]>
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; | |
0499e582 | 406 | unsigned cq_last_tm_flush; |
206aefde | 407 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 408 | |
2b188cc1 JA |
409 | struct { |
410 | spinlock_t completion_lock; | |
e94f141b | 411 | |
89850fce JA |
412 | spinlock_t timeout_lock; |
413 | ||
def596e9 | 414 | /* |
540e32a0 | 415 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
416 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
417 | * For SQPOLL, only the single threaded io_sq_thread() will | |
418 | * manipulate the list, hence no extra locking is needed there. | |
419 | */ | |
540e32a0 | 420 | struct list_head iopoll_list; |
78076bb6 JA |
421 | struct hlist_head *cancel_hash; |
422 | unsigned cancel_hash_bits; | |
915b3dde | 423 | bool poll_multi_queue; |
2b188cc1 | 424 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 425 | |
21b55dbc | 426 | struct io_restriction restrictions; |
3c1a2ead | 427 | |
b13a8918 PB |
428 | /* slow path rsrc auxilary data, used by update/register */ |
429 | struct { | |
430 | struct io_rsrc_node *rsrc_backup_node; | |
431 | struct io_mapped_ubuf *dummy_ubuf; | |
432 | struct io_rsrc_data *file_data; | |
433 | struct io_rsrc_data *buf_data; | |
434 | ||
435 | struct delayed_work rsrc_put_work; | |
436 | struct llist_head rsrc_put_llist; | |
437 | struct list_head rsrc_ref_list; | |
438 | spinlock_t rsrc_ref_lock; | |
439 | }; | |
440 | ||
3c1a2ead | 441 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
442 | struct { |
443 | #if defined(CONFIG_UNIX) | |
444 | struct socket *ring_sock; | |
445 | #endif | |
446 | /* hashed buffered write serialization */ | |
447 | struct io_wq_hash *hash_map; | |
448 | ||
449 | /* Only used for accounting purposes */ | |
450 | struct user_struct *user; | |
451 | struct mm_struct *mm_account; | |
452 | ||
453 | /* ctx exit and cancelation */ | |
9011bf9a PB |
454 | struct llist_head fallback_llist; |
455 | struct delayed_work fallback_work; | |
b986af7e PB |
456 | struct work_struct exit_work; |
457 | struct list_head tctx_list; | |
458 | struct completion ref_comp; | |
e139a1ec PB |
459 | u32 iowq_limits[2]; |
460 | bool iowq_limits_set; | |
b986af7e | 461 | }; |
2b188cc1 JA |
462 | }; |
463 | ||
53e043b2 SM |
464 | struct io_uring_task { |
465 | /* submission side */ | |
09899b19 | 466 | int cached_refs; |
53e043b2 SM |
467 | struct xarray xa; |
468 | struct wait_queue_head wait; | |
ee53fb2b SM |
469 | const struct io_ring_ctx *last; |
470 | struct io_wq *io_wq; | |
53e043b2 | 471 | struct percpu_counter inflight; |
b303fe2e | 472 | atomic_t inflight_tracked; |
53e043b2 | 473 | atomic_t in_idle; |
53e043b2 SM |
474 | |
475 | spinlock_t task_lock; | |
476 | struct io_wq_work_list task_list; | |
53e043b2 | 477 | struct callback_head task_work; |
6294f368 | 478 | bool task_running; |
53e043b2 SM |
479 | }; |
480 | ||
09bb8394 JA |
481 | /* |
482 | * First field must be the file pointer in all the | |
483 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
484 | */ | |
221c5eb2 JA |
485 | struct io_poll_iocb { |
486 | struct file *file; | |
018043be | 487 | struct wait_queue_head *head; |
221c5eb2 | 488 | __poll_t events; |
8c838788 | 489 | bool done; |
221c5eb2 | 490 | bool canceled; |
392edb45 | 491 | struct wait_queue_entry wait; |
221c5eb2 JA |
492 | }; |
493 | ||
9d805892 | 494 | struct io_poll_update { |
018043be | 495 | struct file *file; |
9d805892 PB |
496 | u64 old_user_data; |
497 | u64 new_user_data; | |
498 | __poll_t events; | |
b69de288 JA |
499 | bool update_events; |
500 | bool update_user_data; | |
018043be PB |
501 | }; |
502 | ||
b5dba59e JA |
503 | struct io_close { |
504 | struct file *file; | |
b5dba59e | 505 | int fd; |
7df778be | 506 | u32 file_slot; |
b5dba59e JA |
507 | }; |
508 | ||
ad8a48ac JA |
509 | struct io_timeout_data { |
510 | struct io_kiocb *req; | |
511 | struct hrtimer timer; | |
512 | struct timespec64 ts; | |
513 | enum hrtimer_mode mode; | |
50c1df2b | 514 | u32 flags; |
ad8a48ac JA |
515 | }; |
516 | ||
8ed8d3c3 JA |
517 | struct io_accept { |
518 | struct file *file; | |
519 | struct sockaddr __user *addr; | |
520 | int __user *addr_len; | |
521 | int flags; | |
aaa4db12 | 522 | u32 file_slot; |
09952e3e | 523 | unsigned long nofile; |
8ed8d3c3 JA |
524 | }; |
525 | ||
526 | struct io_sync { | |
527 | struct file *file; | |
528 | loff_t len; | |
529 | loff_t off; | |
530 | int flags; | |
d63d1b5e | 531 | int mode; |
8ed8d3c3 JA |
532 | }; |
533 | ||
fbf23849 JA |
534 | struct io_cancel { |
535 | struct file *file; | |
536 | u64 addr; | |
537 | }; | |
538 | ||
b29472ee JA |
539 | struct io_timeout { |
540 | struct file *file; | |
bfe68a22 PB |
541 | u32 off; |
542 | u32 target_seq; | |
135fcde8 | 543 | struct list_head list; |
90cd7e42 PB |
544 | /* head of the link, used by linked timeouts only */ |
545 | struct io_kiocb *head; | |
89b263f6 JA |
546 | /* for linked completions */ |
547 | struct io_kiocb *prev; | |
b29472ee JA |
548 | }; |
549 | ||
0bdf7a2d PB |
550 | struct io_timeout_rem { |
551 | struct file *file; | |
552 | u64 addr; | |
9c8e11b3 PB |
553 | |
554 | /* timeout update */ | |
555 | struct timespec64 ts; | |
556 | u32 flags; | |
f1042b6c | 557 | bool ltimeout; |
0bdf7a2d PB |
558 | }; |
559 | ||
9adbd45d JA |
560 | struct io_rw { |
561 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
562 | struct kiocb kiocb; | |
563 | u64 addr; | |
564 | u64 len; | |
565 | }; | |
566 | ||
3fbb51c1 JA |
567 | struct io_connect { |
568 | struct file *file; | |
569 | struct sockaddr __user *addr; | |
570 | int addr_len; | |
571 | }; | |
572 | ||
e47293fd JA |
573 | struct io_sr_msg { |
574 | struct file *file; | |
fddaface | 575 | union { |
4af3417a PB |
576 | struct compat_msghdr __user *umsg_compat; |
577 | struct user_msghdr __user *umsg; | |
578 | void __user *buf; | |
fddaface | 579 | }; |
e47293fd | 580 | int msg_flags; |
bcda7baa | 581 | int bgid; |
fddaface | 582 | size_t len; |
bcda7baa | 583 | struct io_buffer *kbuf; |
e47293fd JA |
584 | }; |
585 | ||
15b71abe JA |
586 | struct io_open { |
587 | struct file *file; | |
588 | int dfd; | |
b9445598 | 589 | u32 file_slot; |
15b71abe | 590 | struct filename *filename; |
c12cedf2 | 591 | struct open_how how; |
4022e7af | 592 | unsigned long nofile; |
15b71abe JA |
593 | }; |
594 | ||
269bbe5f | 595 | struct io_rsrc_update { |
05f3fb3c JA |
596 | struct file *file; |
597 | u64 arg; | |
598 | u32 nr_args; | |
599 | u32 offset; | |
600 | }; | |
601 | ||
4840e418 JA |
602 | struct io_fadvise { |
603 | struct file *file; | |
604 | u64 offset; | |
605 | u32 len; | |
606 | u32 advice; | |
607 | }; | |
608 | ||
c1ca757b JA |
609 | struct io_madvise { |
610 | struct file *file; | |
611 | u64 addr; | |
612 | u32 len; | |
613 | u32 advice; | |
614 | }; | |
615 | ||
3e4827b0 JA |
616 | struct io_epoll { |
617 | struct file *file; | |
618 | int epfd; | |
619 | int op; | |
620 | int fd; | |
621 | struct epoll_event event; | |
e47293fd JA |
622 | }; |
623 | ||
7d67af2c PB |
624 | struct io_splice { |
625 | struct file *file_out; | |
7d67af2c PB |
626 | loff_t off_out; |
627 | loff_t off_in; | |
628 | u64 len; | |
dfadddfc | 629 | int splice_fd_in; |
7d67af2c PB |
630 | unsigned int flags; |
631 | }; | |
632 | ||
ddf0322d JA |
633 | struct io_provide_buf { |
634 | struct file *file; | |
635 | __u64 addr; | |
38134ada | 636 | __u32 len; |
ddf0322d JA |
637 | __u32 bgid; |
638 | __u16 nbufs; | |
639 | __u16 bid; | |
640 | }; | |
641 | ||
1d9e1288 BM |
642 | struct io_statx { |
643 | struct file *file; | |
644 | int dfd; | |
645 | unsigned int mask; | |
646 | unsigned int flags; | |
e62753e4 | 647 | const char __user *filename; |
1d9e1288 BM |
648 | struct statx __user *buffer; |
649 | }; | |
650 | ||
36f4fa68 JA |
651 | struct io_shutdown { |
652 | struct file *file; | |
653 | int how; | |
654 | }; | |
655 | ||
80a261fd JA |
656 | struct io_rename { |
657 | struct file *file; | |
658 | int old_dfd; | |
659 | int new_dfd; | |
660 | struct filename *oldpath; | |
661 | struct filename *newpath; | |
662 | int flags; | |
663 | }; | |
664 | ||
14a1143b JA |
665 | struct io_unlink { |
666 | struct file *file; | |
667 | int dfd; | |
668 | int flags; | |
669 | struct filename *filename; | |
670 | }; | |
671 | ||
e34a02dc DK |
672 | struct io_mkdir { |
673 | struct file *file; | |
674 | int dfd; | |
675 | umode_t mode; | |
676 | struct filename *filename; | |
677 | }; | |
678 | ||
7a8721f8 DK |
679 | struct io_symlink { |
680 | struct file *file; | |
681 | int new_dfd; | |
682 | struct filename *oldpath; | |
683 | struct filename *newpath; | |
684 | }; | |
685 | ||
cf30da90 DK |
686 | struct io_hardlink { |
687 | struct file *file; | |
688 | int old_dfd; | |
689 | int new_dfd; | |
690 | struct filename *oldpath; | |
691 | struct filename *newpath; | |
692 | int flags; | |
693 | }; | |
694 | ||
3ca405eb PB |
695 | struct io_completion { |
696 | struct file *file; | |
8c3f9cd1 | 697 | u32 cflags; |
3ca405eb PB |
698 | }; |
699 | ||
f499a021 JA |
700 | struct io_async_connect { |
701 | struct sockaddr_storage address; | |
702 | }; | |
703 | ||
03b1230c JA |
704 | struct io_async_msghdr { |
705 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
706 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
707 | struct iovec *free_iov; | |
03b1230c JA |
708 | struct sockaddr __user *uaddr; |
709 | struct msghdr msg; | |
b537916c | 710 | struct sockaddr_storage addr; |
03b1230c JA |
711 | }; |
712 | ||
f67676d1 JA |
713 | struct io_async_rw { |
714 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
715 | const struct iovec *free_iovec; |
716 | struct iov_iter iter; | |
cd658695 | 717 | struct iov_iter_state iter_state; |
227c0c96 | 718 | size_t bytes_done; |
bcf5a063 | 719 | struct wait_page_queue wpq; |
f67676d1 JA |
720 | }; |
721 | ||
6b47ee6e PB |
722 | enum { |
723 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
724 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
725 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
726 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
727 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 728 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 729 | |
dddca226 | 730 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 731 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
732 | REQ_F_INFLIGHT_BIT, |
733 | REQ_F_CUR_POS_BIT, | |
734 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 735 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 736 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 737 | REQ_F_POLLED_BIT, |
bcda7baa | 738 | REQ_F_BUFFER_SELECTED_BIT, |
e342c807 | 739 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 740 | REQ_F_REISSUE_BIT, |
b8e64b53 | 741 | REQ_F_CREDS_BIT, |
20e60a38 | 742 | REQ_F_REFCOUNT_BIT, |
4d13d1a4 | 743 | REQ_F_ARM_LTIMEOUT_BIT, |
7b29f92d | 744 | /* keep async read/write and isreg together and in order */ |
b191e2df PB |
745 | REQ_F_NOWAIT_READ_BIT, |
746 | REQ_F_NOWAIT_WRITE_BIT, | |
7b29f92d | 747 | REQ_F_ISREG_BIT, |
84557871 JA |
748 | |
749 | /* not a real bit, just to check we're not overflowing the space */ | |
750 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
751 | }; |
752 | ||
753 | enum { | |
754 | /* ctx owns file */ | |
755 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
756 | /* drain existing IO first */ | |
757 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
758 | /* linked sqes */ | |
759 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
760 | /* doesn't sever on completion < 0 */ | |
761 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
762 | /* IOSQE_ASYNC */ | |
763 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
764 | /* IOSQE_BUFFER_SELECT */ |
765 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 766 | |
6b47ee6e | 767 | /* fail rest of links */ |
93d2bcd2 | 768 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 769 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
770 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
771 | /* read/write uses file position */ | |
772 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
773 | /* must not punt to workers */ | |
774 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 775 | /* has or had linked timeout */ |
6b47ee6e | 776 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
777 | /* needs cleanup */ |
778 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
779 | /* already went through poll handler */ |
780 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
781 | /* buffer already selected */ |
782 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
e342c807 PB |
783 | /* completion is deferred through io_comp_state */ |
784 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
785 | /* caller should reissue async */ |
786 | REQ_F_REISSUE = BIT(REQ_F_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); | |
7df778be PB |
1103 | static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags); |
1104 | ||
f1042b6c | 1105 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); |
b9445598 | 1106 | |
2b188cc1 JA |
1107 | static struct kmem_cache *req_cachep; |
1108 | ||
0918682b | 1109 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1110 | |
1111 | struct sock *io_uring_get_socket(struct file *file) | |
1112 | { | |
1113 | #if defined(CONFIG_UNIX) | |
1114 | if (file->f_op == &io_uring_fops) { | |
1115 | struct io_ring_ctx *ctx = file->private_data; | |
1116 | ||
1117 | return ctx->ring_sock->sk; | |
1118 | } | |
1119 | #endif | |
1120 | return NULL; | |
1121 | } | |
1122 | EXPORT_SYMBOL(io_uring_get_socket); | |
1123 | ||
f237c30a PB |
1124 | static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) |
1125 | { | |
1126 | if (!*locked) { | |
1127 | mutex_lock(&ctx->uring_lock); | |
1128 | *locked = true; | |
1129 | } | |
1130 | } | |
1131 | ||
f2f87370 PB |
1132 | #define io_for_each_link(pos, head) \ |
1133 | for (pos = (head); pos; pos = pos->link) | |
1134 | ||
21c843d5 PB |
1135 | /* |
1136 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1137 | * see commit f958d7b528b1 for details. | |
1138 | */ | |
1139 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1140 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1141 | ||
1142 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) | |
1143 | { | |
20e60a38 | 1144 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1145 | return atomic_inc_not_zero(&req->refs); |
1146 | } | |
1147 | ||
21c843d5 PB |
1148 | static inline bool req_ref_put_and_test(struct io_kiocb *req) |
1149 | { | |
20e60a38 PB |
1150 | if (likely(!(req->flags & REQ_F_REFCOUNT))) |
1151 | return true; | |
1152 | ||
21c843d5 PB |
1153 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1154 | return atomic_dec_and_test(&req->refs); | |
1155 | } | |
1156 | ||
1157 | static inline void req_ref_put(struct io_kiocb *req) | |
1158 | { | |
20e60a38 | 1159 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1160 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
1161 | } | |
1162 | ||
1163 | static inline void req_ref_get(struct io_kiocb *req) | |
1164 | { | |
20e60a38 | 1165 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1166 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1167 | atomic_inc(&req->refs); | |
1168 | } | |
1169 | ||
48dcd38d | 1170 | static inline void __io_req_set_refcount(struct io_kiocb *req, int nr) |
20e60a38 PB |
1171 | { |
1172 | if (!(req->flags & REQ_F_REFCOUNT)) { | |
1173 | req->flags |= REQ_F_REFCOUNT; | |
48dcd38d | 1174 | atomic_set(&req->refs, nr); |
20e60a38 PB |
1175 | } |
1176 | } | |
1177 | ||
48dcd38d PB |
1178 | static inline void io_req_set_refcount(struct io_kiocb *req) |
1179 | { | |
1180 | __io_req_set_refcount(req, 1); | |
1181 | } | |
1182 | ||
b895c9a6 | 1183 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1184 | { |
1185 | struct io_ring_ctx *ctx = req->ctx; | |
1186 | ||
269bbe5f | 1187 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1188 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1189 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1190 | } |
1191 | } | |
1192 | ||
f70865db PB |
1193 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1194 | { | |
1195 | bool got = percpu_ref_tryget(ref); | |
1196 | ||
1197 | /* already at zero, wait for ->release() */ | |
1198 | if (!got) | |
1199 | wait_for_completion(compl); | |
1200 | percpu_ref_resurrect(ref); | |
1201 | if (got) | |
1202 | percpu_ref_put(ref); | |
1203 | } | |
1204 | ||
3dd0c97a PB |
1205 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1206 | bool cancel_all) | |
f0baed8e | 1207 | __must_hold(&req->ctx->timeout_lock) |
08d23634 PB |
1208 | { |
1209 | struct io_kiocb *req; | |
1210 | ||
68207680 | 1211 | if (task && head->task != task) |
08d23634 | 1212 | return false; |
3dd0c97a | 1213 | if (cancel_all) |
08d23634 PB |
1214 | return true; |
1215 | ||
1216 | io_for_each_link(req, head) { | |
b05a1bcd | 1217 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1218 | return true; |
08d23634 PB |
1219 | } |
1220 | return false; | |
1221 | } | |
1222 | ||
f0baed8e PB |
1223 | static bool io_match_linked(struct io_kiocb *head) |
1224 | { | |
1225 | struct io_kiocb *req; | |
1226 | ||
1227 | io_for_each_link(req, head) { | |
1228 | if (req->flags & REQ_F_INFLIGHT) | |
1229 | return true; | |
1230 | } | |
1231 | return false; | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * As io_match_task() but protected against racing with linked timeouts. | |
1236 | * User must not hold timeout_lock. | |
1237 | */ | |
1238 | static bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, | |
1239 | bool cancel_all) | |
1240 | { | |
1241 | bool matched; | |
1242 | ||
1243 | if (task && head->task != task) | |
1244 | return false; | |
1245 | if (cancel_all) | |
1246 | return true; | |
1247 | ||
1248 | if (head->flags & REQ_F_LINK_TIMEOUT) { | |
1249 | struct io_ring_ctx *ctx = head->ctx; | |
1250 | ||
1251 | /* protect against races with linked timeouts */ | |
1252 | spin_lock_irq(&ctx->timeout_lock); | |
1253 | matched = io_match_linked(head); | |
1254 | spin_unlock_irq(&ctx->timeout_lock); | |
1255 | } else { | |
1256 | matched = io_match_linked(head); | |
1257 | } | |
1258 | return matched; | |
1259 | } | |
1260 | ||
93d2bcd2 | 1261 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1262 | { |
93d2bcd2 | 1263 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1264 | } |
4a38aed2 | 1265 | |
a8295b98 HX |
1266 | static inline void req_fail_link_node(struct io_kiocb *req, int res) |
1267 | { | |
1268 | req_set_fail(req); | |
1269 | req->result = res; | |
1270 | } | |
1271 | ||
2b188cc1 JA |
1272 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1273 | { | |
1274 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1275 | ||
0f158b4c | 1276 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1277 | } |
1278 | ||
8eb7e2d0 PB |
1279 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1280 | { | |
1281 | return !req->timeout.off; | |
1282 | } | |
1283 | ||
f56165e6 PB |
1284 | static void io_fallback_req_func(struct work_struct *work) |
1285 | { | |
1286 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
1287 | fallback_work.work); | |
1288 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
1289 | struct io_kiocb *req, *tmp; | |
f237c30a | 1290 | bool locked = false; |
f56165e6 PB |
1291 | |
1292 | percpu_ref_get(&ctx->refs); | |
1293 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) | |
f237c30a | 1294 | req->io_task_work.func(req, &locked); |
5636c00d | 1295 | |
f237c30a PB |
1296 | if (locked) { |
1297 | if (ctx->submit_state.compl_nr) | |
1298 | io_submit_flush_completions(ctx); | |
1299 | mutex_unlock(&ctx->uring_lock); | |
1300 | } | |
f56165e6 | 1301 | percpu_ref_put(&ctx->refs); |
f237c30a | 1302 | |
f56165e6 PB |
1303 | } |
1304 | ||
2b188cc1 JA |
1305 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1306 | { | |
1307 | struct io_ring_ctx *ctx; | |
78076bb6 | 1308 | int hash_bits; |
2b188cc1 JA |
1309 | |
1310 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1311 | if (!ctx) | |
1312 | return NULL; | |
1313 | ||
78076bb6 JA |
1314 | /* |
1315 | * Use 5 bits less than the max cq entries, that should give us around | |
1316 | * 32 entries per hash list if totally full and uniformly spread. | |
1317 | */ | |
1318 | hash_bits = ilog2(p->cq_entries); | |
1319 | hash_bits -= 5; | |
1320 | if (hash_bits <= 0) | |
1321 | hash_bits = 1; | |
1322 | ctx->cancel_hash_bits = hash_bits; | |
1323 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1324 | GFP_KERNEL); | |
1325 | if (!ctx->cancel_hash) | |
1326 | goto err; | |
1327 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1328 | ||
6224843d PB |
1329 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1330 | if (!ctx->dummy_ubuf) | |
1331 | goto err; | |
1332 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1333 | ctx->dummy_ubuf->ubuf = -1UL; | |
1334 | ||
21482896 | 1335 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1336 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1337 | goto err; | |
2b188cc1 JA |
1338 | |
1339 | ctx->flags = p->flags; | |
90554200 | 1340 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1341 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1342 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1343 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1344 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1345 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1346 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1347 | mutex_init(&ctx->uring_lock); |
311997b3 | 1348 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1349 | spin_lock_init(&ctx->completion_lock); |
89850fce | 1350 | spin_lock_init(&ctx->timeout_lock); |
540e32a0 | 1351 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1352 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1353 | INIT_LIST_HEAD(&ctx->timeout_list); |
ef9dd637 | 1354 | INIT_LIST_HEAD(&ctx->ltimeout_list); |
d67d2263 BM |
1355 | spin_lock_init(&ctx->rsrc_ref_lock); |
1356 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1357 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1358 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1359 | INIT_LIST_HEAD(&ctx->tctx_list); |
cd0ca2e0 | 1360 | INIT_LIST_HEAD(&ctx->submit_state.free_list); |
d0acdee2 | 1361 | INIT_LIST_HEAD(&ctx->locked_free_list); |
9011bf9a | 1362 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
2b188cc1 | 1363 | return ctx; |
206aefde | 1364 | err: |
6224843d | 1365 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1366 | kfree(ctx->cancel_hash); |
206aefde JA |
1367 | kfree(ctx); |
1368 | return NULL; | |
2b188cc1 JA |
1369 | } |
1370 | ||
8f6ed49a PB |
1371 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1372 | { | |
1373 | struct io_rings *r = ctx->rings; | |
1374 | ||
1375 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1376 | ctx->cq_extra--; | |
1377 | } | |
1378 | ||
9cf7c104 | 1379 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1380 | { |
2bc9930e JA |
1381 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1382 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1383 | |
8f6ed49a | 1384 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1385 | } |
de0617e4 | 1386 | |
9d858b21 | 1387 | return false; |
de0617e4 JA |
1388 | } |
1389 | ||
c97d8a0f PB |
1390 | #define FFS_ASYNC_READ 0x1UL |
1391 | #define FFS_ASYNC_WRITE 0x2UL | |
1392 | #ifdef CONFIG_64BIT | |
1393 | #define FFS_ISREG 0x4UL | |
1394 | #else | |
1395 | #define FFS_ISREG 0x0UL | |
1396 | #endif | |
1397 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
1398 | ||
1399 | static inline bool io_req_ffs_set(struct io_kiocb *req) | |
1400 | { | |
1401 | return IS_ENABLED(CONFIG_64BIT) && (req->flags & REQ_F_FIXED_FILE); | |
1402 | } | |
1403 | ||
ce3d5aae PB |
1404 | static void io_req_track_inflight(struct io_kiocb *req) |
1405 | { | |
ce3d5aae | 1406 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1407 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1408 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1409 | } |
1410 | } | |
1411 | ||
fd08e530 PB |
1412 | static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req) |
1413 | { | |
906c6caa PB |
1414 | if (WARN_ON_ONCE(!req->link)) |
1415 | return NULL; | |
1416 | ||
4d13d1a4 PB |
1417 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
1418 | req->flags |= REQ_F_LINK_TIMEOUT; | |
fd08e530 PB |
1419 | |
1420 | /* linked timeouts should have two refs once prep'ed */ | |
48dcd38d | 1421 | io_req_set_refcount(req); |
4d13d1a4 PB |
1422 | __io_req_set_refcount(req->link, 2); |
1423 | return req->link; | |
fd08e530 PB |
1424 | } |
1425 | ||
1426 | static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) | |
1427 | { | |
4d13d1a4 | 1428 | if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) |
fd08e530 PB |
1429 | return NULL; |
1430 | return __io_prep_linked_timeout(req); | |
1431 | } | |
1432 | ||
1e6fa521 JA |
1433 | static void io_prep_async_work(struct io_kiocb *req) |
1434 | { | |
1435 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1436 | struct io_ring_ctx *ctx = req->ctx; |
1437 | ||
b8e64b53 PB |
1438 | if (!(req->flags & REQ_F_CREDS)) { |
1439 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1440 | req->creds = get_current_cred(); |
b8e64b53 | 1441 | } |
003e8dcc | 1442 | |
e1d675df PB |
1443 | req->work.list.next = NULL; |
1444 | req->work.flags = 0; | |
feaadc4f PB |
1445 | if (req->flags & REQ_F_FORCE_ASYNC) |
1446 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1447 | ||
1e6fa521 JA |
1448 | if (req->flags & REQ_F_ISREG) { |
1449 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1450 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1451 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1452 | if (def->unbound_nonreg_file) |
1453 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1454 | } | |
561fb04a | 1455 | } |
cccf0ee8 | 1456 | |
cbdcb435 | 1457 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1458 | { |
cbdcb435 | 1459 | struct io_kiocb *cur; |
54a91f3b | 1460 | |
44eff40a PB |
1461 | if (req->flags & REQ_F_LINK_TIMEOUT) { |
1462 | struct io_ring_ctx *ctx = req->ctx; | |
1463 | ||
51b348e1 | 1464 | spin_lock_irq(&ctx->timeout_lock); |
44eff40a PB |
1465 | io_for_each_link(cur, req) |
1466 | io_prep_async_work(cur); | |
51b348e1 | 1467 | spin_unlock_irq(&ctx->timeout_lock); |
44eff40a PB |
1468 | } else { |
1469 | io_for_each_link(cur, req) | |
1470 | io_prep_async_work(cur); | |
1471 | } | |
561fb04a JA |
1472 | } |
1473 | ||
f237c30a | 1474 | static void io_queue_async_work(struct io_kiocb *req, bool *locked) |
561fb04a | 1475 | { |
a197f664 | 1476 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1477 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1478 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1479 | |
f237c30a PB |
1480 | /* must not take the lock, NULL it as a precaution */ |
1481 | locked = NULL; | |
1482 | ||
3bfe6106 JA |
1483 | BUG_ON(!tctx); |
1484 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1485 | |
cbdcb435 PB |
1486 | /* init ->work of the whole link before punting */ |
1487 | io_prep_async_link(req); | |
991468dc JA |
1488 | |
1489 | /* | |
1490 | * Not expected to happen, but if we do have a bug where this _can_ | |
1491 | * happen, catch it here and ensure the request is marked as | |
1492 | * canceled. That will make io-wq go through the usual work cancel | |
1493 | * procedure rather than attempt to run this request (or create a new | |
1494 | * worker for it). | |
1495 | */ | |
1496 | if (WARN_ON_ONCE(!same_thread_group(req->task, current))) | |
1497 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1498 | ||
d07f1e8a PB |
1499 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1500 | &req->work, req->flags); | |
ebf93667 | 1501 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1502 | if (link) |
1503 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1504 | } |
1505 | ||
1ee4160c | 1506 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1507 | __must_hold(&req->ctx->completion_lock) |
89850fce | 1508 | __must_hold(&req->ctx->timeout_lock) |
5262f567 | 1509 | { |
e8c2bc1f | 1510 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1511 | |
fd9c7bc5 | 1512 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
2ae2eb9d PB |
1513 | if (status) |
1514 | req_set_fail(req); | |
01cec8c1 PB |
1515 | atomic_set(&req->ctx->cq_timeouts, |
1516 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1517 | list_del_init(&req->timeout.list); |
d4d19c19 | 1518 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
91c2f697 | 1519 | io_put_req_deferred(req); |
5262f567 JA |
1520 | } |
1521 | } | |
1522 | ||
441b8a78 | 1523 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1524 | { |
441b8a78 | 1525 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1526 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1527 | struct io_defer_entry, list); | |
de0617e4 | 1528 | |
9cf7c104 | 1529 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1530 | break; |
27dc8338 | 1531 | list_del_init(&de->list); |
907d1df3 | 1532 | io_req_task_queue(de->req); |
27dc8338 | 1533 | kfree(de); |
441b8a78 | 1534 | } |
04518945 PB |
1535 | } |
1536 | ||
360428f8 | 1537 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
89850fce | 1538 | __must_hold(&ctx->completion_lock) |
de0617e4 | 1539 | { |
441b8a78 | 1540 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
7abdd4c7 | 1541 | struct io_kiocb *req, *tmp; |
f010505b | 1542 | |
79ebeaee | 1543 | spin_lock_irq(&ctx->timeout_lock); |
7abdd4c7 | 1544 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
f010505b | 1545 | u32 events_needed, events_got; |
de0617e4 | 1546 | |
8eb7e2d0 | 1547 | if (io_is_timeout_noseq(req)) |
360428f8 | 1548 | break; |
f010505b MDG |
1549 | |
1550 | /* | |
1551 | * Since seq can easily wrap around over time, subtract | |
1552 | * the last seq at which timeouts were flushed before comparing. | |
1553 | * Assuming not more than 2^31-1 events have happened since, | |
1554 | * these subtractions won't have wrapped, so we can check if | |
1555 | * target is in [last_seq, current_seq] by comparing the two. | |
1556 | */ | |
1557 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1558 | events_got = seq - ctx->cq_last_tm_flush; | |
1559 | if (events_got < events_needed) | |
360428f8 | 1560 | break; |
bfe68a22 | 1561 | |
1ee4160c | 1562 | io_kill_timeout(req, 0); |
f18ee4cf | 1563 | } |
f010505b | 1564 | ctx->cq_last_tm_flush = seq; |
79ebeaee | 1565 | spin_unlock_irq(&ctx->timeout_lock); |
360428f8 | 1566 | } |
5262f567 | 1567 | |
2335f6f5 | 1568 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1569 | { |
2335f6f5 PB |
1570 | if (ctx->off_timeout_used) |
1571 | io_flush_timeouts(ctx); | |
1572 | if (ctx->drain_active) | |
1573 | io_queue_deferred(ctx); | |
1574 | } | |
1575 | ||
1576 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) | |
1577 | { | |
1578 | if (unlikely(ctx->off_timeout_used || ctx->drain_active)) | |
1579 | __io_commit_cqring_flush(ctx); | |
ec30e04b PB |
1580 | /* order cqe stores with ring update */ |
1581 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1582 | } |
1583 | ||
90554200 JA |
1584 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1585 | { | |
1586 | struct io_rings *r = ctx->rings; | |
1587 | ||
a566c556 | 1588 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1589 | } |
1590 | ||
888aae2e PB |
1591 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1592 | { | |
1593 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1594 | } | |
1595 | ||
d068b506 | 1596 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1597 | { |
75b28aff | 1598 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1599 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1600 | |
115e12e5 SB |
1601 | /* |
1602 | * writes to the cq entry need to come after reading head; the | |
1603 | * control dependency is enough as we're using WRITE_ONCE to | |
1604 | * fill the cq entry | |
1605 | */ | |
a566c556 | 1606 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1607 | return NULL; |
1608 | ||
888aae2e | 1609 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1610 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1611 | } |
1612 | ||
f2842ab5 JA |
1613 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1614 | { | |
44c769de | 1615 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1616 | return false; |
7e55a19c SG |
1617 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1618 | return false; | |
44c769de | 1619 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1620 | } |
1621 | ||
2c5d763c JA |
1622 | /* |
1623 | * This should only get called when at least one event has been posted. | |
1624 | * Some applications rely on the eventfd notification count only changing | |
1625 | * IFF a new CQE has been added to the CQ ring. There's no depedency on | |
1626 | * 1:1 relationship between how many times this function is called (and | |
1627 | * hence the eventfd count) and number of CQEs posted to the CQ ring. | |
1628 | */ | |
b41e9852 | 1629 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1630 | { |
5fd46178 JA |
1631 | /* |
1632 | * wake_up_all() may seem excessive, but io_wake_function() and | |
1633 | * io_should_wake() handle the termination of the loop and only | |
1634 | * wake as many waiters as we need to. | |
1635 | */ | |
1636 | if (wq_has_sleeper(&ctx->cq_wait)) | |
1637 | wake_up_all(&ctx->cq_wait); | |
534ca6d6 JA |
1638 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1639 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1640 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1641 | eventfd_signal(ctx->cq_ev_fd, 1); |
3f008385 | 1642 | if (waitqueue_active(&ctx->poll_wait)) |
311997b3 | 1643 | wake_up_interruptible(&ctx->poll_wait); |
1d7bb1d5 JA |
1644 | } |
1645 | ||
80c18e4a PB |
1646 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1647 | { | |
c57a91fb PB |
1648 | /* see waitqueue_active() comment */ |
1649 | smp_mb(); | |
1650 | ||
80c18e4a | 1651 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c57a91fb | 1652 | if (waitqueue_active(&ctx->cq_wait)) |
5fd46178 | 1653 | wake_up_all(&ctx->cq_wait); |
80c18e4a PB |
1654 | } |
1655 | if (io_should_trigger_evfd(ctx)) | |
1656 | eventfd_signal(ctx->cq_ev_fd, 1); | |
3f008385 | 1657 | if (waitqueue_active(&ctx->poll_wait)) |
311997b3 | 1658 | wake_up_interruptible(&ctx->poll_wait); |
80c18e4a PB |
1659 | } |
1660 | ||
c4a2ed72 | 1661 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1662 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1663 | { |
b18032bb | 1664 | bool all_flushed, posted; |
1d7bb1d5 | 1665 | |
a566c556 | 1666 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1667 | return false; |
1d7bb1d5 | 1668 | |
b18032bb | 1669 | posted = false; |
