 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4491 | switch (fa->advice) { |
4492 | case POSIX_FADV_NORMAL: | |
4493 | case POSIX_FADV_RANDOM: | |
4494 | case POSIX_FADV_SEQUENTIAL: | |
4495 | break; | |
4496 | default: | |
4497 | return -EAGAIN; | |
4498 | } | |
4499 | } | |
4840e418 JA |
4500 | |
4501 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4502 | if (ret < 0) | |
93d2bcd2 | 4503 | req_set_fail(req); |
0bdf3398 | 4504 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4505 | return 0; |
4506 | } | |
4507 | ||
eddc7ef5 JA |
4508 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4509 | { | |
2d74d042 | 4510 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4511 | return -EINVAL; |
26578cda | 4512 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
eddc7ef5 | 4513 | return -EINVAL; |
9c280f90 | 4514 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4515 | return -EBADF; |
eddc7ef5 | 4516 | |
1d9e1288 BM |
4517 | req->statx.dfd = READ_ONCE(sqe->fd); |
4518 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4519 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4520 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4521 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4522 | |
4523 | return 0; | |
4524 | } | |
4525 | ||
45d189c6 | 4526 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4527 | { |
1d9e1288 | 4528 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4529 | int ret; |
4530 | ||
59d70013 | 4531 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4532 | return -EAGAIN; |
4533 | ||
e62753e4 BM |
4534 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4535 | ctx->buffer); | |
eddc7ef5 | 4536 | |
eddc7ef5 | 4537 | if (ret < 0) |
93d2bcd2 | 4538 | req_set_fail(req); |
e1e16097 | 4539 | io_req_complete(req, ret); |
eddc7ef5 JA |
4540 | return 0; |
4541 | } | |
4542 | ||
b5dba59e JA |
4543 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4544 | { | |
14587a46 | 4545 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4546 | return -EINVAL; |
b5dba59e | 4547 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
26578cda | 4548 | sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) |
b5dba59e | 4549 | return -EINVAL; |
9c280f90 | 4550 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4551 | return -EBADF; |
b5dba59e JA |
4552 | |
4553 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4554 | return 0; |
b5dba59e JA |
4555 | } |
4556 | ||
889fca73 | 4557 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4558 | { |
9eac1904 | 4559 | struct files_struct *files = current->files; |
3af73b28 | 4560 | struct io_close *close = &req->close; |
9eac1904 | 4561 | struct fdtable *fdt; |
a1fde923 PB |
4562 | struct file *file = NULL; |
4563 | int ret = -EBADF; | |
b5dba59e | 4564 | |
9eac1904 JA |
4565 | spin_lock(&files->file_lock); |
4566 | fdt = files_fdtable(files); | |
4567 | if (close->fd >= fdt->max_fds) { | |
4568 | spin_unlock(&files->file_lock); | |
4569 | goto err; | |
4570 | } | |
4571 | file = fdt->fd[close->fd]; | |
a1fde923 | 4572 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4573 | spin_unlock(&files->file_lock); |
4574 | file = NULL; | |
4575 | goto err; | |
3af73b28 | 4576 | } |
b5dba59e JA |
4577 | |
4578 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4579 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4580 | spin_unlock(&files->file_lock); |
0bf0eefd | 4581 | return -EAGAIN; |
a2100672 | 4582 | } |
b5dba59e | 4583 | |
9eac1904 JA |
4584 | ret = __close_fd_get_file(close->fd, &file); |
4585 | spin_unlock(&files->file_lock); | |
4586 | if (ret < 0) { | |
4587 | if (ret == -ENOENT) | |
4588 | ret = -EBADF; | |
4589 | goto err; | |
4590 | } | |
4591 | ||
3af73b28 | 4592 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4593 | ret = filp_close(file, current->files); |
4594 | err: | |
3af73b28 | 4595 | if (ret < 0) |
93d2bcd2 | 4596 | req_set_fail(req); |
9eac1904 JA |
4597 | if (file) |
4598 | fput(file); | |
889fca73 | 4599 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4600 | return 0; |
b5dba59e JA |
4601 | } |
4602 | ||
1155c76a | 4603 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4604 | { |
4605 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4606 | |
5d17b4a4 JA |
4607 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4608 | return -EINVAL; | |
26578cda PB |
4609 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4610 | sqe->splice_fd_in)) | |
5d17b4a4 JA |
4611 | return -EINVAL; |
4612 | ||
8ed8d3c3 JA |
4613 | req->sync.off = READ_ONCE(sqe->off); |
4614 | req->sync.len = READ_ONCE(sqe->len); | |
4615 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4616 | return 0; |
4617 | } | |
4618 | ||
45d189c6 | 4619 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4620 | { |
8ed8d3c3 JA |
4621 | int ret; |
4622 | ||
ac45abc0 | 4623 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4624 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4625 | return -EAGAIN; |
4626 | ||
9adbd45d | 4627 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4628 | req->sync.flags); |
4629 | if (ret < 0) | |
93d2bcd2 | 4630 | req_set_fail(req); |
e1e16097 | 4631 | io_req_complete(req, ret); |
5d17b4a4 JA |
4632 | return 0; |
4633 | } | |
4634 | ||
469956e8 | 4635 | #if defined(CONFIG_NET) |
02d27d89 PB |
4636 | static int io_setup_async_msg(struct io_kiocb *req, |
4637 | struct io_async_msghdr *kmsg) | |
4638 | { | |
e8c2bc1f JA |
4639 | struct io_async_msghdr *async_msg = req->async_data; |
4640 | ||
4641 | if (async_msg) | |
02d27d89 | 4642 | return -EAGAIN; |
e8c2bc1f | 4643 | if (io_alloc_async_data(req)) { |
257e84a5 | 4644 | kfree(kmsg->free_iov); |
02d27d89 PB |
4645 | return -ENOMEM; |
4646 | } | |
e8c2bc1f | 4647 | async_msg = req->async_data; |
02d27d89 | 4648 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4649 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4650 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4651 | /* if were using fast_iov, set it to the new one */ |
4652 | if (!async_msg->free_iov) | |
4653 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4654 | ||
02d27d89 PB |
4655 | return -EAGAIN; |
4656 | } | |
4657 | ||
2ae523ed PB |
4658 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4659 | struct io_async_msghdr *iomsg) | |
4660 | { | |
2ae523ed | 4661 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4662 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4663 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4664 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4665 | } |
4666 | ||
93642ef8 PB |
4667 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4668 | { | |
4669 | int ret; | |
4670 | ||
93642ef8 PB |
4671 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4672 | if (!ret) | |
4673 | req->flags |= REQ_F_NEED_CLEANUP; | |
4674 | return ret; | |
4675 | } | |
4676 | ||
3529d8c2 | 4677 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4678 | { |
e47293fd | 4679 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4680 | |
d2b6f48b PB |
4681 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4682 | return -EINVAL; | |
4683 | ||
270a5940 | 4684 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4685 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4686 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4687 | if (sr->msg_flags & MSG_DONTWAIT) | |
4688 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4689 | |
d8768362 JA |
4690 | #ifdef CONFIG_COMPAT |
4691 | if (req->ctx->compat) | |
4692 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4693 | #endif | |
93642ef8 | 4694 | return 0; |
03b1230c JA |
4695 | } |
4696 | ||
889fca73 | 4697 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4698 | { |
6b754c8b | 4699 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4700 | struct socket *sock; |
7a7cacba | 4701 | unsigned flags; |
0031275d | 4702 | int min_ret = 0; |
0fa03c62 JA |
4703 | int ret; |
4704 | ||
dba4a925 | 4705 | sock = sock_from_file(req->file); |
7a7cacba | 4706 | if (unlikely(!sock)) |
dba4a925 | 4707 | return -ENOTSOCK; |
3529d8c2 | 4708 | |
257e84a5 PB |
4709 | kmsg = req->async_data; |
4710 | if (!kmsg) { | |
7a7cacba PB |
4711 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4712 | if (ret) | |
4713 | return ret; | |
4714 | kmsg = &iomsg; | |
0fa03c62 | 4715 | } |
0fa03c62 | 4716 | |
04411806 PB |
4717 | flags = req->sr_msg.msg_flags; |
4718 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4719 | flags |= MSG_DONTWAIT; |
0031275d SM |
4720 | if (flags & MSG_WAITALL) |
4721 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4722 | ||
7a7cacba | 4723 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4724 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4725 | return io_setup_async_msg(req, kmsg); |
4726 | if (ret == -ERESTARTSYS) | |
4727 | ret = -EINTR; | |
0fa03c62 | 4728 | |
257e84a5 PB |
4729 | /* fast path, check for non-NULL to avoid function call */ |
4730 | if (kmsg->free_iov) | |
4731 | kfree(kmsg->free_iov); | |
99bc4c38 | 4732 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4733 | if (ret < min_ret) |
93d2bcd2 | 4734 | req_set_fail(req); |
889fca73 | 4735 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4736 | return 0; |
03b1230c | 4737 | } |
aa1fa28f | 4738 | |
889fca73 | 4739 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4740 | { |
7a7cacba PB |
4741 | struct io_sr_msg *sr = &req->sr_msg; |
4742 | struct msghdr msg; | |
4743 | struct iovec iov; | |
fddaface | 4744 | struct socket *sock; |
7a7cacba | 4745 | unsigned flags; |
0031275d | 4746 | int min_ret = 0; |
fddaface JA |
4747 | int ret; |
4748 | ||
dba4a925 | 4749 | sock = sock_from_file(req->file); |
7a7cacba | 4750 | if (unlikely(!sock)) |
dba4a925 | 4751 | return -ENOTSOCK; |
fddaface | 4752 | |
7a7cacba PB |
4753 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4754 | if (unlikely(ret)) | |
14db8411 | 4755 | return ret; |
fddaface | 4756 | |
7a7cacba PB |
4757 | msg.msg_name = NULL; |
4758 | msg.msg_control = NULL; | |
4759 | msg.msg_controllen = 0; | |
4760 | msg.msg_namelen = 0; | |
fddaface | 4761 | |
04411806 PB |
4762 | flags = req->sr_msg.msg_flags; |
4763 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4764 | flags |= MSG_DONTWAIT; |
0031275d SM |
4765 | if (flags & MSG_WAITALL) |
4766 | min_ret = iov_iter_count(&msg.msg_iter); | |
4767 | ||
7a7cacba PB |
4768 | msg.msg_flags = flags; |
4769 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4770 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4771 | return -EAGAIN; |
4772 | if (ret == -ERESTARTSYS) | |
4773 | ret = -EINTR; | |
fddaface | 4774 | |
0031275d | 4775 | if (ret < min_ret) |
93d2bcd2 | 4776 | req_set_fail(req); |
889fca73 | 4777 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4778 | return 0; |
fddaface JA |
4779 | } |
4780 | ||
1400e697 PB |
4781 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4782 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4783 | { |
4784 | struct io_sr_msg *sr = &req->sr_msg; | |
4785 | struct iovec __user *uiov; | |
4786 | size_t iov_len; | |
4787 | int ret; | |
4788 | ||
1400e697 PB |
4789 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4790 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4791 | if (ret) |
4792 | return ret; | |
4793 | ||
4794 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4795 | if (iov_len > 1) | |
4796 | return -EINVAL; | |
5476dfed | 4797 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4798 | return -EFAULT; |
5476dfed | 4799 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4800 | iomsg->free_iov = NULL; |
52de1fe1 | 4801 | } else { |
257e84a5 | 4802 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4803 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4804 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4805 | false); |
52de1fe1 JA |
4806 | if (ret > 0) |
4807 | ret = 0; | |
4808 | } | |
4809 | ||
4810 | return ret; | |
4811 | } | |
4812 | ||
4813 | #ifdef CONFIG_COMPAT | |
4814 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4815 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4816 | { |
52de1fe1 JA |
4817 | struct io_sr_msg *sr = &req->sr_msg; |
4818 | struct compat_iovec __user *uiov; | |
4819 | compat_uptr_t ptr; | |
4820 | compat_size_t len; | |
4821 | int ret; | |
4822 | ||
4af3417a PB |
4823 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4824 | &ptr, &len); | |
52de1fe1 JA |
4825 | if (ret) |
4826 | return ret; | |
4827 | ||
4828 | uiov = compat_ptr(ptr); | |
4829 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4830 | compat_ssize_t clen; | |
4831 | ||
4832 | if (len > 1) | |
4833 | return -EINVAL; | |
4834 | if (!access_ok(uiov, sizeof(*uiov))) | |
4835 | return -EFAULT; | |
4836 | if (__get_user(clen, &uiov->iov_len)) | |
4837 | return -EFAULT; | |
4838 | if (clen < 0) | |
4839 | return -EINVAL; | |
2d280bc8 | 4840 | sr->len = clen; |
257e84a5 | 4841 | iomsg->free_iov = NULL; |
52de1fe1 | 4842 | } else { |
257e84a5 | 4843 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4844 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4845 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4846 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4847 | if (ret < 0) |
4848 | return ret; | |
4849 | } | |
4850 | ||
4851 | return 0; | |
4852 | } | |
4853 | #endif | |
4854 | ||
1400e697 PB |
4855 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4856 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4857 | { |
1400e697 | 4858 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4859 | |
4860 | #ifdef CONFIG_COMPAT | |
4861 | if (req->ctx->compat) | |
1400e697 | 4862 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4863 | #endif |
52de1fe1 | 4864 | |
1400e697 | 4865 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4866 | } |
4867 | ||
bcda7baa | 4868 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4869 | bool needs_lock) |
bcda7baa JA |
4870 | { |
4871 | struct io_sr_msg *sr = &req->sr_msg; | |
4872 | struct io_buffer *kbuf; | |
4873 | ||
bcda7baa JA |
4874 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4875 | if (IS_ERR(kbuf)) | |
4876 | return kbuf; | |
4877 | ||
4878 | sr->kbuf = kbuf; | |
4879 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4880 | return kbuf; |
fddaface JA |
4881 | } |
4882 | ||
7fbb1b54 PB |
4883 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4884 | { | |
4885 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4886 | } | |
4887 | ||
93642ef8 | 4888 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4889 | { |
99bc4c38 | 4890 | int ret; |
3529d8c2 | 4891 | |
93642ef8 PB |
4892 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4893 | if (!ret) | |
4894 | req->flags |= REQ_F_NEED_CLEANUP; | |
4895 | return ret; | |
4896 | } | |
4897 | ||
4898 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4899 | { | |
4900 | struct io_sr_msg *sr = &req->sr_msg; | |
4901 | ||
d2b6f48b PB |
4902 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4903 | return -EINVAL; | |
4904 | ||
270a5940 | 4905 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4906 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4907 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4908 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4909 | if (sr->msg_flags & MSG_DONTWAIT) | |
4910 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4911 | |
d8768362 JA |
4912 | #ifdef CONFIG_COMPAT |
4913 | if (req->ctx->compat) | |
4914 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4915 | #endif | |
93642ef8 | 4916 | return 0; |
aa1fa28f JA |
4917 | } |
4918 | ||
889fca73 | 4919 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4920 | { |
6b754c8b | 4921 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4922 | struct socket *sock; |
7fbb1b54 | 4923 | struct io_buffer *kbuf; |
7a7cacba | 4924 | unsigned flags; |
0031275d | 4925 | int min_ret = 0; |
52de1fe1 | 4926 | int ret, cflags = 0; |
45d189c6 | 4927 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4928 | |
dba4a925 | 4929 | sock = sock_from_file(req->file); |
7a7cacba | 4930 | if (unlikely(!sock)) |
dba4a925 | 4931 | return -ENOTSOCK; |
3529d8c2 | 4932 | |
257e84a5 PB |
4933 | kmsg = req->async_data; |
4934 | if (!kmsg) { | |
7a7cacba PB |
4935 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4936 | if (ret) | |
681fda8d | 4937 | return ret; |
7a7cacba PB |
4938 | kmsg = &iomsg; |
4939 | } | |
03b1230c | 4940 | |
bc02ef33 | 4941 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4942 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4943 | if (IS_ERR(kbuf)) |
52de1fe1 | 4944 | return PTR_ERR(kbuf); |
7a7cacba | 4945 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4946 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4947 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4948 | 1, req->sr_msg.len); |
4949 | } | |
52de1fe1 | 4950 | |
04411806 PB |
4951 | flags = req->sr_msg.msg_flags; |
4952 | if (force_nonblock) | |
7a7cacba | 4953 | flags |= MSG_DONTWAIT; |
0031275d SM |
4954 | if (flags & MSG_WAITALL) |
4955 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4956 | ||
7a7cacba PB |
4957 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4958 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4959 | if (force_nonblock && ret == -EAGAIN) |
4960 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4961 | if (ret == -ERESTARTSYS) |
4962 | ret = -EINTR; | |
03b1230c | 4963 | |
7fbb1b54 PB |
4964 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4965 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4966 | /* fast path, check for non-NULL to avoid function call */ |
4967 | if (kmsg->free_iov) | |
4968 | kfree(kmsg->free_iov); | |
99bc4c38 | 4969 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4970 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4971 | req_set_fail(req); |
889fca73 | 4972 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4973 | return 0; |
0fa03c62 | 4974 | } |
5d17b4a4 | 4975 | |
889fca73 | 4976 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4977 | { |
6b754c8b | 4978 | struct io_buffer *kbuf; |
7a7cacba PB |
4979 | struct io_sr_msg *sr = &req->sr_msg; |
4980 | struct msghdr msg; | |
4981 | void __user *buf = sr->buf; | |
fddaface | 4982 | struct socket *sock; |
7a7cacba PB |
4983 | struct iovec iov; |
4984 | unsigned flags; | |
0031275d | 4985 | int min_ret = 0; |
bcda7baa | 4986 | int ret, cflags = 0; |
45d189c6 | 4987 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4988 | |
dba4a925 | 4989 | sock = sock_from_file(req->file); |
7a7cacba | 4990 | if (unlikely(!sock)) |
dba4a925 | 4991 | return -ENOTSOCK; |
fddaface | 4992 | |
bc02ef33 | 4993 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4994 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4995 | if (IS_ERR(kbuf)) |
4996 | return PTR_ERR(kbuf); | |
7a7cacba | 4997 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4998 | } |
bcda7baa | 4999 | |
7a7cacba | 5000 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
5001 | if (unlikely(ret)) |
5002 | goto out_free; | |
fddaface | 5003 | |
7a7cacba PB |
5004 | msg.msg_name = NULL; |
5005 | msg.msg_control = NULL; | |
5006 | msg.msg_controllen = 0; | |
5007 | msg.msg_namelen = 0; | |
5008 | msg.msg_iocb = NULL; | |
5009 | msg.msg_flags = 0; | |
fddaface | 5010 | |
04411806 PB |
5011 | flags = req->sr_msg.msg_flags; |
5012 | if (force_nonblock) | |
7a7cacba | 5013 | flags |= MSG_DONTWAIT; |
0031275d SM |
5014 | if (flags & MSG_WAITALL) |
5015 | min_ret = iov_iter_count(&msg.msg_iter); | |
5016 | ||
7a7cacba PB |
5017 | ret = sock_recvmsg(sock, &msg, flags); |
5018 | if (force_nonblock && ret == -EAGAIN) | |
5019 | return -EAGAIN; | |
5020 | if (ret == -ERESTARTSYS) | |
5021 | ret = -EINTR; | |
14c32eee | 5022 | out_free: |
7fbb1b54 PB |
5023 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5024 | cflags = io_put_recv_kbuf(req); | |
0031275d | 5025 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 5026 | req_set_fail(req); |
889fca73 | 5027 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5028 | return 0; |
fddaface JA |
5029 | } |
5030 | ||
3529d8c2 | 5031 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5032 | { |
8ed8d3c3 JA |
5033 | struct io_accept *accept = &req->accept; |
5034 | ||
14587a46 | 5035 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5036 | return -EINVAL; |
aaa4db12 | 5037 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5038 | return -EINVAL; |
5039 | ||
d55e5f5b JA |
5040 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5041 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5042 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5043 | accept->nofile = rlimit(RLIMIT_NOFILE); |
a7083ad5 | 5044 | |
aaa4db12 PB |
5045 | accept->file_slot = READ_ONCE(sqe->file_index); |
5046 | if (accept->file_slot && ((req->open.how.flags & O_CLOEXEC) || | |
5047 | (accept->flags & SOCK_CLOEXEC))) | |
5048 | return -EINVAL; | |
a7083ad5 PB |
5049 | if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) |
5050 | return -EINVAL; | |
5051 | if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) | |
5052 | accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; | |
8ed8d3c3 | 5053 | return 0; |
8ed8d3c3 | 5054 | } |
17f2fe35 | 5055 | |
889fca73 | 5056 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5057 | { |
5058 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5059 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5060 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
aaa4db12 | 5061 | bool fixed = !!accept->file_slot; |
a7083ad5 PB |
5062 | struct file *file; |
5063 | int ret, fd; | |
8ed8d3c3 | 5064 | |
e697deed JX |
5065 | if (req->file->f_flags & O_NONBLOCK) |
5066 | req->flags |= REQ_F_NOWAIT; | |
5067 | ||
aaa4db12 PB |
5068 | if (!fixed) { |
5069 | fd = __get_unused_fd_flags(accept->flags, accept->nofile); | |
5070 | if (unlikely(fd < 0)) | |
5071 | return fd; | |
5072 | } | |
a7083ad5 PB |
5073 | file = do_accept(req->file, file_flags, accept->addr, accept->addr_len, |
5074 | accept->flags); | |
5075 | if (IS_ERR(file)) { | |
aaa4db12 PB |
5076 | if (!fixed) |
5077 | put_unused_fd(fd); | |
a7083ad5 PB |
5078 | ret = PTR_ERR(file); |
5079 | if (ret == -EAGAIN && force_nonblock) | |
5080 | return -EAGAIN; | |
ac45abc0 PB |
5081 | if (ret == -ERESTARTSYS) |
5082 | ret = -EINTR; | |
93d2bcd2 | 5083 | req_set_fail(req); |
aaa4db12 | 5084 | } else if (!fixed) { |
a7083ad5 PB |
5085 | fd_install(fd, file); |
5086 | ret = fd; | |
aaa4db12 PB |
5087 | } else { |
5088 | ret = io_install_fixed_file(req, file, issue_flags, | |
5089 | accept->file_slot - 1); | |
ac45abc0 | 5090 | } |
889fca73 | 5091 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5092 | return 0; |
8ed8d3c3 JA |
5093 | } |
5094 | ||
93642ef8 PB |
5095 | static int io_connect_prep_async(struct io_kiocb *req) |
5096 | { | |
5097 | struct io_async_connect *io = req->async_data; | |
5098 | struct io_connect *conn = &req->connect; | |
5099 | ||
5100 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
5101 | } | |
5102 | ||
3529d8c2 | 5103 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5104 | { |
3529d8c2 | 5105 | struct io_connect *conn = &req->connect; |
f499a021 | 5106 | |
14587a46 | 5107 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 | 5108 | return -EINVAL; |
26578cda PB |
5109 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags || |
5110 | sqe->splice_fd_in) | |
3fbb51c1 JA |
5111 | return -EINVAL; |
5112 | ||
3529d8c2 JA |
5113 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5114 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 5115 | return 0; |
f499a021 JA |
5116 | } |
5117 | ||
889fca73 | 5118 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5119 | { |
e8c2bc1f | 5120 | struct io_async_connect __io, *io; |
f8e85cf2 | 5121 | unsigned file_flags; |
3fbb51c1 | 5122 | int ret; |
45d189c6 | 5123 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5124 | |
e8c2bc1f JA |
5125 | if (req->async_data) { |
5126 | io = req->async_data; | |
f499a021 | 5127 | } else { |
3529d8c2 JA |
5128 | ret = move_addr_to_kernel(req->connect.addr, |
5129 | req->connect.addr_len, | |
e8c2bc1f | 5130 | &__io.address); |
f499a021 JA |
5131 | if (ret) |
5132 | goto out; | |
5133 | io = &__io; | |
5134 | } | |
5135 | ||
3fbb51c1 JA |
5136 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5137 | ||
e8c2bc1f | 5138 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5139 | req->connect.addr_len, file_flags); |
87f80d62 | 5140 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5141 | if (req->async_data) |
b7bb4f7d | 5142 | return -EAGAIN; |
e8c2bc1f | 5143 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5144 | ret = -ENOMEM; |
5145 | goto out; | |
5146 | } | |
e8c2bc1f | 5147 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 5148 | return -EAGAIN; |
f499a021 | 5149 | } |
f8e85cf2 JA |
5150 | if (ret == -ERESTARTSYS) |
5151 | ret = -EINTR; | |
f499a021 | 5152 | out: |
4e88d6e7 | 5153 | if (ret < 0) |
93d2bcd2 | 5154 | req_set_fail(req); |
889fca73 | 5155 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5156 | return 0; |
469956e8 Y |
5157 | } |
5158 | #else /* !CONFIG_NET */ | |
99a10081 JA |
5159 | #define IO_NETOP_FN(op) \ |
5160 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
5161 | { \ | |
5162 | return -EOPNOTSUPP; \ | |
5163 | } | |
5164 | ||
5165 | #define IO_NETOP_PREP(op) \ | |
5166 | IO_NETOP_FN(op) \ | |
5167 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
5168 | { \ | |
5169 | return -EOPNOTSUPP; \ | |
5170 | } \ | |
5171 | ||
5172 | #define IO_NETOP_PREP_ASYNC(op) \ | |
5173 | IO_NETOP_PREP(op) \ | |
5174 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
5175 | { \ | |
5176 | return -EOPNOTSUPP; \ | |
5177 | } | |
5178 | ||
5179 | IO_NETOP_PREP_ASYNC(sendmsg); | |
5180 | IO_NETOP_PREP_ASYNC(recvmsg); | |
5181 | IO_NETOP_PREP_ASYNC(connect); | |
5182 | IO_NETOP_PREP(accept); | |
5183 | IO_NETOP_FN(send); | |
5184 | IO_NETOP_FN(recv); | |
469956e8 | 5185 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5186 | |
d7718a9d JA |
5187 | struct io_poll_table { |
5188 | struct poll_table_struct pt; | |
5189 | struct io_kiocb *req; | |
68b11e8b | 5190 | int nr_entries; |
d7718a9d JA |
5191 | int error; |
5192 | }; | |
ce593a6c | 5193 | |
d7718a9d | 5194 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5b0a6acc | 5195 | __poll_t mask, io_req_tw_func_t func) |
d7718a9d | 5196 | { |
d7718a9d JA |
5197 | /* for instances that support it check for an event match first: */ |
5198 | if (mask && !(mask & poll->events)) | |
5199 | return 0; | |
5200 | ||
5201 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5202 | ||
5203 | list_del_init(&poll->wait.entry); | |
5204 | ||
d7718a9d | 5205 | req->result = mask; |
5b0a6acc | 5206 | req->io_task_work.func = func; |
6d816e08 | 5207 | |
d7718a9d | 5208 | /* |
e3aabf95 JA |
5209 | * If this fails, then the task is exiting. When a task exits, the |
5210 | * work gets canceled, so just cancel this request as well instead | |
5211 | * of executing it. We can't safely execute it anyway, as we may not | |
5212 | * have the needed state needed for it anyway. | |
d7718a9d | 5213 | */ |
e09ee510 | 5214 | io_req_task_work_add(req); |
d7718a9d JA |
5215 | return 1; |
5216 | } | |
5217 | ||
74ce6ce4 JA |
5218 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5219 | __acquires(&req->ctx->completion_lock) | |
5220 | { | |
5221 | struct io_ring_ctx *ctx = req->ctx; | |
5222 | ||
316319e8 | 5223 | /* req->task == current here, checking PF_EXITING is safe */ |
e09ee510 PB |
5224 | if (unlikely(req->task->flags & PF_EXITING)) |
5225 | WRITE_ONCE(poll->canceled, true); | |
5226 | ||
74ce6ce4 JA |
5227 | if (!req->result && !READ_ONCE(poll->canceled)) { |
5228 | struct poll_table_struct pt = { ._key = poll->events }; | |
5229 | ||
5230 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5231 | } | |
5232 | ||
79ebeaee | 5233 | spin_lock(&ctx->completion_lock); |
74ce6ce4 JA |
5234 | if (!req->result && !READ_ONCE(poll->canceled)) { |
5235 | add_wait_queue(poll->head, &poll->wait); | |
5236 | return true; | |
5237 | } | |
5238 | ||
5239 | return false; | |
5240 | } | |
5241 | ||
d4e7cd36 | 5242 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5243 | { |
e8c2bc1f | 5244 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5245 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5246 | return req->async_data; |
d4e7cd36 JA |
5247 | return req->apoll->double_poll; |
5248 | } | |
5249 | ||
5250 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5251 | { | |
5252 | if (req->opcode == IORING_OP_POLL_ADD) | |
5253 | return &req->poll; | |
5254 | return &req->apoll->poll; | |
5255 | } | |
5256 | ||
5257 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 5258 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
5259 | { |
5260 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5261 | |
5262 | lockdep_assert_held(&req->ctx->completion_lock); | |
5263 | ||
5264 | if (poll && poll->head) { | |
5265 | struct wait_queue_head *head = poll->head; | |
5266 | ||
79ebeaee | 5267 | spin_lock_irq(&head->lock); |
18bceab1 JA |
5268 | list_del_init(&poll->wait.entry); |
5269 | if (poll->wait.private) | |
de9b4cca | 5270 | req_ref_put(req); |
18bceab1 | 5271 | poll->head = NULL; |
79ebeaee | 5272 | spin_unlock_irq(&head->lock); |
18bceab1 JA |
5273 | } |
5274 | } | |
5275 | ||
31efe48e | 5276 | static bool __io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 5277 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
5278 | { |
5279 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 5280 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 5281 | int error; |
18bceab1 | 5282 | |
e27414be | 5283 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 5284 | error = -ECANCELED; |
88e41cf9 | 5285 | req->poll.events |= EPOLLONESHOT; |
e27414be | 5286 | } else { |