79ebeaee | 1670 | spin_lock(&ctx->completion_lock); |
6c2450ae | 1671 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1672 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1673 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1674 | |
1d7bb1d5 JA |
1675 | if (!cqe && !force) |
1676 | break; | |
6c2450ae PB |
1677 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1678 | struct io_overflow_cqe, list); | |
1679 | if (cqe) | |
1680 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1681 | else | |
8f6ed49a PB |
1682 | io_account_cq_overflow(ctx); |
1683 | ||
b18032bb | 1684 | posted = true; |
6c2450ae PB |
1685 | list_del(&ocqe->list); |
1686 | kfree(ocqe); | |
1d7bb1d5 JA |
1687 | } |
1688 | ||
09e88404 PB |
1689 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1690 | if (all_flushed) { | |
5ed7a37d | 1691 | clear_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1692 | WRITE_ONCE(ctx->rings->sq_flags, |
1693 | ctx->rings->sq_flags & ~IORING_SQ_CQ_OVERFLOW); | |
09e88404 | 1694 | } |
46930143 | 1695 | |
b18032bb JA |
1696 | if (posted) |
1697 | io_commit_cqring(ctx); | |
79ebeaee | 1698 | spin_unlock(&ctx->completion_lock); |
b18032bb JA |
1699 | if (posted) |
1700 | io_cqring_ev_posted(ctx); | |
09e88404 | 1701 | return all_flushed; |
1d7bb1d5 JA |
1702 | } |
1703 | ||
90f67366 | 1704 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx) |
6c503150 | 1705 | { |
ca0a2651 JA |
1706 | bool ret = true; |
1707 | ||
5ed7a37d | 1708 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1709 | /* iopoll syncs against uring_lock, not completion_lock */ |
1710 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1711 | mutex_lock(&ctx->uring_lock); | |
90f67366 | 1712 | ret = __io_cqring_overflow_flush(ctx, false); |
6c503150 PB |
1713 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1714 | mutex_unlock(&ctx->uring_lock); | |
1715 | } | |
ca0a2651 JA |
1716 | |
1717 | return ret; | |
6c503150 PB |
1718 | } |
1719 | ||
6a290a14 PB |
1720 | /* must to be called somewhat shortly after putting a request */ |
1721 | static inline void io_put_task(struct task_struct *task, int nr) | |
1722 | { | |
1723 | struct io_uring_task *tctx = task->io_uring; | |
1724 | ||
e98e49b2 PB |
1725 | if (likely(task == current)) { |
1726 | tctx->cached_refs += nr; | |
1727 | } else { | |
1728 | percpu_counter_sub(&tctx->inflight, nr); | |
1729 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1730 | wake_up(&tctx->wait); | |
1731 | put_task_struct_many(task, nr); | |
1732 | } | |
6a290a14 PB |
1733 | } |
1734 | ||
9a10867a PB |
1735 | static void io_task_refs_refill(struct io_uring_task *tctx) |
1736 | { | |
1737 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
1738 | ||
1739 | percpu_counter_add(&tctx->inflight, refill); | |
1740 | refcount_add(refill, ¤t->usage); | |
1741 | tctx->cached_refs += refill; | |
1742 | } | |
1743 | ||
1744 | static inline void io_get_task_refs(int nr) | |
1745 | { | |
1746 | struct io_uring_task *tctx = current->io_uring; | |
1747 | ||
1748 | tctx->cached_refs -= nr; | |
1749 | if (unlikely(tctx->cached_refs < 0)) | |
1750 | io_task_refs_refill(tctx); | |
1751 | } | |
1752 | ||
abdebba9 PB |
1753 | static __cold void io_uring_drop_tctx_refs(struct task_struct *task) |
1754 | { | |
1755 | struct io_uring_task *tctx = task->io_uring; | |
1756 | unsigned int refs = tctx->cached_refs; | |
1757 | ||
1758 | if (refs) { | |
1759 | tctx->cached_refs = 0; | |
1760 | percpu_counter_sub(&tctx->inflight, refs); | |
1761 | put_task_struct_many(task, refs); | |
1762 | } | |
1763 | } | |
1764 | ||
d4d19c19 PB |
1765 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1766 | long res, unsigned int cflags) | |
2b188cc1 | 1767 | { |
cce4b8b0 | 1768 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1769 | |
cce4b8b0 PB |
1770 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1771 | if (!ocqe) { | |
1772 | /* | |
1773 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1774 | * or cannot allocate an overflow entry, then we need to drop it | |
1775 | * on the floor. | |
1776 | */ | |
8f6ed49a | 1777 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1778 | return false; |
2b188cc1 | 1779 | } |
cce4b8b0 | 1780 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1781 | set_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1782 | WRITE_ONCE(ctx->rings->sq_flags, |
1783 | ctx->rings->sq_flags | IORING_SQ_CQ_OVERFLOW); | |
1784 | ||
cce4b8b0 | 1785 | } |
d4d19c19 | 1786 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1787 | ocqe->cqe.res = res; |
1788 | ocqe->cqe.flags = cflags; | |
1789 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1790 | return true; | |
2b188cc1 JA |
1791 | } |
1792 | ||
d4d19c19 PB |
1793 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1794 | long res, unsigned int cflags) | |
2b188cc1 JA |
1795 | { |
1796 | struct io_uring_cqe *cqe; | |
1797 | ||
d4d19c19 | 1798 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1799 | |
2b188cc1 JA |
1800 | /* |
1801 | * If we can't get a cq entry, userspace overflowed the | |
1802 | * submission (by quite a lot). Increment the overflow count in | |
1803 | * the ring. | |
1804 | */ | |
d068b506 | 1805 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1806 | if (likely(cqe)) { |
d4d19c19 | 1807 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1808 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1809 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1810 | return true; |
2b188cc1 | 1811 | } |
d4d19c19 | 1812 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1813 | } |
1814 | ||
8d13326e | 1815 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1816 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1817 | long res, unsigned int cflags) | |
bcda7baa | 1818 | { |
d4d19c19 | 1819 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1820 | } |
1821 | ||
7a612350 PB |
1822 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1823 | unsigned int cflags) | |
2b188cc1 | 1824 | { |
78e19bbe | 1825 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 | 1826 | |
79ebeaee | 1827 | spin_lock(&ctx->completion_lock); |
d4d19c19 | 1828 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1829 | /* |
1830 | * If we're the last reference to this request, add to our locked | |
1831 | * free_list cache. | |
1832 | */ | |
de9b4cca | 1833 | if (req_ref_put_and_test(req)) { |
7a612350 | 1834 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
0756a869 | 1835 | if (req->flags & IO_DISARM_MASK) |
7a612350 PB |
1836 | io_disarm_next(req); |
1837 | if (req->link) { | |
1838 | io_req_task_queue(req->link); | |
1839 | req->link = NULL; | |
1840 | } | |
1841 | } | |
c7dae4ba JA |
1842 | io_dismantle_req(req); |
1843 | io_put_task(req->task, 1); | |
bb943b82 | 1844 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
d0acdee2 | 1845 | ctx->locked_free_nr++; |
180f829f PB |
1846 | } else { |
1847 | if (!percpu_ref_tryget(&ctx->refs)) | |
1848 | req = NULL; | |
1849 | } | |
7a612350 | 1850 | io_commit_cqring(ctx); |
79ebeaee | 1851 | spin_unlock(&ctx->completion_lock); |
7a612350 | 1852 | |
180f829f PB |
1853 | if (req) { |
1854 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1855 | percpu_ref_put(&ctx->refs); |
180f829f | 1856 | } |
229a7b63 JA |
1857 | } |
1858 | ||
4e3d9ff9 JA |
1859 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1860 | { | |
c854357b | 1861 | return req->flags & IO_REQ_CLEAN_FLAGS; |
4e3d9ff9 JA |
1862 | } |
1863 | ||
a38d68db | 1864 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1865 | unsigned int cflags) |
229a7b63 | 1866 | { |
4e3d9ff9 | 1867 | if (io_req_needs_clean(req)) |
68fb8979 | 1868 | io_clean_op(req); |
a38d68db PB |
1869 | req->result = res; |
1870 | req->compl.cflags = cflags; | |
e342c807 | 1871 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1872 | } |
1873 | ||
889fca73 PB |
1874 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1875 | long res, unsigned cflags) | |
bcda7baa | 1876 | { |
889fca73 PB |
1877 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1878 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1879 | else |
c7dae4ba | 1880 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1881 | } |
1882 | ||
a38d68db | 1883 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1884 | { |
889fca73 | 1885 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1886 | } |
1887 | ||
f41db273 PB |
1888 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1889 | { | |
93d2bcd2 | 1890 | req_set_fail(req); |
f41db273 PB |
1891 | io_req_complete_post(req, res, 0); |
1892 | } | |
1893 | ||
c6d3d9cb PB |
1894 | static void io_req_complete_fail_submit(struct io_kiocb *req) |
1895 | { | |
1896 | /* | |
1897 | * We don't submit, fail them all, for that replace hardlinks with | |
1898 | * normal links. Extra REQ_F_LINK is tolerated. | |
1899 | */ | |
1900 | req->flags &= ~REQ_F_HARDLINK; | |
1901 | req->flags |= REQ_F_LINK; | |
1902 | io_req_complete_failed(req, req->result); | |
1903 | } | |
1904 | ||
864ea921 PB |
1905 | /* |
1906 | * Don't initialise the fields below on every allocation, but do that in | |
1907 | * advance and keep them valid across allocations. | |
1908 | */ | |
1909 | static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
1910 | { | |
1911 | req->ctx = ctx; | |
1912 | req->link = NULL; | |
1913 | req->async_data = NULL; | |
1914 | /* not necessary, but safer to zero */ | |
1915 | req->result = 0; | |
1916 | } | |
1917 | ||
dac7a098 | 1918 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
cd0ca2e0 | 1919 | struct io_submit_state *state) |
dac7a098 | 1920 | { |
79ebeaee | 1921 | spin_lock(&ctx->completion_lock); |
cd0ca2e0 | 1922 | list_splice_init(&ctx->locked_free_list, &state->free_list); |
d0acdee2 | 1923 | ctx->locked_free_nr = 0; |
79ebeaee | 1924 | spin_unlock(&ctx->completion_lock); |
dac7a098 PB |
1925 | } |
1926 | ||
dd78f492 | 1927 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1928 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1929 | { |
c7dae4ba | 1930 | struct io_submit_state *state = &ctx->submit_state; |
dd78f492 | 1931 | int nr; |
0ddf92e8 | 1932 | |
c7dae4ba JA |
1933 | /* |
1934 | * If we have more than a batch's worth of requests in our IRQ side | |
1935 | * locked cache, grab the lock and move them over to our submission | |
1936 | * side cache. | |
1937 | */ | |
d0acdee2 | 1938 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
cd0ca2e0 | 1939 | io_flush_cached_locked_reqs(ctx, state); |
0ddf92e8 | 1940 | |
dd78f492 | 1941 | nr = state->free_reqs; |
cd0ca2e0 PB |
1942 | while (!list_empty(&state->free_list)) { |
1943 | struct io_kiocb *req = list_first_entry(&state->free_list, | |
bb943b82 | 1944 | struct io_kiocb, inflight_entry); |
dd78f492 | 1945 | |
bb943b82 | 1946 | list_del(&req->inflight_entry); |
dd78f492 PB |
1947 | state->reqs[nr++] = req; |
1948 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1949 | break; |
1b4c351f JA |
1950 | } |
1951 | ||
dd78f492 PB |
1952 | state->free_reqs = nr; |
1953 | return nr != 0; | |
0ddf92e8 JA |
1954 | } |
1955 | ||
5d5901a3 PB |
1956 | /* |
1957 | * A request might get retired back into the request caches even before opcode | |
1958 | * handlers and io_issue_sqe() are done with it, e.g. inline completion path. | |
1959 | * Because of that, io_alloc_req() should be called only under ->uring_lock | |
1960 | * and with extra caution to not get a request that is still worked on. | |
1961 | */ | |
e5d1bc0a | 1962 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
5d5901a3 | 1963 | __must_hold(&ctx->uring_lock) |
2b188cc1 | 1964 | { |
e5d1bc0a | 1965 | struct io_submit_state *state = &ctx->submit_state; |
864ea921 PB |
1966 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
1967 | int ret, i; | |
e5d1bc0a | 1968 | |
fe7e3257 | 1969 | BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); |
e5d1bc0a | 1970 | |
864ea921 PB |
1971 | if (likely(state->free_reqs || io_flush_cached_reqs(ctx))) |
1972 | goto got_req; | |
e5d1bc0a | 1973 | |
864ea921 PB |
1974 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1975 | state->reqs); | |
fd6fab2c | 1976 | |
864ea921 PB |
1977 | /* |
1978 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1979 | * retry single alloc to be on the safe side. | |
1980 | */ | |
1981 | if (unlikely(ret <= 0)) { | |
1982 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1983 | if (!state->reqs[0]) | |
1984 | return NULL; | |
1985 | ret = 1; | |
2b188cc1 | 1986 | } |
864ea921 PB |
1987 | |
1988 | for (i = 0; i < ret; i++) | |
1989 | io_preinit_req(state->reqs[i], ctx); | |
1990 | state->free_reqs = ret; | |
e5d1bc0a | 1991 | got_req: |
291b2821 PB |
1992 | state->free_reqs--; |
1993 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1994 | } |
1995 | ||
e1d767f0 | 1996 | static inline void io_put_file(struct file *file) |
8da11c19 | 1997 | { |
e1d767f0 | 1998 | if (file) |
8da11c19 PB |
1999 | fput(file); |
2000 | } | |
2001 | ||
4edf20f9 | 2002 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2003 | { |
094bae49 | 2004 | unsigned int flags = req->flags; |
929a3af9 | 2005 | |
3a0a6902 PB |
2006 | if (io_req_needs_clean(req)) |
2007 | io_clean_op(req); | |
e1d767f0 PB |
2008 | if (!(flags & REQ_F_FIXED_FILE)) |
2009 | io_put_file(req->file); | |
269bbe5f BM |
2010 | if (req->fixed_rsrc_refs) |
2011 | percpu_ref_put(req->fixed_rsrc_refs); | |
99ebe4ef | 2012 | if (req->async_data) { |
094bae49 | 2013 | kfree(req->async_data); |
99ebe4ef PB |
2014 | req->async_data = NULL; |
2015 | } | |
e65ef56d JA |
2016 | } |
2017 | ||
216578e5 | 2018 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2019 | { |
51a4cc11 | 2020 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2021 | |
216578e5 | 2022 | io_dismantle_req(req); |
7c660731 | 2023 | io_put_task(req->task, 1); |
c6ca97b3 | 2024 | |
79ebeaee | 2025 | spin_lock(&ctx->completion_lock); |
bb943b82 | 2026 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
c34b025f | 2027 | ctx->locked_free_nr++; |
79ebeaee | 2028 | spin_unlock(&ctx->completion_lock); |
c34b025f | 2029 | |
ecfc5177 | 2030 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
2031 | } |
2032 | ||
f2f87370 PB |
2033 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2034 | { | |
2035 | struct io_kiocb *nxt = req->link; | |
2036 | ||
2037 | req->link = nxt->link; | |
2038 | nxt->link = NULL; | |
2039 | } | |
2040 | ||
33cc89a9 PB |
2041 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
2042 | __must_hold(&req->ctx->completion_lock) | |
89b263f6 | 2043 | __must_hold(&req->ctx->timeout_lock) |
2665abfd | 2044 | { |
33cc89a9 | 2045 | struct io_kiocb *link = req->link; |
f2f87370 | 2046 | |
b97e736a | 2047 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
c9abd7ad | 2048 | struct io_timeout_data *io = link->async_data; |
7c86ffee | 2049 | |
f2f87370 | 2050 | io_remove_next_linked(req); |
90cd7e42 | 2051 | link->timeout.head = NULL; |
fd9c7bc5 | 2052 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ef9dd637 | 2053 | list_del(&link->timeout.list); |
d4d19c19 PB |
2054 | io_cqring_fill_event(link->ctx, link->user_data, |
2055 | -ECANCELED, 0); | |
91c2f697 | 2056 | io_put_req_deferred(link); |
d4729fbd | 2057 | return true; |
c9abd7ad PB |
2058 | } |
2059 | } | |
d4729fbd | 2060 | return false; |
7c86ffee PB |
2061 | } |
2062 | ||
d148ca4b | 2063 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 2064 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 2065 | { |
33cc89a9 | 2066 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 2067 | |
f2f87370 | 2068 | req->link = NULL; |
f2f87370 | 2069 | while (link) { |
a8295b98 HX |
2070 | long res = -ECANCELED; |
2071 | ||
2072 | if (link->flags & REQ_F_FAIL) | |
2073 | res = link->result; | |
2074 | ||
f2f87370 PB |
2075 | nxt = link->link; |
2076 | link->link = NULL; | |
2665abfd | 2077 | |
f2f87370 | 2078 | trace_io_uring_fail_link(req, link); |
a8295b98 | 2079 | io_cqring_fill_event(link->ctx, link->user_data, res, 0); |
91c2f697 | 2080 | io_put_req_deferred(link); |
f2f87370 | 2081 | link = nxt; |
9e645e11 | 2082 | } |
33cc89a9 | 2083 | } |
9e645e11 | 2084 | |
33cc89a9 PB |
2085 | static bool io_disarm_next(struct io_kiocb *req) |
2086 | __must_hold(&req->ctx->completion_lock) | |
2087 | { | |
2088 | bool posted = false; | |
2089 | ||
0756a869 PB |
2090 | if (req->flags & REQ_F_ARM_LTIMEOUT) { |
2091 | struct io_kiocb *link = req->link; | |
2092 | ||
906c6caa | 2093 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
0756a869 PB |
2094 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
2095 | io_remove_next_linked(req); | |
2096 | io_cqring_fill_event(link->ctx, link->user_data, | |
2097 | -ECANCELED, 0); | |
2098 | io_put_req_deferred(link); | |
2099 | posted = true; | |
2100 | } | |
2101 | } else if (req->flags & REQ_F_LINK_TIMEOUT) { | |
89b263f6 JA |
2102 | struct io_ring_ctx *ctx = req->ctx; |
2103 | ||
2104 | spin_lock_irq(&ctx->timeout_lock); | |
33cc89a9 | 2105 | posted = io_kill_linked_timeout(req); |
89b263f6 JA |
2106 | spin_unlock_irq(&ctx->timeout_lock); |
2107 | } | |
93d2bcd2 | 2108 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 2109 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
2110 | posted |= (req->link != NULL); |
2111 | io_fail_links(req); | |
2112 | } | |
2113 | return posted; | |
9e645e11 JA |
2114 | } |
2115 | ||
3fa5e0f3 | 2116 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2117 | { |
33cc89a9 | 2118 | struct io_kiocb *nxt; |
944e58bf | 2119 | |
9e645e11 JA |
2120 | /* |
2121 | * If LINK is set, we have dependent requests in this chain. If we | |
2122 | * didn't fail this request, queue the first one up, moving any other | |
2123 | * dependencies to the next request. In case of failure, fail the rest | |
2124 | * of the chain. | |
2125 | */ | |
0756a869 | 2126 | if (req->flags & IO_DISARM_MASK) { |
33cc89a9 | 2127 | struct io_ring_ctx *ctx = req->ctx; |
33cc89a9 PB |
2128 | bool posted; |
2129 | ||
79ebeaee | 2130 | spin_lock(&ctx->completion_lock); |
33cc89a9 PB |
2131 | posted = io_disarm_next(req); |
2132 | if (posted) | |
2133 | io_commit_cqring(req->ctx); | |
79ebeaee | 2134 | spin_unlock(&ctx->completion_lock); |
33cc89a9 PB |
2135 | if (posted) |
2136 | io_cqring_ev_posted(ctx); | |
f2f87370 | 2137 | } |
33cc89a9 PB |
2138 | nxt = req->link; |
2139 | req->link = NULL; | |
2140 | return nxt; | |
4d7dd462 | 2141 | } |
9e645e11 | 2142 | |
f2f87370 | 2143 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2144 | { |
cdbff982 | 2145 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2146 | return NULL; |
2147 | return __io_req_find_next(req); | |
2148 | } | |
2149 | ||
f237c30a | 2150 | static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked) |
2c32395d PB |
2151 | { |
2152 | if (!ctx) | |
2153 | return; | |
f237c30a | 2154 | if (*locked) { |
99c8bc52 HX |
2155 | if (ctx->submit_state.compl_nr) |
2156 | io_submit_flush_completions(ctx); | |
2c32395d | 2157 | mutex_unlock(&ctx->uring_lock); |
f237c30a | 2158 | *locked = false; |
2c32395d PB |
2159 | } |
2160 | percpu_ref_put(&ctx->refs); | |
2161 | } | |
2162 | ||
7cbf1722 | 2163 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2164 | { |
f237c30a | 2165 | bool locked = false; |
ebd0df2e | 2166 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
2167 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
2168 | task_work); | |
c40f6379 | 2169 | |
16f72070 | 2170 | while (1) { |
3f18407d PB |
2171 | struct io_wq_work_node *node; |
2172 | ||
8d4ad41e PB |
2173 | if (!tctx->task_list.first && locked && ctx->submit_state.compl_nr) |
2174 | io_submit_flush_completions(ctx); | |
2175 | ||
3f18407d | 2176 | spin_lock_irq(&tctx->task_lock); |
c6538be9 | 2177 | node = tctx->task_list.first; |
3f18407d | 2178 | INIT_WQ_LIST(&tctx->task_list); |
6294f368 PB |
2179 | if (!node) |
2180 | tctx->task_running = false; | |
3f18407d | 2181 | spin_unlock_irq(&tctx->task_lock); |
6294f368 PB |
2182 | if (!node) |
2183 | break; | |
3f18407d | 2184 | |
6294f368 | 2185 | do { |
3f18407d PB |
2186 | struct io_wq_work_node *next = node->next; |
2187 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2188 | io_task_work.node); | |
2189 | ||
2190 | if (req->ctx != ctx) { | |
f237c30a | 2191 | ctx_flush_and_put(ctx, &locked); |
3f18407d | 2192 | ctx = req->ctx; |
126180b9 PB |
2193 | /* if not contended, grab and improve batching */ |
2194 | locked = mutex_trylock(&ctx->uring_lock); | |
3f18407d PB |
2195 | percpu_ref_get(&ctx->refs); |
2196 | } | |
f237c30a | 2197 | req->io_task_work.func(req, &locked); |
3f18407d | 2198 | node = next; |
6294f368 PB |
2199 | } while (node); |
2200 | ||
7cbf1722 | 2201 | cond_resched(); |
3f18407d | 2202 | } |
ebd0df2e | 2203 | |
f237c30a | 2204 | ctx_flush_and_put(ctx, &locked); |
abdebba9 PB |
2205 | |
2206 | /* relaxed read is enough as only the task itself sets ->in_idle */ | |
2207 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2208 | io_uring_drop_tctx_refs(current); | |
7cbf1722 JA |
2209 | } |
2210 | ||
e09ee510 | 2211 | static void io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 2212 | { |
c15b79de | 2213 | struct task_struct *tsk = req->task; |
7cbf1722 | 2214 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 2215 | enum task_work_notify_mode notify; |
e09ee510 | 2216 | struct io_wq_work_node *node; |
0b81e80c | 2217 | unsigned long flags; |
6294f368 | 2218 | bool running; |
7cbf1722 JA |
2219 | |
2220 | WARN_ON_ONCE(!tctx); | |
2221 | ||
0b81e80c | 2222 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2223 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
6294f368 PB |
2224 | running = tctx->task_running; |
2225 | if (!running) | |
2226 | tctx->task_running = true; | |
0b81e80c | 2227 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2228 | |
2229 | /* task_work already pending, we're done */ | |
6294f368 | 2230 | if (running) |
e09ee510 | 2231 | return; |
7cbf1722 | 2232 | |
c15b79de PB |
2233 | /* |
2234 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
2235 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2236 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2237 | * will do the job. | |
2238 | */ | |
2239 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
c15b79de PB |
2240 | if (!task_work_add(tsk, &tctx->task_work, notify)) { |
2241 | wake_up_process(tsk); | |
e09ee510 | 2242 | return; |
c15b79de | 2243 | } |
2215bed9 | 2244 | |
0b81e80c | 2245 | spin_lock_irqsave(&tctx->task_lock, flags); |
6294f368 | 2246 | tctx->task_running = false; |
e09ee510 PB |
2247 | node = tctx->task_list.first; |
2248 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2249 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 2250 | |
e09ee510 PB |
2251 | while (node) { |
2252 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
2253 | node = node->next; | |
2254 | if (llist_add(&req->io_task_work.fallback_node, | |
2255 | &req->ctx->fallback_llist)) | |
2256 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
2257 | } | |
eab30c4d PB |
2258 | } |
2259 | ||
f237c30a | 2260 | static void io_req_task_cancel(struct io_kiocb *req, bool *locked) |
c40f6379 | 2261 | { |
87ceb6a6 | 2262 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2263 | |
b18a1a45 | 2264 | /* not needed for normal modes, but SQPOLL depends on it */ |
f237c30a | 2265 | io_tw_lock(ctx, locked); |
2593553a | 2266 | io_req_complete_failed(req, req->result); |
c40f6379 JA |
2267 | } |
2268 | ||
f237c30a | 2269 | static void io_req_task_submit(struct io_kiocb *req, bool *locked) |
c40f6379 JA |
2270 | { |
2271 | struct io_ring_ctx *ctx = req->ctx; | |
2272 | ||
f237c30a | 2273 | io_tw_lock(ctx, locked); |
316319e8 | 2274 | /* req->task == current here, checking PF_EXITING is safe */ |
af066f31 | 2275 | if (likely(!(req->task->flags & PF_EXITING))) |
c5eef2b9 | 2276 | __io_queue_sqe(req); |
81b6d05c | 2277 | else |
2593553a | 2278 | io_req_complete_failed(req, -EFAULT); |
c40f6379 JA |
2279 | } |
2280 | ||
2c4b8eb6 | 2281 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2282 | { |
2c4b8eb6 | 2283 | req->result = ret; |
5b0a6acc | 2284 | req->io_task_work.func = io_req_task_cancel; |
e09ee510 | 2285 | io_req_task_work_add(req); |
c40f6379 JA |
2286 | } |
2287 | ||
2c4b8eb6 | 2288 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2289 | { |
5b0a6acc | 2290 | req->io_task_work.func = io_req_task_submit; |
e09ee510 | 2291 | io_req_task_work_add(req); |
a3df7698 PB |
2292 | } |
2293 | ||
773af691 JA |
2294 | static void io_req_task_queue_reissue(struct io_kiocb *req) |
2295 | { | |
2296 | req->io_task_work.func = io_queue_async_work; | |
2297 | io_req_task_work_add(req); | |
2298 | } | |
2299 | ||
f2f87370 | 2300 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2301 | { |
9b5f7bd9 | 2302 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2303 | |
2304 | if (nxt) | |
906a8c3f | 2305 | io_req_task_queue(nxt); |
c69f8dbe JL |
2306 | } |
2307 | ||
c3524383 | 2308 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2309 | { |
c3524383 PB |
2310 | io_queue_next(req); |
2311 | __io_free_req(req); | |
2312 | } | |
8766dd51 | 2313 | |
f237c30a PB |
2314 | static void io_free_req_work(struct io_kiocb *req, bool *locked) |
2315 | { | |
2316 | io_free_req(req); | |
2317 | } | |
2318 | ||
2d6500d4 | 2319 | struct req_batch { |
5af1d13e PB |
2320 | struct task_struct *task; |
2321 | int task_refs; | |
1b4c351f | 2322 | int ctx_refs; |
2d6500d4 PB |
2323 | }; |
2324 | ||
5af1d13e PB |
2325 | static inline void io_init_req_batch(struct req_batch *rb) |
2326 | { | |
5af1d13e | 2327 | rb->task_refs = 0; |
9ae72463 | 2328 | rb->ctx_refs = 0; |
5af1d13e PB |
2329 | rb->task = NULL; |
2330 | } | |
2331 | ||
2d6500d4 PB |
2332 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2333 | struct req_batch *rb) | |
2334 | { | |
9ae72463 PB |
2335 | if (rb->ctx_refs) |
2336 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
e98e49b2 | 2337 | if (rb->task) |
e9dbe221 | 2338 | io_put_task(rb->task, rb->task_refs); |
2d6500d4 PB |
2339 | } |
2340 | ||
6ff119a6 PB |
2341 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2342 | struct io_submit_state *state) | |
2d6500d4 | 2343 | { |
f2f87370 | 2344 | io_queue_next(req); |
96670657 | 2345 | io_dismantle_req(req); |
2d6500d4 | 2346 | |
e3bc8e9d | 2347 | if (req->task != rb->task) { |
7c660731 PB |
2348 | if (rb->task) |
2349 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2350 | rb->task = req->task; |
2351 | rb->task_refs = 0; | |
5af1d13e | 2352 | } |
e3bc8e9d | 2353 | rb->task_refs++; |
9ae72463 | 2354 | rb->ctx_refs++; |
5af1d13e | 2355 | |
bd759045 | 2356 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2357 | state->reqs[state->free_reqs++] = req; |
bd759045 | 2358 | else |
cd0ca2e0 | 2359 | list_add(&req->inflight_entry, &state->free_list); |
7a743e22 PB |
2360 | } |
2361 | ||
2a2758f2 | 2362 | static void io_submit_flush_completions(struct io_ring_ctx *ctx) |
a141dd89 | 2363 | __must_hold(&ctx->uring_lock) |
905c172f | 2364 | { |
cd0ca2e0 PB |
2365 | struct io_submit_state *state = &ctx->submit_state; |
2366 | int i, nr = state->compl_nr; | |
905c172f PB |
2367 | struct req_batch rb; |
2368 | ||
79ebeaee | 2369 | spin_lock(&ctx->completion_lock); |
905c172f | 2370 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2371 | struct io_kiocb *req = state->compl_reqs[i]; |
5182ed2e | 2372 | |
d4d19c19 PB |
2373 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2374 | req->compl.cflags); | |
905c172f PB |
2375 | } |
2376 | io_commit_cqring(ctx); | |
79ebeaee | 2377 | spin_unlock(&ctx->completion_lock); |
905c172f | 2378 | io_cqring_ev_posted(ctx); |
5182ed2e PB |
2379 | |
2380 | io_init_req_batch(&rb); | |
905c172f | 2381 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2382 | struct io_kiocb *req = state->compl_reqs[i]; |
905c172f | 2383 | |
91c2f697 | 2384 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2385 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2386 | } |
2387 | ||
2388 | io_req_free_batch_finish(ctx, &rb); | |
cd0ca2e0 | 2389 | state->compl_nr = 0; |
7a743e22 PB |
2390 | } |
2391 | ||
ba816ad6 JA |
2392 | /* |
2393 | * Drop reference to request, return next in chain (if there is one) if this | |
2394 | * was the last reference to this request. | |
2395 | */ | |
0d85035a | 2396 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2397 | { |
9b5f7bd9 PB |
2398 | struct io_kiocb *nxt = NULL; |
2399 | ||
de9b4cca | 2400 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2401 | nxt = io_req_find_next(req); |
4d7dd462 | 2402 | __io_free_req(req); |
2a44f467 | 2403 | } |
9b5f7bd9 | 2404 | return nxt; |
2b188cc1 JA |
2405 | } |
2406 | ||
0d85035a | 2407 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2408 | { |
de9b4cca | 2409 | if (req_ref_put_and_test(req)) |
e65ef56d | 2410 | io_free_req(req); |
2b188cc1 JA |
2411 | } |
2412 | ||
91c2f697 | 2413 | static inline void io_put_req_deferred(struct io_kiocb *req) |
216578e5 | 2414 | { |
91c2f697 | 2415 | if (req_ref_put_and_test(req)) { |
f237c30a | 2416 | req->io_task_work.func = io_free_req_work; |
543af3a1 PB |
2417 | io_req_task_work_add(req); |
2418 | } | |
216578e5 PB |
2419 | } |
2420 | ||
6c503150 | 2421 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2422 | { |
2423 | /* See comment at the top of this file */ | |
2424 | smp_rmb(); | |
e23de15f | 2425 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2426 | } |
2427 | ||
fb5ccc98 PB |
2428 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2429 | { | |
2430 | struct io_rings *rings = ctx->rings; | |
2431 | ||
2432 | /* make sure SQ entry isn't read before tail */ | |
2433 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2434 | } | |
2435 | ||
8ff069bf | 2436 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2437 | { |
8ff069bf | 2438 | unsigned int cflags; |
e94f141b | 2439 | |
bcda7baa JA |
2440 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2441 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2442 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2443 | kfree(kbuf); |
2444 | return cflags; | |
e94f141b JA |
2445 | } |
2446 | ||
8ff069bf | 2447 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2448 | { |
4d954c25 | 2449 | struct io_buffer *kbuf; |
bcda7baa | 2450 | |
ae421d93 PB |
2451 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) |
2452 | return 0; | |
4d954c25 | 2453 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2454 | return io_put_kbuf(req, kbuf); |
2455 | } | |
2456 | ||
4c6e277c JA |
2457 | static inline bool io_run_task_work(void) |
2458 | { | |
ef98eb04 | 2459 | if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) { |
4c6e277c | 2460 | __set_current_state(TASK_RUNNING); |
ef98eb04 | 2461 | tracehook_notify_signal(); |
4c6e277c JA |
2462 | return true; |
2463 | } | |
2464 | ||
2465 | return false; | |
bcda7baa JA |
2466 | } |
2467 | ||
def596e9 JA |
2468 | /* |
2469 | * Find and free completed poll iocbs | |
2470 | */ | |
2471 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
a8576af9 | 2472 | struct list_head *done) |
def596e9 | 2473 | { |
8237e045 | 2474 | struct req_batch rb; |
def596e9 | 2475 | struct io_kiocb *req; |
bbde017a XW |
2476 | |
2477 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2478 | smp_rmb(); | |
def596e9 | 2479 | |
5af1d13e | 2480 | io_init_req_batch(&rb); |
def596e9 | 2481 | while (!list_empty(done)) { |
d21ffe7e | 2482 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2483 | list_del(&req->inflight_entry); |
2484 | ||
ae421d93 PB |
2485 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2486 | io_put_rw_kbuf(req)); | |
def596e9 JA |
2487 | (*nr_events)++; |
2488 | ||
de9b4cca | 2489 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2490 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2491 | } |
def596e9 | 2492 | |
09bb8394 | 2493 | io_commit_cqring(ctx); |
80c18e4a | 2494 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2495 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2496 | } |
2497 | ||
def596e9 | 2498 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
a8576af9 | 2499 | long min) |
def596e9 JA |
2500 | { |
2501 | struct io_kiocb *req, *tmp; | |
2502 | LIST_HEAD(done); | |
2503 | bool spin; | |
def596e9 JA |
2504 | |
2505 | /* | |
2506 | * Only spin for completions if we don't have multiple devices hanging | |
2507 | * off our complete list, and we're under the requested amount. | |
2508 | */ | |
915b3dde | 2509 | spin = !ctx->poll_multi_queue && *nr_events < min; |
def596e9 | 2510 | |
d21ffe7e | 2511 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2512 | struct kiocb *kiocb = &req->rw.kiocb; |
a2416e1e | 2513 | int ret; |
def596e9 JA |
2514 | |
2515 | /* | |
581f9810 BM |
2516 | * Move completed and retryable entries to our local lists. |
2517 | * If we find a request that requires polling, break out | |
2518 | * and complete those lists first, if we have entries there. | |
def596e9 | 2519 | */ |
65a6543d | 2520 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2521 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2522 | continue; |
2523 | } | |
2524 | if (!list_empty(&done)) | |
2525 | break; | |
2526 | ||
2527 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
a2416e1e PB |
2528 | if (unlikely(ret < 0)) |
2529 | return ret; | |
2530 | else if (ret) | |
2531 | spin = false; | |
def596e9 | 2532 | |
3aadc23e PB |
2533 | /* iopoll may have completed current req */ |
2534 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2535 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2536 | } |
2537 | ||
2538 | if (!list_empty(&done)) | |
a8576af9 | 2539 | io_iopoll_complete(ctx, nr_events, &done); |
def596e9 | 2540 | |
a2416e1e | 2541 | return 0; |
def596e9 JA |
2542 | } |
2543 | ||
def596e9 JA |
2544 | /* |
2545 | * We can't just wait for polled events to come to us, we have to actively | |
2546 | * find and complete them. | |
2547 | */ | |
b2edc0a7 | 2548 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2549 | { |
2550 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2551 | return; | |
2552 | ||
2553 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2554 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2555 | unsigned int nr_events = 0; |
2556 | ||
a8576af9 | 2557 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2558 | |
b2edc0a7 PB |
2559 | /* let it sleep and repeat later if can't complete a request */ |
2560 | if (nr_events == 0) | |
2561 | break; | |
08f5439f JA |
2562 | /* |
2563 | * Ensure we allow local-to-the-cpu processing to take place, | |
2564 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2565 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2566 | */ |
3fcee5a6 PB |
2567 | if (need_resched()) { |
2568 | mutex_unlock(&ctx->uring_lock); | |
2569 | cond_resched(); | |
2570 | mutex_lock(&ctx->uring_lock); | |
2571 | } | |
def596e9 JA |
2572 | } |
2573 | mutex_unlock(&ctx->uring_lock); | |
2574 | } | |