5082620f | 5287 | error = mangle_poll(mask); |
e27414be | 5288 | } |
b69de288 JA |
5289 | if (req->poll.events & EPOLLONESHOT) |
5290 | flags = 0; | |
d4d19c19 | 5291 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
88e41cf9 JA |
5292 | req->poll.done = true; |
5293 | flags = 0; | |
5294 | } | |
7b289c38 HX |
5295 | if (flags & IORING_CQE_F_MORE) |
5296 | ctx->cq_extra++; | |
18bceab1 | 5297 | |
88e41cf9 | 5298 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
5299 | } |
5300 | ||
31efe48e XW |
5301 | static inline bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
5302 | __must_hold(&req->ctx->completion_lock) | |
5303 | { | |
5304 | bool done; | |
5305 | ||
5306 | done = __io_poll_complete(req, mask); | |
5307 | io_commit_cqring(req->ctx); | |
5308 | return done; | |
5309 | } | |
5310 | ||
f237c30a | 5311 | static void io_poll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 JA |
5312 | { |
5313 | struct io_ring_ctx *ctx = req->ctx; | |
dd221f46 | 5314 | struct io_kiocb *nxt; |
18bceab1 JA |
5315 | |
5316 | if (io_poll_rewait(req, &req->poll)) { | |
79ebeaee | 5317 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5318 | } else { |
f40b964a | 5319 | bool done; |
18bceab1 | 5320 | |
31efe48e | 5321 | done = __io_poll_complete(req, req->result); |
88e41cf9 | 5322 | if (done) { |
a890d01e | 5323 | io_poll_remove_double(req); |
88e41cf9 | 5324 | hash_del(&req->hash_node); |
f40b964a | 5325 | } else { |
88e41cf9 JA |
5326 | req->result = 0; |
5327 | add_wait_queue(req->poll.head, &req->poll.wait); | |
5328 | } | |
31efe48e | 5329 | io_commit_cqring(ctx); |
79ebeaee | 5330 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5331 | io_cqring_ev_posted(ctx); |
18bceab1 | 5332 | |
88e41cf9 JA |
5333 | if (done) { |
5334 | nxt = io_put_req_find_next(req); | |
5335 | if (nxt) | |
f237c30a | 5336 | io_req_task_submit(nxt, locked); |
88e41cf9 | 5337 | } |
dd221f46 | 5338 | } |
18bceab1 JA |
5339 | } |
5340 | ||
5341 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5342 | int sync, void *key) | |
5343 | { | |
5344 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5345 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 | 5346 | __poll_t mask = key_to_poll(key); |
79ebeaee | 5347 | unsigned long flags; |
18bceab1 JA |
5348 | |
5349 | /* for instances that support it check for an event match first: */ | |
5350 | if (mask && !(mask & poll->events)) | |
5351 | return 0; | |
88e41cf9 JA |
5352 | if (!(poll->events & EPOLLONESHOT)) |
5353 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 5354 | |
8706e04e JA |
5355 | list_del_init(&wait->entry); |
5356 | ||
9ce85ef2 | 5357 | if (poll->head) { |
18bceab1 JA |
5358 | bool done; |
5359 | ||
79ebeaee | 5360 | spin_lock_irqsave(&poll->head->lock, flags); |
807abcb0 | 5361 | done = list_empty(&poll->wait.entry); |
18bceab1 | 5362 | if (!done) |
807abcb0 | 5363 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5364 | /* make sure double remove sees this as being gone */ |
5365 | wait->private = NULL; | |
79ebeaee | 5366 | spin_unlock_irqrestore(&poll->head->lock, flags); |
c8b5e260 JA |
5367 | if (!done) { |
5368 | /* use wait func handler, so it matches the rq type */ | |
5369 | poll->wait.func(&poll->wait, mode, sync, key); | |
5370 | } | |
18bceab1 | 5371 | } |
de9b4cca | 5372 | req_ref_put(req); |
18bceab1 JA |
5373 | return 1; |
5374 | } | |
5375 | ||
5376 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5377 | wait_queue_func_t wake_func) | |
5378 | { | |
5379 | poll->head = NULL; | |
5380 | poll->done = false; | |
5381 | poll->canceled = false; | |
464dca61 JA |
5382 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5383 | /* mask in events that we always want/need */ | |
5384 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5385 | INIT_LIST_HEAD(&poll->wait.entry); |
5386 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5387 | } | |
5388 | ||
5389 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5390 | struct wait_queue_head *head, |
5391 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5392 | { |
5393 | struct io_kiocb *req = pt->req; | |
5394 | ||
5395 | /* | |
68b11e8b PB |
5396 | * The file being polled uses multiple waitqueues for poll handling |
5397 | * (e.g. one for read, one for write). Setup a separate io_poll_iocb | |
5398 | * if this happens. | |
18bceab1 | 5399 | */ |
68b11e8b | 5400 | if (unlikely(pt->nr_entries)) { |
58852d4d PB |
5401 | struct io_poll_iocb *poll_one = poll; |
5402 | ||
23a65db8 PB |
5403 | /* double add on the same waitqueue head, ignore */ |
5404 | if (poll_one->head == head) | |
5405 | return; | |
18bceab1 | 5406 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5407 | if (*poll_ptr) { |
23a65db8 PB |
5408 | if ((*poll_ptr)->head == head) |
5409 | return; | |
18bceab1 JA |
5410 | pt->error = -EINVAL; |
5411 | return; | |
5412 | } | |
ea6a693d JA |
5413 | /* |
5414 | * Can't handle multishot for double wait for now, turn it | |
5415 | * into one-shot mode. | |
5416 | */ | |
7a274727 PB |
5417 | if (!(poll_one->events & EPOLLONESHOT)) |
5418 | poll_one->events |= EPOLLONESHOT; | |
18bceab1 JA |
5419 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5420 | if (!poll) { | |
5421 | pt->error = -ENOMEM; | |
5422 | return; | |
5423 | } | |
58852d4d | 5424 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5425 | req_ref_get(req); |
18bceab1 | 5426 | poll->wait.private = req; |
807abcb0 | 5427 | *poll_ptr = poll; |
18bceab1 JA |
5428 | } |
5429 | ||
68b11e8b | 5430 | pt->nr_entries++; |
18bceab1 | 5431 | poll->head = head; |
a31eb4a2 JX |
5432 | |
5433 | if (poll->events & EPOLLEXCLUSIVE) | |
5434 | add_wait_queue_exclusive(head, &poll->wait); | |
5435 | else | |
5436 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5437 | } |
5438 | ||
5439 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5440 | struct poll_table_struct *p) | |
5441 | { | |
5442 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5443 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5444 | |
807abcb0 | 5445 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5446 | } |
5447 | ||
f237c30a | 5448 | static void io_async_task_func(struct io_kiocb *req, bool *locked) |
d7718a9d | 5449 | { |
d7718a9d JA |
5450 | struct async_poll *apoll = req->apoll; |
5451 | struct io_ring_ctx *ctx = req->ctx; | |
5452 | ||
236daeae | 5453 | trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data); |
d7718a9d | 5454 | |
74ce6ce4 | 5455 | if (io_poll_rewait(req, &apoll->poll)) { |
79ebeaee | 5456 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5457 | return; |
d7718a9d JA |
5458 | } |
5459 | ||
0ea13b44 | 5460 | hash_del(&req->hash_node); |
d4e7cd36 | 5461 | io_poll_remove_double(req); |
79ebeaee | 5462 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5463 | |
0be0b0e3 | 5464 | if (!READ_ONCE(apoll->poll.canceled)) |
f237c30a | 5465 | io_req_task_submit(req, locked); |
0be0b0e3 | 5466 | else |
2593553a | 5467 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5468 | } |
5469 | ||
5470 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5471 | void *key) | |
5472 | { | |
5473 | struct io_kiocb *req = wait->private; | |
5474 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5475 | ||
5476 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5477 | key_to_poll(key)); | |
5478 | ||
5479 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5480 | } | |
5481 | ||
5482 | static void io_poll_req_insert(struct io_kiocb *req) | |
5483 | { | |
5484 | struct io_ring_ctx *ctx = req->ctx; | |
5485 | struct hlist_head *list; | |
5486 | ||
5487 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5488 | hlist_add_head(&req->hash_node, list); | |
5489 | } | |
5490 | ||
5491 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5492 | struct io_poll_iocb *poll, | |
5493 | struct io_poll_table *ipt, __poll_t mask, | |
5494 | wait_queue_func_t wake_func) | |
5495 | __acquires(&ctx->completion_lock) | |
5496 | { | |
5497 | struct io_ring_ctx *ctx = req->ctx; | |
5498 | bool cancel = false; | |
5499 | ||
4d52f338 | 5500 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5501 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5502 | poll->file = req->file; |
18bceab1 | 5503 | poll->wait.private = req; |
d7718a9d JA |
5504 | |
5505 | ipt->pt._key = mask; | |
5506 | ipt->req = req; | |
68b11e8b PB |
5507 | ipt->error = 0; |
5508 | ipt->nr_entries = 0; | |
d7718a9d | 5509 | |
d7718a9d | 5510 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
68b11e8b PB |
5511 | if (unlikely(!ipt->nr_entries) && !ipt->error) |
5512 | ipt->error = -EINVAL; | |
d7718a9d | 5513 | |
79ebeaee | 5514 | spin_lock(&ctx->completion_lock); |
a890d01e | 5515 | if (ipt->error || (mask && (poll->events & EPOLLONESHOT))) |
46fee9ab | 5516 | io_poll_remove_double(req); |
d7718a9d | 5517 | if (likely(poll->head)) { |
79ebeaee | 5518 | spin_lock_irq(&poll->head->lock); |
d7718a9d JA |
5519 | if (unlikely(list_empty(&poll->wait.entry))) { |
5520 | if (ipt->error) | |
5521 | cancel = true; | |
5522 | ipt->error = 0; | |
5523 | mask = 0; | |
5524 | } | |
88e41cf9 | 5525 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5526 | list_del_init(&poll->wait.entry); |
5527 | else if (cancel) | |
5528 | WRITE_ONCE(poll->canceled, true); | |
5529 | else if (!poll->done) /* actually waiting for an event */ | |
5530 | io_poll_req_insert(req); | |
79ebeaee | 5531 | spin_unlock_irq(&poll->head->lock); |
d7718a9d JA |
5532 | } |
5533 | ||
5534 | return mask; | |
5535 | } | |
5536 | ||
59b735ae OL |
5537 | enum { |
5538 | IO_APOLL_OK, | |
5539 | IO_APOLL_ABORTED, | |
5540 | IO_APOLL_READY | |
5541 | }; | |
5542 | ||
5543 | static int io_arm_poll_handler(struct io_kiocb *req) | |
d7718a9d JA |
5544 | { |
5545 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5546 | struct io_ring_ctx *ctx = req->ctx; | |
5547 | struct async_poll *apoll; | |
5548 | struct io_poll_table ipt; | |
b2d9c3da | 5549 | __poll_t ret, mask = EPOLLONESHOT | POLLERR | POLLPRI; |
9dab14b8 | 5550 | int rw; |
d7718a9d JA |
5551 | |
5552 | if (!req->file || !file_can_poll(req->file)) | |
59b735ae | 5553 | return IO_APOLL_ABORTED; |
24c74678 | 5554 | if (req->flags & REQ_F_POLLED) |
59b735ae | 5555 | return IO_APOLL_ABORTED; |
b2d9c3da PB |
5556 | if (!def->pollin && !def->pollout) |
5557 | return IO_APOLL_ABORTED; | |
5558 | ||
5559 | if (def->pollin) { | |
9dab14b8 | 5560 | rw = READ; |
b2d9c3da PB |
5561 | mask |= POLLIN | POLLRDNORM; |
5562 | ||
5563 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5564 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5565 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5566 | mask &= ~POLLIN; | |
5567 | } else { | |
9dab14b8 | 5568 | rw = WRITE; |
b2d9c3da PB |
5569 | mask |= POLLOUT | POLLWRNORM; |
5570 | } | |
5571 | ||
9dab14b8 | 5572 | /* if we can't nonblock try, then no point in arming a poll handler */ |
b191e2df | 5573 | if (!io_file_supports_nowait(req, rw)) |
59b735ae | 5574 | return IO_APOLL_ABORTED; |
d7718a9d JA |
5575 | |
5576 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5577 | if (unlikely(!apoll)) | |
59b735ae | 5578 | return IO_APOLL_ABORTED; |
807abcb0 | 5579 | apoll->double_poll = NULL; |
d7718a9d | 5580 | req->apoll = apoll; |
b2d9c3da | 5581 | req->flags |= REQ_F_POLLED; |
d7718a9d | 5582 | ipt.pt._qproc = io_async_queue_proc; |
48dcd38d | 5583 | io_req_set_refcount(req); |
d7718a9d JA |
5584 | |
5585 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5586 | io_async_wake); | |
79ebeaee | 5587 | spin_unlock(&ctx->completion_lock); |
41a5169c HX |
5588 | if (ret || ipt.error) |
5589 | return ret ? IO_APOLL_READY : IO_APOLL_ABORTED; | |
5590 | ||
236daeae OL |
5591 | trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, |
5592 | mask, apoll->poll.events); | |
59b735ae | 5593 | return IO_APOLL_OK; |
d7718a9d JA |
5594 | } |
5595 | ||
5596 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5597 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5598 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5599 | { |
b41e9852 | 5600 | bool do_complete = false; |
221c5eb2 | 5601 | |
5082620f JA |
5602 | if (!poll->head) |
5603 | return false; | |
79ebeaee | 5604 | spin_lock_irq(&poll->head->lock); |
b2e720ac JA |
5605 | if (do_cancel) |
5606 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5607 | if (!list_empty(&poll->wait.entry)) { |
5608 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5609 | do_complete = true; |
221c5eb2 | 5610 | } |
79ebeaee | 5611 | spin_unlock_irq(&poll->head->lock); |
3bfa5bcb | 5612 | hash_del(&req->hash_node); |
d7718a9d JA |
5613 | return do_complete; |
5614 | } | |
5615 | ||
5d709043 | 5616 | static bool io_poll_remove_one(struct io_kiocb *req) |
e07785b0 | 5617 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5618 | { |
5619 | bool do_complete; | |
5620 | ||
d4e7cd36 | 5621 | io_poll_remove_double(req); |
e31001a3 | 5622 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5623 | |
b41e9852 | 5624 | if (do_complete) { |
d4d19c19 | 5625 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5626 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5627 | req_set_fail(req); |
91c2f697 | 5628 | io_put_req_deferred(req); |
5d709043 | 5629 | } |
b41e9852 | 5630 | return do_complete; |
221c5eb2 JA |
5631 | } |
5632 | ||
76e1b642 JA |
5633 | /* |
5634 | * Returns true if we found and killed one or more poll requests | |
5635 | */ | |
6b81928d | 5636 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5637 | bool cancel_all) |
221c5eb2 | 5638 | { |
78076bb6 | 5639 | struct hlist_node *tmp; |
221c5eb2 | 5640 | struct io_kiocb *req; |
8e2e1faf | 5641 | int posted = 0, i; |
221c5eb2 | 5642 | |
79ebeaee | 5643 | spin_lock(&ctx->completion_lock); |
78076bb6 JA |
5644 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5645 | struct hlist_head *list; | |
5646 | ||
5647 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5648 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5649 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5650 | posted += io_poll_remove_one(req); |
5651 | } | |
221c5eb2 | 5652 | } |
79ebeaee | 5653 | spin_unlock(&ctx->completion_lock); |
b41e9852 | 5654 | |
8e2e1faf JA |
5655 | if (posted) |
5656 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5657 | |
5658 | return posted != 0; | |
221c5eb2 JA |
5659 | } |
5660 | ||
9ba5fac8 PB |
5661 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5662 | bool poll_only) | |
e07785b0 | 5663 | __must_hold(&ctx->completion_lock) |
47f46768 | 5664 | { |
78076bb6 | 5665 | struct hlist_head *list; |
47f46768 JA |
5666 | struct io_kiocb *req; |
5667 | ||
78076bb6 JA |
5668 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5669 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5670 | if (sqe_addr != req->user_data) |
5671 | continue; | |
9ba5fac8 PB |
5672 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5673 | continue; | |
b2cb805f | 5674 | return req; |
47f46768 | 5675 | } |
b2cb805f JA |
5676 | return NULL; |
5677 | } | |
5678 | ||
9ba5fac8 PB |
5679 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5680 | bool poll_only) | |
e07785b0 | 5681 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5682 | { |
5683 | struct io_kiocb *req; | |
5684 | ||
9ba5fac8 | 5685 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5686 | if (!req) |
5687 | return -ENOENT; | |
5688 | if (io_poll_remove_one(req)) | |
5689 | return 0; | |
5690 | ||
5691 | return -EALREADY; | |
47f46768 JA |
5692 | } |
5693 | ||
9096af3e PB |
5694 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5695 | unsigned int flags) | |
5696 | { | |
5697 | u32 events; | |
47f46768 | 5698 | |
9096af3e PB |
5699 | events = READ_ONCE(sqe->poll32_events); |
5700 | #ifdef __BIG_ENDIAN | |
5701 | events = swahw32(events); | |
5702 | #endif | |
5703 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5704 | events |= EPOLLONESHOT; | |
5705 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5706 | } |
5707 | ||
c5de0036 | 5708 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5709 | const struct io_uring_sqe *sqe) |
0969e783 | 5710 | { |
c5de0036 PB |
5711 | struct io_poll_update *upd = &req->poll_update; |
5712 | u32 flags; | |
5713 | ||
0969e783 JA |
5714 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5715 | return -EINVAL; | |
26578cda | 5716 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
c5de0036 PB |
5717 | return -EINVAL; |
5718 | flags = READ_ONCE(sqe->len); | |
5719 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5720 | IORING_POLL_ADD_MULTI)) | |
5721 | return -EINVAL; | |
5722 | /* meaningless without update */ | |
5723 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5724 | return -EINVAL; |
5725 | ||
c5de0036 PB |
5726 | upd->old_user_data = READ_ONCE(sqe->addr); |
5727 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5728 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5729 | |
c5de0036 PB |
5730 | upd->new_user_data = READ_ONCE(sqe->off); |
5731 | if (!upd->update_user_data && upd->new_user_data) | |
5732 | return -EINVAL; | |
5733 | if (upd->update_events) | |
5734 | upd->events = io_poll_parse_events(sqe, flags); | |
5735 | else if (sqe->poll32_events) | |
5736 | return -EINVAL; | |
221c5eb2 | 5737 | |
221c5eb2 JA |
5738 | return 0; |
5739 | } | |
5740 | ||
221c5eb2 JA |
5741 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5742 | void *key) | |
5743 | { | |
c2f2eb7d JA |
5744 | struct io_kiocb *req = wait->private; |
5745 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5746 | |
d7718a9d | 5747 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5748 | } |
5749 | ||
221c5eb2 JA |
5750 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5751 | struct poll_table_struct *p) | |
5752 | { | |
5753 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5754 | ||
e8c2bc1f | 5755 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5756 | } |
5757 | ||
3529d8c2 | 5758 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5759 | { |
5760 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5761 | u32 flags; |
221c5eb2 JA |
5762 | |
5763 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5764 | return -EINVAL; | |
c5de0036 | 5765 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5766 | return -EINVAL; |
5767 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5768 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5769 | return -EINVAL; |
5770 | ||
48dcd38d | 5771 | io_req_set_refcount(req); |
c5de0036 | 5772 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5773 | return 0; |
5774 | } | |
5775 | ||
61e98203 | 5776 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5777 | { |
5778 | struct io_poll_iocb *poll = &req->poll; | |
5779 | struct io_ring_ctx *ctx = req->ctx; | |
5780 | struct io_poll_table ipt; | |
0969e783 | 5781 | __poll_t mask; |
0969e783 | 5782 | |
d7718a9d | 5783 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5784 | |
d7718a9d JA |
5785 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5786 | io_poll_wake); | |
221c5eb2 | 5787 | |
8c838788 | 5788 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5789 | ipt.error = 0; |
e27414be | 5790 | io_poll_complete(req, mask); |
221c5eb2 | 5791 | } |
79ebeaee | 5792 | spin_unlock(&ctx->completion_lock); |
221c5eb2 | 5793 | |
8c838788 JA |
5794 | if (mask) { |
5795 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5796 | if (poll->events & EPOLLONESHOT) |
5797 | io_put_req(req); | |
221c5eb2 | 5798 | } |
8c838788 | 5799 | return ipt.error; |
221c5eb2 JA |
5800 | } |
5801 | ||
c5de0036 | 5802 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5803 | { |
5804 | struct io_ring_ctx *ctx = req->ctx; | |
5805 | struct io_kiocb *preq; | |
cb3b200e | 5806 | bool completing; |
b69de288 JA |
5807 | int ret; |
5808 | ||
79ebeaee | 5809 | spin_lock(&ctx->completion_lock); |
9ba5fac8 | 5810 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5811 | if (!preq) { |
5812 | ret = -ENOENT; | |
5813 | goto err; | |
b69de288 | 5814 | } |
cb3b200e | 5815 | |
c5de0036 PB |
5816 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5817 | completing = true; | |
5818 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5819 | goto err; | |
5820 | } | |
5821 | ||
cb3b200e JA |
5822 | /* |
5823 | * Don't allow racy completion with singleshot, as we cannot safely | |
5824 | * update those. For multishot, if we're racing with completion, just | |
5825 | * let completion re-add it. | |
5826 | */ | |
5827 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5828 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5829 | ret = -EALREADY; | |
5830 | goto err; | |
b69de288 JA |
5831 | } |
5832 | /* we now have a detached poll request. reissue. */ | |
5833 | ret = 0; | |
5834 | err: | |
b69de288 | 5835 | if (ret < 0) { |
79ebeaee | 5836 | spin_unlock(&ctx->completion_lock); |
93d2bcd2 | 5837 | req_set_fail(req); |
b69de288 JA |
5838 | io_req_complete(req, ret); |
5839 | return 0; | |
5840 | } | |
5841 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5842 | if (req->poll_update.update_events) { |
b69de288 | 5843 | preq->poll.events &= ~0xffff; |
9d805892 | 5844 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5845 | preq->poll.events |= IO_POLL_UNMASK; |
5846 | } | |
9d805892 PB |
5847 | if (req->poll_update.update_user_data) |
5848 | preq->user_data = req->poll_update.new_user_data; | |
79ebeaee | 5849 | spin_unlock(&ctx->completion_lock); |
cb3b200e | 5850 | |
b69de288 JA |
5851 | /* complete update request, we're done with it */ |
5852 | io_req_complete(req, ret); | |
5853 | ||
cb3b200e | 5854 | if (!completing) { |
c5de0036 | 5855 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5856 | if (ret < 0) { |
93d2bcd2 | 5857 | req_set_fail(preq); |
cb3b200e JA |
5858 | io_req_complete(preq, ret); |
5859 | } | |
b69de288 JA |
5860 | } |
5861 | return 0; | |
5862 | } | |
5863 | ||
f237c30a | 5864 | static void io_req_task_timeout(struct io_kiocb *req, bool *locked) |
89850fce | 5865 | { |
89850fce | 5866 | req_set_fail(req); |
505657bc | 5867 | io_req_complete_post(req, -ETIME, 0); |
89850fce JA |
5868 | } |
5869 | ||
5262f567 JA |
5870 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5871 | { | |
ad8a48ac JA |
5872 | struct io_timeout_data *data = container_of(timer, |
5873 | struct io_timeout_data, timer); | |
5874 | struct io_kiocb *req = data->req; | |
5875 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5876 | unsigned long flags; |
5877 | ||
89850fce | 5878 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
a71976f3 | 5879 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5880 | atomic_set(&req->ctx->cq_timeouts, |
5881 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
89850fce | 5882 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); |
01cec8c1 | 5883 | |
89850fce JA |
5884 | req->io_task_work.func = io_req_task_timeout; |
5885 | io_req_task_work_add(req); | |
5262f567 JA |
5886 | return HRTIMER_NORESTART; |
5887 | } | |
5888 | ||
fbd15848 PB |
5889 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5890 | __u64 user_data) | |
89850fce | 5891 | __must_hold(&ctx->timeout_lock) |
f254ac04 | 5892 | { |
fbd15848 | 5893 | struct io_timeout_data *io; |
47f46768 | 5894 | struct io_kiocb *req; |
fd9c7bc5 | 5895 | bool found = false; |
f254ac04 | 5896 | |
135fcde8 | 5897 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5898 | found = user_data == req->user_data; |
5899 | if (found) | |
47f46768 | 5900 | break; |
47f46768 | 5901 | } |
fd9c7bc5 PB |
5902 | if (!found) |
5903 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5904 | |
5905 | io = req->async_data; | |
fd9c7bc5 | 5906 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5907 | return ERR_PTR(-EALREADY); |
a71976f3 | 5908 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5909 | return req; |
5910 | } | |
47f46768 | 5911 | |
fbd15848 | 5912 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
ec3c3d0f | 5913 | __must_hold(&ctx->completion_lock) |
89850fce | 5914 | __must_hold(&ctx->timeout_lock) |
fbd15848 PB |
5915 | { |
5916 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5917 | ||
5918 | if (IS_ERR(req)) | |
5919 | return PTR_ERR(req); | |
f254ac04 | 5920 | |
93d2bcd2 | 5921 | req_set_fail(req); |
d4d19c19 | 5922 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
91c2f697 | 5923 | io_put_req_deferred(req); |
f254ac04 JA |
5924 | return 0; |
5925 | } | |
5926 | ||
50c1df2b JA |
5927 | static clockid_t io_timeout_get_clock(struct io_timeout_data *data) |
5928 | { | |
5929 | switch (data->flags & IORING_TIMEOUT_CLOCK_MASK) { | |
5930 | case IORING_TIMEOUT_BOOTTIME: | |
5931 | return CLOCK_BOOTTIME; | |
5932 | case IORING_TIMEOUT_REALTIME: | |
5933 | return CLOCK_REALTIME; | |
5934 | default: | |
5935 | /* can't happen, vetted at prep time */ | |
5936 | WARN_ON_ONCE(1); | |
5937 | fallthrough; | |
5938 | case 0: | |
5939 | return CLOCK_MONOTONIC; | |
5940 | } | |
5941 | } | |
5942 | ||
f1042b6c PB |
5943 | static int io_linked_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5944 | struct timespec64 *ts, enum hrtimer_mode mode) | |
5945 | __must_hold(&ctx->timeout_lock) | |
5946 | { | |
5947 | struct io_timeout_data *io; | |
5948 | struct io_kiocb *req; | |
5949 | bool found = false; | |
5950 | ||
5951 | list_for_each_entry(req, &ctx->ltimeout_list, timeout.list) { | |
5952 | found = user_data == req->user_data; | |
5953 | if (found) | |
5954 | break; | |
5955 | } | |
5956 | if (!found) | |
5957 | return -ENOENT; | |
5958 | ||
5959 | io = req->async_data; | |
5960 | if (hrtimer_try_to_cancel(&io->timer) == -1) | |
5961 | return -EALREADY; | |
5962 | hrtimer_init(&io->timer, io_timeout_get_clock(io), mode); | |
5963 | io->timer.function = io_link_timeout_fn; | |
5964 | hrtimer_start(&io->timer, timespec64_to_ktime(*ts), mode); | |
5965 | return 0; | |
5966 | } | |
5967 | ||
9c8e11b3 PB |
5968 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5969 | struct timespec64 *ts, enum hrtimer_mode mode) | |
89850fce | 5970 | __must_hold(&ctx->timeout_lock) |
47f46768 | 5971 | { |
9c8e11b3 PB |
5972 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5973 | struct io_timeout_data *data; | |
47f46768 | 5974 | |
9c8e11b3 PB |
5975 | if (IS_ERR(req)) |
5976 | return PTR_ERR(req); | |
47f46768 | 5977 | |
9c8e11b3 PB |
5978 | req->timeout.off = 0; /* noseq */ |
5979 | data = req->async_data; | |
5980 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
50c1df2b | 5981 | hrtimer_init(&data->timer, io_timeout_get_clock(data), mode); |
9c8e11b3 PB |
5982 | data->timer.function = io_timeout_fn; |
5983 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5984 | return 0; | |
47f46768 JA |
5985 | } |
5986 | ||
3529d8c2 JA |
5987 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5988 | const struct io_uring_sqe *sqe) | |
b29472ee | 5989 | { |
9c8e11b3 PB |
5990 | struct io_timeout_rem *tr = &req->timeout_rem; |
5991 | ||
b29472ee JA |
5992 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5993 | return -EINVAL; | |
61710e43 DA |
5994 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5995 | return -EINVAL; | |
26578cda | 5996 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->splice_fd_in) |
b29472ee JA |
5997 | return -EINVAL; |
5998 | ||
f1042b6c | 5999 | tr->ltimeout = false; |
9c8e11b3 PB |
6000 | tr->addr = READ_ONCE(sqe->addr); |
6001 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
f1042b6c PB |
6002 | if (tr->flags & IORING_TIMEOUT_UPDATE_MASK) { |
6003 | if (hweight32(tr->flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
6004 | return -EINVAL; | |
6005 | if (tr->flags & IORING_LINK_TIMEOUT_UPDATE) | |
6006 | tr->ltimeout = true; | |
6007 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE_MASK|IORING_TIMEOUT_ABS)) | |
9c8e11b3 PB |
6008 | return -EINVAL; |
6009 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
6010 | return -EFAULT; | |
6011 | } else if (tr->flags) { | |
6012 | /* timeout removal doesn't support flags */ | |
b29472ee | 6013 | return -EINVAL; |
9c8e11b3 | 6014 | } |
b29472ee | 6015 | |
b29472ee JA |
6016 | return 0; |
6017 | } | |
6018 | ||
8662daec PB |
6019 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
6020 | { | |
6021 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
6022 | : HRTIMER_MODE_REL; | |
6023 | } | |
6024 | ||
11365043 JA |
6025 | /* |
6026 | * Remove or update an existing timeout command | |
6027 | */ | |
61e98203 | 6028 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 6029 | { |
9c8e11b3 | 6030 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 6031 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 6032 | int ret; |
11365043 | 6033 | |
ec3c3d0f PB |
6034 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { |
6035 | spin_lock(&ctx->completion_lock); | |
6036 | spin_lock_irq(&ctx->timeout_lock); | |
9c8e11b3 | 6037 | ret = io_timeout_cancel(ctx, tr->addr); |
ec3c3d0f PB |
6038 | spin_unlock_irq(&ctx->timeout_lock); |
6039 | spin_unlock(&ctx->completion_lock); | |
6040 | } else { | |
f1042b6c PB |
6041 | enum hrtimer_mode mode = io_translate_timeout_mode(tr->flags); |
6042 | ||
ec3c3d0f | 6043 | spin_lock_irq(&ctx->timeout_lock); |
f1042b6c PB |
6044 | if (tr->ltimeout) |
6045 | ret = io_linked_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
6046 | else | |
6047 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
ec3c3d0f PB |
6048 | spin_unlock_irq(&ctx->timeout_lock); |
6049 | } | |
11365043 | 6050 | |
4e88d6e7 | 6051 | if (ret < 0) |
93d2bcd2 | 6052 | req_set_fail(req); |
505657bc | 6053 | io_req_complete_post(req, ret, 0); |
11365043 | 6054 | return 0; |
5262f567 JA |
6055 | } |
6056 | ||
3529d8c2 | 6057 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 6058 | bool is_timeout_link) |
5262f567 | 6059 | { |
ad8a48ac | 6060 | struct io_timeout_data *data; |
a41525ab | 6061 | unsigned flags; |
56080b02 | 6062 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 6063 | |