2575 | ||
7668b92a | 2576 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2577 | { |
7668b92a | 2578 | unsigned int nr_events = 0; |
e9979b36 | 2579 | int ret = 0; |
500f9fba | 2580 | |
c7849be9 XW |
2581 | /* |
2582 | * We disallow the app entering submit/complete with polling, but we | |
2583 | * still need to lock the ring to prevent racing with polled issue | |
2584 | * that got punted to a workqueue. | |
2585 | */ | |
2586 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2587 | /* |
2588 | * Don't enter poll loop if we already have events pending. | |
2589 | * If we do, we can potentially be spinning for commands that | |
2590 | * already triggered a CQE (eg in error). | |
2591 | */ | |
5ed7a37d | 2592 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2593 | __io_cqring_overflow_flush(ctx, false); |
2594 | if (io_cqring_events(ctx)) | |
2595 | goto out; | |
def596e9 | 2596 | do { |
500f9fba JA |
2597 | /* |
2598 | * If a submit got punted to a workqueue, we can have the | |
2599 | * application entering polling for a command before it gets | |
2600 | * issued. That app will hold the uring_lock for the duration | |
2601 | * of the poll right here, so we need to take a breather every | |
2602 | * now and then to ensure that the issue has a chance to add | |
2603 | * the poll to the issued list. Otherwise we can spin here | |
2604 | * forever, while the workqueue is stuck trying to acquire the | |
2605 | * very same mutex. | |
2606 | */ | |
e9979b36 | 2607 | if (list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
2608 | u32 tail = ctx->cached_cq_tail; |
2609 | ||
500f9fba | 2610 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2611 | io_run_task_work(); |
500f9fba | 2612 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2613 | |
8f487ef2 PB |
2614 | /* some requests don't go through iopoll_list */ |
2615 | if (tail != ctx->cached_cq_tail || | |
2616 | list_empty(&ctx->iopoll_list)) | |
e9979b36 | 2617 | break; |
500f9fba | 2618 | } |
a8576af9 | 2619 | ret = io_do_iopoll(ctx, &nr_events, min); |
f39c8a5b PB |
2620 | } while (!ret && nr_events < min && !need_resched()); |
2621 | out: | |
500f9fba | 2622 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2623 | return ret; |
2624 | } | |
2625 | ||
491381ce | 2626 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2627 | { |
491381ce JA |
2628 | /* |
2629 | * Tell lockdep we inherited freeze protection from submission | |
2630 | * thread. | |
2631 | */ | |
2632 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2633 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2634 | |
1c98679d PB |
2635 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2636 | sb_end_write(sb); | |
2b188cc1 JA |
2637 | } |
2638 | } | |
2639 | ||
b63534c4 | 2640 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2641 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2642 | { |
ab454438 | 2643 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2644 | |
ab454438 PB |
2645 | if (!rw) |
2646 | return !io_req_prep_async(req); | |
cd658695 | 2647 | iov_iter_restore(&rw->iter, &rw->iter_state); |
ab454438 | 2648 | return true; |
b63534c4 | 2649 | } |
b63534c4 | 2650 | |
3e6a0d3c | 2651 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2652 | { |
355afaeb | 2653 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2654 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2655 | |
355afaeb JA |
2656 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2657 | return false; | |
3e6a0d3c JA |
2658 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2659 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2660 | return false; |
7c977a58 JA |
2661 | /* |
2662 | * If ref is dying, we might be running poll reap from the exit work. | |
2663 | * Don't attempt to reissue from that path, just let it fail with | |
2664 | * -EAGAIN. | |
2665 | */ | |
3e6a0d3c JA |
2666 | if (percpu_ref_is_dying(&ctx->refs)) |
2667 | return false; | |
ef046888 JA |
2668 | /* |
2669 | * Play it safe and assume not safe to re-import and reissue if we're | |
2670 | * not in the original thread group (or in task context). | |
2671 | */ | |
2672 | if (!same_thread_group(req->task, current) || !in_task()) | |
2673 | return false; | |
3e6a0d3c JA |
2674 | return true; |
2675 | } | |
e82ad485 | 2676 | #else |
a1ff1e3f | 2677 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2678 | { |
2679 | return false; | |
2680 | } | |
e82ad485 | 2681 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2682 | { |
b63534c4 JA |
2683 | return false; |
2684 | } | |
3e6a0d3c | 2685 | #endif |
b63534c4 | 2686 | |
8ef12efe | 2687 | static bool __io_complete_rw_common(struct io_kiocb *req, long res) |
a1d7c393 | 2688 | { |
b65c128f PB |
2689 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2690 | kiocb_end_write(req); | |
9532b99b PB |
2691 | if (res != req->result) { |
2692 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2693 | io_rw_should_reissue(req)) { | |
2694 | req->flags |= REQ_F_REISSUE; | |
8ef12efe | 2695 | return true; |
9532b99b | 2696 | } |
93d2bcd2 | 2697 | req_set_fail(req); |
8ef12efe | 2698 | req->result = res; |
9532b99b | 2699 | } |
8ef12efe JA |
2700 | return false; |
2701 | } | |
2702 | ||
f237c30a | 2703 | static void io_req_task_complete(struct io_kiocb *req, bool *locked) |
8ef12efe | 2704 | { |
126180b9 PB |
2705 | unsigned int cflags = io_put_rw_kbuf(req); |
2706 | long res = req->result; | |
2707 | ||
2708 | if (*locked) { | |
2709 | struct io_ring_ctx *ctx = req->ctx; | |
2710 | struct io_submit_state *state = &ctx->submit_state; | |
2711 | ||
2712 | io_req_complete_state(req, res, cflags); | |
2713 | state->compl_reqs[state->compl_nr++] = req; | |
2714 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2715 | io_submit_flush_completions(ctx); | |
2716 | } else { | |
2717 | io_req_complete_post(req, res, cflags); | |
2718 | } | |
8ef12efe JA |
2719 | } |
2720 | ||
2721 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, | |
2722 | unsigned int issue_flags) | |
2723 | { | |
2724 | if (__io_complete_rw_common(req, res)) | |
2725 | return; | |
63637853 | 2726 | __io_req_complete(req, issue_flags, req->result, io_put_rw_kbuf(req)); |
ba816ad6 JA |
2727 | } |
2728 | ||
2729 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2730 | { | |
9adbd45d | 2731 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2732 | |
8ef12efe JA |
2733 | if (__io_complete_rw_common(req, res)) |
2734 | return; | |
2735 | req->result = res; | |
2736 | req->io_task_work.func = io_req_task_complete; | |
2737 | io_req_task_work_add(req); | |
2b188cc1 JA |
2738 | } |
2739 | ||
def596e9 JA |
2740 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2741 | { | |
9adbd45d | 2742 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2743 | |
491381ce JA |
2744 | if (kiocb->ki_flags & IOCB_WRITE) |
2745 | kiocb_end_write(req); | |
9532b99b | 2746 | if (unlikely(res != req->result)) { |
b66ceaf3 PB |
2747 | if (res == -EAGAIN && io_rw_should_reissue(req)) { |
2748 | req->flags |= REQ_F_REISSUE; | |
2749 | return; | |
9532b99b | 2750 | } |
8c130827 | 2751 | } |
bbde017a XW |
2752 | |
2753 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2754 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2755 | smp_wmb(); |
2756 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2757 | } |
2758 | ||
2759 | /* | |
2760 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2761 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2762 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2763 | * accessing the kiocb cookie. |
2764 | */ | |
cb3d8972 | 2765 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2766 | { |
2767 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2768 | const bool in_async = io_wq_current_is_worker(); |
2769 | ||
2770 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2771 | if (unlikely(in_async)) | |
2772 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2773 | |
2774 | /* | |
2775 | * Track whether we have multiple files in our lists. This will impact | |
2776 | * how we do polling eventually, not spinning if we're on potentially | |
2777 | * different devices. | |
2778 | */ | |
540e32a0 | 2779 | if (list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
2780 | ctx->poll_multi_queue = false; |
2781 | } else if (!ctx->poll_multi_queue) { | |
def596e9 | 2782 | struct io_kiocb *list_req; |
915b3dde | 2783 | unsigned int queue_num0, queue_num1; |
def596e9 | 2784 | |
540e32a0 | 2785 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2786 | inflight_entry); |
915b3dde HX |
2787 | |
2788 | if (list_req->file != req->file) { | |
2789 | ctx->poll_multi_queue = true; | |
2790 | } else { | |
2791 | queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie); | |
2792 | queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie); | |
2793 | if (queue_num0 != queue_num1) | |
2794 | ctx->poll_multi_queue = true; | |
2795 | } | |
def596e9 JA |
2796 | } |
2797 | ||
2798 | /* | |
2799 | * For fast devices, IO may have already completed. If it has, add | |
2800 | * it to the front so we find it first. | |
2801 | */ | |
65a6543d | 2802 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2803 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2804 | else |
d21ffe7e | 2805 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2806 | |
cb3d8972 PB |
2807 | if (unlikely(in_async)) { |
2808 | /* | |
2809 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2810 | * in sq thread task context or in io worker task context. If | |
2811 | * current task context is sq thread, we don't need to check | |
2812 | * whether should wake up sq thread. | |
2813 | */ | |
2814 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2815 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2816 | wake_up(&ctx->sq_data->wait); | |
2817 | ||
2818 | mutex_unlock(&ctx->uring_lock); | |
2819 | } | |
def596e9 JA |
2820 | } |
2821 | ||
4503b767 JA |
2822 | static bool io_bdev_nowait(struct block_device *bdev) |
2823 | { | |
9ba0d0c8 | 2824 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2825 | } |
2826 | ||
2b188cc1 JA |
2827 | /* |
2828 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2829 | * any file. For now, just ensure that anything potentially problematic is done | |
2830 | * inline. | |
2831 | */ | |
b191e2df | 2832 | static bool __io_file_supports_nowait(struct file *file, int rw) |
2b188cc1 JA |
2833 | { |
2834 | umode_t mode = file_inode(file)->i_mode; | |
2835 | ||
4503b767 | 2836 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2837 | if (IS_ENABLED(CONFIG_BLOCK) && |
2838 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2839 | return true; |
2840 | return false; | |
2841 | } | |
976517f1 | 2842 | if (S_ISSOCK(mode)) |
2b188cc1 | 2843 | return true; |
4503b767 | 2844 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2845 | if (IS_ENABLED(CONFIG_BLOCK) && |
2846 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2847 | file->f_op != &io_uring_fops) |
2848 | return true; | |
2849 | return false; | |
2850 | } | |
2b188cc1 | 2851 | |
c5b85625 JA |
2852 | /* any ->read/write should understand O_NONBLOCK */ |
2853 | if (file->f_flags & O_NONBLOCK) | |
2854 | return true; | |
2855 | ||
af197f50 JA |
2856 | if (!(file->f_mode & FMODE_NOWAIT)) |
2857 | return false; | |
2858 | ||
2859 | if (rw == READ) | |
2860 | return file->f_op->read_iter != NULL; | |
2861 | ||
2862 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2863 | } |
2864 | ||
b191e2df | 2865 | static bool io_file_supports_nowait(struct io_kiocb *req, int rw) |
7b29f92d | 2866 | { |
b191e2df | 2867 | if (rw == READ && (req->flags & REQ_F_NOWAIT_READ)) |
7b29f92d | 2868 | return true; |
b191e2df | 2869 | else if (rw == WRITE && (req->flags & REQ_F_NOWAIT_WRITE)) |
7b29f92d JA |
2870 | return true; |
2871 | ||
b191e2df | 2872 | return __io_file_supports_nowait(req->file, rw); |
7b29f92d JA |
2873 | } |
2874 | ||
5d329e12 JA |
2875 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2876 | int rw) | |
2b188cc1 | 2877 | { |
def596e9 | 2878 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2879 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2880 | struct file *file = req->file; |
09bb8394 JA |
2881 | unsigned ioprio; |
2882 | int ret; | |
2b188cc1 | 2883 | |
c97d8a0f | 2884 | if (!io_req_ffs_set(req) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2885 | req->flags |= REQ_F_ISREG; |
2886 | ||
2b188cc1 | 2887 | kiocb->ki_pos = READ_ONCE(sqe->off); |
0c840c47 JA |
2888 | if (kiocb->ki_pos == -1) { |
2889 | if (!(file->f_mode & FMODE_STREAM)) { | |
2890 | req->flags |= REQ_F_CUR_POS; | |
2891 | kiocb->ki_pos = file->f_pos; | |
2892 | } else { | |
2893 | kiocb->ki_pos = 0; | |
2894 | } | |
ba04291e | 2895 | } |
2b188cc1 | 2896 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2897 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2898 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2899 | if (unlikely(ret)) | |
2900 | return ret; | |
2b188cc1 | 2901 | |
5d329e12 JA |
2902 | /* |
2903 | * If the file is marked O_NONBLOCK, still allow retry for it if it | |
2904 | * supports async. Otherwise it's impossible to use O_NONBLOCK files | |
2905 | * reliably. If not, or it IOCB_NOWAIT is set, don't retry. | |
2906 | */ | |
2907 | if ((kiocb->ki_flags & IOCB_NOWAIT) || | |
2908 | ((file->f_flags & O_NONBLOCK) && !io_file_supports_nowait(req, rw))) | |
75c668cd PB |
2909 | req->flags |= REQ_F_NOWAIT; |
2910 | ||
2b188cc1 JA |
2911 | ioprio = READ_ONCE(sqe->ioprio); |
2912 | if (ioprio) { | |
2913 | ret = ioprio_check_cap(ioprio); | |
2914 | if (ret) | |
09bb8394 | 2915 | return ret; |
2b188cc1 JA |
2916 | |
2917 | kiocb->ki_ioprio = ioprio; | |
2918 | } else | |
2919 | kiocb->ki_ioprio = get_current_ioprio(); | |
2920 | ||
def596e9 | 2921 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2922 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2923 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2924 | return -EOPNOTSUPP; |
2b188cc1 | 2925 | |
394918eb | 2926 | kiocb->ki_flags |= IOCB_HIPRI | IOCB_ALLOC_CACHE; |
def596e9 | 2927 | kiocb->ki_complete = io_complete_rw_iopoll; |
65a6543d | 2928 | req->iopoll_completed = 0; |
def596e9 | 2929 | } else { |
09bb8394 JA |
2930 | if (kiocb->ki_flags & IOCB_HIPRI) |
2931 | return -EINVAL; | |
def596e9 JA |
2932 | kiocb->ki_complete = io_complete_rw; |
2933 | } | |
9adbd45d | 2934 | |
eae071c9 PB |
2935 | if (req->opcode == IORING_OP_READ_FIXED || |
2936 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2937 | req->imu = NULL; | |
2938 | io_req_set_rsrc_node(req); | |
2939 | } | |
2940 | ||
3529d8c2 JA |
2941 | req->rw.addr = READ_ONCE(sqe->addr); |
2942 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2943 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2944 | return 0; |
2b188cc1 JA |
2945 | } |
2946 | ||
2947 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2948 | { | |
2949 | switch (ret) { | |
2950 | case -EIOCBQUEUED: | |
2951 | break; | |
2952 | case -ERESTARTSYS: | |
2953 | case -ERESTARTNOINTR: | |
2954 | case -ERESTARTNOHAND: | |
2955 | case -ERESTART_RESTARTBLOCK: | |
2956 | /* | |
2957 | * We can't just restart the syscall, since previously | |
2958 | * submitted sqes may already be in progress. Just fail this | |
2959 | * IO with EINTR. | |
2960 | */ | |
2961 | ret = -EINTR; | |
df561f66 | 2962 | fallthrough; |
2b188cc1 JA |
2963 | default: |
2964 | kiocb->ki_complete(kiocb, ret, 0); | |
2965 | } | |
2966 | } | |
2967 | ||
a1d7c393 | 2968 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2969 | unsigned int issue_flags) |
ba816ad6 | 2970 | { |
ba04291e | 2971 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2972 | struct io_async_rw *io = req->async_data; |
ba04291e | 2973 | |
227c0c96 | 2974 | /* add previously done IO, if any */ |
e8c2bc1f | 2975 | if (io && io->bytes_done > 0) { |
227c0c96 | 2976 | if (ret < 0) |
e8c2bc1f | 2977 | ret = io->bytes_done; |
227c0c96 | 2978 | else |
e8c2bc1f | 2979 | ret += io->bytes_done; |
227c0c96 JA |
2980 | } |
2981 | ||
ba04291e JA |
2982 | if (req->flags & REQ_F_CUR_POS) |
2983 | req->file->f_pos = kiocb->ki_pos; | |
b66ceaf3 | 2984 | if (ret >= 0 && (kiocb->ki_complete == io_complete_rw)) |
889fca73 | 2985 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2986 | else |
2987 | io_rw_done(kiocb, ret); | |
97284637 | 2988 | |
b66ceaf3 | 2989 | if (req->flags & REQ_F_REISSUE) { |
97284637 | 2990 | req->flags &= ~REQ_F_REISSUE; |
a7be7c23 | 2991 | if (io_resubmit_prep(req)) { |
773af691 | 2992 | io_req_task_queue_reissue(req); |
8c130827 | 2993 | } else { |
b66ceaf3 PB |
2994 | unsigned int cflags = io_put_rw_kbuf(req); |
2995 | struct io_ring_ctx *ctx = req->ctx; | |
2996 | ||
93d2bcd2 | 2997 | req_set_fail(req); |
14cfbb7a | 2998 | if (!(issue_flags & IO_URING_F_NONBLOCK)) { |
b66ceaf3 PB |
2999 | mutex_lock(&ctx->uring_lock); |
3000 | __io_req_complete(req, issue_flags, ret, cflags); | |
3001 | mutex_unlock(&ctx->uring_lock); | |
3002 | } else { | |
3003 | __io_req_complete(req, issue_flags, ret, cflags); | |
3004 | } | |
97284637 PB |
3005 | } |
3006 | } | |
ba816ad6 JA |
3007 | } |
3008 | ||
eae071c9 PB |
3009 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
3010 | struct io_mapped_ubuf *imu) | |
edafccee | 3011 | { |
9adbd45d | 3012 | size_t len = req->rw.len; |
75769e3f | 3013 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 3014 | size_t offset; |
edafccee | 3015 | |
75769e3f | 3016 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
3017 | return -EFAULT; |
3018 | /* not inside the mapped region */ | |
4751f53d | 3019 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
3020 | return -EFAULT; |
3021 | ||
3022 | /* | |
3023 | * May not be a start of buffer, set size appropriately | |
3024 | * and advance us to the beginning. | |
3025 | */ | |
3026 | offset = buf_addr - imu->ubuf; | |
3027 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3028 | |
3029 | if (offset) { | |
3030 | /* | |
3031 | * Don't use iov_iter_advance() here, as it's really slow for | |
3032 | * using the latter parts of a big fixed buffer - it iterates | |
3033 | * over each segment manually. We can cheat a bit here, because | |
3034 | * we know that: | |
3035 | * | |
3036 | * 1) it's a BVEC iter, we set it up | |
3037 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3038 | * first and last bvec | |
3039 | * | |
3040 | * So just find our index, and adjust the iterator afterwards. | |
3041 | * If the offset is within the first bvec (or the whole first | |
3042 | * bvec, just use iov_iter_advance(). This makes it easier | |
3043 | * since we can just skip the first segment, which may not | |
3044 | * be PAGE_SIZE aligned. | |
3045 | */ | |
3046 | const struct bio_vec *bvec = imu->bvec; | |
3047 | ||
3048 | if (offset <= bvec->bv_len) { | |
3049 | iov_iter_advance(iter, offset); | |
3050 | } else { | |
3051 | unsigned long seg_skip; | |
3052 | ||
3053 | /* skip first vec */ | |
3054 | offset -= bvec->bv_len; | |
3055 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3056 | ||
3057 | iter->bvec = bvec + seg_skip; | |
3058 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3059 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3060 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3061 | } |
3062 | } | |
3063 | ||
847595de | 3064 | return 0; |
edafccee JA |
3065 | } |
3066 | ||
eae071c9 PB |
3067 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
3068 | { | |
3069 | struct io_ring_ctx *ctx = req->ctx; | |
3070 | struct io_mapped_ubuf *imu = req->imu; | |
3071 | u16 index, buf_index = req->buf_index; | |
3072 | ||
3073 | if (likely(!imu)) { | |
3074 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
3075 | return -EFAULT; | |
3076 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
3077 | imu = READ_ONCE(ctx->user_bufs[index]); | |
3078 | req->imu = imu; | |
3079 | } | |
3080 | return __io_import_fixed(req, rw, iter, imu); | |
3081 | } | |
3082 | ||
bcda7baa JA |
3083 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3084 | { | |
3085 | if (needs_lock) | |
3086 | mutex_unlock(&ctx->uring_lock); | |
3087 | } | |
3088 | ||
3089 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3090 | { | |
3091 | /* | |
3092 | * "Normal" inline submissions always hold the uring_lock, since we | |
3093 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3094 | * The only exception is when we've detached the request and issue it | |
3095 | * from an async worker thread, grab the lock for that case. | |
3096 | */ | |
3097 | if (needs_lock) | |
3098 | mutex_lock(&ctx->uring_lock); | |
3099 | } | |
3100 | ||
3101 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3102 | int bgid, struct io_buffer *kbuf, | |
3103 | bool needs_lock) | |
3104 | { | |
3105 | struct io_buffer *head; | |
3106 | ||
3107 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3108 | return kbuf; | |
3109 | ||
3110 | io_ring_submit_lock(req->ctx, needs_lock); | |
3111 | ||
3112 | lockdep_assert_held(&req->ctx->uring_lock); | |
3113 | ||
9e15c3a0 | 3114 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
3115 | if (head) { |
3116 | if (!list_empty(&head->list)) { | |
3117 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3118 | list); | |
3119 | list_del(&kbuf->list); | |
3120 | } else { | |
3121 | kbuf = head; | |
9e15c3a0 | 3122 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
3123 | } |
3124 | if (*len > kbuf->len) | |
3125 | *len = kbuf->len; | |
3126 | } else { | |
3127 | kbuf = ERR_PTR(-ENOBUFS); | |
3128 | } | |
3129 | ||
3130 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3131 | ||
3132 | return kbuf; | |
3133 | } | |
3134 | ||
4d954c25 JA |
3135 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3136 | bool needs_lock) | |
3137 | { | |
3138 | struct io_buffer *kbuf; | |
4f4eeba8 | 3139 | u16 bgid; |
4d954c25 JA |
3140 | |
3141 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3142 | bgid = req->buf_index; |
4d954c25 JA |
3143 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3144 | if (IS_ERR(kbuf)) | |
3145 | return kbuf; | |
3146 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3147 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3148 | return u64_to_user_ptr(kbuf->addr); | |
3149 | } | |
3150 | ||
3151 | #ifdef CONFIG_COMPAT | |
3152 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3153 | bool needs_lock) | |
3154 | { | |
3155 | struct compat_iovec __user *uiov; | |
3156 | compat_ssize_t clen; | |
3157 | void __user *buf; | |
3158 | ssize_t len; | |
3159 | ||
3160 | uiov = u64_to_user_ptr(req->rw.addr); | |
3161 | if (!access_ok(uiov, sizeof(*uiov))) | |
3162 | return -EFAULT; | |
3163 | if (__get_user(clen, &uiov->iov_len)) | |
3164 | return -EFAULT; | |
3165 | if (clen < 0) | |
3166 | return -EINVAL; | |
3167 | ||
3168 | len = clen; | |
3169 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3170 | if (IS_ERR(buf)) | |
3171 | return PTR_ERR(buf); | |
3172 | iov[0].iov_base = buf; | |
3173 | iov[0].iov_len = (compat_size_t) len; | |
3174 | return 0; | |
3175 | } | |
3176 | #endif | |
3177 | ||
3178 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3179 | bool needs_lock) | |
3180 | { | |
3181 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3182 | void __user *buf; | |
3183 | ssize_t len; | |
3184 | ||
3185 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3186 | return -EFAULT; | |
3187 | ||
3188 | len = iov[0].iov_len; | |
3189 | if (len < 0) | |
3190 | return -EINVAL; | |
3191 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3192 | if (IS_ERR(buf)) | |
3193 | return PTR_ERR(buf); | |
3194 | iov[0].iov_base = buf; | |
3195 | iov[0].iov_len = len; | |
3196 | return 0; | |
3197 | } | |
3198 | ||
3199 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3200 | bool needs_lock) | |
3201 | { | |
dddb3e26 JA |
3202 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3203 | struct io_buffer *kbuf; | |
3204 | ||
3205 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3206 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3207 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3208 | return 0; |
dddb3e26 | 3209 | } |
dd201662 | 3210 | if (req->rw.len != 1) |
4d954c25 JA |
3211 | return -EINVAL; |
3212 | ||
3213 | #ifdef CONFIG_COMPAT | |
3214 | if (req->ctx->compat) | |
3215 | return io_compat_import(req, iov, needs_lock); | |
3216 | #endif | |
3217 | ||
3218 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3219 | } | |
3220 | ||
847595de PB |
3221 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3222 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3223 | { |
9adbd45d JA |
3224 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3225 | size_t sqe_len = req->rw.len; | |
847595de | 3226 | u8 opcode = req->opcode; |
4d954c25 | 3227 | ssize_t ret; |
edafccee | 3228 | |
7d009165 | 3229 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3230 | *iovec = NULL; |
9adbd45d | 3231 | return io_import_fixed(req, rw, iter); |
edafccee | 3232 | } |
2b188cc1 | 3233 | |
bcda7baa | 3234 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3235 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3236 | return -EINVAL; |
3237 | ||
3a6820f2 | 3238 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3239 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3240 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3241 | if (IS_ERR(buf)) |
4d954c25 | 3242 | return PTR_ERR(buf); |
3f9d6441 | 3243 | req->rw.len = sqe_len; |
bcda7baa JA |
3244 | } |
3245 | ||
3a6820f2 JA |
3246 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3247 | *iovec = NULL; | |
10fc72e4 | 3248 | return ret; |
3a6820f2 JA |
3249 | } |
3250 | ||
4d954c25 JA |
3251 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3252 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3253 | if (!ret) |
3254 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3255 | *iovec = NULL; |
3256 | return ret; | |
3257 | } | |
3258 | ||
89cd35c5 CH |
3259 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3260 | req->ctx->compat); | |
2b188cc1 JA |
3261 | } |
3262 | ||
0fef9483 JA |
3263 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3264 | { | |
5b09e37e | 3265 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3266 | } |
3267 | ||
31b51510 | 3268 | /* |
32960613 JA |
3269 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3270 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3271 | */ |
4017eb91 | 3272 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3273 | { |
4017eb91 JA |
3274 | struct kiocb *kiocb = &req->rw.kiocb; |
3275 | struct file *file = req->file; | |
32960613 JA |
3276 | ssize_t ret = 0; |
3277 | ||
3278 | /* | |
3279 | * Don't support polled IO through this interface, and we can't | |
3280 | * support non-blocking either. For the latter, this just causes | |
3281 | * the kiocb to be handled from an async context. | |
3282 | */ | |
3283 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3284 | return -EOPNOTSUPP; | |
3285 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3286 | return -EAGAIN; | |
3287 | ||
3288 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3289 | struct iovec iovec; |
32960613 JA |
3290 | ssize_t nr; |
3291 | ||
311ae9e1 PB |
3292 | if (!iov_iter_is_bvec(iter)) { |
3293 | iovec = iov_iter_iovec(iter); | |
3294 | } else { | |
4017eb91 JA |
3295 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3296 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3297 | } |
3298 | ||
32960613 JA |
3299 | if (rw == READ) { |
3300 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3301 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3302 | } else { |
3303 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3304 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3305 | } |
3306 | ||
3307 | if (nr < 0) { | |
3308 | if (!ret) | |
3309 | ret = nr; | |
3310 | break; | |
3311 | } | |
cf15ae0a | 3312 | ret += nr; |
16c8d2df JA |
3313 | if (!iov_iter_is_bvec(iter)) { |
3314 | iov_iter_advance(iter, nr); | |
3315 | } else { | |
16c8d2df | 3316 | req->rw.addr += nr; |
cf15ae0a JA |
3317 | req->rw.len -= nr; |
3318 | if (!req->rw.len) | |
3319 | break; | |
16c8d2df | 3320 | } |
32960613 JA |
3321 | if (nr != iovec.iov_len) |
3322 | break; | |
32960613 JA |
3323 | } |
3324 | ||
3325 | return ret; | |
3326 | } | |
3327 | ||
ff6165b2 JA |
3328 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3329 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3330 | { |
e8c2bc1f | 3331 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3332 | |
ff6165b2 | 3333 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3334 | rw->free_iovec = iovec; |
227c0c96 | 3335 | rw->bytes_done = 0; |
ff6165b2 | 3336 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3337 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3338 | return; |
b64e3444 | 3339 | if (!iovec) { |
ff6165b2 JA |
3340 | unsigned iov_off = 0; |
3341 | ||
3342 | rw->iter.iov = rw->fast_iov; | |
3343 | if (iter->iov != fast_iov) { | |
3344 | iov_off = iter->iov - fast_iov; | |
3345 | rw->iter.iov += iov_off; | |
3346 | } | |
3347 | if (rw->fast_iov != fast_iov) | |
3348 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3349 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3350 | } else { |
3351 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3352 | } |
3353 | } | |
3354 | ||
6cb78689 | 3355 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3356 | { |
e8c2bc1f JA |
3357 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3358 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3359 | return req->async_data == NULL; | |
3d9932a8 XW |
3360 | } |
3361 | ||
ff6165b2 JA |
3362 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3363 | const struct iovec *fast_iov, | |
227c0c96 | 3364 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3365 | { |
26f0505a | 3366 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3367 | return 0; |
e8c2bc1f | 3368 | if (!req->async_data) { |
cd658695 JA |
3369 | struct io_async_rw *iorw; |
3370 | ||
6cb78689 | 3371 | if (io_alloc_async_data(req)) { |
6bf985dc | 3372 | kfree(iovec); |
5d204bcf | 3373 | return -ENOMEM; |
6bf985dc | 3374 | } |
b7bb4f7d | 3375 | |
ff6165b2 | 3376 | io_req_map_rw(req, iovec, fast_iov, iter); |
cd658695 JA |
3377 | iorw = req->async_data; |
3378 | /* we've copied and mapped the iter, ensure state is saved */ | |
3379 | iov_iter_save_state(&iorw->iter, &iorw->iter_state); | |
5d204bcf | 3380 | } |
b7bb4f7d | 3381 | return 0; |
f67676d1 JA |
3382 | } |
3383 | ||
73debe68 | 3384 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3385 | { |
e8c2bc1f | 3386 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3387 | struct iovec *iov = iorw->fast_iov; |
847595de | 3388 | int ret; |
c3e330a4 | 3389 | |
2846c481 | 3390 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3391 | if (unlikely(ret < 0)) |
3392 | return ret; | |
3393 | ||
ab0b196c PB |
3394 | iorw->bytes_done = 0; |
3395 | iorw->free_iovec = iov; | |
3396 | if (iov) | |
3397 | req->flags |= REQ_F_NEED_CLEANUP; | |
cd658695 | 3398 | iov_iter_save_state(&iorw->iter, &iorw->iter_state); |
c3e330a4 PB |
3399 | return 0; |
3400 | } | |
3401 | ||
73debe68 | 3402 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3403 | { |
3529d8c2 JA |
3404 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3405 | return -EBADF; | |
5d329e12 | 3406 | return io_prep_rw(req, sqe, READ); |
f67676d1 JA |
3407 | } |
3408 | ||
c1dd91d1 JA |
3409 | /* |
3410 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3411 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3412 | * This gets called when the page is unlocked, and we generally expect that to | |
3413 | * happen when the page IO is completed and the page is now uptodate. This will | |
3414 | * queue a task_work based retry of the operation, attempting to copy the data | |
3415 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3416 | * do a thread based blocking retry of the operation. That's the unexpected | |
3417 | * slow path. | |
3418 | */ | |
bcf5a063 JA |
3419 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3420 | int sync, void *arg) | |
3421 | { | |
3422 | struct wait_page_queue *wpq; | |
3423 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3424 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3425 | |
3426 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3427 | ||
cdc8fcb4 LT |
3428 | if (!wake_page_match(wpq, key)) |
3429 | return 0; | |
3430 | ||
c8d317aa | 3431 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 | 3432 | list_del_init(&wait->entry); |
921b9054 | 3433 | io_req_task_queue(req); |
bcf5a063 JA |
3434 | return 1; |
3435 | } | |
3436 | ||
c1dd91d1 JA |
3437 | /* |
3438 | * This controls whether a given IO request should be armed for async page | |
3439 | * based retry. If we return false here, the request is handed to the async | |
3440 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3441 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3442 | * will either succeed because the page is now uptodate and unlocked, or it | |
3443 | * will register a callback when the page is unlocked at IO completion. Through | |
3444 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3445 | * That retry will attempt the buffered read again. The retry will generally | |
3446 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3447 | * async worker threads for a blocking retry. | |
3448 | */ | |
227c0c96 | 3449 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3450 | { |
e8c2bc1f JA |
3451 | struct io_async_rw *rw = req->async_data; |
3452 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3453 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3454 | |
bcf5a063 JA |
3455 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3456 | if (req->flags & REQ_F_NOWAIT) | |
3457 | return false; | |
f67676d1 | 3458 | |
227c0c96 | 3459 | /* Only for buffered IO */ |
3b2a4439 | 3460 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3461 | return false; |
3b2a4439 | 3462 | |
bcf5a063 JA |
3463 | /* |
3464 | * just use poll if we can, and don't attempt if the fs doesn't | |
3465 | * support callback based unlocks | |
3466 | */ | |
3467 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3468 | return false; | |
f67676d1 | 3469 | |
3b2a4439 JA |
3470 | wait->wait.func = io_async_buf_func; |
3471 | wait->wait.private = req; | |
3472 | wait->wait.flags = 0; | |
3473 | INIT_LIST_HEAD(&wait->wait.entry); | |
3474 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3475 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3476 | kiocb->ki_waitq = wait; |
3b2a4439 | 3477 | return true; |
bcf5a063 JA |
3478 | } |
3479 | ||
aeab9506 | 3480 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3481 | { |
3482 | if (req->file->f_op->read_iter) | |
3483 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3484 | else if (req->file->f_op->read) |