ad8a48ac | 6064 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 6065 | return -EINVAL; |
26578cda PB |
6066 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1 || |
6067 | sqe->splice_fd_in) | |
a41525ab | 6068 | return -EINVAL; |
56080b02 | 6069 | if (off && is_timeout_link) |
2d28390a | 6070 | return -EINVAL; |
a41525ab | 6071 | flags = READ_ONCE(sqe->timeout_flags); |
50c1df2b JA |
6072 | if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_CLOCK_MASK)) |
6073 | return -EINVAL; | |
6074 | /* more than one clock specified is invalid, obviously */ | |
6075 | if (hweight32(flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
5262f567 | 6076 | return -EINVAL; |
bdf20073 | 6077 | |
ef9dd637 | 6078 | INIT_LIST_HEAD(&req->timeout.list); |
bfe68a22 | 6079 | req->timeout.off = off; |
f18ee4cf PB |
6080 | if (unlikely(off && !req->ctx->off_timeout_used)) |
6081 | req->ctx->off_timeout_used = true; | |
26a61679 | 6082 | |
e8c2bc1f | 6083 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
6084 | return -ENOMEM; |
6085 | ||
e8c2bc1f | 6086 | data = req->async_data; |
ad8a48ac | 6087 | data->req = req; |
50c1df2b | 6088 | data->flags = flags; |
ad8a48ac JA |
6089 | |
6090 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
6091 | return -EFAULT; |
6092 | ||
8662daec | 6093 | data->mode = io_translate_timeout_mode(flags); |
50c1df2b | 6094 | hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode); |
b97e736a PB |
6095 | |
6096 | if (is_timeout_link) { | |
6097 | struct io_submit_link *link = &req->ctx->submit_state.link; | |
6098 | ||
6099 | if (!link->head) | |
6100 | return -EINVAL; | |
6101 | if (link->last->opcode == IORING_OP_LINK_TIMEOUT) | |
6102 | return -EINVAL; | |
4d13d1a4 PB |
6103 | req->timeout.head = link->last; |
6104 | link->last->flags |= REQ_F_ARM_LTIMEOUT; | |
b97e736a | 6105 | } |
ad8a48ac JA |
6106 | return 0; |
6107 | } | |
6108 | ||
61e98203 | 6109 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 6110 | { |
ad8a48ac | 6111 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 6112 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 6113 | struct list_head *entry; |
bfe68a22 | 6114 | u32 tail, off = req->timeout.off; |
ad8a48ac | 6115 | |
89850fce | 6116 | spin_lock_irq(&ctx->timeout_lock); |
93bd25bb | 6117 | |
5262f567 JA |
6118 | /* |
6119 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
6120 | * timeout event to be satisfied. If it isn't set, then this is |
6121 | * a pure timeout request, sequence isn't used. | |
5262f567 | 6122 | */ |
8eb7e2d0 | 6123 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
6124 | entry = ctx->timeout_list.prev; |
6125 | goto add; | |
6126 | } | |
5262f567 | 6127 | |
bfe68a22 PB |
6128 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
6129 | req->timeout.target_seq = tail + off; | |
5262f567 | 6130 | |
f010505b MDG |
6131 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
6132 | * This is safe because ->completion_lock is held, and submissions | |
6133 | * and completions are never mixed in the same ->completion_lock section. | |
6134 | */ | |
6135 | ctx->cq_last_tm_flush = tail; | |
6136 | ||
5262f567 JA |
6137 | /* |
6138 | * Insertion sort, ensuring the first entry in the list is always | |
6139 | * the one we need first. | |
6140 | */ | |
5262f567 | 6141 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
6142 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
6143 | timeout.list); | |
5262f567 | 6144 | |
8eb7e2d0 | 6145 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 6146 | continue; |
bfe68a22 PB |
6147 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
6148 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
6149 | break; |
6150 | } | |
93bd25bb | 6151 | add: |
135fcde8 | 6152 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
6153 | data->timer.function = io_timeout_fn; |
6154 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
89850fce | 6155 | spin_unlock_irq(&ctx->timeout_lock); |
5262f567 JA |
6156 | return 0; |
6157 | } | |
5262f567 | 6158 | |
f458dd84 PB |
6159 | struct io_cancel_data { |
6160 | struct io_ring_ctx *ctx; | |
6161 | u64 user_data; | |
6162 | }; | |
6163 | ||
62755e35 JA |
6164 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
6165 | { | |
6166 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 6167 | struct io_cancel_data *cd = data; |
62755e35 | 6168 | |
f458dd84 | 6169 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
6170 | } |
6171 | ||
f458dd84 PB |
6172 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
6173 | struct io_ring_ctx *ctx) | |
62755e35 | 6174 | { |
f458dd84 | 6175 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 6176 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
6177 | int ret = 0; |
6178 | ||
f458dd84 | 6179 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
6180 | return -ENOENT; |
6181 | ||
f458dd84 | 6182 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
6183 | switch (cancel_ret) { |
6184 | case IO_WQ_CANCEL_OK: | |
6185 | ret = 0; | |
6186 | break; | |
6187 | case IO_WQ_CANCEL_RUNNING: | |
6188 | ret = -EALREADY; | |
6189 | break; | |
6190 | case IO_WQ_CANCEL_NOTFOUND: | |
6191 | ret = -ENOENT; | |
6192 | break; | |
6193 | } | |
6194 | ||
e977d6d3 JA |
6195 | return ret; |
6196 | } | |
6197 | ||
8cb01fac | 6198 | static int io_try_cancel_userdata(struct io_kiocb *req, u64 sqe_addr) |
47f46768 | 6199 | { |
8cb01fac | 6200 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 JA |
6201 | int ret; |
6202 | ||
dadebc35 | 6203 | WARN_ON_ONCE(!io_wq_current_is_worker() && req->task != current); |
8cb01fac | 6204 | |
f458dd84 | 6205 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
df9727af | 6206 | if (ret != -ENOENT) |
8cb01fac | 6207 | return ret; |
505657bc PB |
6208 | |
6209 | spin_lock(&ctx->completion_lock); | |
79ebeaee | 6210 | spin_lock_irq(&ctx->timeout_lock); |
47f46768 | 6211 | ret = io_timeout_cancel(ctx, sqe_addr); |
79ebeaee | 6212 | spin_unlock_irq(&ctx->timeout_lock); |
47f46768 | 6213 | if (ret != -ENOENT) |
505657bc PB |
6214 | goto out; |
6215 | ret = io_poll_cancel(ctx, sqe_addr, false); | |
6216 | out: | |
6217 | spin_unlock(&ctx->completion_lock); | |
6218 | return ret; | |
47f46768 JA |
6219 | } |
6220 | ||
3529d8c2 JA |
6221 | static int io_async_cancel_prep(struct io_kiocb *req, |
6222 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6223 | { |
fbf23849 | 6224 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6225 | return -EINVAL; |
61710e43 DA |
6226 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6227 | return -EINVAL; | |
26578cda PB |
6228 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags || |
6229 | sqe->splice_fd_in) | |
e977d6d3 JA |
6230 | return -EINVAL; |
6231 | ||
fbf23849 JA |
6232 | req->cancel.addr = READ_ONCE(sqe->addr); |
6233 | return 0; | |
6234 | } | |
6235 | ||
61e98203 | 6236 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6237 | { |
6238 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
6239 | u64 sqe_addr = req->cancel.addr; |
6240 | struct io_tctx_node *node; | |
6241 | int ret; | |
6242 | ||
8cb01fac | 6243 | ret = io_try_cancel_userdata(req, sqe_addr); |
58f99373 PB |
6244 | if (ret != -ENOENT) |
6245 | goto done; | |
58f99373 PB |
6246 | |
6247 | /* slow path, try all io-wq's */ | |
6248 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
6249 | ret = -ENOENT; | |
6250 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
6251 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 6252 | |
58f99373 PB |
6253 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
6254 | if (ret != -ENOENT) | |
6255 | break; | |
6256 | } | |
6257 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
58f99373 | 6258 | done: |
58f99373 | 6259 | if (ret < 0) |
93d2bcd2 | 6260 | req_set_fail(req); |
505657bc | 6261 | io_req_complete_post(req, ret, 0); |
5262f567 JA |
6262 | return 0; |
6263 | } | |
6264 | ||
269bbe5f | 6265 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6266 | const struct io_uring_sqe *sqe) |
6267 | { | |
61710e43 DA |
6268 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6269 | return -EINVAL; | |
26578cda | 6270 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
05f3fb3c JA |
6271 | return -EINVAL; |
6272 | ||
269bbe5f BM |
6273 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6274 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6275 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6276 | return -EINVAL; |
269bbe5f | 6277 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6278 | return 0; |
6279 | } | |
6280 | ||
889fca73 | 6281 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6282 | { |
6283 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 6284 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 6285 | int ret; |
fbf23849 | 6286 | |
45d189c6 | 6287 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 6288 | return -EAGAIN; |
05f3fb3c | 6289 | |
269bbe5f BM |
6290 | up.offset = req->rsrc_update.offset; |
6291 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
6292 | up.nr = 0; |
6293 | up.tags = 0; | |
615cee49 | 6294 | up.resv = 0; |
05f3fb3c JA |
6295 | |
6296 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 6297 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 6298 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
6299 | mutex_unlock(&ctx->uring_lock); |
6300 | ||
6301 | if (ret < 0) | |
93d2bcd2 | 6302 | req_set_fail(req); |
889fca73 | 6303 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6304 | return 0; |
6305 | } | |
6306 | ||
bfe76559 | 6307 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6308 | { |
d625c6ee | 6309 | switch (req->opcode) { |
e781573e | 6310 | case IORING_OP_NOP: |
bfe76559 | 6311 | return 0; |
f67676d1 JA |
6312 | case IORING_OP_READV: |
6313 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6314 | case IORING_OP_READ: |
bfe76559 | 6315 | return io_read_prep(req, sqe); |
f67676d1 JA |
6316 | case IORING_OP_WRITEV: |
6317 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6318 | case IORING_OP_WRITE: |
bfe76559 | 6319 | return io_write_prep(req, sqe); |
0969e783 | 6320 | case IORING_OP_POLL_ADD: |
bfe76559 | 6321 | return io_poll_add_prep(req, sqe); |
0969e783 | 6322 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 6323 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 6324 | case IORING_OP_FSYNC: |
1155c76a | 6325 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6326 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6327 | return io_sfr_prep(req, sqe); |
03b1230c | 6328 | case IORING_OP_SENDMSG: |
fddaface | 6329 | case IORING_OP_SEND: |
bfe76559 | 6330 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6331 | case IORING_OP_RECVMSG: |
fddaface | 6332 | case IORING_OP_RECV: |
bfe76559 | 6333 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6334 | case IORING_OP_CONNECT: |
bfe76559 | 6335 | return io_connect_prep(req, sqe); |
2d28390a | 6336 | case IORING_OP_TIMEOUT: |
bfe76559 | 6337 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6338 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6339 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6340 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6341 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6342 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6343 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6344 | case IORING_OP_ACCEPT: |
bfe76559 | 6345 | return io_accept_prep(req, sqe); |
d63d1b5e | 6346 | case IORING_OP_FALLOCATE: |
bfe76559 | 6347 | return io_fallocate_prep(req, sqe); |
15b71abe | 6348 | case IORING_OP_OPENAT: |
bfe76559 | 6349 | return io_openat_prep(req, sqe); |
b5dba59e | 6350 | case IORING_OP_CLOSE: |
bfe76559 | 6351 | return io_close_prep(req, sqe); |
05f3fb3c | 6352 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6353 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6354 | case IORING_OP_STATX: |
bfe76559 | 6355 | return io_statx_prep(req, sqe); |
4840e418 | 6356 | case IORING_OP_FADVISE: |
bfe76559 | 6357 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6358 | case IORING_OP_MADVISE: |
bfe76559 | 6359 | return io_madvise_prep(req, sqe); |
cebdb986 | 6360 | case IORING_OP_OPENAT2: |
bfe76559 | 6361 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6362 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6363 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6364 | case IORING_OP_SPLICE: |
bfe76559 | 6365 | return io_splice_prep(req, sqe); |
ddf0322d | 6366 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6367 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6368 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6369 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6370 | case IORING_OP_TEE: |
bfe76559 | 6371 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6372 | case IORING_OP_SHUTDOWN: |
6373 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6374 | case IORING_OP_RENAMEAT: |
6375 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6376 | case IORING_OP_UNLINKAT: |
6377 | return io_unlinkat_prep(req, sqe); | |
e34a02dc DK |
6378 | case IORING_OP_MKDIRAT: |
6379 | return io_mkdirat_prep(req, sqe); | |
7a8721f8 DK |
6380 | case IORING_OP_SYMLINKAT: |
6381 | return io_symlinkat_prep(req, sqe); | |
cf30da90 DK |
6382 | case IORING_OP_LINKAT: |
6383 | return io_linkat_prep(req, sqe); | |
f67676d1 JA |
6384 | } |
6385 | ||
bfe76559 PB |
6386 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6387 | req->opcode); | |
bd54b6fe | 6388 | return -EINVAL; |
bfe76559 PB |
6389 | } |
6390 | ||
93642ef8 | 6391 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 6392 | { |
b7e298d2 PB |
6393 | if (!io_op_defs[req->opcode].needs_async_setup) |
6394 | return 0; | |
6395 | if (WARN_ON_ONCE(req->async_data)) | |
6396 | return -EFAULT; | |
6397 | if (io_alloc_async_data(req)) | |
6398 | return -EAGAIN; | |
6399 | ||
93642ef8 PB |
6400 | switch (req->opcode) { |
6401 | case IORING_OP_READV: | |
93642ef8 PB |
6402 | return io_rw_prep_async(req, READ); |
6403 | case IORING_OP_WRITEV: | |
93642ef8 PB |
6404 | return io_rw_prep_async(req, WRITE); |
6405 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
6406 | return io_sendmsg_prep_async(req); |
6407 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
6408 | return io_recvmsg_prep_async(req); |
6409 | case IORING_OP_CONNECT: | |
6410 | return io_connect_prep_async(req); | |
6411 | } | |
b7e298d2 PB |
6412 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
6413 | req->opcode); | |
6414 | return -EFAULT; | |
f67676d1 JA |
6415 | } |
6416 | ||
9cf7c104 PB |
6417 | static u32 io_get_sequence(struct io_kiocb *req) |
6418 | { | |
a3dbdf54 | 6419 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 6420 | |
a3dbdf54 PB |
6421 | /* need original cached_sq_head, but it was increased for each req */ |
6422 | io_for_each_link(req, req) | |
6423 | seq--; | |
6424 | return seq; | |
9cf7c104 PB |
6425 | } |
6426 | ||
76cc33d7 | 6427 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 6428 | { |
3c19966d | 6429 | struct io_kiocb *pos; |
a197f664 | 6430 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6431 | struct io_defer_entry *de; |
f67676d1 | 6432 | int ret; |
9cf7c104 | 6433 | u32 seq; |
de0617e4 | 6434 | |
b8ce1b9d PB |
6435 | if (req->flags & REQ_F_FAIL) { |
6436 | io_req_complete_fail_submit(req); | |
6437 | return true; | |
6438 | } | |
6439 | ||
3c19966d PB |
6440 | /* |
6441 | * If we need to drain a request in the middle of a link, drain the | |
6442 | * head request and the next request/link after the current link. | |
6443 | * Considering sequential execution of links, IOSQE_IO_DRAIN will be | |
6444 | * maintained for every request of our link. | |
6445 | */ | |
6446 | if (ctx->drain_next) { | |
6447 | req->flags |= REQ_F_IO_DRAIN; | |
6448 | ctx->drain_next = false; | |
6449 | } | |
6450 | /* not interested in head, start from the first linked */ | |
6451 | io_for_each_link(pos, req->link) { | |
6452 | if (pos->flags & REQ_F_IO_DRAIN) { | |
6453 | ctx->drain_next = true; | |
6454 | req->flags |= REQ_F_IO_DRAIN; | |
6455 | break; | |
6456 | } | |
6457 | } | |
6458 | ||
9d858b21 | 6459 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 | 6460 | if (likely(list_empty_careful(&ctx->defer_list) && |
10c66904 PB |
6461 | !(req->flags & REQ_F_IO_DRAIN))) { |
6462 | ctx->drain_active = false; | |
76cc33d7 | 6463 | return false; |
10c66904 | 6464 | } |
9cf7c104 PB |
6465 | |
6466 | seq = io_get_sequence(req); | |
6467 | /* Still a chance to pass the sequence check */ | |
6468 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6469 | return false; |
de0617e4 | 6470 | |
b7e298d2 | 6471 | ret = io_req_prep_async(req); |
be7053b7 | 6472 | if (ret) |
1b48773f | 6473 | goto fail; |
cbdcb435 | 6474 | io_prep_async_link(req); |
27dc8338 | 6475 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 6476 | if (!de) { |
1b48773f PB |
6477 | ret = -ENOMEM; |
6478 | fail: | |
6479 | io_req_complete_failed(req, ret); | |
76cc33d7 PB |
6480 | return true; |
6481 | } | |
2d28390a | 6482 | |
79ebeaee | 6483 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 6484 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
79ebeaee | 6485 | spin_unlock(&ctx->completion_lock); |
27dc8338 | 6486 | kfree(de); |
f237c30a | 6487 | io_queue_async_work(req, NULL); |
76cc33d7 | 6488 | return true; |
de0617e4 JA |
6489 | } |
6490 | ||
915967f6 | 6491 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6492 | de->req = req; |
9cf7c104 | 6493 | de->seq = seq; |
27dc8338 | 6494 | list_add_tail(&de->list, &ctx->defer_list); |
79ebeaee | 6495 | spin_unlock(&ctx->completion_lock); |
76cc33d7 | 6496 | return true; |
de0617e4 JA |
6497 | } |
6498 | ||
68fb8979 | 6499 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6500 | { |
0e1b6fe3 PB |
6501 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6502 | switch (req->opcode) { | |
6503 | case IORING_OP_READV: | |
6504 | case IORING_OP_READ_FIXED: | |
6505 | case IORING_OP_READ: | |
bcda7baa | 6506 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6507 | break; |
6508 | case IORING_OP_RECVMSG: | |
6509 | case IORING_OP_RECV: | |
bcda7baa | 6510 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6511 | break; |
6512 | } | |
99bc4c38 PB |
6513 | } |
6514 | ||
0e1b6fe3 PB |
6515 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6516 | switch (req->opcode) { | |
6517 | case IORING_OP_READV: | |
6518 | case IORING_OP_READ_FIXED: | |
6519 | case IORING_OP_READ: | |
6520 | case IORING_OP_WRITEV: | |
6521 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6522 | case IORING_OP_WRITE: { |
6523 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
6524 | |
6525 | kfree(io->free_iovec); | |
0e1b6fe3 | 6526 | break; |
e8c2bc1f | 6527 | } |
0e1b6fe3 | 6528 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6529 | case IORING_OP_SENDMSG: { |
6530 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6531 | |
6532 | kfree(io->free_iov); | |
0e1b6fe3 | 6533 | break; |
e8c2bc1f | 6534 | } |
0e1b6fe3 PB |
6535 | case IORING_OP_SPLICE: |
6536 | case IORING_OP_TEE: | |
e1d767f0 PB |
6537 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6538 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6539 | break; |
f3cd4850 JA |
6540 | case IORING_OP_OPENAT: |
6541 | case IORING_OP_OPENAT2: | |
6542 | if (req->open.filename) | |
6543 | putname(req->open.filename); | |
6544 | break; | |
80a261fd JA |
6545 | case IORING_OP_RENAMEAT: |
6546 | putname(req->rename.oldpath); | |
6547 | putname(req->rename.newpath); | |
6548 | break; | |
14a1143b JA |
6549 | case IORING_OP_UNLINKAT: |
6550 | putname(req->unlink.filename); | |
6551 | break; | |
e34a02dc DK |
6552 | case IORING_OP_MKDIRAT: |
6553 | putname(req->mkdir.filename); | |
6554 | break; | |
7a8721f8 DK |
6555 | case IORING_OP_SYMLINKAT: |
6556 | putname(req->symlink.oldpath); | |
6557 | putname(req->symlink.newpath); | |
6558 | break; | |
cf30da90 DK |
6559 | case IORING_OP_LINKAT: |
6560 | putname(req->hardlink.oldpath); | |
6561 | putname(req->hardlink.newpath); | |
6562 | break; | |
0e1b6fe3 | 6563 | } |
99bc4c38 | 6564 | } |
75652a30 JA |
6565 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6566 | kfree(req->apoll->double_poll); | |
6567 | kfree(req->apoll); | |
6568 | req->apoll = NULL; | |
6569 | } | |
3a0a6902 PB |
6570 | if (req->flags & REQ_F_INFLIGHT) { |
6571 | struct io_uring_task *tctx = req->task->io_uring; | |
6572 | ||
6573 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 6574 | } |
c854357b | 6575 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 6576 | put_cred(req->creds); |
c854357b PB |
6577 | |
6578 | req->flags &= ~IO_REQ_CLEAN_FLAGS; | |
99bc4c38 PB |
6579 | } |
6580 | ||
889fca73 | 6581 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6582 | { |
a197f664 | 6583 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6584 | const struct cred *creds = NULL; |
d625c6ee | 6585 | int ret; |
2b188cc1 | 6586 | |
b8e64b53 | 6587 | if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) |
c10d1f98 | 6588 | creds = override_creds(req->creds); |
5730b27e | 6589 | |
d625c6ee | 6590 | switch (req->opcode) { |
2b188cc1 | 6591 | case IORING_OP_NOP: |
889fca73 | 6592 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6593 | break; |
6594 | case IORING_OP_READV: | |
edafccee | 6595 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6596 | case IORING_OP_READ: |
889fca73 | 6597 | ret = io_read(req, issue_flags); |
edafccee | 6598 | break; |
3529d8c2 | 6599 | case IORING_OP_WRITEV: |
edafccee | 6600 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6601 | case IORING_OP_WRITE: |
889fca73 | 6602 | ret = io_write(req, issue_flags); |
2b188cc1 | 6603 | break; |
c992fe29 | 6604 | case IORING_OP_FSYNC: |
45d189c6 | 6605 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6606 | break; |
221c5eb2 | 6607 | case IORING_OP_POLL_ADD: |
61e98203 | 6608 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6609 | break; |
6610 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6611 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6612 | break; |
5d17b4a4 | 6613 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6614 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6615 | break; |
0fa03c62 | 6616 | case IORING_OP_SENDMSG: |
889fca73 | 6617 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6618 | break; |
fddaface | 6619 | case IORING_OP_SEND: |
889fca73 | 6620 | ret = io_send(req, issue_flags); |
0fa03c62 | 6621 | break; |
aa1fa28f | 6622 | case IORING_OP_RECVMSG: |
889fca73 | 6623 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6624 | break; |
fddaface | 6625 | case IORING_OP_RECV: |
889fca73 | 6626 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6627 | break; |
5262f567 | 6628 | case IORING_OP_TIMEOUT: |
61e98203 | 6629 | ret = io_timeout(req, issue_flags); |
5262f567 | 6630 | break; |
11365043 | 6631 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6632 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6633 | break; |
17f2fe35 | 6634 | case IORING_OP_ACCEPT: |
889fca73 | 6635 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6636 | break; |
f8e85cf2 | 6637 | case IORING_OP_CONNECT: |
889fca73 | 6638 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6639 | break; |
62755e35 | 6640 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6641 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6642 | break; |
d63d1b5e | 6643 | case IORING_OP_FALLOCATE: |
45d189c6 | 6644 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6645 | break; |
15b71abe | 6646 | case IORING_OP_OPENAT: |
45d189c6 | 6647 | ret = io_openat(req, issue_flags); |
15b71abe | 6648 | break; |
b5dba59e | 6649 | case IORING_OP_CLOSE: |
889fca73 | 6650 | ret = io_close(req, issue_flags); |
b5dba59e | 6651 | break; |
05f3fb3c | 6652 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6653 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6654 | break; |
eddc7ef5 | 6655 | case IORING_OP_STATX: |
45d189c6 | 6656 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6657 | break; |
4840e418 | 6658 | case IORING_OP_FADVISE: |
45d189c6 | 6659 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6660 | break; |
c1ca757b | 6661 | case IORING_OP_MADVISE: |
45d189c6 | 6662 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6663 | break; |
cebdb986 | 6664 | case IORING_OP_OPENAT2: |
45d189c6 | 6665 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6666 | break; |
3e4827b0 | 6667 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6668 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6669 | break; |
7d67af2c | 6670 | case IORING_OP_SPLICE: |
45d189c6 | 6671 | ret = io_splice(req, issue_flags); |
7d67af2c | 6672 | break; |
ddf0322d | 6673 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6674 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6675 | break; |
067524e9 | 6676 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6677 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6678 | break; |
f2a8d5c7 | 6679 | case IORING_OP_TEE: |
45d189c6 | 6680 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6681 | break; |
36f4fa68 | 6682 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6683 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6684 | break; |
80a261fd | 6685 | case IORING_OP_RENAMEAT: |
45d189c6 | 6686 | ret = io_renameat(req, issue_flags); |
80a261fd | 6687 | break; |
14a1143b | 6688 | case IORING_OP_UNLINKAT: |
45d189c6 | 6689 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6690 | break; |
e34a02dc DK |
6691 | case IORING_OP_MKDIRAT: |
6692 | ret = io_mkdirat(req, issue_flags); | |
6693 | break; | |
7a8721f8 DK |
6694 | case IORING_OP_SYMLINKAT: |
6695 | ret = io_symlinkat(req, issue_flags); | |
6696 | break; | |
cf30da90 DK |
6697 | case IORING_OP_LINKAT: |
6698 | ret = io_linkat(req, issue_flags); | |
6699 | break; | |
2b188cc1 JA |
6700 | default: |
6701 | ret = -EINVAL; | |
6702 | break; | |
6703 | } | |
6704 | ||
5730b27e JA |
6705 | if (creds) |
6706 | revert_creds(creds); | |
def596e9 JA |
6707 | if (ret) |
6708 | return ret; | |
b532576e | 6709 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6710 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6711 | io_iopoll_req_issued(req); | |
def596e9 JA |
6712 | |
6713 | return 0; | |
2b188cc1 JA |
6714 | } |
6715 | ||
ebc11b6c PB |
6716 | static struct io_wq_work *io_wq_free_work(struct io_wq_work *work) |
6717 | { | |
6718 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6719 | ||
6720 | req = io_put_req_find_next(req); | |
6721 | return req ? &req->work : NULL; | |
6722 | } | |
6723 | ||
5280f7e5 | 6724 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6725 | { |
6726 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6727 | struct io_kiocb *timeout; |
561fb04a | 6728 | int ret = 0; |
2b188cc1 | 6729 | |
48dcd38d PB |
6730 | /* one will be dropped by ->io_free_work() after returning to io-wq */ |
6731 | if (!(req->flags & REQ_F_REFCOUNT)) | |
6732 | __io_req_set_refcount(req, 2); | |
6733 | else | |
6734 | req_ref_get(req); | |
5d5901a3 | 6735 | |
6df1db6b PB |
6736 | timeout = io_prep_linked_timeout(req); |
6737 | if (timeout) | |
6738 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6739 | |
dadebc35 | 6740 | /* either cancelled or io-wq is dying, so don't touch tctx->iowq */ |
4014d943 | 6741 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6742 | ret = -ECANCELED; |
31b51510 | 6743 | |
561fb04a | 6744 | if (!ret) { |
561fb04a | 6745 | do { |
889fca73 | 6746 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6747 | /* |
6748 | * We can get EAGAIN for polled IO even though we're | |
6749 | * forcing a sync submission from here, since we can't | |
6750 | * wait for request slots on the block side. | |
6751 | */ | |
6752 | if (ret != -EAGAIN) | |
6753 | break; | |
6754 | cond_resched(); | |
6755 | } while (1); | |
6756 | } | |
31b51510 | 6757 | |
a3df7698 | 6758 | /* avoid locking problems by failing it from a clean context */ |
5d5901a3 | 6759 | if (ret) |
a3df7698 | 6760 | io_req_task_queue_fail(req, ret); |
2b188cc1 JA |
6761 | } |
6762 | ||
aeca241b | 6763 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
042b0d85 | 6764 | unsigned i) |
65e19f54 | 6765 | { |
042b0d85 | 6766 | return &table->files[i]; |
dafecf19 PB |
6767 | } |
6768 | ||
65e19f54 JA |
6769 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6770 | int index) | |
6771 | { | |
aeca241b | 6772 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6773 | |
a04b0ac0 | 6774 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6775 | } |
6776 | ||
a04b0ac0 | 6777 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6778 | { |
6779 | unsigned long file_ptr = (unsigned long) file; | |
6780 | ||
b191e2df | 6781 | if (__io_file_supports_nowait(file, READ)) |
9a321c98 | 6782 | file_ptr |= FFS_ASYNC_READ; |
b191e2df | 6783 | if (__io_file_supports_nowait(file, WRITE)) |
9a321c98 PB |
6784 | file_ptr |= FFS_ASYNC_WRITE; |
6785 | if (S_ISREG(file_inode(file)->i_mode)) | |
6786 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6787 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6788 | } |
6789 | ||
ac177053 PB |
6790 | static inline struct file *io_file_get_fixed(struct io_ring_ctx *ctx, |
6791 | struct io_kiocb *req, int fd) | |
09bb8394 | 6792 | { |
8da11c19 | 6793 | struct file *file; |
ac177053 | 6794 | unsigned long file_ptr; |
09bb8394 | 6795 | |
ac177053 PB |
6796 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
6797 | return NULL; | |
6798 | fd = array_index_nospec(fd, ctx->nr_user_files); | |
6799 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; | |
6800 | file = (struct file *) (file_ptr & FFS_MASK); | |
6801 | file_ptr &= ~FFS_MASK; | |
6802 | /* mask in overlapping REQ_F and FFS bits */ | |
b191e2df | 6803 | req->flags |= (file_ptr << REQ_F_NOWAIT_READ_BIT); |
ac177053 PB |
6804 | io_req_set_rsrc_node(req); |
6805 | return file; | |
6806 | } | |