4017eb91 | 3485 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3486 | else |
3487 | return -EINVAL; | |
f67676d1 JA |
3488 | } |
3489 | ||
7db30437 ML |
3490 | static bool need_read_all(struct io_kiocb *req) |
3491 | { | |
3492 | return req->flags & REQ_F_ISREG || | |
3493 | S_ISBLK(file_inode(req->file)->i_mode); | |
3494 | } | |
3495 | ||
889fca73 | 3496 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3497 | { |
3498 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3499 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3500 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3501 | struct io_async_rw *rw = req->async_data; |
45d189c6 | 3502 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
cd658695 JA |
3503 | struct iov_iter_state __state, *state; |
3504 | ssize_t ret, ret2; | |
ff6165b2 | 3505 | |
2846c481 | 3506 | if (rw) { |
e8c2bc1f | 3507 | iter = &rw->iter; |
cd658695 JA |
3508 | state = &rw->iter_state; |
3509 | /* | |
3510 | * We come here from an earlier attempt, restore our state to | |
3511 | * match in case it doesn't. It's cheap enough that we don't | |
3512 | * need to make this conditional. | |
3513 | */ | |
3514 | iov_iter_restore(iter, state); | |
2846c481 PB |
3515 | iovec = NULL; |
3516 | } else { | |
3517 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3518 | if (ret < 0) | |
3519 | return ret; | |
cd658695 JA |
3520 | state = &__state; |
3521 | iov_iter_save_state(iter, state); | |
2846c481 | 3522 | } |
cd658695 | 3523 | req->result = iov_iter_count(iter); |
2b188cc1 | 3524 | |
fd6c2e4c JA |
3525 | /* Ensure we clear previously set non-block flag */ |
3526 | if (!force_nonblock) | |
29de5f6a | 3527 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3528 | else |
3529 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3530 | ||
24c74678 | 3531 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3532 | if (force_nonblock && !io_file_supports_nowait(req, READ)) { |
6713e7a6 | 3533 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3534 | return ret ?: -EAGAIN; |
6713e7a6 | 3535 | } |
9e645e11 | 3536 | |
cd658695 | 3537 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), req->result); |
5ea5dd45 PB |
3538 | if (unlikely(ret)) { |
3539 | kfree(iovec); | |
3540 | return ret; | |
3541 | } | |
2b188cc1 | 3542 | |
227c0c96 | 3543 | ret = io_iter_do_read(req, iter); |
32960613 | 3544 | |
230d50d4 | 3545 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3546 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3547 | /* IOPOLL retry should happen for io-wq threads */ |
3548 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3549 | goto done; |
75c668cd PB |
3550 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3551 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3552 | goto done; |
f38c7e3a | 3553 | ret = 0; |
230d50d4 JA |
3554 | } else if (ret == -EIOCBQUEUED) { |
3555 | goto out_free; | |
cd658695 | 3556 | } else if (ret <= 0 || ret == req->result || !force_nonblock || |
7db30437 | 3557 | (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { |
7335e3bf | 3558 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3559 | goto done; |
227c0c96 JA |
3560 | } |
3561 | ||
cd658695 JA |
3562 | /* |
3563 | * Don't depend on the iter state matching what was consumed, or being | |
3564 | * untouched in case of error. Restore it and we'll advance it | |
3565 | * manually if we need to. | |
3566 | */ | |
3567 | iov_iter_restore(iter, state); | |
3568 | ||
227c0c96 | 3569 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3570 | if (ret2) |
3571 | return ret2; | |
3572 | ||
fe1cdd55 | 3573 | iovec = NULL; |
e8c2bc1f | 3574 | rw = req->async_data; |
cd658695 JA |
3575 | /* |
3576 | * Now use our persistent iterator and state, if we aren't already. | |
3577 | * We've restored and mapped the iter to match. | |
3578 | */ | |
3579 | if (iter != &rw->iter) { | |
3580 | iter = &rw->iter; | |
3581 | state = &rw->iter_state; | |
3582 | } | |
227c0c96 | 3583 | |
b23df91b | 3584 | do { |
cd658695 JA |
3585 | /* |
3586 | * We end up here because of a partial read, either from | |
3587 | * above or inside this loop. Advance the iter by the bytes | |
3588 | * that were consumed. | |
3589 | */ | |
3590 | iov_iter_advance(iter, ret); | |
3591 | if (!iov_iter_count(iter)) | |
3592 | break; | |
b23df91b | 3593 | rw->bytes_done += ret; |
cd658695 JA |
3594 | iov_iter_save_state(iter, state); |
3595 | ||
b23df91b PB |
3596 | /* if we can retry, do so with the callbacks armed */ |
3597 | if (!io_rw_should_retry(req)) { | |
3598 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3599 | return -EAGAIN; | |
3600 | } | |
3601 | ||
3602 | /* | |
3603 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3604 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3605 | * desired page gets unlocked. We can also get a partial read | |
3606 | * here, and if we do, then just retry at the new offset. | |
3607 | */ | |
3608 | ret = io_iter_do_read(req, iter); | |
3609 | if (ret == -EIOCBQUEUED) | |
3610 | return 0; | |
227c0c96 | 3611 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3612 | kiocb->ki_flags &= ~IOCB_WAITQ; |
cd658695 JA |
3613 | iov_iter_restore(iter, state); |
3614 | } while (ret > 0); | |
227c0c96 | 3615 | done: |
889fca73 | 3616 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3617 | out_free: |
3618 | /* it's faster to check here then delegate to kfree */ | |
3619 | if (iovec) | |
3620 | kfree(iovec); | |
5ea5dd45 | 3621 | return 0; |
2b188cc1 JA |
3622 | } |
3623 | ||
73debe68 | 3624 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3625 | { |
3529d8c2 JA |
3626 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3627 | return -EBADF; | |
5d329e12 | 3628 | return io_prep_rw(req, sqe, WRITE); |
f67676d1 JA |
3629 | } |
3630 | ||
889fca73 | 3631 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3632 | { |
3633 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3634 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3635 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3636 | struct io_async_rw *rw = req->async_data; |
45d189c6 | 3637 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
cd658695 JA |
3638 | struct iov_iter_state __state, *state; |
3639 | ssize_t ret, ret2; | |
2b188cc1 | 3640 | |
2846c481 | 3641 | if (rw) { |
e8c2bc1f | 3642 | iter = &rw->iter; |
cd658695 JA |
3643 | state = &rw->iter_state; |
3644 | iov_iter_restore(iter, state); | |
2846c481 PB |
3645 | iovec = NULL; |
3646 | } else { | |
3647 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3648 | if (ret < 0) | |
3649 | return ret; | |
cd658695 JA |
3650 | state = &__state; |
3651 | iov_iter_save_state(iter, state); | |
2846c481 | 3652 | } |
cd658695 | 3653 | req->result = iov_iter_count(iter); |
2b188cc1 | 3654 | |
fd6c2e4c JA |
3655 | /* Ensure we clear previously set non-block flag */ |
3656 | if (!force_nonblock) | |
a88fc400 PB |
3657 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3658 | else | |
3659 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3660 | |
24c74678 | 3661 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3662 | if (force_nonblock && !io_file_supports_nowait(req, WRITE)) |
f67676d1 | 3663 | goto copy_iov; |
31b51510 | 3664 | |
10d59345 JA |
3665 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3666 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3667 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3668 | goto copy_iov; |
31b51510 | 3669 | |
cd658695 | 3670 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), req->result); |
fa15bafb PB |
3671 | if (unlikely(ret)) |
3672 | goto out_free; | |
4ed734b0 | 3673 | |
fa15bafb PB |
3674 | /* |
3675 | * Open-code file_start_write here to grab freeze protection, | |
3676 | * which will be released by another thread in | |
3677 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3678 | * released so that it doesn't complain about the held lock when | |
3679 | * we return to userspace. | |
3680 | */ | |
3681 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3682 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3683 | __sb_writers_release(file_inode(req->file)->i_sb, |
3684 | SB_FREEZE_WRITE); | |
3685 | } | |
3686 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3687 | |
fa15bafb | 3688 | if (req->file->f_op->write_iter) |
ff6165b2 | 3689 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3690 | else if (req->file->f_op->write) |
4017eb91 | 3691 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3692 | else |
3693 | ret2 = -EINVAL; | |
4ed734b0 | 3694 | |
6ad7f233 PB |
3695 | if (req->flags & REQ_F_REISSUE) { |
3696 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3697 | ret2 = -EAGAIN; |
6ad7f233 | 3698 | } |
230d50d4 | 3699 | |
fa15bafb PB |
3700 | /* |
3701 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3702 | * retry them without IOCB_NOWAIT. | |
3703 | */ | |
3704 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3705 | ret2 = -EAGAIN; | |
75c668cd PB |
3706 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3707 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3708 | goto done; |
fa15bafb | 3709 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3710 | /* IOPOLL retry should happen for io-wq threads */ |
3711 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3712 | goto copy_iov; | |
355afaeb | 3713 | done: |
889fca73 | 3714 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3715 | } else { |
f67676d1 | 3716 | copy_iov: |
cd658695 | 3717 | iov_iter_restore(iter, state); |
227c0c96 | 3718 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3719 | return ret ?: -EAGAIN; |
2b188cc1 | 3720 | } |
31b51510 | 3721 | out_free: |
f261c168 | 3722 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3723 | if (iovec) |
6f2cc166 | 3724 | kfree(iovec); |
2b188cc1 JA |
3725 | return ret; |
3726 | } | |
3727 | ||
80a261fd JA |
3728 | static int io_renameat_prep(struct io_kiocb *req, |
3729 | const struct io_uring_sqe *sqe) | |
3730 | { | |
3731 | struct io_rename *ren = &req->rename; | |
3732 | const char __user *oldf, *newf; | |
3733 | ||
ed7eb259 JA |
3734 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3735 | return -EINVAL; | |
26578cda | 3736 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
ed7eb259 | 3737 | return -EINVAL; |
80a261fd JA |
3738 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3739 | return -EBADF; | |
3740 | ||
3741 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3742 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3743 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3744 | ren->new_dfd = READ_ONCE(sqe->len); | |
3745 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3746 | ||
3747 | ren->oldpath = getname(oldf); | |
3748 | if (IS_ERR(ren->oldpath)) | |
3749 | return PTR_ERR(ren->oldpath); | |
3750 | ||
3751 | ren->newpath = getname(newf); | |
3752 | if (IS_ERR(ren->newpath)) { | |
3753 | putname(ren->oldpath); | |
3754 | return PTR_ERR(ren->newpath); | |
3755 | } | |
3756 | ||
3757 | req->flags |= REQ_F_NEED_CLEANUP; | |
3758 | return 0; | |
3759 | } | |
3760 | ||
45d189c6 | 3761 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3762 | { |
3763 | struct io_rename *ren = &req->rename; | |
3764 | int ret; | |
3765 | ||
45d189c6 | 3766 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3767 | return -EAGAIN; |
3768 | ||
3769 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3770 | ren->newpath, ren->flags); | |
3771 | ||
3772 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3773 | if (ret < 0) | |
93d2bcd2 | 3774 | req_set_fail(req); |
80a261fd JA |
3775 | io_req_complete(req, ret); |
3776 | return 0; | |
3777 | } | |
3778 | ||
14a1143b JA |
3779 | static int io_unlinkat_prep(struct io_kiocb *req, |
3780 | const struct io_uring_sqe *sqe) | |
3781 | { | |
3782 | struct io_unlink *un = &req->unlink; | |
3783 | const char __user *fname; | |
3784 | ||
22634bc5 JA |
3785 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3786 | return -EINVAL; | |
26578cda PB |
3787 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || |
3788 | sqe->splice_fd_in) | |
22634bc5 | 3789 | return -EINVAL; |
14a1143b JA |
3790 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3791 | return -EBADF; | |
3792 | ||
3793 | un->dfd = READ_ONCE(sqe->fd); | |
3794 | ||
3795 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3796 | if (un->flags & ~AT_REMOVEDIR) | |
3797 | return -EINVAL; | |
3798 | ||
3799 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3800 | un->filename = getname(fname); | |
3801 | if (IS_ERR(un->filename)) | |
3802 | return PTR_ERR(un->filename); | |
3803 | ||
3804 | req->flags |= REQ_F_NEED_CLEANUP; | |
3805 | return 0; | |
3806 | } | |
3807 | ||
45d189c6 | 3808 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3809 | { |
3810 | struct io_unlink *un = &req->unlink; | |
3811 | int ret; | |
3812 | ||
45d189c6 | 3813 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3814 | return -EAGAIN; |
3815 | ||
3816 | if (un->flags & AT_REMOVEDIR) | |
3817 | ret = do_rmdir(un->dfd, un->filename); | |
3818 | else | |
3819 | ret = do_unlinkat(un->dfd, un->filename); | |
3820 | ||
3821 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3822 | if (ret < 0) | |
93d2bcd2 | 3823 | req_set_fail(req); |
14a1143b JA |
3824 | io_req_complete(req, ret); |
3825 | return 0; | |
3826 | } | |
3827 | ||
e34a02dc DK |
3828 | static int io_mkdirat_prep(struct io_kiocb *req, |
3829 | const struct io_uring_sqe *sqe) | |
3830 | { | |
3831 | struct io_mkdir *mkd = &req->mkdir; | |
3832 | const char __user *fname; | |
3833 | ||
3834 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3835 | return -EINVAL; | |
3836 | if (sqe->ioprio || sqe->off || sqe->rw_flags || sqe->buf_index || | |
3837 | sqe->splice_fd_in) | |
3838 | return -EINVAL; | |
3839 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3840 | return -EBADF; | |
3841 | ||
3842 | mkd->dfd = READ_ONCE(sqe->fd); | |
3843 | mkd->mode = READ_ONCE(sqe->len); | |
3844 | ||
3845 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3846 | mkd->filename = getname(fname); | |
3847 | if (IS_ERR(mkd->filename)) | |
3848 | return PTR_ERR(mkd->filename); | |
3849 | ||
3850 | req->flags |= REQ_F_NEED_CLEANUP; | |
3851 | return 0; | |
3852 | } | |
3853 | ||
3854 | static int io_mkdirat(struct io_kiocb *req, int issue_flags) | |
3855 | { | |
3856 | struct io_mkdir *mkd = &req->mkdir; | |
3857 | int ret; | |
3858 | ||
3859 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3860 | return -EAGAIN; | |
3861 | ||
3862 | ret = do_mkdirat(mkd->dfd, mkd->filename, mkd->mode); | |
3863 | ||
3864 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3865 | if (ret < 0) | |
3866 | req_set_fail(req); | |
3867 | io_req_complete(req, ret); | |
3868 | return 0; | |
3869 | } | |
3870 | ||
7a8721f8 DK |
3871 | static int io_symlinkat_prep(struct io_kiocb *req, |
3872 | const struct io_uring_sqe *sqe) | |
3873 | { | |
3874 | struct io_symlink *sl = &req->symlink; | |
3875 | const char __user *oldpath, *newpath; | |
3876 | ||
3877 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3878 | return -EINVAL; | |
3879 | if (sqe->ioprio || sqe->len || sqe->rw_flags || sqe->buf_index || | |
3880 | sqe->splice_fd_in) | |
3881 | return -EINVAL; | |
3882 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3883 | return -EBADF; | |
3884 | ||
3885 | sl->new_dfd = READ_ONCE(sqe->fd); | |
3886 | oldpath = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3887 | newpath = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3888 | ||
3889 | sl->oldpath = getname(oldpath); | |
3890 | if (IS_ERR(sl->oldpath)) | |
3891 | return PTR_ERR(sl->oldpath); | |
3892 | ||
3893 | sl->newpath = getname(newpath); | |
3894 | if (IS_ERR(sl->newpath)) { | |
3895 | putname(sl->oldpath); | |
3896 | return PTR_ERR(sl->newpath); | |
3897 | } | |
3898 | ||
3899 | req->flags |= REQ_F_NEED_CLEANUP; | |
3900 | return 0; | |
3901 | } | |
3902 | ||
3903 | static int io_symlinkat(struct io_kiocb *req, int issue_flags) | |
3904 | { | |
3905 | struct io_symlink *sl = &req->symlink; | |
3906 | int ret; | |
3907 | ||
3908 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3909 | return -EAGAIN; | |
3910 | ||
3911 | ret = do_symlinkat(sl->oldpath, sl->new_dfd, sl->newpath); | |
3912 | ||
3913 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3914 | if (ret < 0) | |
3915 | req_set_fail(req); | |
3916 | io_req_complete(req, ret); | |
3917 | return 0; | |
3918 | } | |
3919 | ||
cf30da90 DK |
3920 | static int io_linkat_prep(struct io_kiocb *req, |
3921 | const struct io_uring_sqe *sqe) | |
3922 | { | |
3923 | struct io_hardlink *lnk = &req->hardlink; | |
3924 | const char __user *oldf, *newf; | |
3925 | ||
3926 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3927 | return -EINVAL; | |
3928 | if (sqe->ioprio || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) | |
3929 | return -EINVAL; | |
3930 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3931 | return -EBADF; | |
3932 | ||
3933 | lnk->old_dfd = READ_ONCE(sqe->fd); | |
3934 | lnk->new_dfd = READ_ONCE(sqe->len); | |
3935 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3936 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3937 | lnk->flags = READ_ONCE(sqe->hardlink_flags); | |
3938 | ||
3939 | lnk->oldpath = getname(oldf); | |
3940 | if (IS_ERR(lnk->oldpath)) | |
3941 | return PTR_ERR(lnk->oldpath); | |
3942 | ||
3943 | lnk->newpath = getname(newf); | |
3944 | if (IS_ERR(lnk->newpath)) { | |
3945 | putname(lnk->oldpath); | |
3946 | return PTR_ERR(lnk->newpath); | |
3947 | } | |
3948 | ||
3949 | req->flags |= REQ_F_NEED_CLEANUP; | |
3950 | return 0; | |
3951 | } | |
3952 | ||
3953 | static int io_linkat(struct io_kiocb *req, int issue_flags) | |
3954 | { | |
3955 | struct io_hardlink *lnk = &req->hardlink; | |
3956 | int ret; | |
3957 | ||
3958 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3959 | return -EAGAIN; | |
3960 | ||
3961 | ret = do_linkat(lnk->old_dfd, lnk->oldpath, lnk->new_dfd, | |
3962 | lnk->newpath, lnk->flags); | |
3963 | ||
3964 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3965 | if (ret < 0) | |
3966 | req_set_fail(req); | |
3967 | io_req_complete(req, ret); | |
3968 | return 0; | |
3969 | } | |
3970 | ||
36f4fa68 JA |
3971 | static int io_shutdown_prep(struct io_kiocb *req, |
3972 | const struct io_uring_sqe *sqe) | |
3973 | { | |
3974 | #if defined(CONFIG_NET) | |
3975 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3976 | return -EINVAL; | |
26578cda PB |
3977 | if (unlikely(sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || |
3978 | sqe->buf_index || sqe->splice_fd_in)) | |
36f4fa68 JA |
3979 | return -EINVAL; |
3980 | ||
3981 | req->shutdown.how = READ_ONCE(sqe->len); | |
3982 | return 0; | |
3983 | #else | |
3984 | return -EOPNOTSUPP; | |
3985 | #endif | |
3986 | } | |
3987 | ||
45d189c6 | 3988 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3989 | { |
3990 | #if defined(CONFIG_NET) | |
3991 | struct socket *sock; | |
3992 | int ret; | |
3993 | ||
45d189c6 | 3994 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3995 | return -EAGAIN; |
3996 | ||
48aba79b | 3997 | sock = sock_from_file(req->file); |
36f4fa68 | 3998 | if (unlikely(!sock)) |
48aba79b | 3999 | return -ENOTSOCK; |
36f4fa68 JA |
4000 | |
4001 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 4002 | if (ret < 0) |
93d2bcd2 | 4003 | req_set_fail(req); |
36f4fa68 JA |
4004 | io_req_complete(req, ret); |
4005 | return 0; | |
4006 | #else | |
4007 | return -EOPNOTSUPP; | |
4008 | #endif | |
4009 | } | |
4010 | ||
f2a8d5c7 PB |
4011 | static int __io_splice_prep(struct io_kiocb *req, |
4012 | const struct io_uring_sqe *sqe) | |
7d67af2c | 4013 | { |
fe7e3257 | 4014 | struct io_splice *sp = &req->splice; |
7d67af2c | 4015 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 4016 | |
3232dd02 PB |
4017 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4018 | return -EINVAL; | |
7d67af2c | 4019 | |
7d67af2c PB |
4020 | sp->len = READ_ONCE(sqe->len); |
4021 | sp->flags = READ_ONCE(sqe->splice_flags); | |
7d67af2c PB |
4022 | if (unlikely(sp->flags & ~valid_flags)) |
4023 | return -EINVAL; | |
dfadddfc | 4024 | sp->splice_fd_in = READ_ONCE(sqe->splice_fd_in); |
7d67af2c PB |
4025 | return 0; |
4026 | } | |
4027 | ||
f2a8d5c7 PB |
4028 | static int io_tee_prep(struct io_kiocb *req, |
4029 | const struct io_uring_sqe *sqe) | |
4030 | { | |
4031 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
4032 | return -EINVAL; | |
4033 | return __io_splice_prep(req, sqe); | |
4034 | } | |
4035 | ||
45d189c6 | 4036 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
4037 | { |
4038 | struct io_splice *sp = &req->splice; | |
f2a8d5c7 PB |
4039 | struct file *out = sp->file_out; |
4040 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
dfadddfc | 4041 | struct file *in; |
f2a8d5c7 PB |
4042 | long ret = 0; |
4043 | ||
45d189c6 | 4044 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 | 4045 | return -EAGAIN; |
dfadddfc JA |
4046 | |
4047 | in = io_file_get(req->ctx, req, sp->splice_fd_in, | |
4048 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
4049 | if (!in) { | |
4050 | ret = -EBADF; | |
4051 | goto done; | |
4052 | } | |
4053 | ||
f2a8d5c7 PB |
4054 | if (sp->len) |
4055 | ret = do_tee(in, out, sp->len, flags); | |
4056 | ||
e1d767f0 PB |
4057 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4058 | io_put_file(in); | |
dfadddfc | 4059 | done: |
f2a8d5c7 | 4060 | if (ret != sp->len) |
93d2bcd2 | 4061 | req_set_fail(req); |
e1e16097 | 4062 | io_req_complete(req, ret); |
f2a8d5c7 PB |
4063 | return 0; |
4064 | } | |
4065 | ||
4066 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4067 | { | |
fe7e3257 | 4068 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
4069 | |
4070 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
4071 | sp->off_out = READ_ONCE(sqe->off); | |
4072 | return __io_splice_prep(req, sqe); | |
4073 | } | |
4074 | ||
45d189c6 | 4075 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
4076 | { |
4077 | struct io_splice *sp = &req->splice; | |
7d67af2c PB |
4078 | struct file *out = sp->file_out; |
4079 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
4080 | loff_t *poff_in, *poff_out; | |
dfadddfc | 4081 | struct file *in; |
c9687426 | 4082 | long ret = 0; |
7d67af2c | 4083 | |
45d189c6 | 4084 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 4085 | return -EAGAIN; |
7d67af2c | 4086 | |
dfadddfc JA |
4087 | in = io_file_get(req->ctx, req, sp->splice_fd_in, |
4088 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
4089 | if (!in) { | |
4090 | ret = -EBADF; | |
4091 | goto done; | |
4092 | } | |
4093 | ||
7d67af2c PB |
4094 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; |
4095 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 4096 | |
948a7749 | 4097 | if (sp->len) |
c9687426 | 4098 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 4099 | |
e1d767f0 PB |
4100 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4101 | io_put_file(in); | |
dfadddfc | 4102 | done: |
7d67af2c | 4103 | if (ret != sp->len) |
93d2bcd2 | 4104 | req_set_fail(req); |
e1e16097 | 4105 | io_req_complete(req, ret); |
7d67af2c PB |
4106 | return 0; |
4107 | } | |
4108 | ||
2b188cc1 JA |
4109 | /* |
4110 | * IORING_OP_NOP just posts a completion event, nothing else. | |
4111 | */ | |
889fca73 | 4112 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
4113 | { |
4114 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4115 | |
def596e9 JA |
4116 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4117 | return -EINVAL; | |
4118 | ||
889fca73 | 4119 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4120 | return 0; |
4121 | } | |
4122 | ||
1155c76a | 4123 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4124 | { |
6b06314c | 4125 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4126 | |
6b06314c | 4127 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4128 | return -EINVAL; |
26578cda PB |
4129 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4130 | sqe->splice_fd_in)) | |
c992fe29 CH |
4131 | return -EINVAL; |
4132 | ||
8ed8d3c3 JA |
4133 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4134 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4135 | return -EINVAL; | |
4136 | ||
4137 | req->sync.off = READ_ONCE(sqe->off); | |
4138 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4139 | return 0; |
4140 | } | |
4141 | ||
45d189c6 | 4142 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4143 | { |
8ed8d3c3 | 4144 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4145 | int ret; |
4146 | ||
ac45abc0 | 4147 | /* fsync always requires a blocking context */ |
45d189c6 | 4148 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4149 | return -EAGAIN; |
4150 | ||
9adbd45d | 4151 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4152 | end > 0 ? end : LLONG_MAX, |
4153 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4154 | if (ret < 0) | |
93d2bcd2 | 4155 | req_set_fail(req); |
e1e16097 | 4156 | io_req_complete(req, ret); |
c992fe29 CH |
4157 | return 0; |
4158 | } | |
4159 | ||
d63d1b5e JA |
4160 | static int io_fallocate_prep(struct io_kiocb *req, |
4161 | const struct io_uring_sqe *sqe) | |
4162 | { | |
26578cda PB |
4163 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags || |
4164 | sqe->splice_fd_in) | |
d63d1b5e | 4165 | return -EINVAL; |
3232dd02 PB |
4166 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4167 | return -EINVAL; | |
d63d1b5e JA |
4168 | |
4169 | req->sync.off = READ_ONCE(sqe->off); | |
4170 | req->sync.len = READ_ONCE(sqe->addr); | |
4171 | req->sync.mode = READ_ONCE(sqe->len); | |
4172 | return 0; | |
4173 | } | |
4174 | ||
45d189c6 | 4175 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4176 | { |
ac45abc0 PB |
4177 | int ret; |
4178 | ||
d63d1b5e | 4179 | /* fallocate always requiring blocking context */ |
45d189c6 | 4180 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4181 | return -EAGAIN; |
ac45abc0 PB |
4182 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4183 | req->sync.len); | |
ac45abc0 | 4184 | if (ret < 0) |
93d2bcd2 | 4185 | req_set_fail(req); |
e1e16097 | 4186 | io_req_complete(req, ret); |
5d17b4a4 JA |
4187 | return 0; |
4188 | } | |
4189 | ||
ec65fea5 | 4190 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4191 | { |
f8748881 | 4192 | const char __user *fname; |
15b71abe | 4193 | int ret; |
b7bb4f7d | 4194 | |
d3fddf6d PB |
4195 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4196 | return -EINVAL; | |
b9445598 | 4197 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4198 | return -EINVAL; |
ec65fea5 | 4199 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4200 | return -EBADF; |
03b1230c | 4201 | |
ec65fea5 PB |
4202 | /* open.how should be already initialised */ |
4203 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4204 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4205 | |
25e72d10 PB |
4206 | req->open.dfd = READ_ONCE(sqe->fd); |
4207 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4208 | req->open.filename = getname(fname); |
15b71abe JA |
4209 | if (IS_ERR(req->open.filename)) { |
4210 | ret = PTR_ERR(req->open.filename); | |
4211 | req->open.filename = NULL; | |
4212 | return ret; | |
4213 | } | |
b9445598 PB |
4214 | |
4215 | req->open.file_slot = READ_ONCE(sqe->file_index); | |
4216 | if (req->open.file_slot && (req->open.how.flags & O_CLOEXEC)) | |
4217 | return -EINVAL; | |
4218 | ||
4022e7af | 4219 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4220 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4221 | return 0; |
03b1230c JA |
4222 | } |
4223 | ||
ec65fea5 PB |
4224 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4225 | { | |
d3fddf6d PB |
4226 | u64 mode = READ_ONCE(sqe->len); |
4227 | u64 flags = READ_ONCE(sqe->open_flags); | |
ec65fea5 | 4228 | |
ec65fea5 PB |
4229 | req->open.how = build_open_how(flags, mode); |
4230 | return __io_openat_prep(req, sqe); | |
4231 | } | |
4232 | ||
cebdb986 | 4233 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4234 | { |
cebdb986 | 4235 | struct open_how __user *how; |
cebdb986 | 4236 | size_t len; |
0fa03c62 JA |
4237 | int ret; |
4238 | ||
cebdb986 JA |
4239 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4240 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4241 | if (len < OPEN_HOW_SIZE_VER0) |
4242 | return -EINVAL; | |
3529d8c2 | 4243 | |
cebdb986 JA |
4244 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4245 | len); | |
4246 | if (ret) | |
4247 | return ret; | |
3529d8c2 | 4248 | |
ec65fea5 | 4249 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4250 | } |
4251 | ||
45d189c6 | 4252 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4253 | { |
4254 | struct open_flags op; | |
15b71abe | 4255 | struct file *file; |
b9445598 PB |
4256 | bool resolve_nonblock, nonblock_set; |
4257 | bool fixed = !!req->open.file_slot; | |
15b71abe JA |
4258 | int ret; |
4259 | ||
cebdb986 | 4260 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4261 | if (ret) |
4262 | goto err; | |
3a81fd02 JA |
4263 | nonblock_set = op.open_flag & O_NONBLOCK; |
4264 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4265 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4266 | /* |
4267 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4268 | * it'll always -EAGAIN | |
4269 | */ | |
4270 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4271 | return -EAGAIN; | |
4272 | op.lookup_flags |= LOOKUP_CACHED; | |
4273 | op.open_flag |= O_NONBLOCK; | |
4274 | } | |
15b71abe | 4275 | |
b9445598 PB |
4276 | if (!fixed) { |
4277 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); | |
4278 | if (ret < 0) | |
4279 | goto err; | |
4280 | } | |
15b71abe JA |
4281 | |
4282 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 4283 | if (IS_ERR(file)) { |
944d1444 | 4284 | /* |
12dcb58a PB |
4285 | * We could hang on to this 'fd' on retrying, but seems like |
4286 | * marginal gain for something that is now known to be a slower | |
4287 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 4288 | */ |
b9445598 PB |
4289 | if (!fixed) |
4290 | put_unused_fd(ret); | |
3a81fd02 | 4291 | |
15b71abe | 4292 | ret = PTR_ERR(file); |
12dcb58a PB |
4293 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
4294 | if (ret == -EAGAIN && | |
4295 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
4296 | return -EAGAIN; | |
4297 | goto err; | |
15b71abe | 4298 | } |
12dcb58a PB |
4299 | |
4300 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
4301 | file->f_flags &= ~O_NONBLOCK; | |
4302 | fsnotify_open(file); | |
b9445598 PB |
4303 | |
4304 | if (!fixed) | |
4305 | fd_install(ret, file); | |
4306 | else | |
4307 | ret = io_install_fixed_file(req, file, issue_flags, | |
4308 | req->open.file_slot - 1); | |
15b71abe JA |
4309 | err: |
4310 | putname(req->open.filename); | |
8fef80bf | 4311 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 4312 | if (ret < 0) |
93d2bcd2 | 4313 | req_set_fail(req); |
0bdf3398 | 4314 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
4315 | return 0; |
4316 | } | |
4317 | ||
45d189c6 | 4318 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4319 | { |
e45cff58 | 4320 | return io_openat2(req, issue_flags); |
cebdb986 JA |
4321 | } |
4322 | ||
067524e9 JA |
4323 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4324 | const struct io_uring_sqe *sqe) | |
4325 | { | |
4326 | struct io_provide_buf *p = &req->pbuf; | |
4327 | u64 tmp; | |
4328 | ||
26578cda PB |
4329 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off || |
4330 | sqe->splice_fd_in) | |
067524e9 JA |
4331 | return -EINVAL; |
4332 | ||
4333 | tmp = READ_ONCE(sqe->fd); | |
4334 | if (!tmp || tmp > USHRT_MAX) | |
4335 | return -EINVAL; | |
4336 | ||
4337 | memset(p, 0, sizeof(*p)); | |
4338 | p->nbufs = tmp; | |
4339 | p->bgid = READ_ONCE(sqe->buf_group); | |
4340 | return 0; | |
4341 | } | |
4342 | ||
4343 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4344 | int bgid, unsigned nbufs) | |
4345 | { | |
4346 | unsigned i = 0; | |
4347 | ||
4348 | /* shouldn't happen */ | |
4349 | if (!nbufs) | |
4350 | return 0; | |
4351 | ||
4352 | /* the head kbuf is the list itself */ | |
4353 | while (!list_empty(&buf->list)) { | |
4354 | struct io_buffer *nxt; | |
4355 | ||
4356 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4357 | list_del(&nxt->list); | |
4358 | kfree(nxt); | |
4359 | if (++i == nbufs) | |
4360 | return i; | |
07edfd19 | 4361 | cond_resched(); |
067524e9 JA |
4362 | } |
4363 | i++; | |
4364 | kfree(buf); | |
9e15c3a0 | 4365 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
4366 | |
4367 | return i; | |
4368 | } | |
4369 | ||
889fca73 | 4370 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4371 | { |
4372 | struct io_provide_buf *p = &req->pbuf; | |
4373 | struct io_ring_ctx *ctx = req->ctx; | |
4374 | struct io_buffer *head; | |
4375 | int ret = 0; | |
45d189c6 | 4376 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4377 | |
4378 | io_ring_submit_lock(ctx, !force_nonblock); | |
4379 | ||
4380 | lockdep_assert_held(&ctx->uring_lock); | |
4381 | ||
4382 | ret = -ENOENT; | |
9e15c3a0 | 4383 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
4384 | if (head) |
4385 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 4386 | if (ret < 0) |
93d2bcd2 | 4387 | req_set_fail(req); |
067524e9 | 4388 | |
9fb8cb49 PB |
4389 | /* complete before unlock, IOPOLL may need the lock */ |
4390 | __io_req_complete(req, issue_flags, ret, 0); | |
4391 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
4392 | return 0; |
4393 | } | |
4394 | ||
ddf0322d JA |
4395 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4396 | const struct io_uring_sqe *sqe) | |
4397 | { | |
38134ada | 4398 | unsigned long size, tmp_check; |
ddf0322d JA |
4399 | struct io_provide_buf *p = &req->pbuf; |
4400 | u64 tmp; | |
4401 | ||
26578cda | 4402 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
ddf0322d JA |
4403 | return -EINVAL; |
4404 | ||
4405 | tmp = READ_ONCE(sqe->fd); | |
4406 | if (!tmp || tmp > USHRT_MAX) | |
4407 | return -E2BIG; | |
4408 | p->nbufs = tmp; | |
4409 | p->addr = READ_ONCE(sqe->addr); | |
4410 | p->len = READ_ONCE(sqe->len); | |
4411 | ||
38134ada PB |
4412 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
4413 | &size)) | |
4414 | return -EOVERFLOW; | |
4415 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
4416 | return -EOVERFLOW; | |
4417 | ||
d81269fe PB |
4418 | size = (unsigned long)p->len * p->nbufs; |
4419 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
4420 | return -EFAULT; |
4421 | ||
4422 | p->bgid = READ_ONCE(sqe->buf_group); | |
4423 | tmp = READ_ONCE(sqe->off); | |
4424 | if (tmp > USHRT_MAX) | |
4425 | return -E2BIG; | |
4426 | p->bid = tmp; | |
4427 | return 0; | |
4428 | } | |
4429 | ||
4430 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4431 | { | |
4432 | struct io_buffer *buf; | |