d44f554e | 6807 | |
ac177053 | 6808 | static struct file *io_file_get_normal(struct io_ring_ctx *ctx, |
ac177053 PB |
6809 | struct io_kiocb *req, int fd) |
6810 | { | |
62906e89 | 6811 | struct file *file = fget(fd); |
ac177053 PB |
6812 | |
6813 | trace_io_uring_file_get(ctx, fd); | |
09bb8394 | 6814 | |
ac177053 PB |
6815 | /* we don't allow fixed io_uring files */ |
6816 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6817 | io_req_track_inflight(req); | |
8371adf5 | 6818 | return file; |
09bb8394 JA |
6819 | } |
6820 | ||
ac177053 | 6821 | static inline struct file *io_file_get(struct io_ring_ctx *ctx, |
ac177053 PB |
6822 | struct io_kiocb *req, int fd, bool fixed) |
6823 | { | |
6824 | if (fixed) | |
6825 | return io_file_get_fixed(ctx, req, fd); | |
6826 | else | |
62906e89 | 6827 | return io_file_get_normal(ctx, req, fd); |
ac177053 PB |
6828 | } |
6829 | ||
f237c30a | 6830 | static void io_req_task_link_timeout(struct io_kiocb *req, bool *locked) |
89b263f6 JA |
6831 | { |
6832 | struct io_kiocb *prev = req->timeout.prev; | |
8cb01fac | 6833 | int ret; |
89b263f6 JA |
6834 | |
6835 | if (prev) { | |
8cb01fac | 6836 | ret = io_try_cancel_userdata(req, prev->user_data); |
505657bc | 6837 | io_req_complete_post(req, ret ?: -ETIME, 0); |
89b263f6 | 6838 | io_put_req(prev); |
89b263f6 JA |
6839 | } else { |
6840 | io_req_complete_post(req, -ETIME, 0); | |
6841 | } | |
6842 | } | |
6843 | ||
2665abfd | 6844 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6845 | { |
ad8a48ac JA |
6846 | struct io_timeout_data *data = container_of(timer, |
6847 | struct io_timeout_data, timer); | |
90cd7e42 | 6848 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6849 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6850 | unsigned long flags; |
2665abfd | 6851 | |
89b263f6 | 6852 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
90cd7e42 PB |
6853 | prev = req->timeout.head; |
6854 | req->timeout.head = NULL; | |
2665abfd JA |
6855 | |
6856 | /* | |
6857 | * We don't expect the list to be empty, that will only happen if we | |
6858 | * race with the completion of the linked work. | |
6859 | */ | |
447c19f3 | 6860 | if (prev) { |
f2f87370 | 6861 | io_remove_next_linked(prev); |
447c19f3 PB |
6862 | if (!req_ref_inc_not_zero(prev)) |
6863 | prev = NULL; | |
6864 | } | |
ef9dd637 | 6865 | list_del(&req->timeout.list); |
89b263f6 JA |
6866 | req->timeout.prev = prev; |
6867 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); | |
2665abfd | 6868 | |
89b263f6 JA |
6869 | req->io_task_work.func = io_req_task_link_timeout; |
6870 | io_req_task_work_add(req); | |
2665abfd JA |
6871 | return HRTIMER_NORESTART; |
6872 | } | |
6873 | ||
de968c18 | 6874 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6875 | { |
de968c18 PB |
6876 | struct io_ring_ctx *ctx = req->ctx; |
6877 | ||
89b263f6 | 6878 | spin_lock_irq(&ctx->timeout_lock); |
76a46e06 | 6879 | /* |
f2f87370 PB |
6880 | * If the back reference is NULL, then our linked request finished |
6881 | * before we got a chance to setup the timer | |
76a46e06 | 6882 | */ |
90cd7e42 | 6883 | if (req->timeout.head) { |
e8c2bc1f | 6884 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6885 | |
ad8a48ac JA |
6886 | data->timer.function = io_link_timeout_fn; |
6887 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6888 | data->mode); | |
ef9dd637 | 6889 | list_add_tail(&req->timeout.list, &ctx->ltimeout_list); |
2665abfd | 6890 | } |
89b263f6 | 6891 | spin_unlock_irq(&ctx->timeout_lock); |
2665abfd | 6892 | /* drop submission reference */ |
76a46e06 JA |
6893 | io_put_req(req); |
6894 | } | |
2665abfd | 6895 | |
c5eef2b9 | 6896 | static void __io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 6897 | __must_hold(&req->ctx->uring_lock) |
2b188cc1 | 6898 | { |
906c6caa | 6899 | struct io_kiocb *linked_timeout; |
e0c5c576 | 6900 | int ret; |
2b188cc1 | 6901 | |
59b735ae | 6902 | issue_sqe: |
c5eef2b9 | 6903 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6904 | |
491381ce JA |
6905 | /* |
6906 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6907 | * doesn't support non-blocking read/write attempts | |
6908 | */ | |
1840038e | 6909 | if (likely(!ret)) { |
e342c807 | 6910 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 | 6911 | struct io_ring_ctx *ctx = req->ctx; |
cd0ca2e0 | 6912 | struct io_submit_state *state = &ctx->submit_state; |
e65ef56d | 6913 | |
cd0ca2e0 PB |
6914 | state->compl_reqs[state->compl_nr++] = req; |
6915 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2a2758f2 | 6916 | io_submit_flush_completions(ctx); |
906c6caa | 6917 | return; |
0d63c148 | 6918 | } |
906c6caa PB |
6919 | |
6920 | linked_timeout = io_prep_linked_timeout(req); | |
6921 | if (linked_timeout) | |
6922 | io_queue_linked_timeout(linked_timeout); | |
1840038e | 6923 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
906c6caa PB |
6924 | linked_timeout = io_prep_linked_timeout(req); |
6925 | ||
59b735ae OL |
6926 | switch (io_arm_poll_handler(req)) { |
6927 | case IO_APOLL_READY: | |
906c6caa PB |
6928 | if (linked_timeout) |
6929 | io_unprep_linked_timeout(req); | |
59b735ae OL |
6930 | goto issue_sqe; |
6931 | case IO_APOLL_ABORTED: | |
1840038e PB |
6932 | /* |
6933 | * Queued up for async execution, worker will release | |
6934 | * submit reference when the iocb is actually submitted. | |
6935 | */ | |
f237c30a | 6936 | io_queue_async_work(req, NULL); |
59b735ae | 6937 | break; |
1840038e | 6938 | } |
906c6caa PB |
6939 | |
6940 | if (linked_timeout) | |
6941 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 | 6942 | } else { |
f41db273 | 6943 | io_req_complete_failed(req, ret); |
9e645e11 | 6944 | } |
2b188cc1 JA |
6945 | } |
6946 | ||
441b8a78 | 6947 | static inline void io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 6948 | __must_hold(&req->ctx->uring_lock) |
4fe2c963 | 6949 | { |
10c66904 | 6950 | if (unlikely(req->ctx->drain_active) && io_drain_req(req)) |
76cc33d7 | 6951 | return; |
4fe2c963 | 6952 | |
a8295b98 | 6953 | if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)))) { |
c5eef2b9 | 6954 | __io_queue_sqe(req); |
a8295b98 | 6955 | } else if (req->flags & REQ_F_FAIL) { |
c6d3d9cb | 6956 | io_req_complete_fail_submit(req); |
76cc33d7 PB |
6957 | } else { |
6958 | int ret = io_req_prep_async(req); | |
6959 | ||
6960 | if (unlikely(ret)) | |
6961 | io_req_complete_failed(req, ret); | |
6962 | else | |
f237c30a | 6963 | io_queue_async_work(req, NULL); |
ce35a47a | 6964 | } |
4fe2c963 JL |
6965 | } |
6966 | ||
b16fed66 PB |
6967 | /* |
6968 | * Check SQE restrictions (opcode and flags). | |
6969 | * | |
6970 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6971 | */ | |
6972 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6973 | struct io_kiocb *req, | |
6974 | unsigned int sqe_flags) | |
4fe2c963 | 6975 | { |
4cfb25bf | 6976 | if (likely(!ctx->restricted)) |
b16fed66 PB |
6977 | return true; |
6978 | ||
6979 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6980 | return false; | |
6981 | ||
6982 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6983 | ctx->restrictions.sqe_flags_required) | |
6984 | return false; | |
6985 | ||
6986 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6987 | ctx->restrictions.sqe_flags_required)) | |
6988 | return false; | |
6989 | ||
6990 | return true; | |
4fe2c963 JL |
6991 | } |
6992 | ||
b16fed66 PB |
6993 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6994 | const struct io_uring_sqe *sqe) | |
282cdc86 | 6995 | __must_hold(&ctx->uring_lock) |
b16fed66 PB |
6996 | { |
6997 | struct io_submit_state *state; | |
6998 | unsigned int sqe_flags; | |
003e8dcc | 6999 | int personality, ret = 0; |
b16fed66 | 7000 | |
864ea921 | 7001 | /* req is partially pre-initialised, see io_preinit_req() */ |
b16fed66 PB |
7002 | req->opcode = READ_ONCE(sqe->opcode); |
7003 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
7004 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
7005 | req->user_data = READ_ONCE(sqe->user_data); | |
b16fed66 | 7006 | req->file = NULL; |
b16fed66 | 7007 | req->fixed_rsrc_refs = NULL; |
b16fed66 | 7008 | req->task = current; |
b16fed66 PB |
7009 | |
7010 | /* enforce forwards compatibility on users */ | |
dddca226 | 7011 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 7012 | return -EINVAL; |
b16fed66 PB |
7013 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
7014 | return -EINVAL; | |
4cfb25bf | 7015 | if (!io_check_restriction(ctx, req, sqe_flags)) |
b16fed66 PB |
7016 | return -EACCES; |
7017 | ||
7018 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
7019 | !io_op_defs[req->opcode].buffer_select) | |
7020 | return -EOPNOTSUPP; | |
3c19966d PB |
7021 | if (unlikely(sqe_flags & IOSQE_IO_DRAIN)) |
7022 | ctx->drain_active = true; | |
863e0560 | 7023 | |
003e8dcc JA |
7024 | personality = READ_ONCE(sqe->personality); |
7025 | if (personality) { | |
c10d1f98 PB |
7026 | req->creds = xa_load(&ctx->personalities, personality); |
7027 | if (!req->creds) | |
003e8dcc | 7028 | return -EINVAL; |
c10d1f98 | 7029 | get_cred(req->creds); |
b8e64b53 | 7030 | req->flags |= REQ_F_CREDS; |
003e8dcc | 7031 | } |
b16fed66 PB |
7032 | state = &ctx->submit_state; |
7033 | ||
7034 | /* | |
7035 | * Plug now if we have more than 1 IO left after this, and the target | |
7036 | * is potentially a read/write to block based storage. | |
7037 | */ | |
7038 | if (!state->plug_started && state->ios_left > 1 && | |
7039 | io_op_defs[req->opcode].plug) { | |
7040 | blk_start_plug(&state->plug); | |
7041 | state->plug_started = true; | |
7042 | } | |
7043 | ||
7044 | if (io_op_defs[req->opcode].needs_file) { | |
62906e89 | 7045 | req->file = io_file_get(ctx, req, READ_ONCE(sqe->fd), |
ac177053 | 7046 | (sqe_flags & IOSQE_FIXED_FILE)); |
b16fed66 PB |
7047 | if (unlikely(!req->file)) |
7048 | ret = -EBADF; | |
7049 | } | |
7050 | ||
7051 | state->ios_left--; | |
7052 | return ret; | |
7053 | } | |
7054 | ||
a6b8cadc | 7055 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 7056 | const struct io_uring_sqe *sqe) |
282cdc86 | 7057 | __must_hold(&ctx->uring_lock) |
9e645e11 | 7058 | { |
a1ab7b35 | 7059 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 7060 | int ret; |
9e645e11 | 7061 | |
a6b8cadc PB |
7062 | ret = io_init_req(ctx, req, sqe); |
7063 | if (unlikely(ret)) { | |
7064 | fail_req: | |
a8295b98 | 7065 | /* fail even hard links since we don't submit */ |
de59bc10 | 7066 | if (link->head) { |
a8295b98 HX |
7067 | /* |
7068 | * we can judge a link req is failed or cancelled by if | |
7069 | * REQ_F_FAIL is set, but the head is an exception since | |
7070 | * it may be set REQ_F_FAIL because of other req's failure | |
7071 | * so let's leverage req->result to distinguish if a head | |
7072 | * is set REQ_F_FAIL because of its failure or other req's | |
7073 | * failure so that we can set the correct ret code for it. | |
7074 | * init result here to avoid affecting the normal path. | |
7075 | */ | |
7076 | if (!(link->head->flags & REQ_F_FAIL)) | |
7077 | req_fail_link_node(link->head, -ECANCELED); | |
7078 | } else if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { | |
7079 | /* | |
7080 | * the current req is a normal req, we should return | |
7081 | * error and thus break the submittion loop. | |
7082 | */ | |
7083 | io_req_complete_failed(req, ret); | |
7084 | return ret; | |
de59bc10 | 7085 | } |
a8295b98 HX |
7086 | req_fail_link_node(req, ret); |
7087 | } else { | |
7088 | ret = io_req_prep(req, sqe); | |
7089 | if (unlikely(ret)) | |
7090 | goto fail_req; | |
a6b8cadc | 7091 | } |
441b8a78 | 7092 | |
be7053b7 | 7093 | /* don't need @sqe from now on */ |
236daeae OL |
7094 | trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, |
7095 | req->flags, true, | |
7096 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 7097 | |
9e645e11 JA |
7098 | /* |
7099 | * If we already have a head request, queue this one for async | |
7100 | * submittal once the head completes. If we don't have a head but | |
7101 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
7102 | * submitted sync once the chain is complete. If none of those | |
7103 | * conditions are true (normal request), then just queue it. | |
7104 | */ | |
863e0560 PB |
7105 | if (link->head) { |
7106 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 7107 | |
a8295b98 HX |
7108 | if (!(req->flags & REQ_F_FAIL)) { |
7109 | ret = io_req_prep_async(req); | |
7110 | if (unlikely(ret)) { | |
7111 | req_fail_link_node(req, ret); | |
7112 | if (!(head->flags & REQ_F_FAIL)) | |
7113 | req_fail_link_node(head, -ECANCELED); | |
7114 | } | |
7115 | } | |
9d76377f | 7116 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 7117 | link->last->link = req; |
863e0560 | 7118 | link->last = req; |
32fe525b PB |
7119 | |
7120 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 7121 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 7122 | link->head = NULL; |
5e159204 | 7123 | io_queue_sqe(head); |
32fe525b | 7124 | } |
9e645e11 | 7125 | } else { |
ef4ff581 | 7126 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
7127 | link->head = req; |
7128 | link->last = req; | |
711be031 | 7129 | } else { |
be7053b7 | 7130 | io_queue_sqe(req); |
711be031 | 7131 | } |
9e645e11 | 7132 | } |
2e6e1fde | 7133 | |
1d4240cc | 7134 | return 0; |
9e645e11 JA |
7135 | } |
7136 | ||
9a56a232 JA |
7137 | /* |
7138 | * Batched submission is done, ensure local IO is flushed out. | |
7139 | */ | |
ba88ff11 PB |
7140 | static void io_submit_state_end(struct io_submit_state *state, |
7141 | struct io_ring_ctx *ctx) | |
9a56a232 | 7142 | { |
a1ab7b35 | 7143 | if (state->link.head) |
de59bc10 | 7144 | io_queue_sqe(state->link.head); |
cd0ca2e0 | 7145 | if (state->compl_nr) |
2a2758f2 | 7146 | io_submit_flush_completions(ctx); |
27926b68 JA |
7147 | if (state->plug_started) |
7148 | blk_finish_plug(&state->plug); | |
9a56a232 JA |
7149 | } |
7150 | ||
7151 | /* | |
7152 | * Start submission side cache. | |
7153 | */ | |
7154 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 7155 | unsigned int max_ios) |
9a56a232 | 7156 | { |
27926b68 | 7157 | state->plug_started = false; |
9a56a232 | 7158 | state->ios_left = max_ios; |
a1ab7b35 PB |
7159 | /* set only head, no need to init link_last in advance */ |
7160 | state->link.head = NULL; | |
9a56a232 JA |
7161 | } |
7162 | ||
2b188cc1 JA |
7163 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
7164 | { | |
75b28aff | 7165 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 7166 | |
caf582c6 PB |
7167 | /* |
7168 | * Ensure any loads from the SQEs are done at this point, | |
7169 | * since once we write the new head, the application could | |
7170 | * write new data to them. | |
7171 | */ | |
7172 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
7173 | } |
7174 | ||
2b188cc1 | 7175 | /* |
dd9ae8a0 | 7176 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
7177 | * that is mapped by userspace. This means that care needs to be taken to |
7178 | * ensure that reads are stable, as we cannot rely on userspace always | |
7179 | * being a good citizen. If members of the sqe are validated and then later | |
7180 | * used, it's important that those reads are done through READ_ONCE() to | |
7181 | * prevent a re-load down the line. | |
7182 | */ | |
709b302f | 7183 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 7184 | { |
ea5ab3b5 | 7185 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 7186 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
7187 | |
7188 | /* | |
7189 | * The cached sq head (or cq tail) serves two purposes: | |
7190 | * | |
7191 | * 1) allows us to batch the cost of updating the user visible | |
7192 | * head updates. | |
7193 | * 2) allows the kernel side to track the head on its own, even | |
7194 | * though the application is the one updating it. | |
7195 | */ | |
17d3aeb3 | 7196 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
7197 | if (likely(head < ctx->sq_entries)) |
7198 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
7199 | |
7200 | /* drop invalid entries */ | |
15641e42 PB |
7201 | ctx->cq_extra--; |
7202 | WRITE_ONCE(ctx->rings->sq_dropped, | |
7203 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
7204 | return NULL; |
7205 | } | |
7206 | ||
0f212204 | 7207 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
282cdc86 | 7208 | __must_hold(&ctx->uring_lock) |
6c271ce2 | 7209 | { |
46c4e16a | 7210 | int submitted = 0; |
6c271ce2 | 7211 | |
ee7d46d9 PB |
7212 | /* make sure SQ entry isn't read before tail */ |
7213 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
7214 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
7215 | return -EAGAIN; | |
9a10867a | 7216 | io_get_task_refs(nr); |
6c271ce2 | 7217 | |
ba88ff11 | 7218 | io_submit_state_start(&ctx->submit_state, nr); |
46c4e16a | 7219 | while (submitted < nr) { |
3529d8c2 | 7220 | const struct io_uring_sqe *sqe; |
196be95c | 7221 | struct io_kiocb *req; |
fb5ccc98 | 7222 | |
258b29a9 | 7223 | req = io_alloc_req(ctx); |
196be95c PB |
7224 | if (unlikely(!req)) { |
7225 | if (!submitted) | |
7226 | submitted = -EAGAIN; | |
fb5ccc98 | 7227 | break; |
196be95c | 7228 | } |
4fccfcbb PB |
7229 | sqe = io_get_sqe(ctx); |
7230 | if (unlikely(!sqe)) { | |
0c6e1d7f | 7231 | list_add(&req->inflight_entry, &ctx->submit_state.free_list); |
4fccfcbb PB |
7232 | break; |
7233 | } | |
d3656344 JA |
7234 | /* will complete beyond this point, count as submitted */ |
7235 | submitted++; | |
a1ab7b35 | 7236 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 7237 | break; |
6c271ce2 JA |
7238 | } |
7239 | ||
9466f437 PB |
7240 | if (unlikely(submitted != nr)) { |
7241 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 7242 | int unused = nr - ref_used; |
9466f437 | 7243 | |
09899b19 | 7244 | current->io_uring->cached_refs += unused; |
d8a6df10 | 7245 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 7246 | } |
6c271ce2 | 7247 | |
a1ab7b35 | 7248 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
7249 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
7250 | io_commit_sqring(ctx); | |
7251 | ||
6c271ce2 JA |
7252 | return submitted; |
7253 | } | |
7254 | ||
e4b6d902 PB |
7255 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
7256 | { | |
7257 | return READ_ONCE(sqd->state); | |
7258 | } | |
7259 | ||
23b3628e XW |
7260 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
7261 | { | |
7262 | /* Tell userspace we may need a wakeup call */ | |
79ebeaee | 7263 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7264 | WRITE_ONCE(ctx->rings->sq_flags, |
7265 | ctx->rings->sq_flags | IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7266 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7267 | } |
7268 | ||
7269 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
7270 | { | |
79ebeaee | 7271 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7272 | WRITE_ONCE(ctx->rings->sq_flags, |
7273 | ctx->rings->sq_flags & ~IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7274 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7275 | } |
7276 | ||
08369246 | 7277 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 7278 | { |
c8d1ba58 | 7279 | unsigned int to_submit; |
bdcd3eab | 7280 | int ret = 0; |
6c271ce2 | 7281 | |
c8d1ba58 | 7282 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 7283 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
7284 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
7285 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 7286 | |
906a3c6f | 7287 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 7288 | unsigned nr_events = 0; |
948e1947 PB |
7289 | const struct cred *creds = NULL; |
7290 | ||
7291 | if (ctx->sq_creds != current_cred()) | |
7292 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 7293 | |
c8d1ba58 | 7294 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 7295 | if (!list_empty(&ctx->iopoll_list)) |
a8576af9 | 7296 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 7297 | |
3b763ba1 PB |
7298 | /* |
7299 | * Don't submit if refs are dying, good for io_uring_register(), | |
7300 | * but also it is relied upon by io_ring_exit_work() | |
7301 | */ | |
0298ef96 PB |
7302 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
7303 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 7304 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 7305 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 7306 | |
acfb381d PB |
7307 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
7308 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
7309 | if (creds) |
7310 | revert_creds(creds); | |
acfb381d | 7311 | } |
6c271ce2 | 7312 | |
08369246 XW |
7313 | return ret; |
7314 | } | |
6c271ce2 | 7315 | |
08369246 XW |
7316 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7317 | { | |
7318 | struct io_ring_ctx *ctx; | |
7319 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7320 | |
c9dca27d PB |
7321 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7322 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 7323 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7324 | } |
6c271ce2 | 7325 | |
e4b6d902 PB |
7326 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
7327 | { | |
7328 | bool did_sig = false; | |
7329 | struct ksignal ksig; | |
7330 | ||
7331 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
7332 | signal_pending(current)) { | |
7333 | mutex_unlock(&sqd->lock); | |
7334 | if (signal_pending(current)) | |
7335 | did_sig = get_signal(&ksig); | |
7336 | cond_resched(); | |
7337 | mutex_lock(&sqd->lock); | |
7338 | } | |
e4b6d902 PB |
7339 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
7340 | } | |
7341 | ||
c8d1ba58 JA |
7342 | static int io_sq_thread(void *data) |
7343 | { | |
69fb2131 JA |
7344 | struct io_sq_data *sqd = data; |
7345 | struct io_ring_ctx *ctx; | |
a0d9205f | 7346 | unsigned long timeout = 0; |
37d1e2e3 | 7347 | char buf[TASK_COMM_LEN]; |
08369246 | 7348 | DEFINE_WAIT(wait); |
6c271ce2 | 7349 | |
696ee88a | 7350 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 7351 | set_task_comm(current, buf); |
37d1e2e3 JA |
7352 | |
7353 | if (sqd->sq_cpu != -1) | |
7354 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
7355 | else | |
7356 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
7357 | current->flags |= PF_NO_SETAFFINITY; | |
7358 | ||
09a6f4ef | 7359 | mutex_lock(&sqd->lock); |
e4b6d902 | 7360 | while (1) { |
1a924a80 | 7361 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 7362 | |
e4b6d902 PB |
7363 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
7364 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 7365 | break; |
08369246 XW |
7366 | timeout = jiffies + sqd->sq_thread_idle; |
7367 | } | |
e4b6d902 | 7368 | |
e95eee2d | 7369 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 7370 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 7371 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 7372 | |
08369246 XW |
7373 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
7374 | sqt_spin = true; | |
69fb2131 | 7375 | } |
dd432ea5 PB |
7376 | if (io_run_task_work()) |
7377 | sqt_spin = true; | |
6c271ce2 | 7378 | |
08369246 | 7379 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7380 | cond_resched(); |
08369246 XW |
7381 | if (sqt_spin) |
7382 | timeout = jiffies + sqd->sq_thread_idle; | |
7383 | continue; | |
7384 | } | |
7385 | ||
08369246 | 7386 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
dd432ea5 | 7387 | if (!io_sqd_events_pending(sqd) && !current->task_works) { |
1a924a80 PB |
7388 | bool needs_sched = true; |
7389 | ||
724cb4f9 | 7390 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
7391 | io_ring_set_wakeup_flag(ctx); |
7392 | ||
724cb4f9 HX |
7393 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
7394 | !list_empty_careful(&ctx->iopoll_list)) { | |
7395 | needs_sched = false; | |
7396 | break; | |
7397 | } | |
7398 | if (io_sqring_entries(ctx)) { | |
7399 | needs_sched = false; | |
7400 | break; | |
7401 | } | |
7402 | } | |
7403 | ||
7404 | if (needs_sched) { | |
7405 | mutex_unlock(&sqd->lock); | |
7406 | schedule(); | |
7407 | mutex_lock(&sqd->lock); | |
7408 | } | |
69fb2131 JA |
7409 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7410 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7411 | } |
08369246 XW |
7412 | |
7413 | finish_wait(&sqd->wait, &wait); | |
7414 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 7415 | } |
28cea78a | 7416 | |
78cc687b | 7417 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 7418 | sqd->thread = NULL; |
05962f95 | 7419 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 7420 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 7421 | io_run_task_work(); |
734551df PB |
7422 | mutex_unlock(&sqd->lock); |
7423 | ||
37d1e2e3 JA |
7424 | complete(&sqd->exited); |
7425 | do_exit(0); | |
6c271ce2 JA |
7426 | } |
7427 | ||
bda52162 JA |
7428 | struct io_wait_queue { |
7429 | struct wait_queue_entry wq; | |
7430 | struct io_ring_ctx *ctx; | |
5fd46178 | 7431 | unsigned cq_tail; |
bda52162 JA |
7432 | unsigned nr_timeouts; |
7433 | }; | |
7434 | ||
6c503150 | 7435 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7436 | { |
7437 | struct io_ring_ctx *ctx = iowq->ctx; | |
5fd46178 | 7438 | int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; |
bda52162 JA |
7439 | |
7440 | /* | |
d195a66e | 7441 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7442 | * started waiting. For timeouts, we always want to return to userspace, |
7443 | * regardless of event count. | |
7444 | */ | |
5fd46178 | 7445 | return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
bda52162 JA |
7446 | } |
7447 | ||
7448 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7449 | int wake_flags, void *key) | |
7450 | { | |
7451 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7452 | wq); | |
7453 | ||
6c503150 PB |
7454 | /* |
7455 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7456 | * the task, and the next invocation will do it. | |
7457 | */ | |
5ed7a37d | 7458 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
7459 | return autoremove_wake_function(curr, mode, wake_flags, key); |
7460 | return -1; | |
bda52162 JA |
7461 | } |
7462 | ||
af9c1a44 JA |
7463 | static int io_run_task_work_sig(void) |
7464 | { | |
7465 | if (io_run_task_work()) | |
7466 | return 1; | |
7467 | if (!signal_pending(current)) | |
7468 | return 0; | |
0b8cfa97 | 7469 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 7470 | return -ERESTARTSYS; |
af9c1a44 JA |
7471 | return -EINTR; |
7472 | } | |
7473 | ||
eeb60b9a PB |
7474 | /* when returns >0, the caller should retry */ |
7475 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7476 | struct io_wait_queue *iowq, | |
7477 | signed long *timeout) | |
7478 | { | |
7479 | int ret; | |
7480 | ||
7481 | /* make sure we run task_work before checking for signals */ | |
7482 | ret = io_run_task_work_sig(); | |
7483 | if (ret || io_should_wake(iowq)) | |
7484 | return ret; | |
7485 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 7486 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
7487 | return 1; |
7488 | ||
7489 | *timeout = schedule_timeout(*timeout); | |
7490 | return !*timeout ? -ETIME : 1; | |
7491 | } | |
7492 | ||
2b188cc1 JA |
7493 | /* |
7494 | * Wait until events become available, if we don't already have some. The | |
7495 | * application must reap them itself, as they reside on the shared cq ring. | |
7496 | */ | |
7497 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7498 | const sigset_t __user *sig, size_t sigsz, |
7499 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7500 | { |
90291099 | 7501 | struct io_wait_queue iowq; |
75b28aff | 7502 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7503 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7504 | int ret; | |
2b188cc1 | 7505 | |
b41e9852 | 7506 | do { |
90f67366 | 7507 | io_cqring_overflow_flush(ctx); |
6c503150 | 7508 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7509 | return 0; |
4c6e277c | 7510 | if (!io_run_task_work()) |
b41e9852 | 7511 | break; |
b41e9852 | 7512 | } while (1); |
2b188cc1 JA |
7513 | |
7514 | if (sig) { | |
9e75ad5d AB |
7515 | #ifdef CONFIG_COMPAT |
7516 | if (in_compat_syscall()) | |
7517 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7518 | sigsz); |
9e75ad5d AB |
7519 | else |
7520 | #endif | |
b772434b | 7521 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7522 | |
2b188cc1 JA |
7523 | if (ret) |
7524 | return ret; | |
7525 | } | |
7526 | ||
c73ebb68 | 7527 | if (uts) { |
c1d5a224 PB |
7528 | struct timespec64 ts; |
7529 | ||
c73ebb68 HX |
7530 | if (get_timespec64(&ts, uts)) |
7531 | return -EFAULT; | |
7532 | timeout = timespec64_to_jiffies(&ts); | |
7533 | } | |
7534 | ||
90291099 PB |
7535 | init_waitqueue_func_entry(&iowq.wq, io_wake_function); |
7536 | iowq.wq.private = current; | |
7537 | INIT_LIST_HEAD(&iowq.wq.entry); | |
7538 | iowq.ctx = ctx; | |
bda52162 | 7539 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
5fd46178 | 7540 | iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; |
90291099 | 7541 | |
c826bd7a | 7542 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7543 | do { |
ca0a2651 | 7544 | /* if we can't even flush overflow, don't wait for more */ |
90f67366 | 7545 | if (!io_cqring_overflow_flush(ctx)) { |
ca0a2651 JA |
7546 | ret = -EBUSY; |
7547 | break; | |
7548 | } | |
311997b3 | 7549 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7550 | TASK_INTERRUPTIBLE); |
eeb60b9a | 7551 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
311997b3 | 7552 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7553 | cond_resched(); |
eeb60b9a | 7554 | } while (ret > 0); |
bda52162 | 7555 | |
b7db41c9 | 7556 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7557 | |
75b28aff | 7558 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7559 | } |
7560 | ||