4433 | u64 addr = pbuf->addr; | |
4434 | int i, bid = pbuf->bid; | |
4435 | ||
4436 | for (i = 0; i < pbuf->nbufs; i++) { | |
9990da93 | 4437 | buf = kmalloc(sizeof(*buf), GFP_KERNEL_ACCOUNT); |
ddf0322d JA |
4438 | if (!buf) |
4439 | break; | |
4440 | ||
4441 | buf->addr = addr; | |
d1f82808 | 4442 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
4443 | buf->bid = bid; |
4444 | addr += pbuf->len; | |
4445 | bid++; | |
4446 | if (!*head) { | |
4447 | INIT_LIST_HEAD(&buf->list); | |
4448 | *head = buf; | |
4449 | } else { | |
4450 | list_add_tail(&buf->list, &(*head)->list); | |
4451 | } | |
2d1829ba | 4452 | cond_resched(); |
ddf0322d JA |
4453 | } |
4454 | ||
4455 | return i ? i : -ENOMEM; | |
4456 | } | |
4457 | ||
889fca73 | 4458 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4459 | { |
4460 | struct io_provide_buf *p = &req->pbuf; | |
4461 | struct io_ring_ctx *ctx = req->ctx; | |
4462 | struct io_buffer *head, *list; | |
4463 | int ret = 0; | |
45d189c6 | 4464 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4465 | |
4466 | io_ring_submit_lock(ctx, !force_nonblock); | |
4467 | ||
4468 | lockdep_assert_held(&ctx->uring_lock); | |
4469 | ||
9e15c3a0 | 4470 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4471 | |
4472 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4473 | if (ret >= 0 && !list) { |
4474 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4475 | if (ret < 0) | |
067524e9 | 4476 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4477 | } |
ddf0322d | 4478 | if (ret < 0) |
93d2bcd2 | 4479 | req_set_fail(req); |
9fb8cb49 PB |
4480 | /* complete before unlock, IOPOLL may need the lock */ |
4481 | __io_req_complete(req, issue_flags, ret, 0); | |
4482 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4483 | return 0; |
cebdb986 JA |
4484 | } |
4485 | ||
3e4827b0 JA |
4486 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4487 | const struct io_uring_sqe *sqe) | |
4488 | { | |
4489 | #if defined(CONFIG_EPOLL) | |
26578cda | 4490 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
3e4827b0 | 4491 | return -EINVAL; |
2d74d042 | 4492 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4493 | return -EINVAL; |
3e4827b0 JA |
4494 | |
4495 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4496 | req->epoll.op = READ_ONCE(sqe->len); | |
4497 | req->epoll.fd = READ_ONCE(sqe->off); | |
4498 | ||
4499 | if (ep_op_has_event(req->epoll.op)) { | |
4500 | struct epoll_event __user *ev; | |
4501 | ||
4502 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4503 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4504 | return -EFAULT; | |
4505 | } | |
4506 | ||
4507 | return 0; | |
4508 | #else | |
4509 | return -EOPNOTSUPP; | |
4510 | #endif | |
4511 | } | |
4512 | ||
889fca73 | 4513 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4514 | { |
4515 | #if defined(CONFIG_EPOLL) | |
4516 | struct io_epoll *ie = &req->epoll; | |
4517 | int ret; | |
45d189c6 | 4518 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4519 | |
4520 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4521 | if (force_nonblock && ret == -EAGAIN) | |
4522 | return -EAGAIN; | |
4523 | ||
4524 | if (ret < 0) | |
93d2bcd2 | 4525 | req_set_fail(req); |
889fca73 | 4526 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4527 | return 0; |
4528 | #else | |
4529 | return -EOPNOTSUPP; | |
4530 | #endif | |
4531 | } | |
4532 | ||
c1ca757b JA |
4533 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4534 | { | |
4535 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
26578cda | 4536 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->splice_fd_in) |
c1ca757b | 4537 | return -EINVAL; |
3232dd02 PB |
4538 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4539 | return -EINVAL; | |
c1ca757b JA |
4540 | |
4541 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4542 | req->madvise.len = READ_ONCE(sqe->len); | |
4543 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4544 | return 0; | |
4545 | #else | |
4546 | return -EOPNOTSUPP; | |
4547 | #endif | |
4548 | } | |
4549 | ||
45d189c6 | 4550 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4551 | { |
4552 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4553 | struct io_madvise *ma = &req->madvise; | |
4554 | int ret; | |
4555 | ||
45d189c6 | 4556 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4557 | return -EAGAIN; |
4558 | ||
0726b01e | 4559 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4560 | if (ret < 0) |
93d2bcd2 | 4561 | req_set_fail(req); |
e1e16097 | 4562 | io_req_complete(req, ret); |
c1ca757b JA |
4563 | return 0; |
4564 | #else | |
4565 | return -EOPNOTSUPP; | |
4566 | #endif | |
4567 | } | |
4568 | ||
4840e418 JA |
4569 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4570 | { | |
26578cda | 4571 | if (sqe->ioprio || sqe->buf_index || sqe->addr || sqe->splice_fd_in) |
4840e418 | 4572 | return -EINVAL; |
3232dd02 PB |
4573 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4574 | return -EINVAL; | |
4840e418 JA |
4575 | |
4576 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4577 | req->fadvise.len = READ_ONCE(sqe->len); | |
4578 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4579 | return 0; | |
4580 | } | |
4581 | ||
45d189c6 | 4582 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4583 | { |
4584 | struct io_fadvise *fa = &req->fadvise; | |
4585 | int ret; | |
4586 | ||
45d189c6 | 4587 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4588 | switch (fa->advice) { |
4589 | case POSIX_FADV_NORMAL: | |
4590 | case POSIX_FADV_RANDOM: | |
4591 | case POSIX_FADV_SEQUENTIAL: | |
4592 | break; | |
4593 | default: | |
4594 | return -EAGAIN; | |
4595 | } | |
4596 | } | |
4840e418 JA |
4597 | |
4598 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4599 | if (ret < 0) | |
93d2bcd2 | 4600 | req_set_fail(req); |
0bdf3398 | 4601 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4602 | return 0; |
4603 | } | |
4604 | ||
eddc7ef5 JA |
4605 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4606 | { | |
2d74d042 | 4607 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4608 | return -EINVAL; |
26578cda | 4609 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
eddc7ef5 | 4610 | return -EINVAL; |
9c280f90 | 4611 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4612 | return -EBADF; |
eddc7ef5 | 4613 | |
1d9e1288 BM |
4614 | req->statx.dfd = READ_ONCE(sqe->fd); |
4615 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4616 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4617 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4618 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4619 | |
4620 | return 0; | |
4621 | } | |
4622 | ||
45d189c6 | 4623 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4624 | { |
1d9e1288 | 4625 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4626 | int ret; |
4627 | ||
59d70013 | 4628 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4629 | return -EAGAIN; |
4630 | ||
e62753e4 BM |
4631 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4632 | ctx->buffer); | |
eddc7ef5 | 4633 | |
eddc7ef5 | 4634 | if (ret < 0) |
93d2bcd2 | 4635 | req_set_fail(req); |
e1e16097 | 4636 | io_req_complete(req, ret); |
eddc7ef5 JA |
4637 | return 0; |
4638 | } | |
4639 | ||
b5dba59e JA |
4640 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4641 | { | |
14587a46 | 4642 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4643 | return -EINVAL; |
b5dba59e | 4644 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
7df778be | 4645 | sqe->rw_flags || sqe->buf_index) |
b5dba59e | 4646 | return -EINVAL; |
9c280f90 | 4647 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4648 | return -EBADF; |
b5dba59e JA |
4649 | |
4650 | req->close.fd = READ_ONCE(sqe->fd); | |
7df778be PB |
4651 | req->close.file_slot = READ_ONCE(sqe->file_index); |
4652 | if (req->close.file_slot && req->close.fd) | |
4653 | return -EINVAL; | |
4654 | ||
b5dba59e | 4655 | return 0; |
b5dba59e JA |
4656 | } |
4657 | ||
889fca73 | 4658 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4659 | { |
9eac1904 | 4660 | struct files_struct *files = current->files; |
3af73b28 | 4661 | struct io_close *close = &req->close; |
9eac1904 | 4662 | struct fdtable *fdt; |
a1fde923 PB |
4663 | struct file *file = NULL; |
4664 | int ret = -EBADF; | |
b5dba59e | 4665 | |
7df778be PB |
4666 | if (req->close.file_slot) { |
4667 | ret = io_close_fixed(req, issue_flags); | |
4668 | goto err; | |
4669 | } | |
4670 | ||
9eac1904 JA |
4671 | spin_lock(&files->file_lock); |
4672 | fdt = files_fdtable(files); | |
4673 | if (close->fd >= fdt->max_fds) { | |
4674 | spin_unlock(&files->file_lock); | |
4675 | goto err; | |
4676 | } | |
4677 | file = fdt->fd[close->fd]; | |
a1fde923 | 4678 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4679 | spin_unlock(&files->file_lock); |
4680 | file = NULL; | |
4681 | goto err; | |
3af73b28 | 4682 | } |
b5dba59e JA |
4683 | |
4684 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4685 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4686 | spin_unlock(&files->file_lock); |
0bf0eefd | 4687 | return -EAGAIN; |
a2100672 | 4688 | } |
b5dba59e | 4689 | |
9eac1904 JA |
4690 | ret = __close_fd_get_file(close->fd, &file); |
4691 | spin_unlock(&files->file_lock); | |
4692 | if (ret < 0) { | |
4693 | if (ret == -ENOENT) | |
4694 | ret = -EBADF; | |
4695 | goto err; | |
4696 | } | |
4697 | ||
3af73b28 | 4698 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4699 | ret = filp_close(file, current->files); |
4700 | err: | |
3af73b28 | 4701 | if (ret < 0) |
93d2bcd2 | 4702 | req_set_fail(req); |
9eac1904 JA |
4703 | if (file) |
4704 | fput(file); | |
889fca73 | 4705 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4706 | return 0; |
b5dba59e JA |
4707 | } |
4708 | ||
1155c76a | 4709 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4710 | { |
4711 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4712 | |
5d17b4a4 JA |
4713 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4714 | return -EINVAL; | |
26578cda PB |
4715 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4716 | sqe->splice_fd_in)) | |
5d17b4a4 JA |
4717 | return -EINVAL; |
4718 | ||
8ed8d3c3 JA |
4719 | req->sync.off = READ_ONCE(sqe->off); |
4720 | req->sync.len = READ_ONCE(sqe->len); | |
4721 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4722 | return 0; |
4723 | } | |
4724 | ||
45d189c6 | 4725 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4726 | { |
8ed8d3c3 JA |
4727 | int ret; |
4728 | ||
ac45abc0 | 4729 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4730 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4731 | return -EAGAIN; |
4732 | ||
9adbd45d | 4733 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4734 | req->sync.flags); |
4735 | if (ret < 0) | |
93d2bcd2 | 4736 | req_set_fail(req); |
e1e16097 | 4737 | io_req_complete(req, ret); |
5d17b4a4 JA |
4738 | return 0; |
4739 | } | |
4740 | ||
469956e8 | 4741 | #if defined(CONFIG_NET) |
02d27d89 PB |
4742 | static int io_setup_async_msg(struct io_kiocb *req, |
4743 | struct io_async_msghdr *kmsg) | |
4744 | { | |
e8c2bc1f JA |
4745 | struct io_async_msghdr *async_msg = req->async_data; |
4746 | ||
4747 | if (async_msg) | |
02d27d89 | 4748 | return -EAGAIN; |
e8c2bc1f | 4749 | if (io_alloc_async_data(req)) { |
257e84a5 | 4750 | kfree(kmsg->free_iov); |
02d27d89 PB |
4751 | return -ENOMEM; |
4752 | } | |
e8c2bc1f | 4753 | async_msg = req->async_data; |
02d27d89 | 4754 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4755 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4756 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4757 | /* if were using fast_iov, set it to the new one */ |
4758 | if (!async_msg->free_iov) | |
4759 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4760 | ||
02d27d89 PB |
4761 | return -EAGAIN; |
4762 | } | |
4763 | ||
2ae523ed PB |
4764 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4765 | struct io_async_msghdr *iomsg) | |
4766 | { | |
2ae523ed | 4767 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4768 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4769 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4770 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4771 | } |
4772 | ||
93642ef8 PB |
4773 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4774 | { | |
4775 | int ret; | |
4776 | ||
93642ef8 PB |
4777 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4778 | if (!ret) | |
4779 | req->flags |= REQ_F_NEED_CLEANUP; | |
4780 | return ret; | |
4781 | } | |
4782 | ||
3529d8c2 | 4783 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4784 | { |
e47293fd | 4785 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4786 | |
d2b6f48b PB |
4787 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4788 | return -EINVAL; | |
7cb4bf4c JA |
4789 | if (unlikely(sqe->addr2 || sqe->file_index)) |
4790 | return -EINVAL; | |
d2b6f48b | 4791 | |
270a5940 | 4792 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4793 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4794 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4795 | if (sr->msg_flags & MSG_DONTWAIT) | |
4796 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4797 | |
d8768362 JA |
4798 | #ifdef CONFIG_COMPAT |
4799 | if (req->ctx->compat) | |
4800 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4801 | #endif | |
93642ef8 | 4802 | return 0; |
03b1230c JA |
4803 | } |
4804 | ||
889fca73 | 4805 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4806 | { |
6b754c8b | 4807 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4808 | struct socket *sock; |
7a7cacba | 4809 | unsigned flags; |
0031275d | 4810 | int min_ret = 0; |
0fa03c62 JA |
4811 | int ret; |
4812 | ||
dba4a925 | 4813 | sock = sock_from_file(req->file); |
7a7cacba | 4814 | if (unlikely(!sock)) |
dba4a925 | 4815 | return -ENOTSOCK; |
3529d8c2 | 4816 | |
257e84a5 PB |
4817 | kmsg = req->async_data; |
4818 | if (!kmsg) { | |
7a7cacba PB |
4819 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4820 | if (ret) | |
4821 | return ret; | |
4822 | kmsg = &iomsg; | |
0fa03c62 | 4823 | } |
0fa03c62 | 4824 | |
04411806 PB |
4825 | flags = req->sr_msg.msg_flags; |
4826 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4827 | flags |= MSG_DONTWAIT; |
0031275d SM |
4828 | if (flags & MSG_WAITALL) |
4829 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4830 | ||
7a7cacba | 4831 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4832 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4833 | return io_setup_async_msg(req, kmsg); |
4834 | if (ret == -ERESTARTSYS) | |
4835 | ret = -EINTR; | |
0fa03c62 | 4836 | |
257e84a5 PB |
4837 | /* fast path, check for non-NULL to avoid function call */ |
4838 | if (kmsg->free_iov) | |
4839 | kfree(kmsg->free_iov); | |
99bc4c38 | 4840 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4841 | if (ret < min_ret) |
93d2bcd2 | 4842 | req_set_fail(req); |
889fca73 | 4843 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4844 | return 0; |
03b1230c | 4845 | } |
aa1fa28f | 4846 | |
889fca73 | 4847 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4848 | { |
7a7cacba PB |
4849 | struct io_sr_msg *sr = &req->sr_msg; |
4850 | struct msghdr msg; | |
4851 | struct iovec iov; | |
fddaface | 4852 | struct socket *sock; |
7a7cacba | 4853 | unsigned flags; |
0031275d | 4854 | int min_ret = 0; |
fddaface JA |
4855 | int ret; |
4856 | ||
dba4a925 | 4857 | sock = sock_from_file(req->file); |
7a7cacba | 4858 | if (unlikely(!sock)) |
dba4a925 | 4859 | return -ENOTSOCK; |
fddaface | 4860 | |
7a7cacba PB |
4861 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4862 | if (unlikely(ret)) | |
14db8411 | 4863 | return ret; |
fddaface | 4864 | |
7a7cacba PB |
4865 | msg.msg_name = NULL; |
4866 | msg.msg_control = NULL; | |
4867 | msg.msg_controllen = 0; | |
4868 | msg.msg_namelen = 0; | |
fddaface | 4869 | |
04411806 PB |
4870 | flags = req->sr_msg.msg_flags; |
4871 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4872 | flags |= MSG_DONTWAIT; |
0031275d SM |
4873 | if (flags & MSG_WAITALL) |
4874 | min_ret = iov_iter_count(&msg.msg_iter); | |
4875 | ||
7a7cacba PB |
4876 | msg.msg_flags = flags; |
4877 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4878 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4879 | return -EAGAIN; |
4880 | if (ret == -ERESTARTSYS) | |
4881 | ret = -EINTR; | |
fddaface | 4882 | |
0031275d | 4883 | if (ret < min_ret) |
93d2bcd2 | 4884 | req_set_fail(req); |
889fca73 | 4885 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4886 | return 0; |
fddaface JA |
4887 | } |
4888 | ||
1400e697 PB |
4889 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4890 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4891 | { |
4892 | struct io_sr_msg *sr = &req->sr_msg; | |
4893 | struct iovec __user *uiov; | |
4894 | size_t iov_len; | |
4895 | int ret; | |
4896 | ||
1400e697 PB |
4897 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4898 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4899 | if (ret) |
4900 | return ret; | |
4901 | ||
4902 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4903 | if (iov_len > 1) | |
4904 | return -EINVAL; | |
5476dfed | 4905 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4906 | return -EFAULT; |
5476dfed | 4907 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4908 | iomsg->free_iov = NULL; |
52de1fe1 | 4909 | } else { |
257e84a5 | 4910 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4911 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4912 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4913 | false); |
52de1fe1 JA |
4914 | if (ret > 0) |
4915 | ret = 0; | |
4916 | } | |
4917 | ||
4918 | return ret; | |
4919 | } | |
4920 | ||
4921 | #ifdef CONFIG_COMPAT | |
4922 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4923 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4924 | { |
52de1fe1 JA |
4925 | struct io_sr_msg *sr = &req->sr_msg; |
4926 | struct compat_iovec __user *uiov; | |
4927 | compat_uptr_t ptr; | |
4928 | compat_size_t len; | |
4929 | int ret; | |
4930 | ||
4af3417a PB |
4931 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4932 | &ptr, &len); | |
52de1fe1 JA |
4933 | if (ret) |
4934 | return ret; | |
4935 | ||
4936 | uiov = compat_ptr(ptr); | |
4937 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4938 | compat_ssize_t clen; | |
4939 | ||
4940 | if (len > 1) | |
4941 | return -EINVAL; | |
4942 | if (!access_ok(uiov, sizeof(*uiov))) | |
4943 | return -EFAULT; | |
4944 | if (__get_user(clen, &uiov->iov_len)) | |
4945 | return -EFAULT; | |
4946 | if (clen < 0) | |
4947 | return -EINVAL; | |
2d280bc8 | 4948 | sr->len = clen; |
257e84a5 | 4949 | iomsg->free_iov = NULL; |
52de1fe1 | 4950 | } else { |
257e84a5 | 4951 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4952 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4953 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4954 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4955 | if (ret < 0) |
4956 | return ret; | |
4957 | } | |
4958 | ||
4959 | return 0; | |
4960 | } | |
4961 | #endif | |
4962 | ||
1400e697 PB |
4963 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4964 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4965 | { |
1400e697 | 4966 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4967 | |
4968 | #ifdef CONFIG_COMPAT | |
4969 | if (req->ctx->compat) | |
1400e697 | 4970 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4971 | #endif |
52de1fe1 | 4972 | |
1400e697 | 4973 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4974 | } |
4975 | ||
bcda7baa | 4976 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4977 | bool needs_lock) |
bcda7baa JA |
4978 | { |
4979 | struct io_sr_msg *sr = &req->sr_msg; | |
4980 | struct io_buffer *kbuf; | |
4981 | ||
bcda7baa JA |
4982 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4983 | if (IS_ERR(kbuf)) | |
4984 | return kbuf; | |
4985 | ||
4986 | sr->kbuf = kbuf; | |
4987 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4988 | return kbuf; |
fddaface JA |
4989 | } |
4990 | ||
7fbb1b54 PB |
4991 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4992 | { | |
4993 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4994 | } | |
4995 | ||
93642ef8 | 4996 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4997 | { |
99bc4c38 | 4998 | int ret; |
3529d8c2 | 4999 | |
93642ef8 PB |
5000 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
5001 | if (!ret) | |
5002 | req->flags |= REQ_F_NEED_CLEANUP; | |
5003 | return ret; | |
5004 | } | |
5005 | ||
5006 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5007 | { | |
5008 | struct io_sr_msg *sr = &req->sr_msg; | |
5009 | ||
d2b6f48b PB |
5010 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5011 | return -EINVAL; | |
5012 | ||
270a5940 | 5013 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 5014 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 5015 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
5016 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
5017 | if (sr->msg_flags & MSG_DONTWAIT) | |
5018 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 5019 | |
d8768362 JA |
5020 | #ifdef CONFIG_COMPAT |
5021 | if (req->ctx->compat) | |
5022 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
5023 | #endif | |
93642ef8 | 5024 | return 0; |
aa1fa28f JA |
5025 | } |
5026 | ||
889fca73 | 5027 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 5028 | { |
6b754c8b | 5029 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 5030 | struct socket *sock; |
7fbb1b54 | 5031 | struct io_buffer *kbuf; |
7a7cacba | 5032 | unsigned flags; |
0031275d | 5033 | int min_ret = 0; |
52de1fe1 | 5034 | int ret, cflags = 0; |
45d189c6 | 5035 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 5036 | |
dba4a925 | 5037 | sock = sock_from_file(req->file); |
7a7cacba | 5038 | if (unlikely(!sock)) |
dba4a925 | 5039 | return -ENOTSOCK; |
3529d8c2 | 5040 | |
257e84a5 PB |
5041 | kmsg = req->async_data; |
5042 | if (!kmsg) { | |
7a7cacba PB |
5043 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
5044 | if (ret) | |
681fda8d | 5045 | return ret; |
7a7cacba PB |
5046 | kmsg = &iomsg; |
5047 | } | |
03b1230c | 5048 | |
bc02ef33 | 5049 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 5050 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 5051 | if (IS_ERR(kbuf)) |
52de1fe1 | 5052 | return PTR_ERR(kbuf); |
7a7cacba | 5053 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
5054 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
5055 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
5056 | 1, req->sr_msg.len); |
5057 | } | |
52de1fe1 | 5058 | |
04411806 PB |
5059 | flags = req->sr_msg.msg_flags; |
5060 | if (force_nonblock) | |
7a7cacba | 5061 | flags |= MSG_DONTWAIT; |
0031275d SM |
5062 | if (flags & MSG_WAITALL) |
5063 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
5064 | ||
7a7cacba PB |
5065 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
5066 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
5067 | if (force_nonblock && ret == -EAGAIN) |
5068 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
5069 | if (ret == -ERESTARTSYS) |
5070 | ret = -EINTR; | |
03b1230c | 5071 | |
7fbb1b54 PB |
5072 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5073 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
5074 | /* fast path, check for non-NULL to avoid function call */ |
5075 | if (kmsg->free_iov) | |
5076 | kfree(kmsg->free_iov); | |
99bc4c38 | 5077 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 5078 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 5079 | req_set_fail(req); |
889fca73 | 5080 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 5081 | return 0; |
0fa03c62 | 5082 | } |
5d17b4a4 | 5083 | |
889fca73 | 5084 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 5085 | { |
6b754c8b | 5086 | struct io_buffer *kbuf; |
7a7cacba PB |
5087 | struct io_sr_msg *sr = &req->sr_msg; |
5088 | struct msghdr msg; | |
5089 | void __user *buf = sr->buf; | |
fddaface | 5090 | struct socket *sock; |
7a7cacba PB |
5091 | struct iovec iov; |
5092 | unsigned flags; | |
0031275d | 5093 | int min_ret = 0; |
bcda7baa | 5094 | int ret, cflags = 0; |
45d189c6 | 5095 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 5096 | |
dba4a925 | 5097 | sock = sock_from_file(req->file); |
7a7cacba | 5098 | if (unlikely(!sock)) |
dba4a925 | 5099 | return -ENOTSOCK; |
fddaface | 5100 | |
bc02ef33 | 5101 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 5102 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
5103 | if (IS_ERR(kbuf)) |
5104 | return PTR_ERR(kbuf); | |
7a7cacba | 5105 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 5106 | } |
bcda7baa | 5107 | |
7a7cacba | 5108 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
5109 | if (unlikely(ret)) |
5110 | goto out_free; | |
fddaface | 5111 | |
7a7cacba PB |
5112 | msg.msg_name = NULL; |
5113 | msg.msg_control = NULL; | |
5114 | msg.msg_controllen = 0; | |
5115 | msg.msg_namelen = 0; | |
5116 | msg.msg_iocb = NULL; | |
5117 | msg.msg_flags = 0; | |
fddaface | 5118 | |
04411806 PB |
5119 | flags = req->sr_msg.msg_flags; |
5120 | if (force_nonblock) | |
7a7cacba | 5121 | flags |= MSG_DONTWAIT; |
0031275d SM |
5122 | if (flags & MSG_WAITALL) |
5123 | min_ret = iov_iter_count(&msg.msg_iter); | |
5124 | ||
7a7cacba PB |
5125 | ret = sock_recvmsg(sock, &msg, flags); |
5126 | if (force_nonblock && ret == -EAGAIN) | |
5127 | return -EAGAIN; | |
5128 | if (ret == -ERESTARTSYS) | |
5129 | ret = -EINTR; | |
14c32eee | 5130 | out_free: |
7fbb1b54 PB |
5131 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5132 | cflags = io_put_recv_kbuf(req); | |
0031275d | 5133 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 5134 | req_set_fail(req); |
889fca73 | 5135 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5136 | return 0; |
fddaface JA |
5137 | } |
5138 | ||
3529d8c2 | 5139 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5140 | { |
8ed8d3c3 JA |
5141 | struct io_accept *accept = &req->accept; |
5142 | ||
14587a46 | 5143 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5144 | return -EINVAL; |
aaa4db12 | 5145 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5146 | return -EINVAL; |
5147 | ||
d55e5f5b JA |
5148 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5149 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5150 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5151 | accept->nofile = rlimit(RLIMIT_NOFILE); |
a7083ad5 | 5152 | |
aaa4db12 | 5153 | accept->file_slot = READ_ONCE(sqe->file_index); |
56a8e15e | 5154 | if (accept->file_slot && (accept->flags & SOCK_CLOEXEC)) |
aaa4db12 | 5155 | return -EINVAL; |
a7083ad5 PB |
5156 | if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) |
5157 | return -EINVAL; | |
5158 | if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) | |
5159 | accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; | |
8ed8d3c3 | 5160 | return 0; |
8ed8d3c3 | 5161 | } |
17f2fe35 | 5162 | |
889fca73 | 5163 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5164 | { |
5165 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5166 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5167 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
aaa4db12 | 5168 | bool fixed = !!accept->file_slot; |
a7083ad5 PB |
5169 | struct file *file; |
5170 | int ret, fd; | |
8ed8d3c3 | 5171 | |
e697deed JX |
5172 | if (req->file->f_flags & O_NONBLOCK) |
5173 | req->flags |= REQ_F_NOWAIT; | |
5174 | ||
aaa4db12 PB |
5175 | if (!fixed) { |
5176 | fd = __get_unused_fd_flags(accept->flags, accept->nofile); | |
5177 | if (unlikely(fd < 0)) | |
5178 | return fd; | |
5179 | } | |
a7083ad5 PB |
5180 | file = do_accept(req->file, file_flags, accept->addr, accept->addr_len, |
5181 | accept->flags); | |
5182 | if (IS_ERR(file)) { | |
aaa4db12 PB |
5183 | if (!fixed) |
5184 | put_unused_fd(fd); | |
a7083ad5 PB |
5185 | ret = PTR_ERR(file); |
5186 | if (ret == -EAGAIN && force_nonblock) | |
5187 | return -EAGAIN; | |
ac45abc0 PB |
5188 | if (ret == -ERESTARTSYS) |
5189 | ret = -EINTR; | |
93d2bcd2 | 5190 | req_set_fail(req); |
aaa4db12 | 5191 | } else if (!fixed) { |
a7083ad5 PB |
5192 | fd_install(fd, file); |
5193 | ret = fd; | |
aaa4db12 PB |
5194 | } else { |
5195 | ret = io_install_fixed_file(req, file, issue_flags, | |
5196 | accept->file_slot - 1); | |
ac45abc0 | 5197 | } |
889fca73 | 5198 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5199 | return 0; |
8ed8d3c3 JA |
5200 | } |
5201 | ||
93642ef8 PB |
5202 | static int io_connect_prep_async(struct io_kiocb *req) |
5203 | { | |
5204 | struct io_async_connect *io = req->async_data; | |
5205 | struct io_connect *conn = &req->connect; | |
5206 | ||
5207 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
5208 | } | |
5209 | ||
3529d8c2 | 5210 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5211 | { |
3529d8c2 | 5212 | struct io_connect *conn = &req->connect; |
f499a021 | 5213 | |
14587a46 | 5214 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 | 5215 | return -EINVAL; |
26578cda PB |
5216 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags || |
5217 | sqe->splice_fd_in) | |
3fbb51c1 JA |
5218 | return -EINVAL; |
5219 | ||
3529d8c2 JA |
5220 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5221 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 5222 | return 0; |
f499a021 JA |
5223 | } |
5224 | ||
889fca73 | 5225 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5226 | { |
e8c2bc1f | 5227 | struct io_async_connect __io, *io; |
f8e85cf2 | 5228 | unsigned file_flags; |
3fbb51c1 | 5229 | int ret; |
45d189c6 | 5230 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5231 | |
e8c2bc1f JA |
5232 | if (req->async_data) { |
5233 | io = req->async_data; | |
f499a021 | 5234 | } else { |
3529d8c2 JA |
5235 | ret = move_addr_to_kernel(req->connect.addr, |
5236 | req->connect.addr_len, | |
e8c2bc1f | 5237 | &__io.address); |
f499a021 JA |
5238 | if (ret) |
5239 | goto out; | |
5240 | io = &__io; | |
5241 | } | |
5242 | ||
3fbb51c1 JA |
5243 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5244 | ||
e8c2bc1f | 5245 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5246 | req->connect.addr_len, file_flags); |
87f80d62 | 5247 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5248 | if (req->async_data) |
b7bb4f7d | 5249 | return -EAGAIN; |
e8c2bc1f | 5250 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5251 | ret = -ENOMEM; |
5252 | goto out; | |
5253 | } | |
e8c2bc1f | 5254 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 5255 | return -EAGAIN; |
f499a021 | 5256 | } |
f8e85cf2 JA |
5257 | if (ret == -ERESTARTSYS) |
5258 | ret = -EINTR; | |
f499a021 | 5259 | out: |
4e88d6e7 | 5260 | if (ret < 0) |
93d2bcd2 | 5261 | req_set_fail(req); |
889fca73 | 5262 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5263 | return 0; |
469956e8 Y |
5264 | } |
5265 | #else /* !CONFIG_NET */ | |
99a10081 JA |
5266 | #define IO_NETOP_FN(op) \ |
5267 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
5268 | { \ | |
5269 | return -EOPNOTSUPP; \ | |
5270 | } | |
5271 | ||
5272 | #define IO_NETOP_PREP(op) \ | |
5273 | IO_NETOP_FN(op) \ | |
5274 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
5275 | { \ | |
5276 | return -EOPNOTSUPP; \ | |
5277 | } \ | |
5278 | ||
5279 | #define IO_NETOP_PREP_ASYNC(op) \ | |
5280 | IO_NETOP_PREP(op) \ | |
5281 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
5282 | { \ | |
5283 | return -EOPNOTSUPP; \ | |
5284 | } | |
5285 | ||
5286 | IO_NETOP_PREP_ASYNC(sendmsg); | |
5287 | IO_NETOP_PREP_ASYNC(recvmsg); | |
5288 | IO_NETOP_PREP_ASYNC(connect); | |
5289 | IO_NETOP_PREP(accept); | |
5290 | IO_NETOP_FN(send); | |
5291 | IO_NETOP_FN(recv); | |
469956e8 | 5292 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5293 | |
d7718a9d JA |
5294 | struct io_poll_table { |
5295 | struct poll_table_struct pt; | |
5296 | struct io_kiocb *req; | |
68b11e8b | 5297 | int nr_entries; |
d7718a9d JA |
5298 | int error; |
5299 | }; | |
ce593a6c | 5300 | |
d7718a9d | 5301 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5b0a6acc | 5302 | __poll_t mask, io_req_tw_func_t func) |
d7718a9d | 5303 | { |
d7718a9d JA |
5304 | /* for instances that support it check for an event match first: */ |
5305 | if (mask && !(mask & poll->events)) | |
5306 | return 0; | |
5307 | ||
5308 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5309 | ||
5310 | list_del_init(&poll->wait.entry); | |
5311 | ||
d7718a9d | 5312 | req->result = mask; |
5b0a6acc | 5313 | req->io_task_work.func = func; |
6d816e08 | 5314 | |
d7718a9d | 5315 | /* |
e3aabf95 JA |
5316 | * If this fails, then the task is exiting. When a task exits, the |
5317 | * work gets canceled, so just cancel this request as well instead | |
5318 | * of executing it. We can't safely execute it anyway, as we may not | |
5319 | * have the needed state needed for it anyway. | |
d7718a9d | 5320 | */ |
e09ee510 | 5321 | io_req_task_work_add(req); |
d7718a9d JA |
5322 | return 1; |
5323 | } | |
5324 | ||
74ce6ce4 JA |
5325 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5326 | __acquires(&req->ctx->completion_lock) | |
5327 | { | |
5328 | struct io_ring_ctx *ctx = req->ctx; | |
5329 | ||
316319e8 | 5330 | /* req->task == current here, checking PF_EXITING is safe */ |
e09ee510 PB |
5331 | if (unlikely(req->task->flags & PF_EXITING)) |
5332 | WRITE_ONCE(poll->canceled, true); | |
5333 | ||
74ce6ce4 JA |
5334 | if (!req->result && !READ_ONCE(poll->canceled)) { |
5335 | struct poll_table_struct pt = { ._key = poll->events }; | |
5336 | ||
5337 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5338 | } | |
5339 | ||
79ebeaee | 5340 | spin_lock(&ctx->completion_lock); |
74ce6ce4 JA |
5341 | if (!req->result && !READ_ONCE(poll->canceled)) { |
5342 | add_wait_queue(poll->head, &poll->wait); | |
5343 | return true; | |
5344 | } | |
5345 | ||
5346 | return false; | |
5347 | } | |
5348 | ||
d4e7cd36 | 5349 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5350 | { |
e8c2bc1f | 5351 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5352 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5353 | return req->async_data; |
d4e7cd36 JA |
5354 | return req->apoll->double_poll; |
5355 | } | |
5356 | ||
5357 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5358 | { | |
5359 | if (req->opcode == IORING_OP_POLL_ADD) | |
5360 | return &req->poll; | |
5361 | return &req->apoll->poll; | |
5362 | } | |
5363 | ||
5364 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 5365 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
5366 | { |
5367 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5368 | |
5369 | lockdep_assert_held(&req->ctx->completion_lock); | |
5370 | ||
5371 | if (poll && poll->head) { | |
5372 | struct wait_queue_head *head = poll->head; | |
5373 | ||
79ebeaee | 5374 | spin_lock_irq(&head->lock); |