9123c8ff | 7561 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7562 | { |
9123c8ff | 7563 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7564 | |
846a4ef2 | 7565 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7566 | kfree(table[i]); |
7567 | kfree(table); | |
7568 | } | |
7569 | ||
7570 | static void **io_alloc_page_table(size_t size) | |
7571 | { | |
7572 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7573 | size_t init_size = size; | |
7574 | void **table; | |
7575 | ||
0bea96f5 | 7576 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7577 | if (!table) |
7578 | return NULL; | |
7579 | ||
7580 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 7581 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff | 7582 | |
0bea96f5 | 7583 | table[i] = kzalloc(this_size, GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7584 | if (!table[i]) { |
7585 | io_free_page_table(table, init_size); | |
7586 | return NULL; | |
7587 | } | |
7588 | size -= this_size; | |
7589 | } | |
7590 | return table; | |
05f3fb3c JA |
7591 | } |
7592 | ||
28a9fe25 | 7593 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7594 | { |
28a9fe25 PB |
7595 | percpu_ref_exit(&ref_node->refs); |
7596 | kfree(ref_node); | |
1642b445 PB |
7597 | } |
7598 | ||
b9bd2bea PB |
7599 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
7600 | { | |
7601 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); | |
7602 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; | |
7603 | unsigned long flags; | |
7604 | bool first_add = false; | |
7605 | ||
7606 | spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); | |
7607 | node->done = true; | |
7608 | ||
7609 | while (!list_empty(&ctx->rsrc_ref_list)) { | |
7610 | node = list_first_entry(&ctx->rsrc_ref_list, | |
7611 | struct io_rsrc_node, node); | |
7612 | /* recycle ref nodes in order */ | |
7613 | if (!node->done) | |
7614 | break; | |
7615 | list_del(&node->node); | |
7616 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
7617 | } | |
7618 | spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); | |
7619 | ||
7620 | if (first_add) | |
7621 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
7622 | } | |
7623 | ||
7624 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) | |
7625 | { | |
7626 | struct io_rsrc_node *ref_node; | |
7627 | ||
7628 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); | |
7629 | if (!ref_node) | |
7630 | return NULL; | |
7631 | ||
7632 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, | |
7633 | 0, GFP_KERNEL)) { | |
7634 | kfree(ref_node); | |
7635 | return NULL; | |
7636 | } | |
7637 | INIT_LIST_HEAD(&ref_node->node); | |
7638 | INIT_LIST_HEAD(&ref_node->rsrc_list); | |
7639 | ref_node->done = false; | |
7640 | return ref_node; | |
7641 | } | |
7642 | ||
a7f0ed5a PB |
7643 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7644 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7645 | { |
a7f0ed5a PB |
7646 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7647 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7648 | |
a7f0ed5a PB |
7649 | if (data_to_kill) { |
7650 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7651 | |
a7f0ed5a | 7652 | rsrc_node->rsrc_data = data_to_kill; |
4956b9ea | 7653 | spin_lock_irq(&ctx->rsrc_ref_lock); |
a7f0ed5a | 7654 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); |
4956b9ea | 7655 | spin_unlock_irq(&ctx->rsrc_ref_lock); |
82fbcfa9 | 7656 | |
3e942498 | 7657 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7658 | percpu_ref_kill(&rsrc_node->refs); |
7659 | ctx->rsrc_node = NULL; | |
7660 | } | |
6b06314c | 7661 | |
a7f0ed5a PB |
7662 | if (!ctx->rsrc_node) { |
7663 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7664 | ctx->rsrc_backup_node = NULL; | |
7665 | } | |
8bad28d8 HX |
7666 | } |
7667 | ||
a7f0ed5a | 7668 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7669 | { |
7670 | if (ctx->rsrc_backup_node) | |
7671 | return 0; | |
b895c9a6 | 7672 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7673 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7674 | } |
7675 | ||
40ae0ff7 | 7676 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7677 | { |
7678 | int ret; | |
05589553 | 7679 | |
215c3902 | 7680 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7681 | if (data->quiesce) |
7682 | return -ENXIO; | |
05589553 | 7683 | |
8bad28d8 | 7684 | data->quiesce = true; |
1ffc5422 | 7685 | do { |
a7f0ed5a | 7686 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7687 | if (ret) |
f2303b1f | 7688 | break; |
a7f0ed5a | 7689 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7690 | |
3e942498 PB |
7691 | /* kill initial ref, already quiesced if zero */ |
7692 | if (atomic_dec_and_test(&data->refs)) | |
7693 | break; | |
c018db4a | 7694 | mutex_unlock(&ctx->uring_lock); |
8bad28d8 | 7695 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 | 7696 | ret = wait_for_completion_interruptible(&data->done); |
c018db4a JA |
7697 | if (!ret) { |
7698 | mutex_lock(&ctx->uring_lock); | |
1ffc5422 | 7699 | break; |
c018db4a | 7700 | } |
8bad28d8 | 7701 | |
3e942498 PB |
7702 | atomic_inc(&data->refs); |
7703 | /* wait for all works potentially completing data->done */ | |
7704 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7705 | reinit_completion(&data->done); |
8dd03afe | 7706 | |
1ffc5422 | 7707 | ret = io_run_task_work_sig(); |
8bad28d8 | 7708 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7709 | } while (ret >= 0); |
8bad28d8 | 7710 | data->quiesce = false; |
05f3fb3c | 7711 | |
8bad28d8 | 7712 | return ret; |
d7954b2b BM |
7713 | } |
7714 | ||
2d091d62 PB |
7715 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7716 | { | |
7717 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7718 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7719 | ||
7720 | return &data->tags[table_idx][off]; | |
7721 | } | |
7722 | ||
44b31f2f | 7723 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7724 | { |
2d091d62 PB |
7725 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7726 | ||
7727 | if (data->tags) | |
7728 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7729 | kfree(data); |
7730 | } | |
7731 | ||
d878c816 PB |
7732 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7733 | u64 __user *utags, unsigned nr, | |
7734 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7735 | { |
b895c9a6 | 7736 | struct io_rsrc_data *data; |
2d091d62 | 7737 | int ret = -ENOMEM; |
d878c816 | 7738 | unsigned i; |
1ad555c6 BM |
7739 | |
7740 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7741 | if (!data) | |
d878c816 | 7742 | return -ENOMEM; |
2d091d62 | 7743 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7744 | if (!data->tags) { |
1ad555c6 | 7745 | kfree(data); |
d878c816 PB |
7746 | return -ENOMEM; |
7747 | } | |
2d091d62 PB |
7748 | |
7749 | data->nr = nr; | |
7750 | data->ctx = ctx; | |
7751 | data->do_put = do_put; | |
d878c816 | 7752 | if (utags) { |
2d091d62 | 7753 | ret = -EFAULT; |
d878c816 | 7754 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7755 | u64 *tag_slot = io_get_tag_slot(data, i); |
7756 | ||
7757 | if (copy_from_user(tag_slot, &utags[i], | |
7758 | sizeof(*tag_slot))) | |
2d091d62 | 7759 | goto fail; |
d878c816 | 7760 | } |
1ad555c6 | 7761 | } |
b60c8dce | 7762 | |
3e942498 | 7763 | atomic_set(&data->refs, 1); |
1ad555c6 | 7764 | init_completion(&data->done); |
d878c816 PB |
7765 | *pdata = data; |
7766 | return 0; | |
2d091d62 PB |
7767 | fail: |
7768 | io_rsrc_data_free(data); | |
7769 | return ret; | |
1ad555c6 BM |
7770 | } |
7771 | ||
9123c8ff PB |
7772 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7773 | { | |
0bea96f5 PB |
7774 | table->files = kvcalloc(nr_files, sizeof(table->files[0]), |
7775 | GFP_KERNEL_ACCOUNT); | |
9123c8ff PB |
7776 | return !!table->files; |
7777 | } | |
7778 | ||
042b0d85 | 7779 | static void io_free_file_tables(struct io_file_table *table) |
9123c8ff | 7780 | { |
042b0d85 | 7781 | kvfree(table->files); |
9123c8ff PB |
7782 | table->files = NULL; |
7783 | } | |
7784 | ||
fff4db76 | 7785 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7786 | { |
fff4db76 PB |
7787 | #if defined(CONFIG_UNIX) |
7788 | if (ctx->ring_sock) { | |
7789 | struct sock *sock = ctx->ring_sock->sk; | |
7790 | struct sk_buff *skb; | |
7791 | ||
7792 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7793 | kfree_skb(skb); | |
7794 | } | |
7795 | #else | |
7796 | int i; | |
7797 | ||
7798 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7799 | struct file *file; | |
7800 | ||
7801 | file = io_file_from_index(ctx, i); | |
7802 | if (file) | |
7803 | fput(file); | |
7804 | } | |
7805 | #endif | |
042b0d85 | 7806 | io_free_file_tables(&ctx->file_table); |
44b31f2f | 7807 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7808 | ctx->file_data = NULL; |
7809 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7810 | } |
7811 | ||
d7954b2b BM |
7812 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7813 | { | |
d7954b2b BM |
7814 | int ret; |
7815 | ||
08480400 | 7816 | if (!ctx->file_data) |
d7954b2b | 7817 | return -ENXIO; |
08480400 PB |
7818 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7819 | if (!ret) | |
7820 | __io_sqe_files_unregister(ctx); | |
7821 | return ret; | |
6b06314c JA |
7822 | } |
7823 | ||
37d1e2e3 | 7824 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7825 | __releases(&sqd->lock) |
37d1e2e3 | 7826 | { |
521d6a73 PB |
7827 | WARN_ON_ONCE(sqd->thread == current); |
7828 | ||
9e138a48 PB |
7829 | /* |
7830 | * Do the dance but not conditional clear_bit() because it'd race with | |
7831 | * other threads incrementing park_pending and setting the bit. | |
7832 | */ | |
37d1e2e3 | 7833 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7834 | if (atomic_dec_return(&sqd->park_pending)) |
7835 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7836 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7837 | } |
7838 | ||
86e0d676 | 7839 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7840 | __acquires(&sqd->lock) |
37d1e2e3 | 7841 | { |
521d6a73 PB |
7842 | WARN_ON_ONCE(sqd->thread == current); |
7843 | ||
9e138a48 | 7844 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7845 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7846 | mutex_lock(&sqd->lock); |
05962f95 | 7847 | if (sqd->thread) |
86e0d676 | 7848 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7849 | } |
7850 | ||
7851 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7852 | { | |
521d6a73 | 7853 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7854 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7855 | |
05962f95 | 7856 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7857 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7858 | if (sqd->thread) |
7859 | wake_up_process(sqd->thread); | |
09a6f4ef | 7860 | mutex_unlock(&sqd->lock); |
05962f95 | 7861 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7862 | } |
7863 | ||
534ca6d6 | 7864 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7865 | { |
534ca6d6 | 7866 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7867 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7868 | ||
37d1e2e3 JA |
7869 | io_sq_thread_stop(sqd); |
7870 | kfree(sqd); | |
7871 | } | |
7872 | } | |
7873 | ||
7874 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7875 | { | |
7876 | struct io_sq_data *sqd = ctx->sq_data; | |
7877 | ||
7878 | if (sqd) { | |
05962f95 | 7879 | io_sq_thread_park(sqd); |
521d6a73 | 7880 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7881 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7882 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7883 | |
7884 | io_put_sq_data(sqd); | |
7885 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7886 | } |
7887 | } | |
7888 | ||
aa06165d JA |
7889 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7890 | { | |
7891 | struct io_ring_ctx *ctx_attach; | |
7892 | struct io_sq_data *sqd; | |
7893 | struct fd f; | |
7894 | ||
7895 | f = fdget(p->wq_fd); | |
7896 | if (!f.file) | |
7897 | return ERR_PTR(-ENXIO); | |
7898 | if (f.file->f_op != &io_uring_fops) { | |
7899 | fdput(f); | |
7900 | return ERR_PTR(-EINVAL); | |
7901 | } | |
7902 | ||
7903 | ctx_attach = f.file->private_data; | |
7904 | sqd = ctx_attach->sq_data; | |
7905 | if (!sqd) { | |
7906 | fdput(f); | |
7907 | return ERR_PTR(-EINVAL); | |
7908 | } | |
5c2469e0 JA |
7909 | if (sqd->task_tgid != current->tgid) { |
7910 | fdput(f); | |
7911 | return ERR_PTR(-EPERM); | |
7912 | } | |
aa06165d JA |
7913 | |
7914 | refcount_inc(&sqd->refs); | |
7915 | fdput(f); | |
7916 | return sqd; | |
7917 | } | |
7918 | ||
26984fbf PB |
7919 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7920 | bool *attached) | |
534ca6d6 JA |
7921 | { |
7922 | struct io_sq_data *sqd; | |
7923 | ||
26984fbf | 7924 | *attached = false; |
5c2469e0 JA |
7925 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7926 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7927 | if (!IS_ERR(sqd)) { |
7928 | *attached = true; | |
5c2469e0 | 7929 | return sqd; |
26984fbf | 7930 | } |
5c2469e0 JA |
7931 | /* fall through for EPERM case, setup new sqd/task */ |
7932 | if (PTR_ERR(sqd) != -EPERM) | |
7933 | return sqd; | |
7934 | } | |
aa06165d | 7935 | |
534ca6d6 JA |
7936 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7937 | if (!sqd) | |
7938 | return ERR_PTR(-ENOMEM); | |
7939 | ||
9e138a48 | 7940 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7941 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7942 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7943 | mutex_init(&sqd->lock); |
534ca6d6 | 7944 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7945 | init_completion(&sqd->exited); |
534ca6d6 JA |
7946 | return sqd; |
7947 | } | |
7948 | ||
6b06314c | 7949 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7950 | /* |
7951 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7952 | * the io_uring can be safely unregistered on process exit, even if we have | |
7953 | * loops in the file referencing. | |
7954 | */ | |
7955 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7956 | { | |
7957 | struct sock *sk = ctx->ring_sock->sk; | |
7958 | struct scm_fp_list *fpl; | |
7959 | struct sk_buff *skb; | |
08a45173 | 7960 | int i, nr_files; |
6b06314c | 7961 | |
6b06314c JA |
7962 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7963 | if (!fpl) | |
7964 | return -ENOMEM; | |
7965 | ||
7966 | skb = alloc_skb(0, GFP_KERNEL); | |
7967 | if (!skb) { | |
7968 | kfree(fpl); | |
7969 | return -ENOMEM; | |
7970 | } | |
7971 | ||
7972 | skb->sk = sk; | |
6b06314c | 7973 | |
08a45173 | 7974 | nr_files = 0; |
62e398be | 7975 | fpl->user = get_uid(current_user()); |
6b06314c | 7976 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7977 | struct file *file = io_file_from_index(ctx, i + offset); |
7978 | ||
7979 | if (!file) | |
08a45173 | 7980 | continue; |
65e19f54 | 7981 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7982 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7983 | nr_files++; | |
6b06314c JA |
7984 | } |
7985 | ||
08a45173 JA |
7986 | if (nr_files) { |
7987 | fpl->max = SCM_MAX_FD; | |
7988 | fpl->count = nr_files; | |
7989 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7990 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7991 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7992 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7993 | |
08a45173 JA |
7994 | for (i = 0; i < nr_files; i++) |
7995 | fput(fpl->fp[i]); | |
7996 | } else { | |
7997 | kfree_skb(skb); | |
7998 | kfree(fpl); | |
7999 | } | |
6b06314c JA |
8000 | |
8001 | return 0; | |
8002 | } | |
8003 | ||
8004 | /* | |
8005 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
8006 | * causes regular reference counting to break down. We rely on the UNIX | |
8007 | * garbage collection to take care of this problem for us. | |
8008 | */ | |
8009 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8010 | { | |
8011 | unsigned left, total; | |
8012 | int ret = 0; | |
8013 | ||
8014 | total = 0; | |
8015 | left = ctx->nr_user_files; | |
8016 | while (left) { | |
8017 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
8018 | |
8019 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
8020 | if (ret) | |
8021 | break; | |
8022 | left -= this_files; | |
8023 | total += this_files; | |
8024 | } | |
8025 | ||
8026 | if (!ret) | |
8027 | return 0; | |
8028 | ||
8029 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
8030 | struct file *file = io_file_from_index(ctx, total); |
8031 | ||
8032 | if (file) | |
8033 | fput(file); | |
6b06314c JA |
8034 | total++; |
8035 | } | |
8036 | ||
8037 | return ret; | |
8038 | } | |
8039 | #else | |
8040 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8041 | { | |
8042 | return 0; | |
8043 | } | |
8044 | #endif | |
8045 | ||
47e90392 | 8046 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 8047 | { |
50238531 | 8048 | struct file *file = prsrc->file; |
05f3fb3c JA |
8049 | #if defined(CONFIG_UNIX) |
8050 | struct sock *sock = ctx->ring_sock->sk; | |
8051 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
8052 | struct sk_buff *skb; | |
8053 | int i; | |
8054 | ||
8055 | __skb_queue_head_init(&list); | |
8056 | ||
8057 | /* | |
8058 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
8059 | * remove this entry and rearrange the file array. | |
8060 | */ | |
8061 | skb = skb_dequeue(head); | |
8062 | while (skb) { | |
8063 | struct scm_fp_list *fp; | |
8064 | ||
8065 | fp = UNIXCB(skb).fp; | |
8066 | for (i = 0; i < fp->count; i++) { | |
8067 | int left; | |
8068 | ||
8069 | if (fp->fp[i] != file) | |
8070 | continue; | |
8071 | ||
8072 | unix_notinflight(fp->user, fp->fp[i]); | |
8073 | left = fp->count - 1 - i; | |
8074 | if (left) { | |
8075 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
8076 | left * sizeof(struct file *)); | |
8077 | } | |
8078 | fp->count--; | |
8079 | if (!fp->count) { | |
8080 | kfree_skb(skb); | |
8081 | skb = NULL; | |
8082 | } else { | |
8083 | __skb_queue_tail(&list, skb); | |
8084 | } | |
8085 | fput(file); | |
8086 | file = NULL; | |
8087 | break; | |
8088 | } | |
8089 | ||
8090 | if (!file) | |
8091 | break; | |
8092 | ||
8093 | __skb_queue_tail(&list, skb); | |
8094 | ||
8095 | skb = skb_dequeue(head); | |
8096 | } | |
8097 | ||
8098 | if (skb_peek(&list)) { | |
8099 | spin_lock_irq(&head->lock); | |
8100 | while ((skb = __skb_dequeue(&list)) != NULL) | |
8101 | __skb_queue_tail(head, skb); | |
8102 | spin_unlock_irq(&head->lock); | |
8103 | } | |
8104 | #else | |
8105 | fput(file); | |
8106 | #endif | |
8107 | } | |
8108 | ||
b895c9a6 | 8109 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 8110 | { |
b895c9a6 | 8111 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
8112 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
8113 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 8114 | |
269bbe5f BM |
8115 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
8116 | list_del(&prsrc->list); | |
b60c8dce PB |
8117 | |
8118 | if (prsrc->tag) { | |
8119 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
8120 | |
8121 | io_ring_submit_lock(ctx, lock_ring); | |
79ebeaee | 8122 | spin_lock(&ctx->completion_lock); |
b60c8dce | 8123 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 8124 | ctx->cq_extra++; |
b60c8dce | 8125 | io_commit_cqring(ctx); |
79ebeaee | 8126 | spin_unlock(&ctx->completion_lock); |
b60c8dce PB |
8127 | io_cqring_ev_posted(ctx); |
8128 | io_ring_submit_unlock(ctx, lock_ring); | |
8129 | } | |
8130 | ||
40ae0ff7 | 8131 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 8132 | kfree(prsrc); |
65e19f54 | 8133 | } |
05589553 | 8134 | |
28a9fe25 | 8135 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
8136 | if (atomic_dec_and_test(&rsrc_data->refs)) |
8137 | complete(&rsrc_data->done); | |
2faf852d | 8138 | } |
65e19f54 | 8139 | |
269bbe5f | 8140 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
8141 | { |
8142 | struct io_ring_ctx *ctx; | |
8143 | struct llist_node *node; | |
8144 | ||
269bbe5f BM |
8145 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
8146 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
8147 | |
8148 | while (node) { | |
b895c9a6 | 8149 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
8150 | struct llist_node *next = node->next; |
8151 | ||
b895c9a6 | 8152 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 8153 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
8154 | node = next; |
8155 | } | |
8156 | } | |
8157 | ||
6b06314c | 8158 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 8159 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
8160 | { |
8161 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 8162 | struct file *file; |
f3baed39 | 8163 | int fd, ret; |
846a4ef2 | 8164 | unsigned i; |
6b06314c | 8165 | |
05f3fb3c | 8166 | if (ctx->file_data) |
6b06314c JA |
8167 | return -EBUSY; |
8168 | if (!nr_args) | |
8169 | return -EINVAL; | |
8170 | if (nr_args > IORING_MAX_FIXED_FILES) | |
8171 | return -EMFILE; | |
3a1b8a4e PB |
8172 | if (nr_args > rlimit(RLIMIT_NOFILE)) |
8173 | return -EMFILE; | |
a7f0ed5a | 8174 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
8175 | if (ret) |
8176 | return ret; | |
d878c816 PB |
8177 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
8178 | &ctx->file_data); | |
8179 | if (ret) | |
8180 | return ret; | |
6b06314c | 8181 | |
f3baed39 | 8182 | ret = -ENOMEM; |
aeca241b | 8183 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 8184 | goto out_free; |
65e19f54 | 8185 | |
08a45173 | 8186 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 8187 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
8188 | ret = -EFAULT; |
8189 | goto out_fput; | |
8190 | } | |
08a45173 | 8191 | /* allow sparse sets */ |
792e3582 PB |
8192 | if (fd == -1) { |
8193 | ret = -EINVAL; | |
2d091d62 | 8194 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 8195 | goto out_fput; |
08a45173 | 8196 | continue; |
792e3582 | 8197 | } |
6b06314c | 8198 | |
05f3fb3c | 8199 | file = fget(fd); |
6b06314c | 8200 | ret = -EBADF; |
792e3582 | 8201 | if (unlikely(!file)) |
600cf3f8 | 8202 | goto out_fput; |
05f3fb3c | 8203 | |
6b06314c JA |
8204 | /* |
8205 | * Don't allow io_uring instances to be registered. If UNIX | |
8206 | * isn't enabled, then this causes a reference cycle and this | |
8207 | * instance can never get freed. If UNIX is enabled we'll | |
8208 | * handle it just fine, but there's still no point in allowing | |
8209 | * a ring fd as it doesn't support regular read/write anyway. | |
8210 | */ | |
05f3fb3c JA |
8211 | if (file->f_op == &io_uring_fops) { |
8212 | fput(file); | |
600cf3f8 | 8213 | goto out_fput; |
6b06314c | 8214 | } |
aeca241b | 8215 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
8216 | } |
8217 | ||
6b06314c | 8218 | ret = io_sqe_files_scm(ctx); |
05589553 | 8219 | if (ret) { |
08480400 | 8220 | __io_sqe_files_unregister(ctx); |
05589553 XW |
8221 | return ret; |
8222 | } | |
6b06314c | 8223 | |
a7f0ed5a | 8224 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 8225 | return ret; |
600cf3f8 PB |
8226 | out_fput: |
8227 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8228 | file = io_file_from_index(ctx, i); | |
8229 | if (file) | |
8230 | fput(file); | |
8231 | } | |
042b0d85 | 8232 | io_free_file_tables(&ctx->file_table); |
600cf3f8 | 8233 | ctx->nr_user_files = 0; |
600cf3f8 | 8234 | out_free: |
44b31f2f | 8235 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 8236 | ctx->file_data = NULL; |
6b06314c JA |
8237 | return ret; |
8238 | } | |
8239 | ||
c3a31e60 JA |
8240 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
8241 | int index) | |
8242 | { | |
8243 | #if defined(CONFIG_UNIX) | |
8244 | struct sock *sock = ctx->ring_sock->sk; | |
8245 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
8246 | struct sk_buff *skb; | |
8247 | ||
8248 | /* | |
8249 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
8250 | * file set. If there's no room, fall back to allocating a new skb | |
8251 | * and filling it in. | |
8252 | */ | |
8253 | spin_lock_irq(&head->lock); | |
8254 | skb = skb_peek(head); | |
8255 | if (skb) { | |
8256 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
8257 | ||
8258 | if (fpl->count < SCM_MAX_FD) { | |
8259 | __skb_unlink(skb, head); | |
8260 | spin_unlock_irq(&head->lock); | |
8261 | fpl->fp[fpl->count] = get_file(file); | |
8262 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
8263 | fpl->count++; | |
8264 | spin_lock_irq(&head->lock); | |
8265 | __skb_queue_head(head, skb); | |
8266 | } else { | |
8267 | skb = NULL; | |
8268 | } | |
8269 | } | |
8270 | spin_unlock_irq(&head->lock); | |
8271 | ||
8272 | if (skb) { | |
8273 | fput(file); | |
8274 | return 0; | |
8275 | } | |
8276 | ||
8277 | return __io_sqe_files_scm(ctx, 1, index); | |
8278 | #else | |
8279 | return 0; | |
8280 | #endif | |
8281 | } | |
8282 | ||
b9445598 PB |
8283 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
8284 | unsigned int issue_flags, u32 slot_index) | |
8285 | { | |
8286 | struct io_ring_ctx *ctx = req->ctx; | |
8287 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; | |
8288 | struct io_fixed_file *file_slot; | |
8289 | int ret = -EBADF; | |
8290 | ||
8291 | io_ring_submit_lock(ctx, !force_nonblock); | |
8292 | if (file->f_op == &io_uring_fops) | |
8293 | goto err; | |
8294 | ret = -ENXIO; | |
8295 | if (!ctx->file_data) | |
8296 | goto err; | |
8297 | ret = -EINVAL; | |
8298 | if (slot_index >= ctx->nr_user_files) | |
8299 | goto err; | |
8300 | ||
8301 | slot_index = array_index_nospec(slot_index, ctx->nr_user_files); | |
8302 | file_slot = io_fixed_file_slot(&ctx->file_table, slot_index); | |
8303 | ret = -EBADF; | |
8304 | if (file_slot->file_ptr) | |
8305 | goto err; | |
8306 | ||
8307 | *io_get_tag_slot(ctx->file_data, slot_index) = 0; | |
8308 | io_fixed_file_set(file_slot, file); | |
8309 | ret = io_sqe_file_register(ctx, file, slot_index); | |
8310 | if (ret) { | |
8311 | file_slot->file_ptr = 0; | |
8312 | goto err; | |
8313 | } | |
8314 | ||
8315 | ret = 0; | |
8316 | err: | |
8317 | io_ring_submit_unlock(ctx, !force_nonblock); | |
8318 | if (ret) | |
8319 | fput(file); | |
8320 | return ret; | |
8321 | } | |
8322 | ||
b60c8dce | 8323 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 8324 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 8325 | { |
269bbe5f | 8326 | struct io_rsrc_put *prsrc; |
05f3fb3c | 8327 | |
269bbe5f BM |
8328 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
8329 | if (!prsrc) | |
a5318d3c | 8330 | return -ENOMEM; |
05f3fb3c | 8331 | |
2d091d62 | 8332 | prsrc->tag = *io_get_tag_slot(data, idx); |
50238531 | 8333 | prsrc->rsrc = rsrc; |
e7c78371 | 8334 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 8335 | return 0; |
05f3fb3c JA |
8336 | } |
8337 | ||
8338 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 8339 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
8340 | unsigned nr_args) |
8341 | { | |
c3bdad02 | 8342 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 8343 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 8344 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
8345 | struct io_fixed_file *file_slot; |
8346 | struct file *file; | |
98f0b3b4 PB |
8347 | int fd, i, err = 0; |
8348 | unsigned int done; | |
05589553 | 8349 | bool needs_switch = false; |
c3a31e60 | 8350 | |
98f0b3b4 PB |
8351 | if (!ctx->file_data) |
8352 | return -ENXIO; | |
8353 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
8354 | return -EINVAL; |
8355 | ||
67973b93 | 8356 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
8357 | u64 tag = 0; |
8358 | ||
8359 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
8360 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
8361 | err = -EFAULT; |
8362 | break; | |
8363 | } | |
c3bdad02 PB |
8364 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
8365 | err = -EINVAL; | |
8366 | break; | |
8367 | } | |
4e0377a1 | 8368 | if (fd == IORING_REGISTER_FILES_SKIP) |
8369 | continue; | |
8370 | ||
67973b93 | 8371 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 8372 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 8373 | |
a04b0ac0 PB |
8374 | if (file_slot->file_ptr) { |
8375 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
8376 | err = io_queue_rsrc_removal(data, up->offset + done, |
8377 | ctx->rsrc_node, file); | |
a5318d3c HD |
8378 | if (err) |
8379 | break; | |
a04b0ac0 | 8380 | file_slot->file_ptr = 0; |
05589553 | 8381 | needs_switch = true; |
c3a31e60 JA |
8382 | } |
8383 | if (fd != -1) { | |
c3a31e60 JA |
8384 | file = fget(fd); |
8385 | if (!file) { | |
8386 | err = -EBADF; | |
8387 | break; | |
8388 | } | |
8389 | /* | |
8390 | * Don't allow io_uring instances to be registered. If | |
8391 | * UNIX isn't enabled, then this causes a reference | |
8392 | * cycle and this instance can never get freed. If UNIX | |
8393 | * is enabled we'll handle it just fine, but there's | |
8394 | * still no point in allowing a ring fd as it doesn't | |
8395 | * support regular read/write anyway. | |
8396 | */ | |
8397 | if (file->f_op == &io_uring_fops) { | |
8398 | fput(file); | |
8399 | err = -EBADF; | |
8400 | break; | |
8401 | } | |
2d091d62 | 8402 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 8403 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 8404 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8405 | if (err) { |
a04b0ac0 | 8406 | file_slot->file_ptr = 0; |
f3bd9dae | 8407 | fput(file); |
c3a31e60 | 8408 | break; |
f3bd9dae | 8409 | } |
c3a31e60 | 8410 | } |
05f3fb3c JA |
8411 | } |
8412 | ||
a7f0ed5a PB |
8413 | if (needs_switch) |
8414 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
8415 | return done ? done : err; |
8416 | } | |
05589553 | 8417 | |
685fe7fe JA |
8418 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
8419 | struct task_struct *task) | |