18bceab1 JA |
5375 | list_del_init(&poll->wait.entry); |
5376 | if (poll->wait.private) | |
de9b4cca | 5377 | req_ref_put(req); |
18bceab1 | 5378 | poll->head = NULL; |
79ebeaee | 5379 | spin_unlock_irq(&head->lock); |
18bceab1 JA |
5380 | } |
5381 | } | |
5382 | ||
31efe48e | 5383 | static bool __io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 5384 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
5385 | { |
5386 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 5387 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 5388 | int error; |
18bceab1 | 5389 | |
e27414be | 5390 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 5391 | error = -ECANCELED; |
88e41cf9 | 5392 | req->poll.events |= EPOLLONESHOT; |
e27414be | 5393 | } else { |
5082620f | 5394 | error = mangle_poll(mask); |
e27414be | 5395 | } |
b69de288 JA |
5396 | if (req->poll.events & EPOLLONESHOT) |
5397 | flags = 0; | |
a62682f9 HX |
5398 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5399 | req->poll.events |= EPOLLONESHOT; | |
88e41cf9 | 5400 | flags = 0; |
a62682f9 | 5401 | } |
7b289c38 HX |
5402 | if (flags & IORING_CQE_F_MORE) |
5403 | ctx->cq_extra++; | |
18bceab1 | 5404 | |
88e41cf9 | 5405 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
5406 | } |
5407 | ||
31efe48e XW |
5408 | static inline bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
5409 | __must_hold(&req->ctx->completion_lock) | |
5410 | { | |
5411 | bool done; | |
5412 | ||
5413 | done = __io_poll_complete(req, mask); | |
5414 | io_commit_cqring(req->ctx); | |
5415 | return done; | |
5416 | } | |
5417 | ||
f237c30a | 5418 | static void io_poll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 JA |
5419 | { |
5420 | struct io_ring_ctx *ctx = req->ctx; | |
dd221f46 | 5421 | struct io_kiocb *nxt; |
18bceab1 JA |
5422 | |
5423 | if (io_poll_rewait(req, &req->poll)) { | |
79ebeaee | 5424 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5425 | } else { |
f40b964a | 5426 | bool done; |
18bceab1 | 5427 | |
5b7aa38d HX |
5428 | if (req->poll.done) { |
5429 | spin_unlock(&ctx->completion_lock); | |
5430 | return; | |
5431 | } | |
31efe48e | 5432 | done = __io_poll_complete(req, req->result); |
88e41cf9 | 5433 | if (done) { |
a890d01e | 5434 | io_poll_remove_double(req); |
88e41cf9 | 5435 | hash_del(&req->hash_node); |
bd99c71b | 5436 | req->poll.done = true; |
f40b964a | 5437 | } else { |
88e41cf9 JA |
5438 | req->result = 0; |
5439 | add_wait_queue(req->poll.head, &req->poll.wait); | |
5440 | } | |
31efe48e | 5441 | io_commit_cqring(ctx); |
79ebeaee | 5442 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5443 | io_cqring_ev_posted(ctx); |
18bceab1 | 5444 | |
88e41cf9 JA |
5445 | if (done) { |
5446 | nxt = io_put_req_find_next(req); | |
5447 | if (nxt) | |
f237c30a | 5448 | io_req_task_submit(nxt, locked); |
88e41cf9 | 5449 | } |
dd221f46 | 5450 | } |
18bceab1 JA |
5451 | } |
5452 | ||
5453 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5454 | int sync, void *key) | |
5455 | { | |
5456 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5457 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 | 5458 | __poll_t mask = key_to_poll(key); |
79ebeaee | 5459 | unsigned long flags; |
18bceab1 JA |
5460 | |
5461 | /* for instances that support it check for an event match first: */ | |
5462 | if (mask && !(mask & poll->events)) | |
5463 | return 0; | |
88e41cf9 JA |
5464 | if (!(poll->events & EPOLLONESHOT)) |
5465 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 5466 | |
8706e04e JA |
5467 | list_del_init(&wait->entry); |
5468 | ||
9ce85ef2 | 5469 | if (poll->head) { |
18bceab1 JA |
5470 | bool done; |
5471 | ||
79ebeaee | 5472 | spin_lock_irqsave(&poll->head->lock, flags); |
807abcb0 | 5473 | done = list_empty(&poll->wait.entry); |
18bceab1 | 5474 | if (!done) |
807abcb0 | 5475 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5476 | /* make sure double remove sees this as being gone */ |
5477 | wait->private = NULL; | |
79ebeaee | 5478 | spin_unlock_irqrestore(&poll->head->lock, flags); |
c8b5e260 JA |
5479 | if (!done) { |
5480 | /* use wait func handler, so it matches the rq type */ | |
5481 | poll->wait.func(&poll->wait, mode, sync, key); | |
5482 | } | |
18bceab1 | 5483 | } |
de9b4cca | 5484 | req_ref_put(req); |
18bceab1 JA |
5485 | return 1; |
5486 | } | |
5487 | ||
5488 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5489 | wait_queue_func_t wake_func) | |
5490 | { | |
5491 | poll->head = NULL; | |
5492 | poll->done = false; | |
5493 | poll->canceled = false; | |
464dca61 JA |
5494 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5495 | /* mask in events that we always want/need */ | |
5496 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5497 | INIT_LIST_HEAD(&poll->wait.entry); |
5498 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5499 | } | |
5500 | ||
5501 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5502 | struct wait_queue_head *head, |
5503 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5504 | { |
5505 | struct io_kiocb *req = pt->req; | |
5506 | ||
5507 | /* | |
68b11e8b PB |
5508 | * The file being polled uses multiple waitqueues for poll handling |
5509 | * (e.g. one for read, one for write). Setup a separate io_poll_iocb | |
5510 | * if this happens. | |
18bceab1 | 5511 | */ |
68b11e8b | 5512 | if (unlikely(pt->nr_entries)) { |
58852d4d PB |
5513 | struct io_poll_iocb *poll_one = poll; |
5514 | ||
23a65db8 PB |
5515 | /* double add on the same waitqueue head, ignore */ |
5516 | if (poll_one->head == head) | |
5517 | return; | |
18bceab1 | 5518 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5519 | if (*poll_ptr) { |
23a65db8 PB |
5520 | if ((*poll_ptr)->head == head) |
5521 | return; | |
18bceab1 JA |
5522 | pt->error = -EINVAL; |
5523 | return; | |
5524 | } | |
ea6a693d JA |
5525 | /* |
5526 | * Can't handle multishot for double wait for now, turn it | |
5527 | * into one-shot mode. | |
5528 | */ | |
7a274727 PB |
5529 | if (!(poll_one->events & EPOLLONESHOT)) |
5530 | poll_one->events |= EPOLLONESHOT; | |
18bceab1 JA |
5531 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5532 | if (!poll) { | |
5533 | pt->error = -ENOMEM; | |
5534 | return; | |
5535 | } | |
58852d4d | 5536 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5537 | req_ref_get(req); |
18bceab1 | 5538 | poll->wait.private = req; |
807abcb0 | 5539 | *poll_ptr = poll; |
18bceab1 JA |
5540 | } |
5541 | ||
68b11e8b | 5542 | pt->nr_entries++; |
18bceab1 | 5543 | poll->head = head; |
a31eb4a2 JX |
5544 | |
5545 | if (poll->events & EPOLLEXCLUSIVE) | |
5546 | add_wait_queue_exclusive(head, &poll->wait); | |
5547 | else | |
5548 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5549 | } |
5550 | ||
5551 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5552 | struct poll_table_struct *p) | |
5553 | { | |
5554 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5555 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5556 | |
807abcb0 | 5557 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5558 | } |
5559 | ||
f237c30a | 5560 | static void io_async_task_func(struct io_kiocb *req, bool *locked) |
d7718a9d | 5561 | { |
d7718a9d JA |
5562 | struct async_poll *apoll = req->apoll; |
5563 | struct io_ring_ctx *ctx = req->ctx; | |
5564 | ||
236daeae | 5565 | trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data); |
d7718a9d | 5566 | |
74ce6ce4 | 5567 | if (io_poll_rewait(req, &apoll->poll)) { |
79ebeaee | 5568 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5569 | return; |
d7718a9d JA |
5570 | } |
5571 | ||
0ea13b44 | 5572 | hash_del(&req->hash_node); |
d4e7cd36 | 5573 | io_poll_remove_double(req); |
bd99c71b | 5574 | apoll->poll.done = true; |
79ebeaee | 5575 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5576 | |
0be0b0e3 | 5577 | if (!READ_ONCE(apoll->poll.canceled)) |
f237c30a | 5578 | io_req_task_submit(req, locked); |
0be0b0e3 | 5579 | else |
2593553a | 5580 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5581 | } |
5582 | ||
5583 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5584 | void *key) | |
5585 | { | |
5586 | struct io_kiocb *req = wait->private; | |
5587 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5588 | ||
5589 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5590 | key_to_poll(key)); | |
5591 | ||
5592 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5593 | } | |
5594 | ||
5595 | static void io_poll_req_insert(struct io_kiocb *req) | |
5596 | { | |
5597 | struct io_ring_ctx *ctx = req->ctx; | |
5598 | struct hlist_head *list; | |
5599 | ||
5600 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5601 | hlist_add_head(&req->hash_node, list); | |
5602 | } | |
5603 | ||
5604 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5605 | struct io_poll_iocb *poll, | |
5606 | struct io_poll_table *ipt, __poll_t mask, | |
5607 | wait_queue_func_t wake_func) | |
5608 | __acquires(&ctx->completion_lock) | |
5609 | { | |
5610 | struct io_ring_ctx *ctx = req->ctx; | |
5611 | bool cancel = false; | |
5612 | ||
4d52f338 | 5613 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5614 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5615 | poll->file = req->file; |
18bceab1 | 5616 | poll->wait.private = req; |
d7718a9d JA |
5617 | |
5618 | ipt->pt._key = mask; | |
5619 | ipt->req = req; | |
68b11e8b PB |
5620 | ipt->error = 0; |
5621 | ipt->nr_entries = 0; | |
d7718a9d | 5622 | |
d7718a9d | 5623 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
68b11e8b PB |
5624 | if (unlikely(!ipt->nr_entries) && !ipt->error) |
5625 | ipt->error = -EINVAL; | |
d7718a9d | 5626 | |
79ebeaee | 5627 | spin_lock(&ctx->completion_lock); |
a890d01e | 5628 | if (ipt->error || (mask && (poll->events & EPOLLONESHOT))) |
46fee9ab | 5629 | io_poll_remove_double(req); |
d7718a9d | 5630 | if (likely(poll->head)) { |
79ebeaee | 5631 | spin_lock_irq(&poll->head->lock); |
d7718a9d JA |
5632 | if (unlikely(list_empty(&poll->wait.entry))) { |
5633 | if (ipt->error) | |
5634 | cancel = true; | |
5635 | ipt->error = 0; | |
5636 | mask = 0; | |
5637 | } | |
88e41cf9 | 5638 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5639 | list_del_init(&poll->wait.entry); |
5640 | else if (cancel) | |
5641 | WRITE_ONCE(poll->canceled, true); | |
5642 | else if (!poll->done) /* actually waiting for an event */ | |
5643 | io_poll_req_insert(req); | |
79ebeaee | 5644 | spin_unlock_irq(&poll->head->lock); |
d7718a9d JA |
5645 | } |
5646 | ||
5647 | return mask; | |
5648 | } | |
5649 | ||
59b735ae OL |
5650 | enum { |
5651 | IO_APOLL_OK, | |
5652 | IO_APOLL_ABORTED, | |
5653 | IO_APOLL_READY | |
5654 | }; | |
5655 | ||
5656 | static int io_arm_poll_handler(struct io_kiocb *req) | |
d7718a9d JA |
5657 | { |
5658 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5659 | struct io_ring_ctx *ctx = req->ctx; | |
5660 | struct async_poll *apoll; | |
5661 | struct io_poll_table ipt; | |
b2d9c3da | 5662 | __poll_t ret, mask = EPOLLONESHOT | POLLERR | POLLPRI; |
9dab14b8 | 5663 | int rw; |
d7718a9d JA |
5664 | |
5665 | if (!req->file || !file_can_poll(req->file)) | |
59b735ae | 5666 | return IO_APOLL_ABORTED; |
24c74678 | 5667 | if (req->flags & REQ_F_POLLED) |
59b735ae | 5668 | return IO_APOLL_ABORTED; |
b2d9c3da PB |
5669 | if (!def->pollin && !def->pollout) |
5670 | return IO_APOLL_ABORTED; | |
5671 | ||
5672 | if (def->pollin) { | |
9dab14b8 | 5673 | rw = READ; |
b2d9c3da PB |
5674 | mask |= POLLIN | POLLRDNORM; |
5675 | ||
5676 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5677 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5678 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5679 | mask &= ~POLLIN; | |
5680 | } else { | |
9dab14b8 | 5681 | rw = WRITE; |
b2d9c3da PB |
5682 | mask |= POLLOUT | POLLWRNORM; |
5683 | } | |
5684 | ||
9dab14b8 | 5685 | /* if we can't nonblock try, then no point in arming a poll handler */ |
b191e2df | 5686 | if (!io_file_supports_nowait(req, rw)) |
59b735ae | 5687 | return IO_APOLL_ABORTED; |
d7718a9d JA |
5688 | |
5689 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5690 | if (unlikely(!apoll)) | |
59b735ae | 5691 | return IO_APOLL_ABORTED; |
807abcb0 | 5692 | apoll->double_poll = NULL; |
d7718a9d | 5693 | req->apoll = apoll; |
b2d9c3da | 5694 | req->flags |= REQ_F_POLLED; |
d7718a9d | 5695 | ipt.pt._qproc = io_async_queue_proc; |
48dcd38d | 5696 | io_req_set_refcount(req); |
d7718a9d JA |
5697 | |
5698 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5699 | io_async_wake); | |
79ebeaee | 5700 | spin_unlock(&ctx->completion_lock); |
41a5169c HX |
5701 | if (ret || ipt.error) |
5702 | return ret ? IO_APOLL_READY : IO_APOLL_ABORTED; | |
5703 | ||
236daeae OL |
5704 | trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, |
5705 | mask, apoll->poll.events); | |
59b735ae | 5706 | return IO_APOLL_OK; |
d7718a9d JA |
5707 | } |
5708 | ||
5709 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5710 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5711 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5712 | { |
b41e9852 | 5713 | bool do_complete = false; |
221c5eb2 | 5714 | |
5082620f JA |
5715 | if (!poll->head) |
5716 | return false; | |
79ebeaee | 5717 | spin_lock_irq(&poll->head->lock); |
b2e720ac JA |
5718 | if (do_cancel) |
5719 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5720 | if (!list_empty(&poll->wait.entry)) { |
5721 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5722 | do_complete = true; |
221c5eb2 | 5723 | } |
79ebeaee | 5724 | spin_unlock_irq(&poll->head->lock); |
3bfa5bcb | 5725 | hash_del(&req->hash_node); |
d7718a9d JA |
5726 | return do_complete; |
5727 | } | |
5728 | ||
5d709043 | 5729 | static bool io_poll_remove_one(struct io_kiocb *req) |
e07785b0 | 5730 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5731 | { |
5732 | bool do_complete; | |
5733 | ||
d4e7cd36 | 5734 | io_poll_remove_double(req); |
e31001a3 | 5735 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5736 | |
b41e9852 | 5737 | if (do_complete) { |
d4d19c19 | 5738 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5739 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5740 | req_set_fail(req); |
91c2f697 | 5741 | io_put_req_deferred(req); |
5d709043 | 5742 | } |
b41e9852 | 5743 | return do_complete; |
221c5eb2 JA |
5744 | } |
5745 | ||
76e1b642 JA |
5746 | /* |
5747 | * Returns true if we found and killed one or more poll requests | |
5748 | */ | |
6b81928d | 5749 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5750 | bool cancel_all) |
221c5eb2 | 5751 | { |
78076bb6 | 5752 | struct hlist_node *tmp; |
221c5eb2 | 5753 | struct io_kiocb *req; |
8e2e1faf | 5754 | int posted = 0, i; |
221c5eb2 | 5755 | |
79ebeaee | 5756 | spin_lock(&ctx->completion_lock); |
78076bb6 JA |
5757 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5758 | struct hlist_head *list; | |
5759 | ||
5760 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5761 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
f0baed8e | 5762 | if (io_match_task_safe(req, tsk, cancel_all)) |
f3606e3a JA |
5763 | posted += io_poll_remove_one(req); |
5764 | } | |
221c5eb2 | 5765 | } |
79ebeaee | 5766 | spin_unlock(&ctx->completion_lock); |
b41e9852 | 5767 | |
8e2e1faf JA |
5768 | if (posted) |
5769 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5770 | |
5771 | return posted != 0; | |
221c5eb2 JA |
5772 | } |
5773 | ||
9ba5fac8 PB |
5774 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5775 | bool poll_only) | |
e07785b0 | 5776 | __must_hold(&ctx->completion_lock) |
47f46768 | 5777 | { |
78076bb6 | 5778 | struct hlist_head *list; |
47f46768 JA |
5779 | struct io_kiocb *req; |
5780 | ||
78076bb6 JA |
5781 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5782 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5783 | if (sqe_addr != req->user_data) |
5784 | continue; | |
9ba5fac8 PB |
5785 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5786 | continue; | |
b2cb805f | 5787 | return req; |
47f46768 | 5788 | } |
b2cb805f JA |
5789 | return NULL; |
5790 | } | |
5791 | ||
9ba5fac8 PB |
5792 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5793 | bool poll_only) | |
e07785b0 | 5794 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5795 | { |
5796 | struct io_kiocb *req; | |
5797 | ||
9ba5fac8 | 5798 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5799 | if (!req) |
5800 | return -ENOENT; | |
5801 | if (io_poll_remove_one(req)) | |
5802 | return 0; | |
5803 | ||
5804 | return -EALREADY; | |
47f46768 JA |
5805 | } |
5806 | ||
9096af3e PB |
5807 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5808 | unsigned int flags) | |
5809 | { | |
5810 | u32 events; | |
47f46768 | 5811 | |
9096af3e PB |
5812 | events = READ_ONCE(sqe->poll32_events); |
5813 | #ifdef __BIG_ENDIAN | |
5814 | events = swahw32(events); | |
5815 | #endif | |
5816 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5817 | events |= EPOLLONESHOT; | |
5818 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5819 | } |
5820 | ||
c5de0036 | 5821 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5822 | const struct io_uring_sqe *sqe) |
0969e783 | 5823 | { |
c5de0036 PB |
5824 | struct io_poll_update *upd = &req->poll_update; |
5825 | u32 flags; | |
5826 | ||
0969e783 JA |
5827 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5828 | return -EINVAL; | |
26578cda | 5829 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
c5de0036 PB |
5830 | return -EINVAL; |
5831 | flags = READ_ONCE(sqe->len); | |
5832 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5833 | IORING_POLL_ADD_MULTI)) | |
5834 | return -EINVAL; | |
5835 | /* meaningless without update */ | |
5836 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5837 | return -EINVAL; |
5838 | ||
c5de0036 PB |
5839 | upd->old_user_data = READ_ONCE(sqe->addr); |
5840 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5841 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5842 | |
c5de0036 PB |
5843 | upd->new_user_data = READ_ONCE(sqe->off); |
5844 | if (!upd->update_user_data && upd->new_user_data) | |
5845 | return -EINVAL; | |
5846 | if (upd->update_events) | |
5847 | upd->events = io_poll_parse_events(sqe, flags); | |
5848 | else if (sqe->poll32_events) | |
5849 | return -EINVAL; | |
221c5eb2 | 5850 | |
221c5eb2 JA |
5851 | return 0; |
5852 | } | |
5853 | ||
221c5eb2 JA |
5854 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5855 | void *key) | |
5856 | { | |
c2f2eb7d JA |
5857 | struct io_kiocb *req = wait->private; |
5858 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5859 | |
d7718a9d | 5860 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5861 | } |
5862 | ||
221c5eb2 JA |
5863 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5864 | struct poll_table_struct *p) | |
5865 | { | |
5866 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5867 | ||
e8c2bc1f | 5868 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5869 | } |
5870 | ||
3529d8c2 | 5871 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5872 | { |
5873 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5874 | u32 flags; |
221c5eb2 JA |
5875 | |
5876 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5877 | return -EINVAL; | |
c5de0036 | 5878 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5879 | return -EINVAL; |
5880 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5881 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5882 | return -EINVAL; |
5883 | ||
48dcd38d | 5884 | io_req_set_refcount(req); |
c5de0036 | 5885 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5886 | return 0; |
5887 | } | |
5888 | ||
61e98203 | 5889 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5890 | { |
5891 | struct io_poll_iocb *poll = &req->poll; | |
5892 | struct io_ring_ctx *ctx = req->ctx; | |
5893 | struct io_poll_table ipt; | |
0969e783 | 5894 | __poll_t mask; |
5b7aa38d | 5895 | bool done; |
0969e783 | 5896 | |
d7718a9d | 5897 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5898 | |
d7718a9d JA |
5899 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5900 | io_poll_wake); | |
221c5eb2 | 5901 | |
8c838788 | 5902 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5903 | ipt.error = 0; |
5b7aa38d | 5904 | done = io_poll_complete(req, mask); |
221c5eb2 | 5905 | } |
79ebeaee | 5906 | spin_unlock(&ctx->completion_lock); |
221c5eb2 | 5907 | |
8c838788 JA |
5908 | if (mask) { |
5909 | io_cqring_ev_posted(ctx); | |
5b7aa38d | 5910 | if (done) |
88e41cf9 | 5911 | io_put_req(req); |
221c5eb2 | 5912 | } |
8c838788 | 5913 | return ipt.error; |
221c5eb2 JA |
5914 | } |
5915 | ||
c5de0036 | 5916 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5917 | { |
5918 | struct io_ring_ctx *ctx = req->ctx; | |
5919 | struct io_kiocb *preq; | |
cb3b200e | 5920 | bool completing; |
b69de288 JA |
5921 | int ret; |
5922 | ||
79ebeaee | 5923 | spin_lock(&ctx->completion_lock); |
9ba5fac8 | 5924 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5925 | if (!preq) { |
5926 | ret = -ENOENT; | |
5927 | goto err; | |
b69de288 | 5928 | } |
cb3b200e | 5929 | |
c5de0036 PB |
5930 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5931 | completing = true; | |
5932 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5933 | goto err; | |
5934 | } | |
5935 | ||
cb3b200e JA |
5936 | /* |
5937 | * Don't allow racy completion with singleshot, as we cannot safely | |
5938 | * update those. For multishot, if we're racing with completion, just | |
5939 | * let completion re-add it. | |
5940 | */ | |
4470ec55 | 5941 | io_poll_remove_double(preq); |
cb3b200e JA |
5942 | completing = !__io_poll_remove_one(preq, &preq->poll, false); |
5943 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5944 | ret = -EALREADY; | |
5945 | goto err; | |
b69de288 JA |
5946 | } |
5947 | /* we now have a detached poll request. reissue. */ | |
5948 | ret = 0; | |
5949 | err: | |
b69de288 | 5950 | if (ret < 0) { |
79ebeaee | 5951 | spin_unlock(&ctx->completion_lock); |
93d2bcd2 | 5952 | req_set_fail(req); |
b69de288 JA |
5953 | io_req_complete(req, ret); |
5954 | return 0; | |
5955 | } | |
5956 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5957 | if (req->poll_update.update_events) { |
b69de288 | 5958 | preq->poll.events &= ~0xffff; |
9d805892 | 5959 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5960 | preq->poll.events |= IO_POLL_UNMASK; |
5961 | } | |
9d805892 PB |
5962 | if (req->poll_update.update_user_data) |
5963 | preq->user_data = req->poll_update.new_user_data; | |
79ebeaee | 5964 | spin_unlock(&ctx->completion_lock); |
cb3b200e | 5965 | |
b69de288 JA |
5966 | /* complete update request, we're done with it */ |
5967 | io_req_complete(req, ret); | |
5968 | ||
cb3b200e | 5969 | if (!completing) { |
c5de0036 | 5970 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5971 | if (ret < 0) { |
93d2bcd2 | 5972 | req_set_fail(preq); |
cb3b200e JA |
5973 | io_req_complete(preq, ret); |
5974 | } | |
b69de288 JA |
5975 | } |
5976 | return 0; | |
5977 | } | |
5978 | ||
f237c30a | 5979 | static void io_req_task_timeout(struct io_kiocb *req, bool *locked) |
89850fce | 5980 | { |
89850fce | 5981 | req_set_fail(req); |
505657bc | 5982 | io_req_complete_post(req, -ETIME, 0); |
89850fce JA |
5983 | } |
5984 | ||
5262f567 JA |
5985 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5986 | { | |
ad8a48ac JA |
5987 | struct io_timeout_data *data = container_of(timer, |
5988 | struct io_timeout_data, timer); | |
5989 | struct io_kiocb *req = data->req; | |
5990 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5991 | unsigned long flags; |
5992 | ||
89850fce | 5993 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
a71976f3 | 5994 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5995 | atomic_set(&req->ctx->cq_timeouts, |
5996 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
89850fce | 5997 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); |
01cec8c1 | 5998 | |
89850fce JA |
5999 | req->io_task_work.func = io_req_task_timeout; |
6000 | io_req_task_work_add(req); | |
5262f567 JA |
6001 | return HRTIMER_NORESTART; |
6002 | } | |
6003 | ||
fbd15848 PB |
6004 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
6005 | __u64 user_data) | |
89850fce | 6006 | __must_hold(&ctx->timeout_lock) |
f254ac04 | 6007 | { |
fbd15848 | 6008 | struct io_timeout_data *io; |
47f46768 | 6009 | struct io_kiocb *req; |
fd9c7bc5 | 6010 | bool found = false; |
f254ac04 | 6011 | |
135fcde8 | 6012 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
6013 | found = user_data == req->user_data; |
6014 | if (found) | |
47f46768 | 6015 | break; |
47f46768 | 6016 | } |
fd9c7bc5 PB |
6017 | if (!found) |
6018 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
6019 | |
6020 | io = req->async_data; | |
fd9c7bc5 | 6021 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 6022 | return ERR_PTR(-EALREADY); |
a71976f3 | 6023 | list_del_init(&req->timeout.list); |
fbd15848 PB |
6024 | return req; |
6025 | } | |
47f46768 | 6026 | |
fbd15848 | 6027 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
ec3c3d0f | 6028 | __must_hold(&ctx->completion_lock) |
89850fce | 6029 | __must_hold(&ctx->timeout_lock) |
fbd15848 PB |
6030 | { |
6031 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
6032 | ||
6033 | if (IS_ERR(req)) | |
6034 | return PTR_ERR(req); | |
f254ac04 | 6035 | |
93d2bcd2 | 6036 | req_set_fail(req); |
d4d19c19 | 6037 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
91c2f697 | 6038 | io_put_req_deferred(req); |
f254ac04 JA |
6039 | return 0; |
6040 | } | |
6041 | ||
50c1df2b JA |
6042 | static clockid_t io_timeout_get_clock(struct io_timeout_data *data) |
6043 | { | |
6044 | switch (data->flags & IORING_TIMEOUT_CLOCK_MASK) { | |
6045 | case IORING_TIMEOUT_BOOTTIME: | |
6046 | return CLOCK_BOOTTIME; | |
6047 | case IORING_TIMEOUT_REALTIME: | |
6048 | return CLOCK_REALTIME; | |
6049 | default: | |
6050 | /* can't happen, vetted at prep time */ | |
6051 | WARN_ON_ONCE(1); | |
6052 | fallthrough; | |
6053 | case 0: | |
6054 | return CLOCK_MONOTONIC; | |
6055 | } | |
6056 | } | |
6057 | ||
f1042b6c PB |
6058 | static int io_linked_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
6059 | struct timespec64 *ts, enum hrtimer_mode mode) | |
6060 | __must_hold(&ctx->timeout_lock) | |
6061 | { | |
6062 | struct io_timeout_data *io; | |
6063 | struct io_kiocb *req; | |
6064 | bool found = false; | |
6065 | ||
6066 | list_for_each_entry(req, &ctx->ltimeout_list, timeout.list) { | |
6067 | found = user_data == req->user_data; | |
6068 | if (found) | |
6069 | break; | |
6070 | } | |
6071 | if (!found) | |
6072 | return -ENOENT; | |
6073 | ||
6074 | io = req->async_data; | |
6075 | if (hrtimer_try_to_cancel(&io->timer) == -1) | |
6076 | return -EALREADY; | |
6077 | hrtimer_init(&io->timer, io_timeout_get_clock(io), mode); | |
6078 | io->timer.function = io_link_timeout_fn; | |
6079 | hrtimer_start(&io->timer, timespec64_to_ktime(*ts), mode); | |
6080 | return 0; | |
6081 | } | |
6082 | ||
9c8e11b3 PB |
6083 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
6084 | struct timespec64 *ts, enum hrtimer_mode mode) | |
89850fce | 6085 | __must_hold(&ctx->timeout_lock) |
47f46768 | 6086 | { |
9c8e11b3 PB |
6087 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
6088 | struct io_timeout_data *data; | |
47f46768 | 6089 | |
9c8e11b3 PB |
6090 | if (IS_ERR(req)) |
6091 | return PTR_ERR(req); | |
47f46768 | 6092 | |
9c8e11b3 PB |
6093 | req->timeout.off = 0; /* noseq */ |
6094 | data = req->async_data; | |
6095 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
50c1df2b | 6096 | hrtimer_init(&data->timer, io_timeout_get_clock(data), mode); |
9c8e11b3 PB |
6097 | data->timer.function = io_timeout_fn; |
6098 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
6099 | return 0; | |
47f46768 JA |
6100 | } |
6101 | ||
3529d8c2 JA |
6102 | static int io_timeout_remove_prep(struct io_kiocb *req, |
6103 | const struct io_uring_sqe *sqe) | |
b29472ee | 6104 | { |
9c8e11b3 PB |
6105 | struct io_timeout_rem *tr = &req->timeout_rem; |
6106 | ||
b29472ee JA |
6107 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
6108 | return -EINVAL; | |
61710e43 DA |
6109 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6110 | return -EINVAL; | |
26578cda | 6111 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->splice_fd_in) |
b29472ee JA |
6112 | return -EINVAL; |
6113 | ||
f1042b6c | 6114 | tr->ltimeout = false; |
9c8e11b3 PB |
6115 | tr->addr = READ_ONCE(sqe->addr); |
6116 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
f1042b6c PB |
6117 | if (tr->flags & IORING_TIMEOUT_UPDATE_MASK) { |
6118 | if (hweight32(tr->flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
6119 | return -EINVAL; | |
6120 | if (tr->flags & IORING_LINK_TIMEOUT_UPDATE) | |
6121 | tr->ltimeout = true; | |
6122 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE_MASK|IORING_TIMEOUT_ABS)) | |
9c8e11b3 PB |
6123 | return -EINVAL; |
6124 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
6125 | return -EFAULT; | |
6126 | } else if (tr->flags) { | |
6127 | /* timeout removal doesn't support flags */ | |
b29472ee | 6128 | return -EINVAL; |
9c8e11b3 | 6129 | } |
b29472ee | 6130 | |
b29472ee JA |
6131 | return 0; |
6132 | } | |
6133 | ||
8662daec PB |
6134 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
6135 | { | |
6136 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
6137 | : HRTIMER_MODE_REL; | |
6138 | } | |
6139 | ||
11365043 JA |
6140 | /* |
6141 | * Remove or update an existing timeout command | |
6142 | */ | |
61e98203 | 6143 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 6144 | { |
9c8e11b3 | 6145 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 6146 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 6147 | int ret; |
11365043 | 6148 | |
ec3c3d0f PB |
6149 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { |
6150 | spin_lock(&ctx->completion_lock); | |
6151 | spin_lock_irq(&ctx->timeout_lock); | |
9c8e11b3 | 6152 | ret = io_timeout_cancel(ctx, tr->addr); |
ec3c3d0f PB |
6153 | spin_unlock_irq(&ctx->timeout_lock); |
6154 | spin_unlock(&ctx->completion_lock); | |
6155 | } else { | |
f1042b6c PB |
6156 | enum hrtimer_mode mode = io_translate_timeout_mode(tr->flags); |
6157 | ||
ec3c3d0f | 6158 | spin_lock_irq(&ctx->timeout_lock); |
f1042b6c PB |
6159 | if (tr->ltimeout) |
6160 | ret = io_linked_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
6161 | else | |
6162 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
ec3c3d0f PB |
6163 | spin_unlock_irq(&ctx->timeout_lock); |
6164 | } | |
11365043 | 6165 | |
4e88d6e7 | 6166 | if (ret < 0) |
93d2bcd2 | 6167 | req_set_fail(req); |
505657bc | 6168 | io_req_complete_post(req, ret, 0); |
11365043 | 6169 | return 0; |
5262f567 JA |
6170 | } |
6171 | ||
3529d8c2 | 6172 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 6173 | bool is_timeout_link) |
5262f567 | 6174 | { |
ad8a48ac | 6175 | struct io_timeout_data *data; |
a41525ab | 6176 | unsigned flags; |
56080b02 | 6177 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 6178 | |
ad8a48ac | 6179 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 6180 | return -EINVAL; |
26578cda PB |
6181 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1 || |
6182 | sqe->splice_fd_in) | |
a41525ab | 6183 | return -EINVAL; |
56080b02 | 6184 | if (off && is_timeout_link) |
2d28390a | 6185 | return -EINVAL; |
a41525ab | 6186 | flags = READ_ONCE(sqe->timeout_flags); |
50c1df2b JA |
6187 | if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_CLOCK_MASK)) |
6188 | return -EINVAL; | |
6189 | /* more than one clock specified is invalid, obviously */ | |
6190 | if (hweight32(flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
5262f567 | 6191 | return -EINVAL; |
bdf20073 | 6192 | |
ef9dd637 | 6193 | INIT_LIST_HEAD(&req->timeout.list); |
bfe68a22 | 6194 | req->timeout.off = off; |
f18ee4cf PB |
6195 | if (unlikely(off && !req->ctx->off_timeout_used)) |
6196 | req->ctx->off_timeout_used = true; | |
26a61679 | 6197 | |
e8c2bc1f | 6198 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
6199 | return -ENOMEM; |
6200 | ||
e8c2bc1f | 6201 | data = req->async_data; |
ad8a48ac | 6202 | data->req = req; |
50c1df2b | 6203 | data->flags = flags; |
ad8a48ac JA |
6204 | |
6205 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
6206 | return -EFAULT; |
6207 | ||
7abdd4c7 | 6208 | INIT_LIST_HEAD(&req->timeout.list); |
8662daec | 6209 | data->mode = io_translate_timeout_mode(flags); |
50c1df2b | 6210 | hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode); |
b97e736a PB |
6211 | |
6212 | if (is_timeout_link) { | |
6213 | struct io_submit_link *link = &req->ctx->submit_state.link; | |
6214 | ||
6215 | if (!link->head) | |
6216 | return -EINVAL; | |
6217 | if (link->last->opcode == IORING_OP_LINK_TIMEOUT) | |
6218 | return -EINVAL; | |
4d13d1a4 PB |
6219 | req->timeout.head = link->last; |
6220 | link->last->flags |= REQ_F_ARM_LTIMEOUT; | |
b97e736a | 6221 | } |
ad8a48ac JA |
6222 | return 0; |
6223 | } | |
6224 | ||
61e98203 | 6225 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 6226 | { |
ad8a48ac | 6227 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 6228 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 6229 | struct list_head *entry; |
bfe68a22 | 6230 | u32 tail, off = req->timeout.off; |
ad8a48ac | 6231 | |
89850fce | 6232 | spin_lock_irq(&ctx->timeout_lock); |
93bd25bb | 6233 | |
5262f567 JA |
6234 | /* |
6235 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
6236 | * timeout event to be satisfied. If it isn't set, then this is |
6237 | * a pure timeout request, sequence isn't used. | |
5262f567 | 6238 | */ |
8eb7e2d0 | 6239 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
6240 | entry = ctx->timeout_list.prev; |
6241 | goto add; | |
6242 | } | |
5262f567 | 6243 | |
bfe68a22 PB |
6244 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
6245 | req->timeout.target_seq = tail + off; | |
5262f567 | 6246 | |
f010505b MDG |
6247 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
6248 | * This is safe because ->completion_lock is held, and submissions | |
6249 | * and completions are never mixed in the same ->completion_lock section. | |
6250 | */ | |
6251 | ctx->cq_last_tm_flush = tail; | |
6252 | ||
5262f567 JA |
6253 | /* |
6254 | * Insertion sort, ensuring the first entry in the list is always | |
6255 | * the one we need first. | |
6256 | */ | |
5262f567 | 6257 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
6258 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
6259 | timeout.list); | |
5262f567 | 6260 | |
8eb7e2d0 | 6261 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 6262 | continue; |
bfe68a22 PB |
6263 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
6264 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
6265 | break; |
6266 | } | |
93bd25bb | 6267 | add: |
135fcde8 | 6268 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