24369c2e | 8420 | { |
e941894e | 8421 | struct io_wq_hash *hash; |
24369c2e | 8422 | struct io_wq_data data; |
24369c2e | 8423 | unsigned int concurrency; |
24369c2e | 8424 | |
362a9e65 | 8425 | mutex_lock(&ctx->uring_lock); |
e941894e JA |
8426 | hash = ctx->hash_map; |
8427 | if (!hash) { | |
8428 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
362a9e65 YY |
8429 | if (!hash) { |
8430 | mutex_unlock(&ctx->uring_lock); | |
e941894e | 8431 | return ERR_PTR(-ENOMEM); |
362a9e65 | 8432 | } |
e941894e JA |
8433 | refcount_set(&hash->refs, 1); |
8434 | init_waitqueue_head(&hash->wait); | |
8435 | ctx->hash_map = hash; | |
24369c2e | 8436 | } |
362a9e65 | 8437 | mutex_unlock(&ctx->uring_lock); |
24369c2e | 8438 | |
e941894e | 8439 | data.hash = hash; |
685fe7fe | 8440 | data.task = task; |
ebc11b6c | 8441 | data.free_work = io_wq_free_work; |
f5fa38c5 | 8442 | data.do_work = io_wq_submit_work; |
24369c2e | 8443 | |
d25e3a3d JA |
8444 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
8445 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 8446 | |
5aa75ed5 | 8447 | return io_wq_create(concurrency, &data); |
24369c2e PB |
8448 | } |
8449 | ||
5aa75ed5 JA |
8450 | static int io_uring_alloc_task_context(struct task_struct *task, |
8451 | struct io_ring_ctx *ctx) | |
0f212204 JA |
8452 | { |
8453 | struct io_uring_task *tctx; | |
d8a6df10 | 8454 | int ret; |
0f212204 | 8455 | |
09899b19 | 8456 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
8457 | if (unlikely(!tctx)) |
8458 | return -ENOMEM; | |
8459 | ||
d8a6df10 JA |
8460 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8461 | if (unlikely(ret)) { | |
8462 | kfree(tctx); | |
8463 | return ret; | |
8464 | } | |
8465 | ||
685fe7fe | 8466 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
8467 | if (IS_ERR(tctx->io_wq)) { |
8468 | ret = PTR_ERR(tctx->io_wq); | |
8469 | percpu_counter_destroy(&tctx->inflight); | |
8470 | kfree(tctx); | |
8471 | return ret; | |
8472 | } | |
8473 | ||
0f212204 JA |
8474 | xa_init(&tctx->xa); |
8475 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 8476 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 8477 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 8478 | task->io_uring = tctx; |
7cbf1722 JA |
8479 | spin_lock_init(&tctx->task_lock); |
8480 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 8481 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
8482 | return 0; |
8483 | } | |
8484 | ||
8485 | void __io_uring_free(struct task_struct *tsk) | |
8486 | { | |
8487 | struct io_uring_task *tctx = tsk->io_uring; | |
8488 | ||
8489 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 8490 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 8491 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 8492 | |
d8a6df10 | 8493 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8494 | kfree(tctx); |
8495 | tsk->io_uring = NULL; | |
8496 | } | |
8497 | ||
7e84e1c7 SG |
8498 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8499 | struct io_uring_params *p) | |
2b188cc1 JA |
8500 | { |
8501 | int ret; | |
8502 | ||
d25e3a3d JA |
8503 | /* Retain compatibility with failing for an invalid attach attempt */ |
8504 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
8505 | IORING_SETUP_ATTACH_WQ) { | |
8506 | struct fd f; | |
8507 | ||
8508 | f = fdget(p->wq_fd); | |
8509 | if (!f.file) | |
8510 | return -ENXIO; | |
0cc936f7 JA |
8511 | if (f.file->f_op != &io_uring_fops) { |
8512 | fdput(f); | |
f2a48dd0 | 8513 | return -EINVAL; |
0cc936f7 JA |
8514 | } |
8515 | fdput(f); | |
d25e3a3d | 8516 | } |
6c271ce2 | 8517 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 8518 | struct task_struct *tsk; |
534ca6d6 | 8519 | struct io_sq_data *sqd; |
26984fbf | 8520 | bool attached; |
534ca6d6 | 8521 | |
26984fbf | 8522 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8523 | if (IS_ERR(sqd)) { |
8524 | ret = PTR_ERR(sqd); | |
8525 | goto err; | |
8526 | } | |
69fb2131 | 8527 | |
7c30f36a | 8528 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8529 | ctx->sq_data = sqd; |
917257da JA |
8530 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8531 | if (!ctx->sq_thread_idle) | |
8532 | ctx->sq_thread_idle = HZ; | |
8533 | ||
78d7f6ba | 8534 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8535 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8536 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8537 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8538 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8539 | io_sq_thread_unpark(sqd); |
8540 | ||
de75a3d3 PB |
8541 | if (ret < 0) |
8542 | goto err; | |
8543 | if (attached) | |
5aa75ed5 | 8544 | return 0; |
aa06165d | 8545 | |
6c271ce2 | 8546 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8547 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8548 | |
917257da | 8549 | ret = -EINVAL; |
f2a48dd0 | 8550 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8551 | goto err_sqpoll; |
37d1e2e3 | 8552 | sqd->sq_cpu = cpu; |
6c271ce2 | 8553 | } else { |
37d1e2e3 | 8554 | sqd->sq_cpu = -1; |
6c271ce2 | 8555 | } |
37d1e2e3 JA |
8556 | |
8557 | sqd->task_pid = current->pid; | |
5c2469e0 | 8558 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8559 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8560 | if (IS_ERR(tsk)) { | |
8561 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8562 | goto err_sqpoll; |
6c271ce2 | 8563 | } |
97a73a0f | 8564 | |
46fe18b1 | 8565 | sqd->thread = tsk; |
97a73a0f | 8566 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8567 | wake_up_new_task(tsk); |
0f212204 JA |
8568 | if (ret) |
8569 | goto err; | |
6c271ce2 JA |
8570 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8571 | /* Can't have SQ_AFF without SQPOLL */ | |
8572 | ret = -EINVAL; | |
8573 | goto err; | |
8574 | } | |
8575 | ||
2b188cc1 | 8576 | return 0; |
f2a48dd0 PB |
8577 | err_sqpoll: |
8578 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8579 | err: |
37d1e2e3 | 8580 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8581 | return ret; |
8582 | } | |
8583 | ||
a087e2b5 BM |
8584 | static inline void __io_unaccount_mem(struct user_struct *user, |
8585 | unsigned long nr_pages) | |
2b188cc1 JA |
8586 | { |
8587 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8588 | } | |
8589 | ||
a087e2b5 BM |
8590 | static inline int __io_account_mem(struct user_struct *user, |
8591 | unsigned long nr_pages) | |
2b188cc1 JA |
8592 | { |
8593 | unsigned long page_limit, cur_pages, new_pages; | |
8594 | ||
8595 | /* Don't allow more pages than we can safely lock */ | |
8596 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8597 | ||
8598 | do { | |
8599 | cur_pages = atomic_long_read(&user->locked_vm); | |
8600 | new_pages = cur_pages + nr_pages; | |
8601 | if (new_pages > page_limit) | |
8602 | return -ENOMEM; | |
8603 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8604 | new_pages) != cur_pages); | |
8605 | ||
8606 | return 0; | |
8607 | } | |
8608 | ||
26bfa89e | 8609 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8610 | { |
62e398be | 8611 | if (ctx->user) |
a087e2b5 | 8612 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8613 | |
26bfa89e JA |
8614 | if (ctx->mm_account) |
8615 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8616 | } |
8617 | ||
26bfa89e | 8618 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8619 | { |
30975825 BM |
8620 | int ret; |
8621 | ||
62e398be | 8622 | if (ctx->user) { |
30975825 BM |
8623 | ret = __io_account_mem(ctx->user, nr_pages); |
8624 | if (ret) | |
8625 | return ret; | |
8626 | } | |
8627 | ||
26bfa89e JA |
8628 | if (ctx->mm_account) |
8629 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8630 | |
8631 | return 0; | |
8632 | } | |
8633 | ||
2b188cc1 JA |
8634 | static void io_mem_free(void *ptr) |
8635 | { | |
52e04ef4 MR |
8636 | struct page *page; |
8637 | ||
8638 | if (!ptr) | |
8639 | return; | |
2b188cc1 | 8640 | |
52e04ef4 | 8641 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8642 | if (put_page_testzero(page)) |
8643 | free_compound_page(page); | |
8644 | } | |
8645 | ||
8646 | static void *io_mem_alloc(size_t size) | |
8647 | { | |
8648 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8649 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8650 | |
8651 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8652 | } | |
8653 | ||
75b28aff HV |
8654 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8655 | size_t *sq_offset) | |
8656 | { | |
8657 | struct io_rings *rings; | |
8658 | size_t off, sq_array_size; | |
8659 | ||
8660 | off = struct_size(rings, cqes, cq_entries); | |
8661 | if (off == SIZE_MAX) | |
8662 | return SIZE_MAX; | |
8663 | ||
8664 | #ifdef CONFIG_SMP | |
8665 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8666 | if (off == 0) | |
8667 | return SIZE_MAX; | |
8668 | #endif | |
8669 | ||
b36200f5 DV |
8670 | if (sq_offset) |
8671 | *sq_offset = off; | |
8672 | ||
75b28aff HV |
8673 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8674 | if (sq_array_size == SIZE_MAX) | |
8675 | return SIZE_MAX; | |
8676 | ||
8677 | if (check_add_overflow(off, sq_array_size, &off)) | |
8678 | return SIZE_MAX; | |
8679 | ||
75b28aff HV |
8680 | return off; |
8681 | } | |
8682 | ||
41edf1a5 | 8683 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8684 | { |
41edf1a5 | 8685 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8686 | unsigned int i; |
8687 | ||
6224843d PB |
8688 | if (imu != ctx->dummy_ubuf) { |
8689 | for (i = 0; i < imu->nr_bvecs; i++) | |
8690 | unpin_user_page(imu->bvec[i].bv_page); | |
8691 | if (imu->acct_pages) | |
8692 | io_unaccount_mem(ctx, imu->acct_pages); | |
8693 | kvfree(imu); | |
8694 | } | |
41edf1a5 | 8695 | *slot = NULL; |
7f61a1e9 PB |
8696 | } |
8697 | ||
bd54b6fe | 8698 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8699 | { |
634d00df PB |
8700 | io_buffer_unmap(ctx, &prsrc->buf); |
8701 | prsrc->buf = NULL; | |
bd54b6fe | 8702 | } |
edafccee | 8703 | |
bd54b6fe BM |
8704 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8705 | { | |
8706 | unsigned int i; | |
edafccee | 8707 | |
7f61a1e9 PB |
8708 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8709 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8710 | kfree(ctx->user_bufs); |
bb6659cc | 8711 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8712 | ctx->user_bufs = NULL; |
bd54b6fe | 8713 | ctx->buf_data = NULL; |
edafccee | 8714 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8715 | } |
8716 | ||
0a96bbe4 | 8717 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8718 | { |
bd54b6fe | 8719 | int ret; |
edafccee | 8720 | |
bd54b6fe | 8721 | if (!ctx->buf_data) |
edafccee JA |
8722 | return -ENXIO; |
8723 | ||
bd54b6fe BM |
8724 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8725 | if (!ret) | |
8726 | __io_sqe_buffers_unregister(ctx); | |
8727 | return ret; | |
edafccee JA |
8728 | } |
8729 | ||
8730 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8731 | void __user *arg, unsigned index) | |
8732 | { | |
8733 | struct iovec __user *src; | |
8734 | ||
8735 | #ifdef CONFIG_COMPAT | |
8736 | if (ctx->compat) { | |
8737 | struct compat_iovec __user *ciovs; | |
8738 | struct compat_iovec ciov; | |
8739 | ||
8740 | ciovs = (struct compat_iovec __user *) arg; | |
8741 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8742 | return -EFAULT; | |
8743 | ||
d55e5f5b | 8744 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8745 | dst->iov_len = ciov.iov_len; |
8746 | return 0; | |
8747 | } | |
8748 | #endif | |
8749 | src = (struct iovec __user *) arg; | |
8750 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8751 | return -EFAULT; | |
8752 | return 0; | |
8753 | } | |
8754 | ||
de293938 JA |
8755 | /* |
8756 | * Not super efficient, but this is just a registration time. And we do cache | |
8757 | * the last compound head, so generally we'll only do a full search if we don't | |
8758 | * match that one. | |
8759 | * | |
8760 | * We check if the given compound head page has already been accounted, to | |
8761 | * avoid double accounting it. This allows us to account the full size of the | |
8762 | * page, not just the constituent pages of a huge page. | |
8763 | */ | |
8764 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8765 | int nr_pages, struct page *hpage) | |
8766 | { | |
8767 | int i, j; | |
8768 | ||
8769 | /* check current page array */ | |
8770 | for (i = 0; i < nr_pages; i++) { | |
8771 | if (!PageCompound(pages[i])) | |
8772 | continue; | |
8773 | if (compound_head(pages[i]) == hpage) | |
8774 | return true; | |
8775 | } | |
8776 | ||
8777 | /* check previously registered pages */ | |
8778 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8779 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8780 | |
8781 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8782 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8783 | continue; | |
8784 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8785 | return true; | |
8786 | } | |
8787 | } | |
8788 | ||
8789 | return false; | |
8790 | } | |
8791 | ||
8792 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8793 | int nr_pages, struct io_mapped_ubuf *imu, | |
8794 | struct page **last_hpage) | |
8795 | { | |
8796 | int i, ret; | |
8797 | ||
216e5835 | 8798 | imu->acct_pages = 0; |
de293938 JA |
8799 | for (i = 0; i < nr_pages; i++) { |
8800 | if (!PageCompound(pages[i])) { | |
8801 | imu->acct_pages++; | |
8802 | } else { | |
8803 | struct page *hpage; | |
8804 | ||
8805 | hpage = compound_head(pages[i]); | |
8806 | if (hpage == *last_hpage) | |
8807 | continue; | |
8808 | *last_hpage = hpage; | |
8809 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8810 | continue; | |
8811 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8812 | } | |
8813 | } | |
8814 | ||
8815 | if (!imu->acct_pages) | |
8816 | return 0; | |
8817 | ||
26bfa89e | 8818 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8819 | if (ret) |
8820 | imu->acct_pages = 0; | |
8821 | return ret; | |
8822 | } | |
8823 | ||
0a96bbe4 | 8824 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8825 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8826 | struct page **last_hpage) |
edafccee | 8827 | { |
41edf1a5 | 8828 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8829 | struct vm_area_struct **vmas = NULL; |
8830 | struct page **pages = NULL; | |
0a96bbe4 BM |
8831 | unsigned long off, start, end, ubuf; |
8832 | size_t size; | |
8833 | int ret, pret, nr_pages, i; | |
8834 | ||
6224843d PB |
8835 | if (!iov->iov_base) { |
8836 | *pimu = ctx->dummy_ubuf; | |
8837 | return 0; | |
8838 | } | |
8839 | ||
0a96bbe4 BM |
8840 | ubuf = (unsigned long) iov->iov_base; |
8841 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8842 | start = ubuf >> PAGE_SHIFT; | |
8843 | nr_pages = end - start; | |
8844 | ||
41edf1a5 | 8845 | *pimu = NULL; |
0a96bbe4 BM |
8846 | ret = -ENOMEM; |
8847 | ||
8848 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8849 | if (!pages) | |
8850 | goto done; | |
8851 | ||
8852 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8853 | GFP_KERNEL); | |
8854 | if (!vmas) | |
8855 | goto done; | |
edafccee | 8856 | |
41edf1a5 | 8857 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8858 | if (!imu) |
0a96bbe4 BM |
8859 | goto done; |
8860 | ||
8861 | ret = 0; | |
8862 | mmap_read_lock(current->mm); | |
8863 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8864 | pages, vmas); | |
8865 | if (pret == nr_pages) { | |
8866 | /* don't support file backed memory */ | |
8867 | for (i = 0; i < nr_pages; i++) { | |
8868 | struct vm_area_struct *vma = vmas[i]; | |
8869 | ||
40dad765 PB |
8870 | if (vma_is_shmem(vma)) |
8871 | continue; | |
0a96bbe4 BM |
8872 | if (vma->vm_file && |
8873 | !is_file_hugepages(vma->vm_file)) { | |
8874 | ret = -EOPNOTSUPP; | |
8875 | break; | |
8876 | } | |
8877 | } | |
8878 | } else { | |
8879 | ret = pret < 0 ? pret : -EFAULT; | |
8880 | } | |
8881 | mmap_read_unlock(current->mm); | |
8882 | if (ret) { | |
8883 | /* | |
8884 | * if we did partial map, or found file backed vmas, | |
8885 | * release any pages we did get | |
8886 | */ | |
8887 | if (pret > 0) | |
8888 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8889 | goto done; |
8890 | } | |
8891 | ||
8892 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8893 | if (ret) { | |
8894 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8895 | goto done; |
8896 | } | |
8897 | ||
8898 | off = ubuf & ~PAGE_MASK; | |
8899 | size = iov->iov_len; | |
8900 | for (i = 0; i < nr_pages; i++) { | |
8901 | size_t vec_len; | |
8902 | ||
8903 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8904 | imu->bvec[i].bv_page = pages[i]; | |
8905 | imu->bvec[i].bv_len = vec_len; | |
8906 | imu->bvec[i].bv_offset = off; | |
8907 | off = 0; | |
8908 | size -= vec_len; | |
8909 | } | |
8910 | /* store original address for later verification */ | |
8911 | imu->ubuf = ubuf; | |
4751f53d | 8912 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8913 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8914 | *pimu = imu; |
0a96bbe4 BM |
8915 | ret = 0; |
8916 | done: | |
41edf1a5 PB |
8917 | if (ret) |
8918 | kvfree(imu); | |
0a96bbe4 BM |
8919 | kvfree(pages); |
8920 | kvfree(vmas); | |
8921 | return ret; | |
8922 | } | |
8923 | ||
2b358604 | 8924 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8925 | { |
87094465 PB |
8926 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8927 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8928 | } |
edafccee | 8929 | |
2b358604 BM |
8930 | static int io_buffer_validate(struct iovec *iov) |
8931 | { | |
50e96989 PB |
8932 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8933 | ||
2b358604 BM |
8934 | /* |
8935 | * Don't impose further limits on the size and buffer | |
8936 | * constraints here, we'll -EINVAL later when IO is | |
8937 | * submitted if they are wrong. | |
8938 | */ | |
6224843d PB |
8939 | if (!iov->iov_base) |
8940 | return iov->iov_len ? -EFAULT : 0; | |
8941 | if (!iov->iov_len) | |
2b358604 | 8942 | return -EFAULT; |
edafccee | 8943 | |
2b358604 BM |
8944 | /* arbitrary limit, but we need something */ |
8945 | if (iov->iov_len > SZ_1G) | |
8946 | return -EFAULT; | |
edafccee | 8947 | |
50e96989 PB |
8948 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8949 | return -EOVERFLOW; | |
8950 | ||
2b358604 BM |
8951 | return 0; |
8952 | } | |
edafccee | 8953 | |
2b358604 | 8954 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 8955 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 8956 | { |
bd54b6fe BM |
8957 | struct page *last_hpage = NULL; |
8958 | struct io_rsrc_data *data; | |
2b358604 BM |
8959 | int i, ret; |
8960 | struct iovec iov; | |
edafccee | 8961 | |
87094465 PB |
8962 | if (ctx->user_bufs) |
8963 | return -EBUSY; | |
489809e2 | 8964 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 8965 | return -EINVAL; |
bd54b6fe | 8966 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8967 | if (ret) |
8968 | return ret; | |
d878c816 PB |
8969 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
8970 | if (ret) | |
8971 | return ret; | |
bd54b6fe BM |
8972 | ret = io_buffers_map_alloc(ctx, nr_args); |
8973 | if (ret) { | |
bb6659cc | 8974 | io_rsrc_data_free(data); |
bd54b6fe BM |
8975 | return ret; |
8976 | } | |
edafccee | 8977 | |
87094465 | 8978 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
8979 | ret = io_copy_iov(ctx, &iov, arg, i); |
8980 | if (ret) | |
0a96bbe4 | 8981 | break; |
2b358604 BM |
8982 | ret = io_buffer_validate(&iov); |
8983 | if (ret) | |
0a96bbe4 | 8984 | break; |
2d091d62 | 8985 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
8986 | ret = -EINVAL; |
8987 | break; | |
8988 | } | |
edafccee | 8989 | |
41edf1a5 PB |
8990 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
8991 | &last_hpage); | |
0a96bbe4 BM |
8992 | if (ret) |
8993 | break; | |
edafccee | 8994 | } |
0a96bbe4 | 8995 | |
bd54b6fe | 8996 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8997 | |
bd54b6fe BM |
8998 | ctx->buf_data = data; |
8999 | if (ret) | |
9000 | __io_sqe_buffers_unregister(ctx); | |
9001 | else | |
9002 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
9003 | return ret; |
9004 | } | |
9005 | ||
634d00df PB |
9006 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
9007 | struct io_uring_rsrc_update2 *up, | |
9008 | unsigned int nr_args) | |
9009 | { | |
9010 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
9011 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
9012 | struct page *last_hpage = NULL; |
9013 | bool needs_switch = false; | |
9014 | __u32 done; | |
9015 | int i, err; | |
9016 | ||
9017 | if (!ctx->buf_data) | |
9018 | return -ENXIO; | |
9019 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
9020 | return -EINVAL; | |
9021 | ||
9022 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
9023 | struct io_mapped_ubuf *imu; |
9024 | int offset = up->offset + done; | |
634d00df PB |
9025 | u64 tag = 0; |
9026 | ||
9027 | err = io_copy_iov(ctx, &iov, iovs, done); | |
9028 | if (err) | |
9029 | break; | |
9030 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
9031 | err = -EFAULT; | |
9032 | break; | |
9033 | } | |
0b8c0e7c PB |
9034 | err = io_buffer_validate(&iov); |
9035 | if (err) | |
9036 | break; | |
cf3770e7 CIK |
9037 | if (!iov.iov_base && tag) { |
9038 | err = -EINVAL; | |
9039 | break; | |
9040 | } | |
0b8c0e7c PB |
9041 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
9042 | if (err) | |
9043 | break; | |
634d00df | 9044 | |
0b8c0e7c | 9045 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 9046 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
9047 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
9048 | ctx->rsrc_node, ctx->user_bufs[i]); | |
9049 | if (unlikely(err)) { | |
9050 | io_buffer_unmap(ctx, &imu); | |
634d00df | 9051 | break; |
0b8c0e7c | 9052 | } |
634d00df PB |
9053 | ctx->user_bufs[i] = NULL; |
9054 | needs_switch = true; | |
9055 | } | |
9056 | ||
0b8c0e7c | 9057 | ctx->user_bufs[i] = imu; |
2d091d62 | 9058 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
9059 | } |
9060 | ||
9061 | if (needs_switch) | |
9062 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
9063 | return done ? done : err; | |
9064 | } | |
9065 | ||
9b402849 JA |
9066 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
9067 | { | |
9068 | __s32 __user *fds = arg; | |
9069 | int fd; | |
9070 | ||
9071 | if (ctx->cq_ev_fd) | |
9072 | return -EBUSY; | |
9073 | ||
9074 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
9075 | return -EFAULT; | |
9076 | ||
9077 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
9078 | if (IS_ERR(ctx->cq_ev_fd)) { | |
9079 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
fe7e3257 | 9080 | |
9b402849 JA |
9081 | ctx->cq_ev_fd = NULL; |
9082 | return ret; | |
9083 | } | |
9084 | ||
9085 | return 0; | |
9086 | } | |
9087 | ||
9088 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
9089 | { | |
9090 | if (ctx->cq_ev_fd) { | |
9091 | eventfd_ctx_put(ctx->cq_ev_fd); | |
9092 | ctx->cq_ev_fd = NULL; | |
9093 | return 0; | |
9094 | } | |
9095 | ||
9096 | return -ENXIO; | |
9097 | } | |
9098 | ||
5a2e745d JA |
9099 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
9100 | { | |
9e15c3a0 JA |
9101 | struct io_buffer *buf; |
9102 | unsigned long index; | |
9103 | ||
9104 | xa_for_each(&ctx->io_buffers, index, buf) | |
9105 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
9106 | } |
9107 | ||
7255834e | 9108 | static void io_req_cache_free(struct list_head *list) |
1b4c351f | 9109 | { |
68e68ee6 | 9110 | struct io_kiocb *req, *nxt; |
1b4c351f | 9111 | |
bb943b82 PB |
9112 | list_for_each_entry_safe(req, nxt, list, inflight_entry) { |
9113 | list_del(&req->inflight_entry); | |
1b4c351f JA |
9114 | kmem_cache_free(req_cachep, req); |
9115 | } | |
9116 | } | |
9117 | ||
4010fec4 | 9118 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 9119 | { |
cd0ca2e0 | 9120 | struct io_submit_state *state = &ctx->submit_state; |
bf019da7 | 9121 | |
9a4fdbd8 JA |
9122 | mutex_lock(&ctx->uring_lock); |
9123 | ||
cd0ca2e0 PB |
9124 | if (state->free_reqs) { |
9125 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); | |
9126 | state->free_reqs = 0; | |
8e5c66c4 | 9127 | } |
9a4fdbd8 | 9128 | |
cd0ca2e0 PB |
9129 | io_flush_cached_locked_reqs(ctx, state); |
9130 | io_req_cache_free(&state->free_list); | |
9a4fdbd8 JA |
9131 | mutex_unlock(&ctx->uring_lock); |
9132 | } | |
9133 | ||
43597aac | 9134 | static void io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 9135 | { |
43597aac | 9136 | if (data && !atomic_dec_and_test(&data->refs)) |
bd54b6fe | 9137 | wait_for_completion(&data->done); |
bd54b6fe | 9138 | } |
04fc6c80 | 9139 | |
2b188cc1 JA |
9140 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
9141 | { | |
37d1e2e3 | 9142 | io_sq_thread_finish(ctx); |
2aede0e4 | 9143 | |
37d1e2e3 | 9144 | if (ctx->mm_account) { |
2aede0e4 JA |
9145 | mmdrop(ctx->mm_account); |
9146 | ctx->mm_account = NULL; | |
30975825 | 9147 | } |
def596e9 | 9148 | |
43597aac PB |
9149 | /* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */ |
9150 | io_wait_rsrc_data(ctx->buf_data); | |
9151 | io_wait_rsrc_data(ctx->file_data); | |
9152 | ||
8bad28d8 | 9153 | mutex_lock(&ctx->uring_lock); |
43597aac | 9154 | if (ctx->buf_data) |
bd54b6fe | 9155 | __io_sqe_buffers_unregister(ctx); |
43597aac | 9156 | if (ctx->file_data) |
08480400 | 9157 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
9158 | if (ctx->rings) |
9159 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 9160 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 9161 | io_eventfd_unregister(ctx); |
5a2e745d | 9162 | io_destroy_buffers(ctx); |
07db298a PB |
9163 | if (ctx->sq_creds) |
9164 | put_cred(ctx->sq_creds); | |
def596e9 | 9165 | |
a7f0ed5a PB |
9166 | /* there are no registered resources left, nobody uses it */ |
9167 | if (ctx->rsrc_node) | |
9168 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 9169 | if (ctx->rsrc_backup_node) |
b895c9a6 | 9170 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
9171 | flush_delayed_work(&ctx->rsrc_put_work); |
9172 | ||
9173 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
9174 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 9175 | |
2b188cc1 | 9176 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
9177 | if (ctx->ring_sock) { |
9178 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 9179 | sock_release(ctx->ring_sock); |
355e8d26 | 9180 | } |
2b188cc1 | 9181 | #endif |
ef9dd637 | 9182 | WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list)); |
2b188cc1 | 9183 | |
75b28aff | 9184 | io_mem_free(ctx->rings); |
2b188cc1 | 9185 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
9186 | |
9187 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 9188 | free_uid(ctx->user); |
4010fec4 | 9189 | io_req_caches_free(ctx); |
e941894e JA |
9190 | if (ctx->hash_map) |
9191 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 9192 | kfree(ctx->cancel_hash); |
6224843d | 9193 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
9194 | kfree(ctx); |
9195 | } | |
9196 | ||
9197 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
9198 | { | |
9199 | struct io_ring_ctx *ctx = file->private_data; | |
9200 | __poll_t mask = 0; | |
9201 | ||
311997b3 | 9202 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
9203 | /* |
9204 | * synchronizes with barrier from wq_has_sleeper call in | |
9205 | * io_commit_cqring | |
9206 | */ | |
2b188cc1 | 9207 | smp_rmb(); |
90554200 | 9208 | if (!io_sqring_full(ctx)) |
2b188cc1 | 9209 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
9210 | |
9211 | /* | |
9212 | * Don't flush cqring overflow list here, just do a simple check. | |
9213 | * Otherwise there could possible be ABBA deadlock: | |
9214 | * CPU0 CPU1 | |
9215 | * ---- ---- | |
9216 | * lock(&ctx->uring_lock); | |
9217 | * lock(&ep->mtx); | |
9218 | * lock(&ctx->uring_lock); | |
9219 | * lock(&ep->mtx); | |
9220 | * | |
9221 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
9222 | * pushs them to do the flush. | |
9223 | */ | |
5ed7a37d | 9224 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
9225 | mask |= EPOLLIN | EPOLLRDNORM; |
9226 | ||
9227 | return mask; | |
9228 | } | |
9229 | ||
9230 | static int io_uring_fasync(int fd, struct file *file, int on) | |
9231 | { | |
9232 | struct io_ring_ctx *ctx = file->private_data; | |
9233 | ||
9234 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
9235 | } | |
9236 | ||
0bead8cd | 9237 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 9238 | { |
4379bf8b | 9239 | const struct cred *creds; |
071698e1 | 9240 | |
61cf9370 | 9241 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
9242 | if (creds) { |
9243 | put_cred(creds); | |
0bead8cd | 9244 | return 0; |
1e6fa521 | 9245 | } |
0bead8cd YD |
9246 | |
9247 | return -EINVAL; | |
9248 | } | |
9249 | ||
d56d938b PB |
9250 | struct io_tctx_exit { |
9251 | struct callback_head task_work; | |
9252 | struct completion completion; | |
baf186c4 | 9253 | struct io_ring_ctx *ctx; |
d56d938b PB |
9254 | }; |
9255 | ||
9256 | static void io_tctx_exit_cb(struct callback_head *cb) | |
9257 | { | |
9258 | struct io_uring_task *tctx = current->io_uring; | |
9259 | struct io_tctx_exit *work; | |
9260 | ||
9261 | work = container_of(cb, struct io_tctx_exit, task_work); | |
9262 | /* | |
9263 | * When @in_idle, we're in cancellation and it's racy to remove the | |
9264 | * node. It'll be removed by the end of cancellation, just ignore it. | |
9265 | */ | |
9266 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 9267 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
9268 | complete(&work->completion); |
9269 | } | |
9270 | ||
28090c13 PB |
9271 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
9272 | { | |
9273 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
9274 | ||
9275 | return req->ctx == data; | |