6269 | data->timer.function = io_timeout_fn; |
6270 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
89850fce | 6271 | spin_unlock_irq(&ctx->timeout_lock); |
5262f567 JA |
6272 | return 0; |
6273 | } | |
5262f567 | 6274 | |
f458dd84 PB |
6275 | struct io_cancel_data { |
6276 | struct io_ring_ctx *ctx; | |
6277 | u64 user_data; | |
6278 | }; | |
6279 | ||
62755e35 JA |
6280 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
6281 | { | |
6282 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 6283 | struct io_cancel_data *cd = data; |
62755e35 | 6284 | |
f458dd84 | 6285 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
6286 | } |
6287 | ||
f458dd84 PB |
6288 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
6289 | struct io_ring_ctx *ctx) | |
62755e35 | 6290 | { |
f458dd84 | 6291 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 6292 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
6293 | int ret = 0; |
6294 | ||
f458dd84 | 6295 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
6296 | return -ENOENT; |
6297 | ||
f458dd84 | 6298 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
6299 | switch (cancel_ret) { |
6300 | case IO_WQ_CANCEL_OK: | |
6301 | ret = 0; | |
6302 | break; | |
6303 | case IO_WQ_CANCEL_RUNNING: | |
6304 | ret = -EALREADY; | |
6305 | break; | |
6306 | case IO_WQ_CANCEL_NOTFOUND: | |
6307 | ret = -ENOENT; | |
6308 | break; | |
6309 | } | |
6310 | ||
e977d6d3 JA |
6311 | return ret; |
6312 | } | |
6313 | ||
8cb01fac | 6314 | static int io_try_cancel_userdata(struct io_kiocb *req, u64 sqe_addr) |
47f46768 | 6315 | { |
8cb01fac | 6316 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 JA |
6317 | int ret; |
6318 | ||
dadebc35 | 6319 | WARN_ON_ONCE(!io_wq_current_is_worker() && req->task != current); |
8cb01fac | 6320 | |
f458dd84 | 6321 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
df9727af | 6322 | if (ret != -ENOENT) |
8cb01fac | 6323 | return ret; |
505657bc PB |
6324 | |
6325 | spin_lock(&ctx->completion_lock); | |
79ebeaee | 6326 | spin_lock_irq(&ctx->timeout_lock); |
47f46768 | 6327 | ret = io_timeout_cancel(ctx, sqe_addr); |
79ebeaee | 6328 | spin_unlock_irq(&ctx->timeout_lock); |
47f46768 | 6329 | if (ret != -ENOENT) |
505657bc PB |
6330 | goto out; |
6331 | ret = io_poll_cancel(ctx, sqe_addr, false); | |
6332 | out: | |
6333 | spin_unlock(&ctx->completion_lock); | |
6334 | return ret; | |
47f46768 JA |
6335 | } |
6336 | ||
3529d8c2 JA |
6337 | static int io_async_cancel_prep(struct io_kiocb *req, |
6338 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6339 | { |
fbf23849 | 6340 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6341 | return -EINVAL; |
61710e43 DA |
6342 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6343 | return -EINVAL; | |
26578cda PB |
6344 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags || |
6345 | sqe->splice_fd_in) | |
e977d6d3 JA |
6346 | return -EINVAL; |
6347 | ||
fbf23849 JA |
6348 | req->cancel.addr = READ_ONCE(sqe->addr); |
6349 | return 0; | |
6350 | } | |
6351 | ||
61e98203 | 6352 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6353 | { |
6354 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
6355 | u64 sqe_addr = req->cancel.addr; |
6356 | struct io_tctx_node *node; | |
6357 | int ret; | |
6358 | ||
8cb01fac | 6359 | ret = io_try_cancel_userdata(req, sqe_addr); |
58f99373 PB |
6360 | if (ret != -ENOENT) |
6361 | goto done; | |
58f99373 PB |
6362 | |
6363 | /* slow path, try all io-wq's */ | |
6364 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
6365 | ret = -ENOENT; | |
6366 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
6367 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 6368 | |
58f99373 PB |
6369 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
6370 | if (ret != -ENOENT) | |
6371 | break; | |
6372 | } | |
6373 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
58f99373 | 6374 | done: |
58f99373 | 6375 | if (ret < 0) |
93d2bcd2 | 6376 | req_set_fail(req); |
505657bc | 6377 | io_req_complete_post(req, ret, 0); |
5262f567 JA |
6378 | return 0; |
6379 | } | |
6380 | ||
269bbe5f | 6381 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6382 | const struct io_uring_sqe *sqe) |
6383 | { | |
61710e43 DA |
6384 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6385 | return -EINVAL; | |
26578cda | 6386 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
05f3fb3c JA |
6387 | return -EINVAL; |
6388 | ||
269bbe5f BM |
6389 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6390 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6391 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6392 | return -EINVAL; |
269bbe5f | 6393 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6394 | return 0; |
6395 | } | |
6396 | ||
889fca73 | 6397 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6398 | { |
6399 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 6400 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 6401 | int ret; |
fbf23849 | 6402 | |
269bbe5f BM |
6403 | up.offset = req->rsrc_update.offset; |
6404 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
6405 | up.nr = 0; |
6406 | up.tags = 0; | |
615cee49 | 6407 | up.resv = 0; |
c9747fa5 | 6408 | up.resv2 = 0; |
05f3fb3c | 6409 | |
cdb31c29 | 6410 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); |
fdecb662 | 6411 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 6412 | &up, req->rsrc_update.nr_args); |
cdb31c29 | 6413 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); |
05f3fb3c JA |
6414 | |
6415 | if (ret < 0) | |
93d2bcd2 | 6416 | req_set_fail(req); |
889fca73 | 6417 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6418 | return 0; |
6419 | } | |
6420 | ||
bfe76559 | 6421 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6422 | { |
d625c6ee | 6423 | switch (req->opcode) { |
e781573e | 6424 | case IORING_OP_NOP: |
bfe76559 | 6425 | return 0; |
f67676d1 JA |
6426 | case IORING_OP_READV: |
6427 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6428 | case IORING_OP_READ: |
bfe76559 | 6429 | return io_read_prep(req, sqe); |
f67676d1 JA |
6430 | case IORING_OP_WRITEV: |
6431 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6432 | case IORING_OP_WRITE: |
bfe76559 | 6433 | return io_write_prep(req, sqe); |
0969e783 | 6434 | case IORING_OP_POLL_ADD: |
bfe76559 | 6435 | return io_poll_add_prep(req, sqe); |
0969e783 | 6436 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 6437 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 6438 | case IORING_OP_FSYNC: |
1155c76a | 6439 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6440 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6441 | return io_sfr_prep(req, sqe); |
03b1230c | 6442 | case IORING_OP_SENDMSG: |
fddaface | 6443 | case IORING_OP_SEND: |
bfe76559 | 6444 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6445 | case IORING_OP_RECVMSG: |
fddaface | 6446 | case IORING_OP_RECV: |
bfe76559 | 6447 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6448 | case IORING_OP_CONNECT: |
bfe76559 | 6449 | return io_connect_prep(req, sqe); |
2d28390a | 6450 | case IORING_OP_TIMEOUT: |
bfe76559 | 6451 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6452 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6453 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6454 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6455 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6456 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6457 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6458 | case IORING_OP_ACCEPT: |
bfe76559 | 6459 | return io_accept_prep(req, sqe); |
d63d1b5e | 6460 | case IORING_OP_FALLOCATE: |
bfe76559 | 6461 | return io_fallocate_prep(req, sqe); |
15b71abe | 6462 | case IORING_OP_OPENAT: |
bfe76559 | 6463 | return io_openat_prep(req, sqe); |
b5dba59e | 6464 | case IORING_OP_CLOSE: |
bfe76559 | 6465 | return io_close_prep(req, sqe); |
05f3fb3c | 6466 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6467 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6468 | case IORING_OP_STATX: |
bfe76559 | 6469 | return io_statx_prep(req, sqe); |
4840e418 | 6470 | case IORING_OP_FADVISE: |
bfe76559 | 6471 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6472 | case IORING_OP_MADVISE: |
bfe76559 | 6473 | return io_madvise_prep(req, sqe); |
cebdb986 | 6474 | case IORING_OP_OPENAT2: |
bfe76559 | 6475 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6476 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6477 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6478 | case IORING_OP_SPLICE: |
bfe76559 | 6479 | return io_splice_prep(req, sqe); |
ddf0322d | 6480 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6481 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6482 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6483 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6484 | case IORING_OP_TEE: |
bfe76559 | 6485 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6486 | case IORING_OP_SHUTDOWN: |
6487 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6488 | case IORING_OP_RENAMEAT: |
6489 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6490 | case IORING_OP_UNLINKAT: |
6491 | return io_unlinkat_prep(req, sqe); | |
e34a02dc DK |
6492 | case IORING_OP_MKDIRAT: |
6493 | return io_mkdirat_prep(req, sqe); | |
7a8721f8 DK |
6494 | case IORING_OP_SYMLINKAT: |
6495 | return io_symlinkat_prep(req, sqe); | |
cf30da90 DK |
6496 | case IORING_OP_LINKAT: |
6497 | return io_linkat_prep(req, sqe); | |
f67676d1 JA |
6498 | } |
6499 | ||
bfe76559 PB |
6500 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6501 | req->opcode); | |
bd54b6fe | 6502 | return -EINVAL; |
bfe76559 PB |
6503 | } |
6504 | ||
93642ef8 | 6505 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 6506 | { |
b7e298d2 PB |
6507 | if (!io_op_defs[req->opcode].needs_async_setup) |
6508 | return 0; | |
6509 | if (WARN_ON_ONCE(req->async_data)) | |
6510 | return -EFAULT; | |
6511 | if (io_alloc_async_data(req)) | |
6512 | return -EAGAIN; | |
6513 | ||
93642ef8 PB |
6514 | switch (req->opcode) { |
6515 | case IORING_OP_READV: | |
93642ef8 PB |
6516 | return io_rw_prep_async(req, READ); |
6517 | case IORING_OP_WRITEV: | |
93642ef8 PB |
6518 | return io_rw_prep_async(req, WRITE); |
6519 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
6520 | return io_sendmsg_prep_async(req); |
6521 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
6522 | return io_recvmsg_prep_async(req); |
6523 | case IORING_OP_CONNECT: | |
6524 | return io_connect_prep_async(req); | |
6525 | } | |
b7e298d2 PB |
6526 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
6527 | req->opcode); | |
6528 | return -EFAULT; | |
f67676d1 JA |
6529 | } |
6530 | ||
9cf7c104 PB |
6531 | static u32 io_get_sequence(struct io_kiocb *req) |
6532 | { | |
a3dbdf54 | 6533 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 6534 | |
a3dbdf54 PB |
6535 | /* need original cached_sq_head, but it was increased for each req */ |
6536 | io_for_each_link(req, req) | |
6537 | seq--; | |
6538 | return seq; | |
9cf7c104 PB |
6539 | } |
6540 | ||
76cc33d7 | 6541 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 6542 | { |
3c19966d | 6543 | struct io_kiocb *pos; |
a197f664 | 6544 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6545 | struct io_defer_entry *de; |
f67676d1 | 6546 | int ret; |
9cf7c104 | 6547 | u32 seq; |
de0617e4 | 6548 | |
b8ce1b9d PB |
6549 | if (req->flags & REQ_F_FAIL) { |
6550 | io_req_complete_fail_submit(req); | |
6551 | return true; | |
6552 | } | |
6553 | ||
3c19966d PB |
6554 | /* |
6555 | * If we need to drain a request in the middle of a link, drain the | |
6556 | * head request and the next request/link after the current link. | |
6557 | * Considering sequential execution of links, IOSQE_IO_DRAIN will be | |
6558 | * maintained for every request of our link. | |
6559 | */ | |
6560 | if (ctx->drain_next) { | |
6561 | req->flags |= REQ_F_IO_DRAIN; | |
6562 | ctx->drain_next = false; | |
6563 | } | |
6564 | /* not interested in head, start from the first linked */ | |
6565 | io_for_each_link(pos, req->link) { | |
6566 | if (pos->flags & REQ_F_IO_DRAIN) { | |
6567 | ctx->drain_next = true; | |
6568 | req->flags |= REQ_F_IO_DRAIN; | |
6569 | break; | |
6570 | } | |
6571 | } | |
6572 | ||
9d858b21 | 6573 | /* Still need defer if there is pending req in defer list. */ |
9632410b | 6574 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 6575 | if (likely(list_empty_careful(&ctx->defer_list) && |
10c66904 | 6576 | !(req->flags & REQ_F_IO_DRAIN))) { |
9632410b | 6577 | spin_unlock(&ctx->completion_lock); |
10c66904 | 6578 | ctx->drain_active = false; |
76cc33d7 | 6579 | return false; |
10c66904 | 6580 | } |
9632410b | 6581 | spin_unlock(&ctx->completion_lock); |
9cf7c104 PB |
6582 | |
6583 | seq = io_get_sequence(req); | |
6584 | /* Still a chance to pass the sequence check */ | |
6585 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6586 | return false; |
de0617e4 | 6587 | |
b7e298d2 | 6588 | ret = io_req_prep_async(req); |
be7053b7 | 6589 | if (ret) |
1b48773f | 6590 | goto fail; |
cbdcb435 | 6591 | io_prep_async_link(req); |
27dc8338 | 6592 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 6593 | if (!de) { |
1b48773f PB |
6594 | ret = -ENOMEM; |
6595 | fail: | |
6596 | io_req_complete_failed(req, ret); | |
76cc33d7 PB |
6597 | return true; |
6598 | } | |
2d28390a | 6599 | |
79ebeaee | 6600 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 6601 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
79ebeaee | 6602 | spin_unlock(&ctx->completion_lock); |
27dc8338 | 6603 | kfree(de); |
f237c30a | 6604 | io_queue_async_work(req, NULL); |
76cc33d7 | 6605 | return true; |
de0617e4 JA |
6606 | } |
6607 | ||
915967f6 | 6608 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6609 | de->req = req; |
9cf7c104 | 6610 | de->seq = seq; |
27dc8338 | 6611 | list_add_tail(&de->list, &ctx->defer_list); |
79ebeaee | 6612 | spin_unlock(&ctx->completion_lock); |
76cc33d7 | 6613 | return true; |
de0617e4 JA |
6614 | } |
6615 | ||
68fb8979 | 6616 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6617 | { |
0e1b6fe3 PB |
6618 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6619 | switch (req->opcode) { | |
6620 | case IORING_OP_READV: | |
6621 | case IORING_OP_READ_FIXED: | |
6622 | case IORING_OP_READ: | |
bcda7baa | 6623 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6624 | break; |
6625 | case IORING_OP_RECVMSG: | |
6626 | case IORING_OP_RECV: | |
bcda7baa | 6627 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6628 | break; |
6629 | } | |
99bc4c38 PB |
6630 | } |
6631 | ||
0e1b6fe3 PB |
6632 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6633 | switch (req->opcode) { | |
6634 | case IORING_OP_READV: | |
6635 | case IORING_OP_READ_FIXED: | |
6636 | case IORING_OP_READ: | |
6637 | case IORING_OP_WRITEV: | |
6638 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6639 | case IORING_OP_WRITE: { |
6640 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
6641 | |
6642 | kfree(io->free_iovec); | |
0e1b6fe3 | 6643 | break; |
e8c2bc1f | 6644 | } |
0e1b6fe3 | 6645 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6646 | case IORING_OP_SENDMSG: { |
6647 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6648 | |
6649 | kfree(io->free_iov); | |
0e1b6fe3 | 6650 | break; |
e8c2bc1f | 6651 | } |
f3cd4850 JA |
6652 | case IORING_OP_OPENAT: |
6653 | case IORING_OP_OPENAT2: | |
6654 | if (req->open.filename) | |
6655 | putname(req->open.filename); | |
6656 | break; | |
80a261fd JA |
6657 | case IORING_OP_RENAMEAT: |
6658 | putname(req->rename.oldpath); | |
6659 | putname(req->rename.newpath); | |
6660 | break; | |
14a1143b JA |
6661 | case IORING_OP_UNLINKAT: |
6662 | putname(req->unlink.filename); | |
6663 | break; | |
e34a02dc DK |
6664 | case IORING_OP_MKDIRAT: |
6665 | putname(req->mkdir.filename); | |
6666 | break; | |
7a8721f8 DK |
6667 | case IORING_OP_SYMLINKAT: |
6668 | putname(req->symlink.oldpath); | |
6669 | putname(req->symlink.newpath); | |
6670 | break; | |
cf30da90 DK |
6671 | case IORING_OP_LINKAT: |
6672 | putname(req->hardlink.oldpath); | |
6673 | putname(req->hardlink.newpath); | |
6674 | break; | |
0e1b6fe3 | 6675 | } |
99bc4c38 | 6676 | } |
75652a30 JA |
6677 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6678 | kfree(req->apoll->double_poll); | |
6679 | kfree(req->apoll); | |
6680 | req->apoll = NULL; | |
6681 | } | |
3a0a6902 PB |
6682 | if (req->flags & REQ_F_INFLIGHT) { |
6683 | struct io_uring_task *tctx = req->task->io_uring; | |
6684 | ||
6685 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 6686 | } |
c854357b | 6687 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 6688 | put_cred(req->creds); |
c854357b PB |
6689 | |
6690 | req->flags &= ~IO_REQ_CLEAN_FLAGS; | |
99bc4c38 PB |
6691 | } |
6692 | ||
889fca73 | 6693 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6694 | { |
a197f664 | 6695 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6696 | const struct cred *creds = NULL; |
d625c6ee | 6697 | int ret; |
2b188cc1 | 6698 | |
b8e64b53 | 6699 | if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) |
c10d1f98 | 6700 | creds = override_creds(req->creds); |
5730b27e | 6701 | |
d625c6ee | 6702 | switch (req->opcode) { |
2b188cc1 | 6703 | case IORING_OP_NOP: |
889fca73 | 6704 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6705 | break; |
6706 | case IORING_OP_READV: | |
edafccee | 6707 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6708 | case IORING_OP_READ: |
889fca73 | 6709 | ret = io_read(req, issue_flags); |
edafccee | 6710 | break; |
3529d8c2 | 6711 | case IORING_OP_WRITEV: |
edafccee | 6712 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6713 | case IORING_OP_WRITE: |
889fca73 | 6714 | ret = io_write(req, issue_flags); |
2b188cc1 | 6715 | break; |
c992fe29 | 6716 | case IORING_OP_FSYNC: |
45d189c6 | 6717 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6718 | break; |
221c5eb2 | 6719 | case IORING_OP_POLL_ADD: |
61e98203 | 6720 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6721 | break; |
6722 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6723 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6724 | break; |
5d17b4a4 | 6725 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6726 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6727 | break; |
0fa03c62 | 6728 | case IORING_OP_SENDMSG: |
889fca73 | 6729 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6730 | break; |
fddaface | 6731 | case IORING_OP_SEND: |
889fca73 | 6732 | ret = io_send(req, issue_flags); |
0fa03c62 | 6733 | break; |
aa1fa28f | 6734 | case IORING_OP_RECVMSG: |
889fca73 | 6735 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6736 | break; |
fddaface | 6737 | case IORING_OP_RECV: |
889fca73 | 6738 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6739 | break; |
5262f567 | 6740 | case IORING_OP_TIMEOUT: |
61e98203 | 6741 | ret = io_timeout(req, issue_flags); |
5262f567 | 6742 | break; |
11365043 | 6743 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6744 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6745 | break; |
17f2fe35 | 6746 | case IORING_OP_ACCEPT: |
889fca73 | 6747 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6748 | break; |
f8e85cf2 | 6749 | case IORING_OP_CONNECT: |
889fca73 | 6750 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6751 | break; |
62755e35 | 6752 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6753 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6754 | break; |
d63d1b5e | 6755 | case IORING_OP_FALLOCATE: |
45d189c6 | 6756 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6757 | break; |
15b71abe | 6758 | case IORING_OP_OPENAT: |
45d189c6 | 6759 | ret = io_openat(req, issue_flags); |
15b71abe | 6760 | break; |
b5dba59e | 6761 | case IORING_OP_CLOSE: |
889fca73 | 6762 | ret = io_close(req, issue_flags); |
b5dba59e | 6763 | break; |
05f3fb3c | 6764 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6765 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6766 | break; |
eddc7ef5 | 6767 | case IORING_OP_STATX: |
45d189c6 | 6768 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6769 | break; |
4840e418 | 6770 | case IORING_OP_FADVISE: |
45d189c6 | 6771 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6772 | break; |
c1ca757b | 6773 | case IORING_OP_MADVISE: |
45d189c6 | 6774 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6775 | break; |
cebdb986 | 6776 | case IORING_OP_OPENAT2: |
45d189c6 | 6777 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6778 | break; |
3e4827b0 | 6779 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6780 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6781 | break; |
7d67af2c | 6782 | case IORING_OP_SPLICE: |
45d189c6 | 6783 | ret = io_splice(req, issue_flags); |
7d67af2c | 6784 | break; |
ddf0322d | 6785 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6786 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6787 | break; |
067524e9 | 6788 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6789 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6790 | break; |
f2a8d5c7 | 6791 | case IORING_OP_TEE: |
45d189c6 | 6792 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6793 | break; |
36f4fa68 | 6794 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6795 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6796 | break; |
80a261fd | 6797 | case IORING_OP_RENAMEAT: |
45d189c6 | 6798 | ret = io_renameat(req, issue_flags); |
80a261fd | 6799 | break; |
14a1143b | 6800 | case IORING_OP_UNLINKAT: |
45d189c6 | 6801 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6802 | break; |
e34a02dc DK |
6803 | case IORING_OP_MKDIRAT: |
6804 | ret = io_mkdirat(req, issue_flags); | |
6805 | break; | |
7a8721f8 DK |
6806 | case IORING_OP_SYMLINKAT: |
6807 | ret = io_symlinkat(req, issue_flags); | |
6808 | break; | |
cf30da90 DK |
6809 | case IORING_OP_LINKAT: |
6810 | ret = io_linkat(req, issue_flags); | |
6811 | break; | |
2b188cc1 JA |
6812 | default: |
6813 | ret = -EINVAL; | |
6814 | break; | |
6815 | } | |
6816 | ||
5730b27e JA |
6817 | if (creds) |
6818 | revert_creds(creds); | |
def596e9 JA |
6819 | if (ret) |
6820 | return ret; | |
b532576e | 6821 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6822 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6823 | io_iopoll_req_issued(req); | |
def596e9 JA |
6824 | |
6825 | return 0; | |
2b188cc1 JA |
6826 | } |
6827 | ||
ebc11b6c PB |
6828 | static struct io_wq_work *io_wq_free_work(struct io_wq_work *work) |
6829 | { | |
6830 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6831 | ||
6832 | req = io_put_req_find_next(req); | |
6833 | return req ? &req->work : NULL; | |
6834 | } | |
6835 | ||
5280f7e5 | 6836 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6837 | { |
6838 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6839 | struct io_kiocb *timeout; |
561fb04a | 6840 | int ret = 0; |
2b188cc1 | 6841 | |
48dcd38d PB |
6842 | /* one will be dropped by ->io_free_work() after returning to io-wq */ |
6843 | if (!(req->flags & REQ_F_REFCOUNT)) | |
6844 | __io_req_set_refcount(req, 2); | |
6845 | else | |
6846 | req_ref_get(req); | |
5d5901a3 | 6847 | |
6df1db6b PB |
6848 | timeout = io_prep_linked_timeout(req); |
6849 | if (timeout) | |
6850 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6851 | |
dadebc35 | 6852 | /* either cancelled or io-wq is dying, so don't touch tctx->iowq */ |
4014d943 | 6853 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6854 | ret = -ECANCELED; |
31b51510 | 6855 | |
561fb04a | 6856 | if (!ret) { |
561fb04a | 6857 | do { |
889fca73 | 6858 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6859 | /* |
6860 | * We can get EAGAIN for polled IO even though we're | |
6861 | * forcing a sync submission from here, since we can't | |
6862 | * wait for request slots on the block side. | |
6863 | */ | |
6864 | if (ret != -EAGAIN) | |
6865 | break; | |
6866 | cond_resched(); | |
6867 | } while (1); | |
6868 | } | |
31b51510 | 6869 | |
a3df7698 | 6870 | /* avoid locking problems by failing it from a clean context */ |
5d5901a3 | 6871 | if (ret) |
a3df7698 | 6872 | io_req_task_queue_fail(req, ret); |
2b188cc1 JA |
6873 | } |
6874 | ||
aeca241b | 6875 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
042b0d85 | 6876 | unsigned i) |
65e19f54 | 6877 | { |
042b0d85 | 6878 | return &table->files[i]; |
dafecf19 PB |
6879 | } |
6880 | ||
65e19f54 JA |
6881 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6882 | int index) | |
6883 | { | |
aeca241b | 6884 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6885 | |
a04b0ac0 | 6886 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6887 | } |
6888 | ||
a04b0ac0 | 6889 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6890 | { |
6891 | unsigned long file_ptr = (unsigned long) file; | |
6892 | ||
b191e2df | 6893 | if (__io_file_supports_nowait(file, READ)) |
9a321c98 | 6894 | file_ptr |= FFS_ASYNC_READ; |
b191e2df | 6895 | if (__io_file_supports_nowait(file, WRITE)) |
9a321c98 PB |
6896 | file_ptr |= FFS_ASYNC_WRITE; |
6897 | if (S_ISREG(file_inode(file)->i_mode)) | |
6898 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6899 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6900 | } |
6901 | ||
ac177053 PB |
6902 | static inline struct file *io_file_get_fixed(struct io_ring_ctx *ctx, |
6903 | struct io_kiocb *req, int fd) | |
09bb8394 | 6904 | { |
8da11c19 | 6905 | struct file *file; |
ac177053 | 6906 | unsigned long file_ptr; |
09bb8394 | 6907 | |
ac177053 PB |
6908 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
6909 | return NULL; | |
6910 | fd = array_index_nospec(fd, ctx->nr_user_files); | |
6911 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; | |
6912 | file = (struct file *) (file_ptr & FFS_MASK); | |
6913 | file_ptr &= ~FFS_MASK; | |
6914 | /* mask in overlapping REQ_F and FFS bits */ | |
b191e2df | 6915 | req->flags |= (file_ptr << REQ_F_NOWAIT_READ_BIT); |
ac177053 PB |
6916 | io_req_set_rsrc_node(req); |
6917 | return file; | |
6918 | } | |
d44f554e | 6919 | |
ac177053 | 6920 | static struct file *io_file_get_normal(struct io_ring_ctx *ctx, |
ac177053 PB |
6921 | struct io_kiocb *req, int fd) |
6922 | { | |
62906e89 | 6923 | struct file *file = fget(fd); |
ac177053 PB |
6924 | |
6925 | trace_io_uring_file_get(ctx, fd); | |
09bb8394 | 6926 | |
ac177053 PB |
6927 | /* we don't allow fixed io_uring files */ |
6928 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6929 | io_req_track_inflight(req); | |
8371adf5 | 6930 | return file; |
09bb8394 JA |
6931 | } |
6932 | ||
ac177053 | 6933 | static inline struct file *io_file_get(struct io_ring_ctx *ctx, |
ac177053 PB |
6934 | struct io_kiocb *req, int fd, bool fixed) |
6935 | { | |
6936 | if (fixed) | |
6937 | return io_file_get_fixed(ctx, req, fd); | |
6938 | else | |
62906e89 | 6939 | return io_file_get_normal(ctx, req, fd); |
ac177053 PB |
6940 | } |
6941 | ||
f237c30a | 6942 | static void io_req_task_link_timeout(struct io_kiocb *req, bool *locked) |
89b263f6 JA |
6943 | { |
6944 | struct io_kiocb *prev = req->timeout.prev; | |
fc590dd1 | 6945 | int ret = -ENOENT; |
89b263f6 JA |
6946 | |
6947 | if (prev) { | |
fc590dd1 PB |
6948 | if (!(req->task->flags & PF_EXITING)) |
6949 | ret = io_try_cancel_userdata(req, prev->user_data); | |
505657bc | 6950 | io_req_complete_post(req, ret ?: -ETIME, 0); |
89b263f6 | 6951 | io_put_req(prev); |
89b263f6 JA |
6952 | } else { |
6953 | io_req_complete_post(req, -ETIME, 0); | |
6954 | } | |
6955 | } | |
6956 | ||
2665abfd | 6957 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6958 | { |
ad8a48ac JA |
6959 | struct io_timeout_data *data = container_of(timer, |
6960 | struct io_timeout_data, timer); | |
90cd7e42 | 6961 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6962 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6963 | unsigned long flags; |
2665abfd | 6964 | |
89b263f6 | 6965 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
90cd7e42 PB |
6966 | prev = req->timeout.head; |
6967 | req->timeout.head = NULL; | |
2665abfd JA |
6968 | |
6969 | /* | |
6970 | * We don't expect the list to be empty, that will only happen if we | |
6971 | * race with the completion of the linked work. | |
6972 | */ | |
447c19f3 | 6973 | if (prev) { |
f2f87370 | 6974 | io_remove_next_linked(prev); |
447c19f3 PB |
6975 | if (!req_ref_inc_not_zero(prev)) |
6976 | prev = NULL; | |
6977 | } | |
ef9dd637 | 6978 | list_del(&req->timeout.list); |
89b263f6 JA |
6979 | req->timeout.prev = prev; |
6980 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); | |
2665abfd | 6981 | |
89b263f6 JA |
6982 | req->io_task_work.func = io_req_task_link_timeout; |
6983 | io_req_task_work_add(req); | |
2665abfd JA |
6984 | return HRTIMER_NORESTART; |
6985 | } | |
6986 | ||
de968c18 | 6987 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6988 | { |
de968c18 PB |
6989 | struct io_ring_ctx *ctx = req->ctx; |
6990 | ||
89b263f6 | 6991 | spin_lock_irq(&ctx->timeout_lock); |
76a46e06 | 6992 | /* |
f2f87370 PB |
6993 | * If the back reference is NULL, then our linked request finished |
6994 | * before we got a chance to setup the timer | |
76a46e06 | 6995 | */ |
90cd7e42 | 6996 | if (req->timeout.head) { |
e8c2bc1f | 6997 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6998 | |
ad8a48ac JA |
6999 | data->timer.function = io_link_timeout_fn; |
7000 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
7001 | data->mode); | |
ef9dd637 | 7002 | list_add_tail(&req->timeout.list, &ctx->ltimeout_list); |
2665abfd | 7003 | } |
89b263f6 | 7004 | spin_unlock_irq(&ctx->timeout_lock); |
2665abfd | 7005 | /* drop submission reference */ |
76a46e06 JA |
7006 | io_put_req(req); |
7007 | } | |
2665abfd | 7008 | |
c5eef2b9 | 7009 | static void __io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 7010 | __must_hold(&req->ctx->uring_lock) |
2b188cc1 | 7011 | { |
906c6caa | 7012 | struct io_kiocb *linked_timeout; |
e0c5c576 | 7013 | int ret; |
2b188cc1 | 7014 | |
59b735ae | 7015 | issue_sqe: |
c5eef2b9 | 7016 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 7017 | |
491381ce JA |
7018 | /* |
7019 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
7020 | * doesn't support non-blocking read/write attempts | |
7021 | */ | |
1840038e | 7022 | if (likely(!ret)) { |
e342c807 | 7023 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 | 7024 | struct io_ring_ctx *ctx = req->ctx; |
cd0ca2e0 | 7025 | struct io_submit_state *state = &ctx->submit_state; |
e65ef56d | 7026 | |
cd0ca2e0 PB |
7027 | state->compl_reqs[state->compl_nr++] = req; |
7028 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2a2758f2 | 7029 | io_submit_flush_completions(ctx); |
906c6caa | 7030 | return; |
0d63c148 | 7031 | } |
906c6caa PB |
7032 | |
7033 | linked_timeout = io_prep_linked_timeout(req); | |
7034 | if (linked_timeout) | |
7035 | io_queue_linked_timeout(linked_timeout); | |
1840038e | 7036 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
906c6caa PB |
7037 | linked_timeout = io_prep_linked_timeout(req); |
7038 | ||
59b735ae OL |
7039 | switch (io_arm_poll_handler(req)) { |
7040 | case IO_APOLL_READY: | |
906c6caa | 7041 | if (linked_timeout) |
4ea672ab | 7042 | io_queue_linked_timeout(linked_timeout); |
59b735ae OL |
7043 | goto issue_sqe; |
7044 | case IO_APOLL_ABORTED: | |
1840038e PB |
7045 | /* |
7046 | * Queued up for async execution, worker will release | |
7047 | * submit reference when the iocb is actually submitted. | |
7048 | */ | |
f237c30a | 7049 | io_queue_async_work(req, NULL); |
59b735ae | 7050 | break; |
1840038e | 7051 | } |
906c6caa PB |
7052 | |
7053 | if (linked_timeout) | |
7054 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 | 7055 | } else { |
f41db273 | 7056 | io_req_complete_failed(req, ret); |
9e645e11 | 7057 | } |
2b188cc1 JA |
7058 | } |
7059 | ||
441b8a78 | 7060 | static inline void io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 7061 | __must_hold(&req->ctx->uring_lock) |
4fe2c963 | 7062 | { |
10c66904 | 7063 | if (unlikely(req->ctx->drain_active) && io_drain_req(req)) |
76cc33d7 | 7064 | return; |
4fe2c963 | 7065 | |
a8295b98 | 7066 | if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)))) { |
c5eef2b9 | 7067 | __io_queue_sqe(req); |
a8295b98 | 7068 | } else if (req->flags & REQ_F_FAIL) { |
c6d3d9cb | 7069 | io_req_complete_fail_submit(req); |
76cc33d7 PB |
7070 | } else { |
7071 | int ret = io_req_prep_async(req); | |
7072 | ||
7073 | if (unlikely(ret)) | |
7074 | io_req_complete_failed(req, ret); | |
7075 | else | |
f237c30a | 7076 | io_queue_async_work(req, NULL); |
ce35a47a | 7077 | } |
4fe2c963 JL |
7078 | } |
7079 | ||
b16fed66 PB |
7080 | /* |
7081 | * Check SQE restrictions (opcode and flags). | |
7082 | * | |
7083 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
7084 | */ | |
7085 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
7086 | struct io_kiocb *req, | |
7087 | unsigned int sqe_flags) | |
4fe2c963 | 7088 | { |
4cfb25bf | 7089 | if (likely(!ctx->restricted)) |
b16fed66 PB |
7090 | return true; |
7091 | ||
7092 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
7093 | return false; | |
7094 | ||
7095 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
7096 | ctx->restrictions.sqe_flags_required) | |
7097 | return false; | |
7098 | ||
7099 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
7100 | ctx->restrictions.sqe_flags_required)) | |
7101 | return false; | |
7102 | ||
7103 | return true; | |
4fe2c963 JL |
7104 | } |
7105 | ||
b16fed66 PB |
7106 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
7107 | const struct io_uring_sqe *sqe) | |
282cdc86 | 7108 | __must_hold(&ctx->uring_lock) |
b16fed66 PB |
7109 | { |
7110 | struct io_submit_state *state; | |
7111 | unsigned int sqe_flags; | |
003e8dcc | 7112 | int personality, ret = 0; |
b16fed66 | 7113 | |
864ea921 | 7114 | /* req is partially pre-initialised, see io_preinit_req() */ |
b16fed66 PB |
7115 | req->opcode = READ_ONCE(sqe->opcode); |
7116 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
7117 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
7118 | req->user_data = READ_ONCE(sqe->user_data); | |
b16fed66 | 7119 | req->file = NULL; |
b16fed66 | 7120 | req->fixed_rsrc_refs = NULL; |
b16fed66 | 7121 | req->task = current; |
b16fed66 PB |
7122 | |
7123 | /* enforce forwards compatibility on users */ | |
dddca226 | 7124 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 7125 | return -EINVAL; |
b16fed66 PB |
7126 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
7127 | return -EINVAL; | |
4cfb25bf | 7128 | if (!io_check_restriction(ctx, req, sqe_flags)) |
b16fed66 PB |
7129 | return -EACCES; |
7130 | ||
7131 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
7132 | !io_op_defs[req->opcode].buffer_select) | |
7133 | return -EOPNOTSUPP; | |
3c19966d PB |
7134 | if (unlikely(sqe_flags & IOSQE_IO_DRAIN)) |