9276 | } | |
9277 | ||
85faa7b8 JA |
9278 | static void io_ring_exit_work(struct work_struct *work) |
9279 | { | |
d56d938b | 9280 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 9281 | unsigned long timeout = jiffies + HZ * 60 * 5; |
58d3be2c | 9282 | unsigned long interval = HZ / 20; |
d56d938b PB |
9283 | struct io_tctx_exit exit; |
9284 | struct io_tctx_node *node; | |
9285 | int ret; | |
85faa7b8 | 9286 | |
56952e91 JA |
9287 | /* |
9288 | * If we're doing polled IO and end up having requests being | |
9289 | * submitted async (out-of-line), then completions can come in while | |
9290 | * we're waiting for refs to drop. We need to reap these manually, | |
9291 | * as nobody else will be looking for them. | |
9292 | */ | |
b2edc0a7 | 9293 | do { |
3dd0c97a | 9294 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
9295 | if (ctx->sq_data) { |
9296 | struct io_sq_data *sqd = ctx->sq_data; | |
9297 | struct task_struct *tsk; | |
9298 | ||
9299 | io_sq_thread_park(sqd); | |
9300 | tsk = sqd->thread; | |
9301 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
9302 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
9303 | io_cancel_ctx_cb, ctx, true); | |
9304 | io_sq_thread_unpark(sqd); | |
9305 | } | |
b5bb3a24 | 9306 | |
58d3be2c PB |
9307 | if (WARN_ON_ONCE(time_after(jiffies, timeout))) { |
9308 | /* there is little hope left, don't run it too often */ | |
9309 | interval = HZ * 60; | |
9310 | } | |
9311 | } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); | |
d56d938b | 9312 | |
7f00651a PB |
9313 | init_completion(&exit.completion); |
9314 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
9315 | exit.ctx = ctx; | |
89b5066e PB |
9316 | /* |
9317 | * Some may use context even when all refs and requests have been put, | |
9318 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 9319 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
9320 | * this lock/unlock section also waits them to finish. |
9321 | */ | |
d56d938b PB |
9322 | mutex_lock(&ctx->uring_lock); |
9323 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
9324 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
9325 | ||
d56d938b PB |
9326 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
9327 | ctx_node); | |
7f00651a PB |
9328 | /* don't spin on a single task if cancellation failed */ |
9329 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
9330 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
9331 | if (WARN_ON_ONCE(ret)) | |
9332 | continue; | |
9333 | wake_up_process(node->task); | |
9334 | ||
9335 | mutex_unlock(&ctx->uring_lock); | |
9336 | wait_for_completion(&exit.completion); | |
d56d938b PB |
9337 | mutex_lock(&ctx->uring_lock); |
9338 | } | |
9339 | mutex_unlock(&ctx->uring_lock); | |
79ebeaee JA |
9340 | spin_lock(&ctx->completion_lock); |
9341 | spin_unlock(&ctx->completion_lock); | |
d56d938b | 9342 | |
85faa7b8 JA |
9343 | io_ring_ctx_free(ctx); |
9344 | } | |
9345 | ||
80c4cbdb PB |
9346 | /* Returns true if we found and killed one or more timeouts */ |
9347 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 9348 | bool cancel_all) |
80c4cbdb PB |
9349 | { |
9350 | struct io_kiocb *req, *tmp; | |
9351 | int canceled = 0; | |
9352 | ||
79ebeaee JA |
9353 | spin_lock(&ctx->completion_lock); |
9354 | spin_lock_irq(&ctx->timeout_lock); | |
80c4cbdb | 9355 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
3dd0c97a | 9356 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
9357 | io_kill_timeout(req, -ECANCELED); |
9358 | canceled++; | |
9359 | } | |
9360 | } | |
79ebeaee | 9361 | spin_unlock_irq(&ctx->timeout_lock); |
51520426 PB |
9362 | if (canceled != 0) |
9363 | io_commit_cqring(ctx); | |
79ebeaee | 9364 | spin_unlock(&ctx->completion_lock); |
80c4cbdb PB |
9365 | if (canceled != 0) |
9366 | io_cqring_ev_posted(ctx); | |
9367 | return canceled != 0; | |
9368 | } | |
9369 | ||
2b188cc1 JA |
9370 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
9371 | { | |
61cf9370 MWO |
9372 | unsigned long index; |
9373 | struct creds *creds; | |
9374 | ||
2b188cc1 JA |
9375 | mutex_lock(&ctx->uring_lock); |
9376 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 9377 | if (ctx->rings) |
6c2450ae | 9378 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
9379 | xa_for_each(&ctx->personalities, index, creds) |
9380 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
9381 | mutex_unlock(&ctx->uring_lock); |
9382 | ||
3dd0c97a PB |
9383 | io_kill_timeouts(ctx, NULL, true); |
9384 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 9385 | |
15dff286 | 9386 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 9387 | io_iopoll_try_reap_events(ctx); |
309fc03a | 9388 | |
85faa7b8 | 9389 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
9390 | /* |
9391 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
9392 | * if we're exiting a ton of rings at the same time. It just adds | |
9393 | * noise and overhead, there's no discernable change in runtime | |
9394 | * over using system_wq. | |
9395 | */ | |
9396 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
9397 | } |
9398 | ||
9399 | static int io_uring_release(struct inode *inode, struct file *file) | |
9400 | { | |
9401 | struct io_ring_ctx *ctx = file->private_data; | |
9402 | ||
9403 | file->private_data = NULL; | |
9404 | io_ring_ctx_wait_and_kill(ctx); | |
9405 | return 0; | |
9406 | } | |
9407 | ||
f6edbabb PB |
9408 | struct io_task_cancel { |
9409 | struct task_struct *task; | |
3dd0c97a | 9410 | bool all; |
f6edbabb | 9411 | }; |
f254ac04 | 9412 | |
f6edbabb | 9413 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 9414 | { |
9a472ef7 | 9415 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 9416 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
9417 | bool ret; |
9418 | ||
3dd0c97a | 9419 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
9420 | struct io_ring_ctx *ctx = req->ctx; |
9421 | ||
9422 | /* protect against races with linked timeouts */ | |
79ebeaee | 9423 | spin_lock(&ctx->completion_lock); |
3dd0c97a | 9424 | ret = io_match_task(req, cancel->task, cancel->all); |
79ebeaee | 9425 | spin_unlock(&ctx->completion_lock); |
9a472ef7 | 9426 | } else { |
3dd0c97a | 9427 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
9428 | } |
9429 | return ret; | |
b711d4ea JA |
9430 | } |
9431 | ||
e1915f76 | 9432 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 9433 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 9434 | { |
e1915f76 | 9435 | struct io_defer_entry *de; |
b7ddce3c PB |
9436 | LIST_HEAD(list); |
9437 | ||
79ebeaee | 9438 | spin_lock(&ctx->completion_lock); |
b7ddce3c | 9439 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { |
3dd0c97a | 9440 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
9441 | list_cut_position(&list, &ctx->defer_list, &de->list); |
9442 | break; | |
9443 | } | |
9444 | } | |
79ebeaee | 9445 | spin_unlock(&ctx->completion_lock); |
e1915f76 PB |
9446 | if (list_empty(&list)) |
9447 | return false; | |
b7ddce3c PB |
9448 | |
9449 | while (!list_empty(&list)) { | |
9450 | de = list_first_entry(&list, struct io_defer_entry, list); | |
9451 | list_del_init(&de->list); | |
f41db273 | 9452 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
9453 | kfree(de); |
9454 | } | |
e1915f76 | 9455 | return true; |
b7ddce3c PB |
9456 | } |
9457 | ||
1b00764f PB |
9458 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
9459 | { | |
9460 | struct io_tctx_node *node; | |
9461 | enum io_wq_cancel cret; | |
9462 | bool ret = false; | |
9463 | ||
9464 | mutex_lock(&ctx->uring_lock); | |
9465 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
9466 | struct io_uring_task *tctx = node->task->io_uring; | |
9467 | ||
9468 | /* | |
9469 | * io_wq will stay alive while we hold uring_lock, because it's | |
9470 | * killed after ctx nodes, which requires to take the lock. | |
9471 | */ | |
9472 | if (!tctx || !tctx->io_wq) | |
9473 | continue; | |
9474 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
9475 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9476 | } | |
9477 | mutex_unlock(&ctx->uring_lock); | |
9478 | ||
9479 | return ret; | |
9480 | } | |
9481 | ||
9936c7c2 PB |
9482 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
9483 | struct task_struct *task, | |
3dd0c97a | 9484 | bool cancel_all) |
9936c7c2 | 9485 | { |
3dd0c97a | 9486 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 9487 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
9488 | |
9489 | while (1) { | |
9490 | enum io_wq_cancel cret; | |
9491 | bool ret = false; | |
9492 | ||
1b00764f PB |
9493 | if (!task) { |
9494 | ret |= io_uring_try_cancel_iowq(ctx); | |
9495 | } else if (tctx && tctx->io_wq) { | |
9496 | /* | |
9497 | * Cancels requests of all rings, not only @ctx, but | |
9498 | * it's fine as the task is in exit/exec. | |
9499 | */ | |
5aa75ed5 | 9500 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
9501 | &cancel, true); |
9502 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9503 | } | |
9504 | ||
9505 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 9506 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 9507 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
9508 | while (!list_empty_careful(&ctx->iopoll_list)) { |
9509 | io_iopoll_try_reap_events(ctx); | |
9510 | ret = true; | |
9511 | } | |
9512 | } | |
9513 | ||
3dd0c97a PB |
9514 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
9515 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
9516 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
9517 | if (task) |
9518 | ret |= io_run_task_work(); | |
9936c7c2 PB |
9519 | if (!ret) |
9520 | break; | |
9521 | cond_resched(); | |
9522 | } | |
9523 | } | |
9524 | ||
eef51daa | 9525 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 9526 | { |
236434c3 | 9527 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 9528 | struct io_tctx_node *node; |
a528b04e | 9529 | int ret; |
236434c3 MWO |
9530 | |
9531 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9532 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9533 | if (unlikely(ret)) |
9534 | return ret; | |
236434c3 | 9535 | tctx = current->io_uring; |
0f212204 | 9536 | } |
cf27f3b1 PB |
9537 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9538 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9539 | if (!node) | |
9540 | return -ENOMEM; | |
9541 | node->ctx = ctx; | |
9542 | node->task = current; | |
13bf43f5 | 9543 | |
cf27f3b1 PB |
9544 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9545 | node, GFP_KERNEL)); | |
9546 | if (ret) { | |
9547 | kfree(node); | |
9548 | return ret; | |
0f212204 | 9549 | } |
cf27f3b1 PB |
9550 | |
9551 | mutex_lock(&ctx->uring_lock); | |
9552 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9553 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9554 | } |
cf27f3b1 | 9555 | tctx->last = ctx; |
0f212204 JA |
9556 | return 0; |
9557 | } | |
9558 | ||
cf27f3b1 PB |
9559 | /* |
9560 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9561 | */ | |
eef51daa | 9562 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9563 | { |
9564 | struct io_uring_task *tctx = current->io_uring; | |
9565 | ||
9566 | if (likely(tctx && tctx->last == ctx)) | |
9567 | return 0; | |
eef51daa | 9568 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9569 | } |
9570 | ||
0f212204 JA |
9571 | /* |
9572 | * Remove this io_uring_file -> task mapping. | |
9573 | */ | |
eef51daa | 9574 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9575 | { |
9576 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9577 | struct io_tctx_node *node; |
2941267b | 9578 | |
eebd2e37 PB |
9579 | if (!tctx) |
9580 | return; | |
13bf43f5 PB |
9581 | node = xa_erase(&tctx->xa, index); |
9582 | if (!node) | |
2941267b | 9583 | return; |
0f212204 | 9584 | |
13bf43f5 PB |
9585 | WARN_ON_ONCE(current != node->task); |
9586 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9587 | ||
9588 | mutex_lock(&node->ctx->uring_lock); | |
9589 | list_del(&node->ctx_node); | |
9590 | mutex_unlock(&node->ctx->uring_lock); | |
9591 | ||
baf186c4 | 9592 | if (tctx->last == node->ctx) |
0f212204 | 9593 | tctx->last = NULL; |
13bf43f5 | 9594 | kfree(node); |
0f212204 JA |
9595 | } |
9596 | ||
8452d4a6 | 9597 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9598 | { |
ba5ef6dc | 9599 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9600 | struct io_tctx_node *node; |
de7f1d9e PB |
9601 | unsigned long index; |
9602 | ||
13bf43f5 | 9603 | xa_for_each(&tctx->xa, index, node) |
eef51daa | 9604 | io_uring_del_tctx_node(index); |
b16ef427 ME |
9605 | if (wq) { |
9606 | /* | |
9607 | * Must be after io_uring_del_task_file() (removes nodes under | |
9608 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9609 | */ | |
ba5ef6dc | 9610 | io_wq_put_and_exit(wq); |
dadebc35 | 9611 | tctx->io_wq = NULL; |
b16ef427 | 9612 | } |
de7f1d9e PB |
9613 | } |
9614 | ||
3f48cf18 | 9615 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9616 | { |
3f48cf18 PB |
9617 | if (tracked) |
9618 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9619 | return percpu_counter_sum(&tctx->inflight); |
9620 | } | |
9621 | ||
09899b19 PB |
9622 | static void io_uring_drop_tctx_refs(struct task_struct *task) |
9623 | { | |
9624 | struct io_uring_task *tctx = task->io_uring; | |
9625 | unsigned int refs = tctx->cached_refs; | |
9626 | ||
e9dbe221 PB |
9627 | if (refs) { |
9628 | tctx->cached_refs = 0; | |
9629 | percpu_counter_sub(&tctx->inflight, refs); | |
9630 | put_task_struct_many(task, refs); | |
9631 | } | |
09899b19 PB |
9632 | } |
9633 | ||
78cc687b PB |
9634 | /* |
9635 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
9636 | * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation. | |
9637 | */ | |
9638 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9639 | { |
521d6a73 | 9640 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9641 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9642 | s64 inflight; |
9643 | DEFINE_WAIT(wait); | |
fdaf083c | 9644 | |
78cc687b PB |
9645 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9646 | ||
6d042ffb PO |
9647 | if (!current->io_uring) |
9648 | return; | |
17a91051 PB |
9649 | if (tctx->io_wq) |
9650 | io_wq_exit_start(tctx->io_wq); | |
9651 | ||
0e9ddb39 PB |
9652 | atomic_inc(&tctx->in_idle); |
9653 | do { | |
e9dbe221 | 9654 | io_uring_drop_tctx_refs(current); |
0e9ddb39 | 9655 | /* read completions before cancelations */ |
78cc687b | 9656 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9657 | if (!inflight) |
9658 | break; | |
fdaf083c | 9659 | |
78cc687b PB |
9660 | if (!sqd) { |
9661 | struct io_tctx_node *node; | |
9662 | unsigned long index; | |
0f212204 | 9663 | |
78cc687b PB |
9664 | xa_for_each(&tctx->xa, index, node) { |
9665 | /* sqpoll task will cancel all its requests */ | |
9666 | if (node->ctx->sq_data) | |
9667 | continue; | |
9668 | io_uring_try_cancel_requests(node->ctx, current, | |
9669 | cancel_all); | |
9670 | } | |
9671 | } else { | |
9672 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9673 | io_uring_try_cancel_requests(ctx, current, | |
9674 | cancel_all); | |
9675 | } | |
17a91051 | 9676 | |
0f212204 | 9677 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
e9dbe221 | 9678 | io_uring_drop_tctx_refs(current); |
0f212204 | 9679 | /* |
a1bb3cd5 PB |
9680 | * If we've seen completions, retry without waiting. This |
9681 | * avoids a race where a completion comes in before we did | |
9682 | * prepare_to_wait(). | |
0f212204 | 9683 | */ |
3dd0c97a | 9684 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9685 | schedule(); |
f57555ed | 9686 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9687 | } while (1); |
fdaf083c | 9688 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9689 | |
8452d4a6 | 9690 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9691 | if (cancel_all) { |
3f48cf18 PB |
9692 | /* for exec all current's requests should be gone, kill tctx */ |
9693 | __io_uring_free(current); | |
9694 | } | |
44e728b8 PB |
9695 | } |
9696 | ||
f552a27a | 9697 | void __io_uring_cancel(bool cancel_all) |
78cc687b | 9698 | { |
f552a27a | 9699 | io_uring_cancel_generic(cancel_all, NULL); |
78cc687b PB |
9700 | } |
9701 | ||
6c5c240e RP |
9702 | static void *io_uring_validate_mmap_request(struct file *file, |
9703 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9704 | { |
2b188cc1 | 9705 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9706 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9707 | struct page *page; |
9708 | void *ptr; | |
9709 | ||
9710 | switch (offset) { | |
9711 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9712 | case IORING_OFF_CQ_RING: |
9713 | ptr = ctx->rings; | |
2b188cc1 JA |
9714 | break; |
9715 | case IORING_OFF_SQES: | |
9716 | ptr = ctx->sq_sqes; | |
9717 | break; | |
2b188cc1 | 9718 | default: |
6c5c240e | 9719 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9720 | } |
9721 | ||
9722 | page = virt_to_head_page(ptr); | |
a50b854e | 9723 | if (sz > page_size(page)) |
6c5c240e RP |
9724 | return ERR_PTR(-EINVAL); |
9725 | ||
9726 | return ptr; | |
9727 | } | |
9728 | ||
9729 | #ifdef CONFIG_MMU | |
9730 | ||
9731 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9732 | { | |
9733 | size_t sz = vma->vm_end - vma->vm_start; | |
9734 | unsigned long pfn; | |
9735 | void *ptr; | |
9736 | ||
9737 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9738 | if (IS_ERR(ptr)) | |
9739 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9740 | |
9741 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9742 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9743 | } | |
9744 | ||
6c5c240e RP |
9745 | #else /* !CONFIG_MMU */ |
9746 | ||
9747 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9748 | { | |
9749 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9750 | } | |
9751 | ||
9752 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9753 | { | |
9754 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9755 | } | |
9756 | ||
9757 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9758 | unsigned long addr, unsigned long len, | |
9759 | unsigned long pgoff, unsigned long flags) | |
9760 | { | |
9761 | void *ptr; | |
9762 | ||
9763 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9764 | if (IS_ERR(ptr)) | |
9765 | return PTR_ERR(ptr); | |
9766 | ||
9767 | return (unsigned long) ptr; | |
9768 | } | |
9769 | ||
9770 | #endif /* !CONFIG_MMU */ | |
9771 | ||
d9d05217 | 9772 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9773 | { |
9774 | DEFINE_WAIT(wait); | |
9775 | ||
9776 | do { | |
9777 | if (!io_sqring_full(ctx)) | |
9778 | break; | |
90554200 JA |
9779 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9780 | ||
9781 | if (!io_sqring_full(ctx)) | |
9782 | break; | |
90554200 JA |
9783 | schedule(); |
9784 | } while (!signal_pending(current)); | |
9785 | ||
9786 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9787 | return 0; |
90554200 JA |
9788 | } |
9789 | ||
c73ebb68 HX |
9790 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9791 | struct __kernel_timespec __user **ts, | |
9792 | const sigset_t __user **sig) | |
9793 | { | |
9794 | struct io_uring_getevents_arg arg; | |
9795 | ||
9796 | /* | |
9797 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9798 | * is just a pointer to the sigset_t. | |
9799 | */ | |
9800 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9801 | *sig = (const sigset_t __user *) argp; | |
9802 | *ts = NULL; | |
9803 | return 0; | |
9804 | } | |
9805 | ||
9806 | /* | |
9807 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9808 | * timespec and sigset_t pointers if good. | |
9809 | */ | |
9810 | if (*argsz != sizeof(arg)) | |
9811 | return -EINVAL; | |
9812 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9813 | return -EFAULT; | |
9814 | *sig = u64_to_user_ptr(arg.sigmask); | |
9815 | *argsz = arg.sigmask_sz; | |
9816 | *ts = u64_to_user_ptr(arg.ts); | |
9817 | return 0; | |
9818 | } | |
9819 | ||
2b188cc1 | 9820 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9821 | u32, min_complete, u32, flags, const void __user *, argp, |
9822 | size_t, argsz) | |
2b188cc1 JA |
9823 | { |
9824 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9825 | int submitted = 0; |
9826 | struct fd f; | |
33f993da | 9827 | long ret; |
2b188cc1 | 9828 | |
4c6e277c | 9829 | io_run_task_work(); |
b41e9852 | 9830 | |
33f993da PB |
9831 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9832 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9833 | return -EINVAL; |
9834 | ||
9835 | f = fdget(fd); | |
33f993da | 9836 | if (unlikely(!f.file)) |
2b188cc1 JA |
9837 | return -EBADF; |
9838 | ||
9839 | ret = -EOPNOTSUPP; | |
33f993da | 9840 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9841 | goto out_fput; |
9842 | ||
9843 | ret = -ENXIO; | |
9844 | ctx = f.file->private_data; | |
33f993da | 9845 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9846 | goto out_fput; |
9847 | ||
7e84e1c7 | 9848 | ret = -EBADFD; |
33f993da | 9849 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9850 | goto out; |
9851 | ||
6c271ce2 JA |
9852 | /* |
9853 | * For SQ polling, the thread will do all submissions and completions. | |
9854 | * Just return the requested submit count, and wake the thread if | |
9855 | * we were asked to. | |
9856 | */ | |
b2a9eada | 9857 | ret = 0; |
6c271ce2 | 9858 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
90f67366 | 9859 | io_cqring_overflow_flush(ctx); |
89448c47 | 9860 | |
21f96522 JA |
9861 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9862 | ret = -EOWNERDEAD; | |
04147488 | 9863 | goto out; |
21f96522 | 9864 | } |
6c271ce2 | 9865 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9866 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9867 | if (flags & IORING_ENTER_SQ_WAIT) { |
9868 | ret = io_sqpoll_wait_sq(ctx); | |
9869 | if (ret) | |
9870 | goto out; | |
9871 | } | |
6c271ce2 | 9872 | submitted = to_submit; |
b2a9eada | 9873 | } else if (to_submit) { |
eef51daa | 9874 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
9875 | if (unlikely(ret)) |
9876 | goto out; | |
2b188cc1 | 9877 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9878 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9879 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9880 | |
9881 | if (submitted != to_submit) | |
9882 | goto out; | |
2b188cc1 JA |
9883 | } |
9884 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9885 | const sigset_t __user *sig; |
9886 | struct __kernel_timespec __user *ts; | |
9887 | ||
9888 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9889 | if (unlikely(ret)) | |
9890 | goto out; | |
9891 | ||
2b188cc1 JA |
9892 | min_complete = min(min_complete, ctx->cq_entries); |
9893 | ||
32b2244a XW |
9894 | /* |
9895 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9896 | * space applications don't need to do io completion events | |
9897 | * polling again, they can rely on io_sq_thread to do polling | |
9898 | * work, which can reduce cpu usage and uring_lock contention. | |
9899 | */ | |
9900 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9901 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9902 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9903 | } else { |
c73ebb68 | 9904 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9905 | } |
2b188cc1 JA |
9906 | } |
9907 | ||
7c504e65 | 9908 | out: |
6805b32e | 9909 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9910 | out_fput: |
9911 | fdput(f); | |
9912 | return submitted ? submitted : ret; | |
9913 | } | |
9914 | ||
bebdb65e | 9915 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9916 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9917 | const struct cred *cred) | |
87ce955b | 9918 | { |
87ce955b JA |
9919 | struct user_namespace *uns = seq_user_ns(m); |
9920 | struct group_info *gi; | |
9921 | kernel_cap_t cap; | |
9922 | unsigned __capi; | |
9923 | int g; | |
9924 | ||
9925 | seq_printf(m, "%5d\n", id); | |
9926 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9927 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9928 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9929 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9930 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9931 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9932 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9933 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9934 | seq_puts(m, "\n\tGroups:\t"); | |
9935 | gi = cred->group_info; | |
9936 | for (g = 0; g < gi->ngroups; g++) { | |
9937 | seq_put_decimal_ull(m, g ? " " : "", | |
9938 | from_kgid_munged(uns, gi->gid[g])); | |
9939 | } | |
9940 | seq_puts(m, "\n\tCapEff:\t"); | |
9941 | cap = cred->cap_effective; | |
9942 | CAP_FOR_EACH_U32(__capi) | |
9943 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9944 | seq_putc(m, '\n'); | |
9945 | return 0; | |
9946 | } | |
9947 | ||
9948 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9949 | { | |
dbbe9c64 | 9950 | struct io_sq_data *sq = NULL; |
fad8e0de | 9951 | bool has_lock; |
87ce955b JA |
9952 | int i; |
9953 | ||
fad8e0de JA |
9954 | /* |
9955 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9956 | * since fdinfo case grabs it in the opposite direction of normal use | |
9957 | * cases. If we fail to get the lock, we just don't iterate any | |
9958 | * structures that could be going away outside the io_uring mutex. | |
9959 | */ | |
9960 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9961 | ||
5f3f26f9 | 9962 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9963 | sq = ctx->sq_data; |
5f3f26f9 JA |
9964 | if (!sq->thread) |
9965 | sq = NULL; | |
9966 | } | |
dbbe9c64 JQ |
9967 | |
9968 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9969 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9970 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9971 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9972 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9973 | |
87ce955b JA |
9974 | if (f) |
9975 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9976 | else | |
9977 | seq_printf(m, "%5u: <none>\n", i); | |
9978 | } | |
9979 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9980 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9981 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9982 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9983 | |
4751f53d | 9984 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9985 | } |
61cf9370 MWO |
9986 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9987 | unsigned long index; | |
9988 | const struct cred *cred; | |
9989 | ||
87ce955b | 9990 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9991 | xa_for_each(&ctx->personalities, index, cred) |
9992 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9993 | } |
d7718a9d | 9994 | seq_printf(m, "PollList:\n"); |
79ebeaee | 9995 | spin_lock(&ctx->completion_lock); |
d7718a9d JA |
9996 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
9997 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9998 | struct io_kiocb *req; | |
9999 | ||
10000 | hlist_for_each_entry(req, list, hash_node) | |
10001 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
10002 | req->task->task_works != NULL); | |
10003 | } | |
79ebeaee | 10004 | spin_unlock(&ctx->completion_lock); |
fad8e0de JA |
10005 | if (has_lock) |
10006 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
10007 | } |
10008 | ||
10009 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
10010 | { | |
10011 | struct io_ring_ctx *ctx = f->private_data; | |
10012 | ||
10013 | if (percpu_ref_tryget(&ctx->refs)) { | |
10014 | __io_uring_show_fdinfo(ctx, m); | |
10015 | percpu_ref_put(&ctx->refs); | |
10016 | } | |
10017 | } | |
bebdb65e | 10018 | #endif |
87ce955b | 10019 | |
2b188cc1 JA |
10020 | static const struct file_operations io_uring_fops = { |
10021 | .release = io_uring_release, | |
10022 | .mmap = io_uring_mmap, | |
6c5c240e RP |
10023 | #ifndef CONFIG_MMU |
10024 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
10025 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
10026 | #endif | |
2b188cc1 JA |
10027 | .poll = io_uring_poll, |
10028 | .fasync = io_uring_fasync, | |
bebdb65e | 10029 | #ifdef CONFIG_PROC_FS |
87ce955b | 10030 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 10031 | #endif |
2b188cc1 JA |
10032 | }; |
10033 | ||
10034 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
10035 | struct io_uring_params *p) | |
10036 | { | |
75b28aff HV |
10037 | struct io_rings *rings; |
10038 | size_t size, sq_array_offset; | |
2b188cc1 | 10039 | |
bd740481 JA |
10040 | /* make sure these are sane, as we already accounted them */ |
10041 | ctx->sq_entries = p->sq_entries; | |
10042 | ctx->cq_entries = p->cq_entries; | |
10043 | ||
75b28aff HV |
10044 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
10045 | if (size == SIZE_MAX) | |
10046 | return -EOVERFLOW; | |
10047 | ||
10048 | rings = io_mem_alloc(size); | |
10049 | if (!rings) | |
2b188cc1 JA |
10050 | return -ENOMEM; |
10051 | ||
75b28aff HV |
10052 | ctx->rings = rings; |
10053 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
10054 | rings->sq_ring_mask = p->sq_entries - 1; | |
10055 | rings->cq_ring_mask = p->cq_entries - 1; | |
10056 | rings->sq_ring_entries = p->sq_entries; | |
10057 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
10058 | |
10059 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
10060 | if (size == SIZE_MAX) { |
10061 | io_mem_free(ctx->rings); | |
10062 | ctx->rings = NULL; | |
2b188cc1 | 10063 | return -EOVERFLOW; |
eb065d30 | 10064 | } |
2b188cc1 JA |
10065 | |
10066 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
10067 | if (!ctx->sq_sqes) { |
10068 | io_mem_free(ctx->rings); | |
10069 | ctx->rings = NULL; | |
2b188cc1 | 10070 | return -ENOMEM; |
eb065d30 | 10071 | } |
2b188cc1 | 10072 | |
2b188cc1 JA |
10073 | return 0; |
10074 | } | |
10075 | ||
9faadcc8 PB |
10076 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
10077 | { | |
10078 | int ret, fd; | |
10079 | ||
10080 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
10081 | if (fd < 0) | |
10082 | return fd; | |
10083 | ||
eef51daa | 10084 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
10085 | if (ret) { |
10086 | put_unused_fd(fd); | |
10087 | return ret; | |
10088 | } | |
10089 | fd_install(fd, file); | |
10090 | return fd; | |
10091 | } | |
10092 | ||
2b188cc1 JA |
10093 | /* |
10094 | * Allocate an anonymous fd, this is what constitutes the application | |
10095 | * visible backing of an io_uring instance. The application mmaps this | |
10096 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
10097 | * we have to tie this fd to a socket for file garbage collection purposes. | |
10098 | */ | |
9faadcc8 | 10099 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
10100 | { |
10101 | struct file *file; | |
9faadcc8 | 10102 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
10103 | int ret; |
10104 | ||
2b188cc1 JA |
10105 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
10106 | &ctx->ring_sock); | |
10107 | if (ret) | |
9faadcc8 | 10108 | return ERR_PTR(ret); |
2b188cc1 JA |
10109 | #endif |
10110 | ||
2b188cc1 JA |
10111 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
10112 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 10113 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
10114 | if (IS_ERR(file)) { |
10115 | sock_release(ctx->ring_sock); | |
10116 | ctx->ring_sock = NULL; | |
10117 | } else { | |
10118 | ctx->ring_sock->file = file; | |
0f212204 | 10119 | } |
2b188cc1 | 10120 | #endif |
9faadcc8 | 10121 | return file; |
2b188cc1 JA |
10122 | } |
10123 | ||
7f13657d XW |
10124 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
10125 | struct io_uring_params __user *params) | |
2b188cc1 | 10126 | { |
2b188cc1 | 10127 | struct io_ring_ctx *ctx; |
9faadcc8 | 10128 | struct file *file; |
2b188cc1 JA |
10129 | int ret; |
10130 | ||
8110c1a6 | 10131 | if (!entries) |
2b188cc1 | 10132 | return -EINVAL; |
8110c1a6 JA |
10133 | if (entries > IORING_MAX_ENTRIES) { |
10134 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
10135 | return -EINVAL; | |
10136 | entries = IORING_MAX_ENTRIES; | |
10137 | } | |
2b188cc1 JA |
10138 | |
10139 | /* | |
10140 | * Use twice as many entries for the CQ ring. It's possible for the | |
10141 | * application to drive a higher depth than the size of the SQ ring, | |
10142 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
10143 | * some flexibility in overcommitting a bit. If the application has |
10144 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
10145 | * of CQ ring entries manually. | |
2b188cc1 JA |
10146 | */ |
10147 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
10148 | if (p->flags & IORING_SETUP_CQSIZE) { |
10149 | /* | |
10150 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
10151 | * to a power-of-two, if it isn't already. We do NOT impose | |
10152 | * any cq vs sq ring sizing. | |
10153 | */ | |
eb2667b3 | 10154 | if (!p->cq_entries) |
33a107f0 | 10155 | return -EINVAL; |
8110c1a6 JA |
10156 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
10157 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
10158 | return -EINVAL; | |
10159 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
10160 | } | |
eb2667b3 JQ |
10161 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
10162 | if (p->cq_entries < p->sq_entries) | |
10163 | return -EINVAL; | |
33a107f0 JA |
10164 | } else { |
10165 | p->cq_entries = 2 * p->sq_entries; | |
10166 | } | |
2b188cc1 | 10167 | |
2b188cc1 | 10168 | ctx = io_ring_ctx_alloc(p); |
62e398be | 10169 | if (!ctx) |
2b188cc1 | 10170 | return -ENOMEM; |
2b188cc1 | 10171 | ctx->compat = in_compat_syscall(); |
62e398be JA |
10172 | if (!capable(CAP_IPC_LOCK)) |
10173 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
10174 | |
10175 | /* | |
10176 | * This is just grabbed for accounting purposes. When a process exits, | |
10177 | * the mm is exited and dropped before the files, hence we need to hang | |
10178 | * on to this mm purely for the purposes of being able to unaccount | |
10179 | * memory (locked/pinned vm). It's not used for anything else. | |
10180 | */ | |
6b7898eb | 10181 | mmgrab(current->mm); |
2aede0e4 | 10182 | ctx->mm_account = current->mm; |
6b7898eb | 10183 | |
2b188cc1 JA |
10184 | ret = io_allocate_scq_urings(ctx, p); |
10185 | if (ret) | |
10186 | goto err; | |
10187 | ||
7e84e1c7 | 10188 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
10189 | if (ret) |
10190 | goto err; | |
eae071c9 | 10191 | /* always set a rsrc node */ |
47b228ce PB |
10192 | ret = io_rsrc_node_switch_start(ctx); |
10193 | if (ret) | |
10194 | goto err; | |
eae071c9 | 10195 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 10196 | |
2b188cc1 | 10197 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
10198 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
10199 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
10200 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
10201 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
10202 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
10203 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
10204 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
10205 | |
10206 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
10207 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
10208 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
10209 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
10210 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
10211 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
10212 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 10213 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 10214 | |
7f13657d XW |
10215 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
10216 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 10217 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 10218 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
10219 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
10220 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
10221 | |
10222 | if (copy_to_user(params, p, sizeof(*p))) { | |
10223 | ret = -EFAULT; | |
10224 | goto err; | |
10225 | } | |
d1719f70 | 10226 | |
9faadcc8 PB |
10227 | file = io_uring_get_file(ctx); |
10228 | if (IS_ERR(file)) { | |
10229 | ret = PTR_ERR(file); | |
10230 | goto err; | |
10231 | } | |
10232 | ||
044c1ab3 JA |
10233 | /* |
10234 | * Install ring fd as the very last thing, so we don't risk someone | |
10235 | * having closed it before we finish setup | |
10236 | */ | |
9faadcc8 PB |
10237 | ret = io_uring_install_fd(ctx, file); |
10238 | if (ret < 0) { | |
10239 | /* fput will clean it up */ | |
10240 | fput(file); | |
10241 | return ret; | |
10242 | } | |
044c1ab3 | 10243 | |
c826bd7a | 10244 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
10245 | return ret; |
10246 | err: | |
10247 | io_ring_ctx_wait_and_kill(ctx); | |
10248 | return ret; | |
10249 | } | |
10250 | ||
10251 | /* | |
10252 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
10253 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
10254 | * params structure passed in. | |
10255 | */ | |
10256 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
10257 | { | |
10258 | struct io_uring_params p; | |
2b188cc1 JA |
10259 | int i; |
10260 | ||
10261 | if (copy_from_user(&p, params, sizeof(p))) | |
10262 | return -EFAULT; | |
10263 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
10264 | if (p.resv[i]) | |
10265 | return -EINVAL; | |
10266 | } | |
10267 | ||
6c271ce2 | 10268 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 10269 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
10270 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
10271 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
10272 | return -EINVAL; |
10273 | ||
7f13657d | 10274 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
10275 | } |
10276 | ||
10277 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
10278 | struct io_uring_params __user *, params) | |
10279 | { | |
10280 | return io_uring_setup(entries, params); | |
10281 | } | |
10282 | ||
66f4af93 JA |
10283 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
10284 | { | |
10285 | struct io_uring_probe *p; | |
10286 | size_t size; | |
10287 | int i, ret; | |
10288 | ||
10289 | size = struct_size(p, ops, nr_args); | |
10290 | if (size == SIZE_MAX) | |
10291 | return -EOVERFLOW; | |
10292 | p = kzalloc(size, GFP_KERNEL); | |
10293 | if (!p) | |
10294 | return -ENOMEM; | |
10295 | ||
10296 | ret = -EFAULT; | |
10297 | if (copy_from_user(p, arg, size)) | |
10298 | goto out; | |
10299 | ret = -EINVAL; | |
10300 | if (memchr_inv(p, 0, size)) | |
10301 | goto out; | |
10302 | ||
10303 | p->last_op = IORING_OP_LAST - 1; | |
10304 | if (nr_args > IORING_OP_LAST) | |
10305 | nr_args = IORING_OP_LAST; | |
10306 | ||
10307 | for (i = 0; i < nr_args; i++) { | |
10308 | p->ops[i].op = i; | |
10309 | if (!io_op_defs[i].not_supported) | |
10310 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
10311 | } | |
10312 | p->ops_len = i; | |
10313 | ||
10314 | ret = 0; | |
10315 | if (copy_to_user(arg, p, size)) | |
10316 | ret = -EFAULT; | |
10317 | out: | |
10318 | kfree(p); | |
10319 | return ret; | |
10320 | } | |
10321 | ||
071698e1 JA |
10322 | static int io_register_personality(struct io_ring_ctx *ctx) |
10323 | { | |
4379bf8b | 10324 | const struct cred *creds; |
61cf9370 | 10325 | u32 id; |
1e6fa521 | 10326 | int ret; |
071698e1 | 10327 | |
4379bf8b | 10328 | creds = get_current_cred(); |
1e6fa521 | 10329 | |
61cf9370 MWO |
10330 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
10331 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
a30f895a JA |
10332 | if (ret < 0) { |
10333 | put_cred(creds); | |
10334 | return ret; | |
10335 | } | |
10336 | return id; | |
071698e1 JA |
10337 | } |
10338 | ||
21b55dbc SG |
10339 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
10340 | unsigned int nr_args) | |
10341 | { | |
10342 | struct io_uring_restriction *res; | |
10343 | size_t size; | |
10344 | int i, ret; | |
10345 | ||
7e84e1c7 SG |
10346 | /* Restrictions allowed only if rings started disabled */ |
10347 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10348 | return -EBADFD; | |
10349 | ||
21b55dbc | 10350 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 10351 | if (ctx->restrictions.registered) |
21b55dbc SG |
10352 | return -EBUSY; |
10353 | ||
10354 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
10355 | return -EINVAL; | |
10356 | ||
10357 | size = array_size(nr_args, sizeof(*res)); | |
10358 | if (size == SIZE_MAX) | |
10359 | return -EOVERFLOW; | |
10360 | ||
10361 | res = memdup_user(arg, size); | |
10362 | if (IS_ERR(res)) | |
10363 | return PTR_ERR(res); | |
10364 | ||
10365 | ret = 0; | |
10366 | ||
10367 | for (i = 0; i < nr_args; i++) { | |
10368 | switch (res[i].opcode) { | |
10369 | case IORING_RESTRICTION_REGISTER_OP: | |
10370 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
10371 | ret = -EINVAL; | |
10372 | goto out; | |
10373 | } | |
10374 | ||
10375 | __set_bit(res[i].register_op, | |
10376 | ctx->restrictions.register_op); | |
10377 | break; | |
10378 | case IORING_RESTRICTION_SQE_OP: | |
10379 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
10380 | ret = -EINVAL; | |
10381 | goto out; | |
10382 | } | |
10383 | ||
10384 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
10385 | break; | |
10386 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
10387 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
10388 | break; | |
10389 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
10390 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
10391 | break; | |
10392 | default: | |
10393 | ret = -EINVAL; | |
10394 | goto out; | |
10395 | } | |
10396 | } | |
10397 | ||
10398 | out: | |
10399 | /* Reset all restrictions if an error happened */ | |
10400 | if (ret != 0) | |
10401 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
10402 | else | |
7e84e1c7 | 10403 | ctx->restrictions.registered = true; |
21b55dbc SG |
10404 | |
10405 | kfree(res); | |
10406 | return ret; | |
10407 | } | |
10408 | ||
7e84e1c7 SG |
10409 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10410 | { | |
10411 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10412 | return -EBADFD; | |
10413 | ||
10414 | if (ctx->restrictions.registered) | |
10415 | ctx->restricted = 1; | |
10416 | ||
0298ef96 PB |
10417 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
10418 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
10419 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
10420 | return 0; |
10421 | } | |
10422 | ||
fdecb662 | 10423 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 10424 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
10425 | unsigned nr_args) |
10426 | { | |
10427 | __u32 tmp; | |
10428 | int err; | |
10429 | ||
c3bdad02 PB |
10430 | if (up->resv) |
10431 | return -EINVAL; | |
98f0b3b4 PB |
10432 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
10433 | return -EOVERFLOW; | |
10434 | err = io_rsrc_node_switch_start(ctx); | |
10435 | if (err) | |
10436 | return err; | |
10437 | ||
fdecb662 PB |
10438 | switch (type) { |
10439 | case IORING_RSRC_FILE: | |
98f0b3b4 | 10440 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
10441 | case IORING_RSRC_BUFFER: |
10442 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
10443 | } |
10444 | return -EINVAL; | |
10445 | } | |
10446 | ||
c3bdad02 PB |
10447 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
10448 | unsigned nr_args) | |
98f0b3b4 | 10449 | { |
c3bdad02 | 10450 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
10451 | |
10452 | if (!nr_args) | |
10453 | return -EINVAL; | |
c3bdad02 PB |
10454 | memset(&up, 0, sizeof(up)); |
10455 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
10456 | return -EFAULT; | |
10457 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
10458 | } | |
10459 | ||
10460 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 10461 | unsigned size, unsigned type) |
c3bdad02 PB |
10462 | { |
10463 | struct io_uring_rsrc_update2 up; | |
10464 | ||
10465 | if (size != sizeof(up)) | |
10466 | return -EINVAL; | |
98f0b3b4 PB |
10467 | if (copy_from_user(&up, arg, sizeof(up))) |
10468 | return -EFAULT; | |
992da01a | 10469 | if (!up.nr || up.resv) |
98f0b3b4 | 10470 | return -EINVAL; |
992da01a | 10471 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
10472 | } |
10473 | ||
792e3582 | 10474 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 10475 | unsigned int size, unsigned int type) |
792e3582 PB |
10476 | { |
10477 | struct io_uring_rsrc_register rr; | |
10478 | ||
10479 | /* keep it extendible */ | |
10480 | if (size != sizeof(rr)) | |
10481 | return -EINVAL; | |
10482 | ||
10483 | memset(&rr, 0, sizeof(rr)); | |
10484 | if (copy_from_user(&rr, arg, size)) | |
10485 | return -EFAULT; | |
992da01a | 10486 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
10487 | return -EINVAL; |
10488 | ||
992da01a | 10489 | switch (type) { |
792e3582 PB |
10490 | case IORING_RSRC_FILE: |
10491 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
10492 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
10493 | case IORING_RSRC_BUFFER: |
10494 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
10495 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
10496 | } |
10497 | return -EINVAL; | |
10498 | } | |
10499 | ||
fe76421d JA |
10500 | static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg, |
10501 | unsigned len) | |
10502 | { | |
10503 | struct io_uring_task *tctx = current->io_uring; | |
10504 | cpumask_var_t new_mask; | |
10505 | int ret; | |
10506 | ||
10507 | if (!tctx || !tctx->io_wq) | |
10508 | return -EINVAL; | |
10509 | ||
10510 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
10511 | return -ENOMEM; | |
10512 | ||
10513 | cpumask_clear(new_mask); | |
10514 | if (len > cpumask_size()) | |
10515 | len = cpumask_size(); | |
10516 | ||
10517 | if (copy_from_user(new_mask, arg, len)) { | |
10518 | free_cpumask_var(new_mask); | |
10519 | return -EFAULT; | |
10520 | } | |
10521 | ||
10522 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
10523 | free_cpumask_var(new_mask); | |
10524 | return ret; | |
10525 | } | |
10526 | ||
10527 | static int io_unregister_iowq_aff(struct io_ring_ctx *ctx) | |
10528 | { | |
10529 | struct io_uring_task *tctx = current->io_uring; | |
10530 | ||
10531 | if (!tctx || !tctx->io_wq) | |
10532 | return -EINVAL; | |
10533 | ||
10534 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
10535 | } | |
10536 | ||
2e480058 JA |
10537 | static int io_register_iowq_max_workers(struct io_ring_ctx *ctx, |
10538 | void __user *arg) | |
10539 | { | |
fa84693b JA |
10540 | struct io_uring_task *tctx = NULL; |
10541 | struct io_sq_data *sqd = NULL; | |
2e480058 JA |
10542 | __u32 new_count[2]; |
10543 | int i, ret; | |
10544 | ||
2e480058 JA |
10545 | if (copy_from_user(new_count, arg, sizeof(new_count))) |
10546 | return -EFAULT; | |
10547 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
10548 | if (new_count[i] > INT_MAX) | |
10549 | return -EINVAL; | |
10550 | ||
fa84693b JA |
10551 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
10552 | sqd = ctx->sq_data; | |
10553 | if (sqd) { | |
009ad9f0 JA |
10554 | /* |
10555 | * Observe the correct sqd->lock -> ctx->uring_lock | |
10556 | * ordering. Fine to drop uring_lock here, we hold | |
10557 | * a ref to the ctx. | |
10558 | */ | |
10559 | mutex_unlock(&ctx->uring_lock); | |
fa84693b | 10560 | mutex_lock(&sqd->lock); |
009ad9f0 | 10561 | mutex_lock(&ctx->uring_lock); |
fa84693b JA |
10562 | tctx = sqd->thread->io_uring; |
10563 | } | |
10564 | } else { | |
10565 | tctx = current->io_uring; | |
10566 | } | |
10567 | ||
10568 | ret = -EINVAL; | |
10569 | if (!tctx || !tctx->io_wq) | |
10570 | goto err; | |
10571 | ||
2e480058 JA |
10572 | ret = io_wq_max_workers(tctx->io_wq, new_count); |
10573 | if (ret) | |
fa84693b JA |
10574 | goto err; |
10575 | ||
10576 | if (sqd) | |
10577 | mutex_unlock(&sqd->lock); | |
2e480058 JA |
10578 | |
10579 | if (copy_to_user(arg, new_count, sizeof(new_count))) | |
10580 | return -EFAULT; | |
10581 | ||
10582 | return 0; | |
fa84693b JA |
10583 | err: |
10584 | if (sqd) | |
10585 | mutex_unlock(&sqd->lock); | |
10586 | return ret; | |
2e480058 JA |
10587 | } |
10588 | ||
071698e1 JA |
10589 | static bool io_register_op_must_quiesce(int op) |
10590 | { | |
10591 | switch (op) { | |
bd54b6fe BM |
10592 | case IORING_REGISTER_BUFFERS: |
10593 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 10594 | case IORING_REGISTER_FILES: |
071698e1 JA |
10595 | case IORING_UNREGISTER_FILES: |
10596 | case IORING_REGISTER_FILES_UPDATE: | |
10597 | case IORING_REGISTER_PROBE: | |
10598 | case IORING_REGISTER_PERSONALITY: | |
10599 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
10600 | case IORING_REGISTER_FILES2: |
10601 | case IORING_REGISTER_FILES_UPDATE2: | |
10602 | case IORING_REGISTER_BUFFERS2: | |
10603 | case IORING_REGISTER_BUFFERS_UPDATE: | |
fe76421d JA |
10604 | case IORING_REGISTER_IOWQ_AFF: |
10605 | case IORING_UNREGISTER_IOWQ_AFF: | |
2e480058 | 10606 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
071698e1 JA |
10607 | return false; |
10608 | default: | |
10609 | return true; | |
10610 | } | |
10611 | } | |
10612 | ||
e73c5c7c PB |
10613 | static int io_ctx_quiesce(struct io_ring_ctx *ctx) |
10614 | { | |
10615 | long ret; | |
10616 | ||
10617 | percpu_ref_kill(&ctx->refs); | |
10618 | ||
10619 | /* | |
10620 | * Drop uring mutex before waiting for references to exit. If another | |
10621 | * thread is currently inside io_uring_enter() it might need to grab the | |
10622 | * uring_lock to make progress. If we hold it here across the drain | |
10623 | * wait, then we can deadlock. It's safe to drop the mutex here, since | |
10624 | * no new references will come in after we've killed the percpu ref. | |
10625 | */ | |
10626 | mutex_unlock(&ctx->uring_lock); | |
10627 | do { | |
10628 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10629 | if (!ret) | |
10630 | break; | |
10631 | ret = io_run_task_work_sig(); | |
10632 | } while (ret >= 0); | |
10633 | mutex_lock(&ctx->uring_lock); | |
10634 | ||
10635 | if (ret) | |
10636 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); | |
10637 | return ret; | |
10638 | } | |
10639 | ||
edafccee JA |
10640 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10641 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10642 | __releases(ctx->uring_lock) |
10643 | __acquires(ctx->uring_lock) | |
edafccee JA |
10644 | { |
10645 | int ret; | |
10646 | ||
35fa71a0 JA |
10647 | /* |
10648 | * We're inside the ring mutex, if the ref is already dying, then | |
10649 | * someone else killed the ctx or is already going through | |
10650 | * io_uring_register(). | |
10651 | */ | |
10652 | if (percpu_ref_is_dying(&ctx->refs)) | |
10653 | return -ENXIO; | |
10654 | ||
75c4021a PB |
10655 | if (ctx->restricted) { |
10656 | if (opcode >= IORING_REGISTER_LAST) | |
10657 | return -EINVAL; | |
10658 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10659 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10660 | return -EACCES; | |
10661 | } | |
10662 | ||
071698e1 | 10663 | if (io_register_op_must_quiesce(opcode)) { |
e73c5c7c PB |
10664 | ret = io_ctx_quiesce(ctx); |
10665 | if (ret) | |
f70865db | 10666 | return ret; |
05f3fb3c | 10667 | } |
edafccee JA |
10668 | |
10669 | switch (opcode) { | |
10670 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10671 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10672 | break; |
10673 | case IORING_UNREGISTER_BUFFERS: | |
10674 | ret = -EINVAL; | |
10675 | if (arg || nr_args) | |
10676 | break; | |
0a96bbe4 | 10677 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10678 | break; |
6b06314c | 10679 | case IORING_REGISTER_FILES: |
792e3582 | 10680 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10681 | break; |
10682 | case IORING_UNREGISTER_FILES: | |
10683 | ret = -EINVAL; | |
10684 | if (arg || nr_args) | |
10685 | break; | |
10686 | ret = io_sqe_files_unregister(ctx); | |
10687 | break; | |
c3a31e60 | 10688 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10689 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10690 | break; |
9b402849 | 10691 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10692 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10693 | ret = -EINVAL; |
10694 | if (nr_args != 1) | |
10695 | break; | |
10696 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10697 | if (ret) |
10698 | break; | |
10699 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10700 | ctx->eventfd_async = 1; | |
10701 | else | |
10702 | ctx->eventfd_async = 0; | |
9b402849 JA |
10703 | break; |
10704 | case IORING_UNREGISTER_EVENTFD: | |
10705 | ret = -EINVAL; | |
10706 | if (arg || nr_args) | |
10707 | break; | |
10708 | ret = io_eventfd_unregister(ctx); | |
10709 | break; | |
66f4af93 JA |
10710 | case IORING_REGISTER_PROBE: |
10711 | ret = -EINVAL; | |
10712 | if (!arg || nr_args > 256) | |
10713 | break; | |
10714 | ret = io_probe(ctx, arg, nr_args); | |
10715 | break; | |
071698e1 JA |
10716 | case IORING_REGISTER_PERSONALITY: |
10717 | ret = -EINVAL; | |
10718 | if (arg || nr_args) | |
10719 | break; | |
10720 | ret = io_register_personality(ctx); | |
10721 | break; | |
10722 | case IORING_UNREGISTER_PERSONALITY: | |
10723 | ret = -EINVAL; | |
10724 | if (arg) | |
10725 | break; | |
10726 | ret = io_unregister_personality(ctx, nr_args); | |
10727 | break; | |
7e84e1c7 SG |
10728 | case IORING_REGISTER_ENABLE_RINGS: |
10729 | ret = -EINVAL; | |
10730 | if (arg || nr_args) | |
10731 | break; | |
10732 | ret = io_register_enable_rings(ctx); | |
10733 | break; | |
21b55dbc SG |
10734 | case IORING_REGISTER_RESTRICTIONS: |
10735 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10736 | break; | |
992da01a PB |
10737 | case IORING_REGISTER_FILES2: |
10738 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10739 | break; | |
10740 | case IORING_REGISTER_FILES_UPDATE2: | |
10741 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10742 | IORING_RSRC_FILE); | |
10743 | break; | |
10744 | case IORING_REGISTER_BUFFERS2: | |
10745 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10746 | break; |
992da01a PB |
10747 | case IORING_REGISTER_BUFFERS_UPDATE: |
10748 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10749 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10750 | break; |
fe76421d JA |
10751 | case IORING_REGISTER_IOWQ_AFF: |
10752 | ret = -EINVAL; | |
10753 | if (!arg || !nr_args) | |
10754 | break; | |
10755 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
10756 | break; | |
10757 | case IORING_UNREGISTER_IOWQ_AFF: | |
10758 | ret = -EINVAL; | |
10759 | if (arg || nr_args) | |
10760 | break; | |
10761 | ret = io_unregister_iowq_aff(ctx); | |
10762 | break; | |
2e480058 JA |
10763 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
10764 | ret = -EINVAL; | |
10765 | if (!arg || nr_args != 2) | |
10766 | break; | |
10767 | ret = io_register_iowq_max_workers(ctx, arg); | |
10768 | break; | |
edafccee JA |
10769 | default: |
10770 | ret = -EINVAL; | |
10771 | break; | |
10772 | } | |
10773 | ||
071698e1 | 10774 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10775 | /* bring the ctx back to life */ |
05f3fb3c | 10776 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10777 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10778 | } |
edafccee JA |
10779 | return ret; |
10780 | } | |
10781 | ||
10782 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10783 | void __user *, arg, unsigned int, nr_args) | |
10784 | { | |
10785 | struct io_ring_ctx *ctx; | |
10786 | long ret = -EBADF; | |
10787 | struct fd f; | |
10788 | ||
10789 | f = fdget(fd); | |
10790 | if (!f.file) | |
10791 | return -EBADF; | |
10792 | ||
10793 | ret = -EOPNOTSUPP; | |
10794 | if (f.file->f_op != &io_uring_fops) | |
10795 | goto out_fput; | |
10796 | ||
10797 | ctx = f.file->private_data; | |
10798 | ||
b6c23dd5 PB |
10799 | io_run_task_work(); |
10800 | ||
edafccee JA |
10801 | mutex_lock(&ctx->uring_lock); |
10802 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10803 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10804 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10805 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10806 | out_fput: |
10807 | fdput(f); | |
10808 | return ret; | |
10809 | } | |
10810 | ||
2b188cc1 JA |
10811 | static int __init io_uring_init(void) |
10812 | { | |
d7f62e82 SM |
10813 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10814 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10815 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10816 | } while (0) | |
10817 | ||
10818 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10819 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10820 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10821 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10822 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10823 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10824 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10825 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10826 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10827 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10828 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10829 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10830 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10831 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10832 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10833 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10834 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10835 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10836 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10837 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10838 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10839 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10840 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10841 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10842 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10843 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10844 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10845 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10846 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 10847 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 10848 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 10849 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
b9445598 | 10850 | BUILD_BUG_SQE_ELEM(44, __u32, file_index); |
d7f62e82 | 10851 | |
b0d658ec PB |
10852 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10853 | sizeof(struct io_uring_rsrc_update)); | |
10854 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10855 | sizeof(struct io_uring_rsrc_update2)); | |
90499ad0 PB |
10856 | |
10857 | /* ->buf_index is u16 */ | |
10858 | BUILD_BUG_ON(IORING_MAX_REG_BUFFERS >= (1u << 16)); | |
10859 | ||
b0d658ec PB |
10860 | /* should fit into one byte */ |
10861 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10862 | ||
d3656344 | 10863 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10864 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
16340eab | 10865 | |
91f245d5 JA |
10866 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10867 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10868 | return 0; |
10869 | }; | |
10870 | __initcall(io_uring_init); |