7135 | ctx->drain_active = true; | |
863e0560 | 7136 | |
003e8dcc JA |
7137 | personality = READ_ONCE(sqe->personality); |
7138 | if (personality) { | |
c10d1f98 PB |
7139 | req->creds = xa_load(&ctx->personalities, personality); |
7140 | if (!req->creds) | |
003e8dcc | 7141 | return -EINVAL; |
c10d1f98 | 7142 | get_cred(req->creds); |
b8e64b53 | 7143 | req->flags |= REQ_F_CREDS; |
003e8dcc | 7144 | } |
b16fed66 PB |
7145 | state = &ctx->submit_state; |
7146 | ||
7147 | /* | |
7148 | * Plug now if we have more than 1 IO left after this, and the target | |
7149 | * is potentially a read/write to block based storage. | |
7150 | */ | |
7151 | if (!state->plug_started && state->ios_left > 1 && | |
7152 | io_op_defs[req->opcode].plug) { | |
7153 | blk_start_plug(&state->plug); | |
7154 | state->plug_started = true; | |
7155 | } | |
7156 | ||
7157 | if (io_op_defs[req->opcode].needs_file) { | |
62906e89 | 7158 | req->file = io_file_get(ctx, req, READ_ONCE(sqe->fd), |
ac177053 | 7159 | (sqe_flags & IOSQE_FIXED_FILE)); |
b16fed66 PB |
7160 | if (unlikely(!req->file)) |
7161 | ret = -EBADF; | |
7162 | } | |
7163 | ||
7164 | state->ios_left--; | |
7165 | return ret; | |
7166 | } | |
7167 | ||
a6b8cadc | 7168 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 7169 | const struct io_uring_sqe *sqe) |
282cdc86 | 7170 | __must_hold(&ctx->uring_lock) |
9e645e11 | 7171 | { |
a1ab7b35 | 7172 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 7173 | int ret; |
9e645e11 | 7174 | |
a6b8cadc PB |
7175 | ret = io_init_req(ctx, req, sqe); |
7176 | if (unlikely(ret)) { | |
7177 | fail_req: | |
a8295b98 | 7178 | /* fail even hard links since we don't submit */ |
de59bc10 | 7179 | if (link->head) { |
a8295b98 HX |
7180 | /* |
7181 | * we can judge a link req is failed or cancelled by if | |
7182 | * REQ_F_FAIL is set, but the head is an exception since | |
7183 | * it may be set REQ_F_FAIL because of other req's failure | |
7184 | * so let's leverage req->result to distinguish if a head | |
7185 | * is set REQ_F_FAIL because of its failure or other req's | |
7186 | * failure so that we can set the correct ret code for it. | |
7187 | * init result here to avoid affecting the normal path. | |
7188 | */ | |
7189 | if (!(link->head->flags & REQ_F_FAIL)) | |
7190 | req_fail_link_node(link->head, -ECANCELED); | |
7191 | } else if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { | |
7192 | /* | |
7193 | * the current req is a normal req, we should return | |
7194 | * error and thus break the submittion loop. | |
7195 | */ | |
7196 | io_req_complete_failed(req, ret); | |
7197 | return ret; | |
de59bc10 | 7198 | } |
a8295b98 HX |
7199 | req_fail_link_node(req, ret); |
7200 | } else { | |
7201 | ret = io_req_prep(req, sqe); | |
7202 | if (unlikely(ret)) | |
7203 | goto fail_req; | |
a6b8cadc | 7204 | } |
441b8a78 | 7205 | |
be7053b7 | 7206 | /* don't need @sqe from now on */ |
236daeae OL |
7207 | trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, |
7208 | req->flags, true, | |
7209 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 7210 | |
9e645e11 JA |
7211 | /* |
7212 | * If we already have a head request, queue this one for async | |
7213 | * submittal once the head completes. If we don't have a head but | |
7214 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
7215 | * submitted sync once the chain is complete. If none of those | |
7216 | * conditions are true (normal request), then just queue it. | |
7217 | */ | |
863e0560 PB |
7218 | if (link->head) { |
7219 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 7220 | |
a8295b98 HX |
7221 | if (!(req->flags & REQ_F_FAIL)) { |
7222 | ret = io_req_prep_async(req); | |
7223 | if (unlikely(ret)) { | |
7224 | req_fail_link_node(req, ret); | |
7225 | if (!(head->flags & REQ_F_FAIL)) | |
7226 | req_fail_link_node(head, -ECANCELED); | |
7227 | } | |
7228 | } | |
9d76377f | 7229 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 7230 | link->last->link = req; |
863e0560 | 7231 | link->last = req; |
32fe525b PB |
7232 | |
7233 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 7234 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 7235 | link->head = NULL; |
5e159204 | 7236 | io_queue_sqe(head); |
32fe525b | 7237 | } |
9e645e11 | 7238 | } else { |
ef4ff581 | 7239 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
7240 | link->head = req; |
7241 | link->last = req; | |
711be031 | 7242 | } else { |
be7053b7 | 7243 | io_queue_sqe(req); |
711be031 | 7244 | } |
9e645e11 | 7245 | } |
2e6e1fde | 7246 | |
1d4240cc | 7247 | return 0; |
9e645e11 JA |
7248 | } |
7249 | ||
9a56a232 JA |
7250 | /* |
7251 | * Batched submission is done, ensure local IO is flushed out. | |
7252 | */ | |
ba88ff11 PB |
7253 | static void io_submit_state_end(struct io_submit_state *state, |
7254 | struct io_ring_ctx *ctx) | |
9a56a232 | 7255 | { |
a1ab7b35 | 7256 | if (state->link.head) |
de59bc10 | 7257 | io_queue_sqe(state->link.head); |
cd0ca2e0 | 7258 | if (state->compl_nr) |
2a2758f2 | 7259 | io_submit_flush_completions(ctx); |
27926b68 JA |
7260 | if (state->plug_started) |
7261 | blk_finish_plug(&state->plug); | |
9a56a232 JA |
7262 | } |
7263 | ||
7264 | /* | |
7265 | * Start submission side cache. | |
7266 | */ | |
7267 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 7268 | unsigned int max_ios) |
9a56a232 | 7269 | { |
27926b68 | 7270 | state->plug_started = false; |
9a56a232 | 7271 | state->ios_left = max_ios; |
a1ab7b35 PB |
7272 | /* set only head, no need to init link_last in advance */ |
7273 | state->link.head = NULL; | |
9a56a232 JA |
7274 | } |
7275 | ||
2b188cc1 JA |
7276 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
7277 | { | |
75b28aff | 7278 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 7279 | |
caf582c6 PB |
7280 | /* |
7281 | * Ensure any loads from the SQEs are done at this point, | |
7282 | * since once we write the new head, the application could | |
7283 | * write new data to them. | |
7284 | */ | |
7285 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
7286 | } |
7287 | ||
2b188cc1 | 7288 | /* |
dd9ae8a0 | 7289 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
7290 | * that is mapped by userspace. This means that care needs to be taken to |
7291 | * ensure that reads are stable, as we cannot rely on userspace always | |
7292 | * being a good citizen. If members of the sqe are validated and then later | |
7293 | * used, it's important that those reads are done through READ_ONCE() to | |
7294 | * prevent a re-load down the line. | |
7295 | */ | |
709b302f | 7296 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 7297 | { |
ea5ab3b5 | 7298 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 7299 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
7300 | |
7301 | /* | |
7302 | * The cached sq head (or cq tail) serves two purposes: | |
7303 | * | |
7304 | * 1) allows us to batch the cost of updating the user visible | |
7305 | * head updates. | |
7306 | * 2) allows the kernel side to track the head on its own, even | |
7307 | * though the application is the one updating it. | |
7308 | */ | |
17d3aeb3 | 7309 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
7310 | if (likely(head < ctx->sq_entries)) |
7311 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
7312 | |
7313 | /* drop invalid entries */ | |
15641e42 PB |
7314 | ctx->cq_extra--; |
7315 | WRITE_ONCE(ctx->rings->sq_dropped, | |
7316 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
7317 | return NULL; |
7318 | } | |
7319 | ||
0f212204 | 7320 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
282cdc86 | 7321 | __must_hold(&ctx->uring_lock) |
6c271ce2 | 7322 | { |
46c4e16a | 7323 | int submitted = 0; |
6c271ce2 | 7324 | |
ee7d46d9 PB |
7325 | /* make sure SQ entry isn't read before tail */ |
7326 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
7327 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
7328 | return -EAGAIN; | |
9a10867a | 7329 | io_get_task_refs(nr); |
6c271ce2 | 7330 | |
ba88ff11 | 7331 | io_submit_state_start(&ctx->submit_state, nr); |
46c4e16a | 7332 | while (submitted < nr) { |
3529d8c2 | 7333 | const struct io_uring_sqe *sqe; |
196be95c | 7334 | struct io_kiocb *req; |
fb5ccc98 | 7335 | |
258b29a9 | 7336 | req = io_alloc_req(ctx); |
196be95c PB |
7337 | if (unlikely(!req)) { |
7338 | if (!submitted) | |
7339 | submitted = -EAGAIN; | |
fb5ccc98 | 7340 | break; |
196be95c | 7341 | } |
4fccfcbb PB |
7342 | sqe = io_get_sqe(ctx); |
7343 | if (unlikely(!sqe)) { | |
0c6e1d7f | 7344 | list_add(&req->inflight_entry, &ctx->submit_state.free_list); |
4fccfcbb PB |
7345 | break; |
7346 | } | |
d3656344 JA |
7347 | /* will complete beyond this point, count as submitted */ |
7348 | submitted++; | |
a1ab7b35 | 7349 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 7350 | break; |
6c271ce2 JA |
7351 | } |
7352 | ||
9466f437 PB |
7353 | if (unlikely(submitted != nr)) { |
7354 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 7355 | int unused = nr - ref_used; |
9466f437 | 7356 | |
09899b19 | 7357 | current->io_uring->cached_refs += unused; |
d8a6df10 | 7358 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 7359 | } |
6c271ce2 | 7360 | |
a1ab7b35 | 7361 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
7362 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
7363 | io_commit_sqring(ctx); | |
7364 | ||
6c271ce2 JA |
7365 | return submitted; |
7366 | } | |
7367 | ||
e4b6d902 PB |
7368 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
7369 | { | |
7370 | return READ_ONCE(sqd->state); | |
7371 | } | |
7372 | ||
23b3628e XW |
7373 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
7374 | { | |
7375 | /* Tell userspace we may need a wakeup call */ | |
79ebeaee | 7376 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7377 | WRITE_ONCE(ctx->rings->sq_flags, |
7378 | ctx->rings->sq_flags | IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7379 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7380 | } |
7381 | ||
7382 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
7383 | { | |
79ebeaee | 7384 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7385 | WRITE_ONCE(ctx->rings->sq_flags, |
7386 | ctx->rings->sq_flags & ~IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7387 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7388 | } |
7389 | ||
08369246 | 7390 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 7391 | { |
c8d1ba58 | 7392 | unsigned int to_submit; |
bdcd3eab | 7393 | int ret = 0; |
6c271ce2 | 7394 | |
c8d1ba58 | 7395 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 7396 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
7397 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
7398 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 7399 | |
906a3c6f | 7400 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 7401 | unsigned nr_events = 0; |
948e1947 PB |
7402 | const struct cred *creds = NULL; |
7403 | ||
7404 | if (ctx->sq_creds != current_cred()) | |
7405 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 7406 | |
c8d1ba58 | 7407 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 7408 | if (!list_empty(&ctx->iopoll_list)) |
a8576af9 | 7409 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 7410 | |
3b763ba1 PB |
7411 | /* |
7412 | * Don't submit if refs are dying, good for io_uring_register(), | |
7413 | * but also it is relied upon by io_ring_exit_work() | |
7414 | */ | |
0298ef96 PB |
7415 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
7416 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 7417 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 7418 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 7419 | |
acfb381d PB |
7420 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
7421 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
7422 | if (creds) |
7423 | revert_creds(creds); | |
acfb381d | 7424 | } |
6c271ce2 | 7425 | |
08369246 XW |
7426 | return ret; |
7427 | } | |
6c271ce2 | 7428 | |
08369246 XW |
7429 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7430 | { | |
7431 | struct io_ring_ctx *ctx; | |
7432 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7433 | |
c9dca27d PB |
7434 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7435 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 7436 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7437 | } |
6c271ce2 | 7438 | |
e4b6d902 PB |
7439 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
7440 | { | |
7441 | bool did_sig = false; | |
7442 | struct ksignal ksig; | |
7443 | ||
7444 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
7445 | signal_pending(current)) { | |
7446 | mutex_unlock(&sqd->lock); | |
7447 | if (signal_pending(current)) | |
7448 | did_sig = get_signal(&ksig); | |
7449 | cond_resched(); | |
7450 | mutex_lock(&sqd->lock); | |
7451 | } | |
e4b6d902 PB |
7452 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
7453 | } | |
7454 | ||
c8d1ba58 JA |
7455 | static int io_sq_thread(void *data) |
7456 | { | |
69fb2131 JA |
7457 | struct io_sq_data *sqd = data; |
7458 | struct io_ring_ctx *ctx; | |
a0d9205f | 7459 | unsigned long timeout = 0; |
37d1e2e3 | 7460 | char buf[TASK_COMM_LEN]; |
08369246 | 7461 | DEFINE_WAIT(wait); |
6c271ce2 | 7462 | |
696ee88a | 7463 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 7464 | set_task_comm(current, buf); |
37d1e2e3 JA |
7465 | |
7466 | if (sqd->sq_cpu != -1) | |
7467 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
7468 | else | |
7469 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
7470 | current->flags |= PF_NO_SETAFFINITY; | |
7471 | ||
09a6f4ef | 7472 | mutex_lock(&sqd->lock); |
e4b6d902 | 7473 | while (1) { |
1a924a80 | 7474 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 7475 | |
e4b6d902 PB |
7476 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
7477 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 7478 | break; |
08369246 XW |
7479 | timeout = jiffies + sqd->sq_thread_idle; |
7480 | } | |
e4b6d902 | 7481 | |
e95eee2d | 7482 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 7483 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 7484 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 7485 | |
08369246 XW |
7486 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
7487 | sqt_spin = true; | |
69fb2131 | 7488 | } |
dd432ea5 PB |
7489 | if (io_run_task_work()) |
7490 | sqt_spin = true; | |
6c271ce2 | 7491 | |
08369246 | 7492 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7493 | cond_resched(); |
08369246 XW |
7494 | if (sqt_spin) |
7495 | timeout = jiffies + sqd->sq_thread_idle; | |
7496 | continue; | |
7497 | } | |
7498 | ||
08369246 | 7499 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
dd432ea5 | 7500 | if (!io_sqd_events_pending(sqd) && !current->task_works) { |
1a924a80 PB |
7501 | bool needs_sched = true; |
7502 | ||
724cb4f9 | 7503 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
7504 | io_ring_set_wakeup_flag(ctx); |
7505 | ||
724cb4f9 HX |
7506 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
7507 | !list_empty_careful(&ctx->iopoll_list)) { | |
7508 | needs_sched = false; | |
7509 | break; | |
7510 | } | |
7511 | if (io_sqring_entries(ctx)) { | |
7512 | needs_sched = false; | |
7513 | break; | |
7514 | } | |
7515 | } | |
7516 | ||
7517 | if (needs_sched) { | |
7518 | mutex_unlock(&sqd->lock); | |
7519 | schedule(); | |
7520 | mutex_lock(&sqd->lock); | |
7521 | } | |
69fb2131 JA |
7522 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7523 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7524 | } |
08369246 XW |
7525 | |
7526 | finish_wait(&sqd->wait, &wait); | |
7527 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 7528 | } |
28cea78a | 7529 | |
78cc687b | 7530 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 7531 | sqd->thread = NULL; |
05962f95 | 7532 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 7533 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 7534 | io_run_task_work(); |
734551df PB |
7535 | mutex_unlock(&sqd->lock); |
7536 | ||
37d1e2e3 JA |
7537 | complete(&sqd->exited); |
7538 | do_exit(0); | |
6c271ce2 JA |
7539 | } |
7540 | ||
bda52162 JA |
7541 | struct io_wait_queue { |
7542 | struct wait_queue_entry wq; | |
7543 | struct io_ring_ctx *ctx; | |
5fd46178 | 7544 | unsigned cq_tail; |
bda52162 JA |
7545 | unsigned nr_timeouts; |
7546 | }; | |
7547 | ||
6c503150 | 7548 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7549 | { |
7550 | struct io_ring_ctx *ctx = iowq->ctx; | |
5fd46178 | 7551 | int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; |
bda52162 JA |
7552 | |
7553 | /* | |
d195a66e | 7554 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7555 | * started waiting. For timeouts, we always want to return to userspace, |
7556 | * regardless of event count. | |
7557 | */ | |
5fd46178 | 7558 | return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
bda52162 JA |
7559 | } |
7560 | ||
7561 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7562 | int wake_flags, void *key) | |
7563 | { | |
7564 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7565 | wq); | |
7566 | ||
6c503150 PB |
7567 | /* |
7568 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7569 | * the task, and the next invocation will do it. | |
7570 | */ | |
5ed7a37d | 7571 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
7572 | return autoremove_wake_function(curr, mode, wake_flags, key); |
7573 | return -1; | |
bda52162 JA |
7574 | } |
7575 | ||
af9c1a44 JA |
7576 | static int io_run_task_work_sig(void) |
7577 | { | |
7578 | if (io_run_task_work()) | |
7579 | return 1; | |
7580 | if (!signal_pending(current)) | |
7581 | return 0; | |
0b8cfa97 | 7582 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 7583 | return -ERESTARTSYS; |
af9c1a44 JA |
7584 | return -EINTR; |
7585 | } | |
7586 | ||
eeb60b9a PB |
7587 | /* when returns >0, the caller should retry */ |
7588 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7589 | struct io_wait_queue *iowq, | |
68668891 | 7590 | ktime_t timeout) |
eeb60b9a PB |
7591 | { |
7592 | int ret; | |
7593 | ||
7594 | /* make sure we run task_work before checking for signals */ | |
7595 | ret = io_run_task_work_sig(); | |
7596 | if (ret || io_should_wake(iowq)) | |
7597 | return ret; | |
7598 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 7599 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
7600 | return 1; |
7601 | ||
68668891 JA |
7602 | if (!schedule_hrtimeout(&timeout, HRTIMER_MODE_ABS)) |
7603 | return -ETIME; | |
7604 | return 1; | |
eeb60b9a PB |
7605 | } |
7606 | ||
2b188cc1 JA |
7607 | /* |
7608 | * Wait until events become available, if we don't already have some. The | |
7609 | * application must reap them itself, as they reside on the shared cq ring. | |
7610 | */ | |
7611 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7612 | const sigset_t __user *sig, size_t sigsz, |
7613 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7614 | { |
90291099 | 7615 | struct io_wait_queue iowq; |
75b28aff | 7616 | struct io_rings *rings = ctx->rings; |
68668891 | 7617 | ktime_t timeout = KTIME_MAX; |
c1d5a224 | 7618 | int ret; |
2b188cc1 | 7619 | |
b41e9852 | 7620 | do { |
90f67366 | 7621 | io_cqring_overflow_flush(ctx); |
6c503150 | 7622 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7623 | return 0; |
4c6e277c | 7624 | if (!io_run_task_work()) |
b41e9852 | 7625 | break; |
b41e9852 | 7626 | } while (1); |
2b188cc1 | 7627 | |
44df58d4 XW |
7628 | if (uts) { |
7629 | struct timespec64 ts; | |
7630 | ||
7631 | if (get_timespec64(&ts, uts)) | |
7632 | return -EFAULT; | |
68668891 | 7633 | timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns()); |
44df58d4 XW |
7634 | } |
7635 | ||
2b188cc1 | 7636 | if (sig) { |
9e75ad5d AB |
7637 | #ifdef CONFIG_COMPAT |
7638 | if (in_compat_syscall()) | |
7639 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7640 | sigsz); |
9e75ad5d AB |
7641 | else |
7642 | #endif | |
b772434b | 7643 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7644 | |
2b188cc1 JA |
7645 | if (ret) |
7646 | return ret; | |
7647 | } | |
7648 | ||
90291099 PB |
7649 | init_waitqueue_func_entry(&iowq.wq, io_wake_function); |
7650 | iowq.wq.private = current; | |
7651 | INIT_LIST_HEAD(&iowq.wq.entry); | |
7652 | iowq.ctx = ctx; | |
bda52162 | 7653 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
5fd46178 | 7654 | iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; |
90291099 | 7655 | |
c826bd7a | 7656 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7657 | do { |
ca0a2651 | 7658 | /* if we can't even flush overflow, don't wait for more */ |
90f67366 | 7659 | if (!io_cqring_overflow_flush(ctx)) { |
ca0a2651 JA |
7660 | ret = -EBUSY; |
7661 | break; | |
7662 | } | |
311997b3 | 7663 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7664 | TASK_INTERRUPTIBLE); |
68668891 | 7665 | ret = io_cqring_wait_schedule(ctx, &iowq, timeout); |
311997b3 | 7666 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7667 | cond_resched(); |
eeb60b9a | 7668 | } while (ret > 0); |
bda52162 | 7669 | |
b7db41c9 | 7670 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7671 | |
75b28aff | 7672 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7673 | } |
7674 | ||
9123c8ff | 7675 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7676 | { |
9123c8ff | 7677 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7678 | |
846a4ef2 | 7679 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7680 | kfree(table[i]); |
7681 | kfree(table); | |
7682 | } | |
7683 | ||
7684 | static void **io_alloc_page_table(size_t size) | |
7685 | { | |
7686 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7687 | size_t init_size = size; | |
7688 | void **table; | |
7689 | ||
0bea96f5 | 7690 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7691 | if (!table) |
7692 | return NULL; | |
7693 | ||
7694 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 7695 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff | 7696 | |
0bea96f5 | 7697 | table[i] = kzalloc(this_size, GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7698 | if (!table[i]) { |
7699 | io_free_page_table(table, init_size); | |
7700 | return NULL; | |
7701 | } | |
7702 | size -= this_size; | |
7703 | } | |
7704 | return table; | |
05f3fb3c JA |
7705 | } |
7706 | ||
28a9fe25 | 7707 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7708 | { |
28a9fe25 PB |
7709 | percpu_ref_exit(&ref_node->refs); |
7710 | kfree(ref_node); | |
1642b445 PB |
7711 | } |
7712 | ||
b9bd2bea PB |
7713 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
7714 | { | |
7715 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); | |
7716 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; | |
7717 | unsigned long flags; | |
7718 | bool first_add = false; | |
817ab1d1 | 7719 | unsigned long delay = HZ; |
b9bd2bea PB |
7720 | |
7721 | spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); | |
7722 | node->done = true; | |
7723 | ||
817ab1d1 DY |
7724 | /* if we are mid-quiesce then do not delay */ |
7725 | if (node->rsrc_data->quiesce) | |
7726 | delay = 0; | |
7727 | ||
b9bd2bea PB |
7728 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7729 | node = list_first_entry(&ctx->rsrc_ref_list, | |
7730 | struct io_rsrc_node, node); | |
7731 | /* recycle ref nodes in order */ | |
7732 | if (!node->done) | |
7733 | break; | |
7734 | list_del(&node->node); | |
7735 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
7736 | } | |
7737 | spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); | |
7738 | ||
7739 | if (first_add) | |
817ab1d1 | 7740 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
b9bd2bea PB |
7741 | } |
7742 | ||
7743 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) | |
7744 | { | |
7745 | struct io_rsrc_node *ref_node; | |
7746 | ||
7747 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); | |
7748 | if (!ref_node) | |
7749 | return NULL; | |
7750 | ||
7751 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, | |
7752 | 0, GFP_KERNEL)) { | |
7753 | kfree(ref_node); | |
7754 | return NULL; | |
7755 | } | |
7756 | INIT_LIST_HEAD(&ref_node->node); | |
7757 | INIT_LIST_HEAD(&ref_node->rsrc_list); | |
7758 | ref_node->done = false; | |
7759 | return ref_node; | |
7760 | } | |
7761 | ||
a7f0ed5a PB |
7762 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7763 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7764 | { |
a7f0ed5a PB |
7765 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7766 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7767 | |
a7f0ed5a PB |
7768 | if (data_to_kill) { |
7769 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7770 | |
a7f0ed5a | 7771 | rsrc_node->rsrc_data = data_to_kill; |
4956b9ea | 7772 | spin_lock_irq(&ctx->rsrc_ref_lock); |
a7f0ed5a | 7773 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); |
4956b9ea | 7774 | spin_unlock_irq(&ctx->rsrc_ref_lock); |
82fbcfa9 | 7775 | |
3e942498 | 7776 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7777 | percpu_ref_kill(&rsrc_node->refs); |
7778 | ctx->rsrc_node = NULL; | |
7779 | } | |
6b06314c | 7780 | |
a7f0ed5a PB |
7781 | if (!ctx->rsrc_node) { |
7782 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7783 | ctx->rsrc_backup_node = NULL; | |
7784 | } | |
8bad28d8 HX |
7785 | } |
7786 | ||
a7f0ed5a | 7787 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7788 | { |
7789 | if (ctx->rsrc_backup_node) | |
7790 | return 0; | |
b895c9a6 | 7791 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7792 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7793 | } |
7794 | ||
40ae0ff7 | 7795 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7796 | { |
7797 | int ret; | |
05589553 | 7798 | |
215c3902 | 7799 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7800 | if (data->quiesce) |
7801 | return -ENXIO; | |
05589553 | 7802 | |
8bad28d8 | 7803 | data->quiesce = true; |
1ffc5422 | 7804 | do { |
a7f0ed5a | 7805 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7806 | if (ret) |
f2303b1f | 7807 | break; |
a7f0ed5a | 7808 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7809 | |
3e942498 PB |
7810 | /* kill initial ref, already quiesced if zero */ |
7811 | if (atomic_dec_and_test(&data->refs)) | |
7812 | break; | |
c018db4a | 7813 | mutex_unlock(&ctx->uring_lock); |
8bad28d8 | 7814 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 | 7815 | ret = wait_for_completion_interruptible(&data->done); |
c018db4a JA |
7816 | if (!ret) { |
7817 | mutex_lock(&ctx->uring_lock); | |
2bcee8e2 DY |
7818 | if (atomic_read(&data->refs) > 0) { |
7819 | /* | |
7820 | * it has been revived by another thread while | |
7821 | * we were unlocked | |
7822 | */ | |
7823 | mutex_unlock(&ctx->uring_lock); | |
7824 | } else { | |
7825 | break; | |
7826 | } | |
c018db4a | 7827 | } |
8bad28d8 | 7828 | |
3e942498 PB |
7829 | atomic_inc(&data->refs); |
7830 | /* wait for all works potentially completing data->done */ | |
7831 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7832 | reinit_completion(&data->done); |
8dd03afe | 7833 | |
1ffc5422 | 7834 | ret = io_run_task_work_sig(); |
8bad28d8 | 7835 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7836 | } while (ret >= 0); |
8bad28d8 | 7837 | data->quiesce = false; |
05f3fb3c | 7838 | |
8bad28d8 | 7839 | return ret; |
d7954b2b BM |
7840 | } |
7841 | ||
2d091d62 PB |
7842 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7843 | { | |
7844 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7845 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7846 | ||
7847 | return &data->tags[table_idx][off]; | |
7848 | } | |
7849 | ||
44b31f2f | 7850 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7851 | { |
2d091d62 PB |
7852 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7853 | ||
7854 | if (data->tags) | |
7855 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7856 | kfree(data); |
7857 | } | |
7858 | ||
d878c816 PB |
7859 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7860 | u64 __user *utags, unsigned nr, | |
7861 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7862 | { |
b895c9a6 | 7863 | struct io_rsrc_data *data; |
2d091d62 | 7864 | int ret = -ENOMEM; |
d878c816 | 7865 | unsigned i; |
1ad555c6 BM |
7866 | |
7867 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7868 | if (!data) | |
d878c816 | 7869 | return -ENOMEM; |
2d091d62 | 7870 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7871 | if (!data->tags) { |
1ad555c6 | 7872 | kfree(data); |
d878c816 PB |
7873 | return -ENOMEM; |
7874 | } | |
2d091d62 PB |
7875 | |
7876 | data->nr = nr; | |
7877 | data->ctx = ctx; | |
7878 | data->do_put = do_put; | |
d878c816 | 7879 | if (utags) { |
2d091d62 | 7880 | ret = -EFAULT; |
d878c816 | 7881 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7882 | u64 *tag_slot = io_get_tag_slot(data, i); |
7883 | ||
7884 | if (copy_from_user(tag_slot, &utags[i], | |
7885 | sizeof(*tag_slot))) | |
2d091d62 | 7886 | goto fail; |
d878c816 | 7887 | } |
1ad555c6 | 7888 | } |
b60c8dce | 7889 | |
3e942498 | 7890 | atomic_set(&data->refs, 1); |
1ad555c6 | 7891 | init_completion(&data->done); |
d878c816 PB |
7892 | *pdata = data; |
7893 | return 0; | |
2d091d62 PB |
7894 | fail: |
7895 | io_rsrc_data_free(data); | |
7896 | return ret; | |
1ad555c6 BM |
7897 | } |
7898 | ||
9123c8ff PB |
7899 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7900 | { | |
0bea96f5 PB |
7901 | table->files = kvcalloc(nr_files, sizeof(table->files[0]), |
7902 | GFP_KERNEL_ACCOUNT); | |
9123c8ff PB |
7903 | return !!table->files; |
7904 | } | |
7905 | ||
042b0d85 | 7906 | static void io_free_file_tables(struct io_file_table *table) |
9123c8ff | 7907 | { |
042b0d85 | 7908 | kvfree(table->files); |
9123c8ff PB |
7909 | table->files = NULL; |
7910 | } | |
7911 | ||
fff4db76 | 7912 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7913 | { |
fff4db76 PB |
7914 | #if defined(CONFIG_UNIX) |
7915 | if (ctx->ring_sock) { | |
7916 | struct sock *sock = ctx->ring_sock->sk; | |
7917 | struct sk_buff *skb; | |
7918 | ||
7919 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7920 | kfree_skb(skb); | |
7921 | } | |
7922 | #else | |
7923 | int i; | |
7924 | ||
7925 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7926 | struct file *file; | |
7927 | ||
7928 | file = io_file_from_index(ctx, i); | |
7929 | if (file) | |
7930 | fput(file); | |
7931 | } | |
7932 | #endif | |
042b0d85 | 7933 | io_free_file_tables(&ctx->file_table); |
44b31f2f | 7934 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7935 | ctx->file_data = NULL; |
7936 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7937 | } |
7938 | ||
d7954b2b BM |
7939 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7940 | { | |
d7954b2b BM |
7941 | int ret; |
7942 | ||
08480400 | 7943 | if (!ctx->file_data) |
d7954b2b | 7944 | return -ENXIO; |
08480400 PB |
7945 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7946 | if (!ret) | |
7947 | __io_sqe_files_unregister(ctx); | |
7948 | return ret; | |
6b06314c JA |
7949 | } |
7950 | ||
37d1e2e3 | 7951 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7952 | __releases(&sqd->lock) |
37d1e2e3 | 7953 | { |
521d6a73 PB |
7954 | WARN_ON_ONCE(sqd->thread == current); |
7955 | ||
9e138a48 PB |
7956 | /* |
7957 | * Do the dance but not conditional clear_bit() because it'd race with | |
7958 | * other threads incrementing park_pending and setting the bit. | |
7959 | */ | |
37d1e2e3 | 7960 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7961 | if (atomic_dec_return(&sqd->park_pending)) |
7962 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7963 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7964 | } |
7965 | ||
86e0d676 | 7966 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7967 | __acquires(&sqd->lock) |
37d1e2e3 | 7968 | { |
521d6a73 PB |
7969 | WARN_ON_ONCE(sqd->thread == current); |
7970 | ||
9e138a48 | 7971 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7972 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7973 | mutex_lock(&sqd->lock); |
05962f95 | 7974 | if (sqd->thread) |
86e0d676 | 7975 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7976 | } |
7977 | ||
7978 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7979 | { | |
521d6a73 | 7980 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7981 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7982 | |
05962f95 | 7983 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7984 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7985 | if (sqd->thread) |
7986 | wake_up_process(sqd->thread); | |
09a6f4ef | 7987 | mutex_unlock(&sqd->lock); |
05962f95 | 7988 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7989 | } |
7990 | ||
534ca6d6 | 7991 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7992 | { |
534ca6d6 | 7993 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7994 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7995 | ||
37d1e2e3 JA |
7996 | io_sq_thread_stop(sqd); |
7997 | kfree(sqd); | |
7998 | } | |
7999 | } | |
8000 | ||
8001 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
8002 | { | |
8003 | struct io_sq_data *sqd = ctx->sq_data; | |
8004 | ||
8005 | if (sqd) { | |
05962f95 | 8006 | io_sq_thread_park(sqd); |
521d6a73 | 8007 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 8008 | io_sqd_update_thread_idle(sqd); |
05962f95 | 8009 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
8010 | |
8011 | io_put_sq_data(sqd); | |
8012 | ctx->sq_data = NULL; | |
534ca6d6 JA |
8013 | } |
8014 | } | |
8015 | ||
aa06165d JA |
8016 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
8017 | { | |
8018 | struct io_ring_ctx *ctx_attach; | |
8019 | struct io_sq_data *sqd; | |
8020 | struct fd f; | |
8021 | ||
8022 | f = fdget(p->wq_fd); | |
8023 | if (!f.file) | |
8024 | return ERR_PTR(-ENXIO); | |
8025 | if (f.file->f_op != &io_uring_fops) { | |
8026 | fdput(f); | |
8027 | return ERR_PTR(-EINVAL); | |
8028 | } | |
8029 | ||
8030 | ctx_attach = f.file->private_data; | |
8031 | sqd = ctx_attach->sq_data; | |
8032 | if (!sqd) { | |
8033 | fdput(f); | |
8034 | return ERR_PTR(-EINVAL); | |
8035 | } | |
5c2469e0 JA |
8036 | if (sqd->task_tgid != current->tgid) { |
8037 | fdput(f); | |
8038 | return ERR_PTR(-EPERM); | |
8039 | } | |
aa06165d JA |
8040 | |
8041 | refcount_inc(&sqd->refs); | |
8042 | fdput(f); | |
8043 | return sqd; | |
8044 | } | |
8045 | ||
26984fbf PB |
8046 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
8047 | bool *attached) | |
534ca6d6 JA |
8048 | { |
8049 | struct io_sq_data *sqd; | |
8050 | ||
26984fbf | 8051 | *attached = false; |
5c2469e0 JA |
8052 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
8053 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
8054 | if (!IS_ERR(sqd)) { |
8055 | *attached = true; | |
5c2469e0 | 8056 | return sqd; |
26984fbf | 8057 | } |
5c2469e0 JA |
8058 | /* fall through for EPERM case, setup new sqd/task */ |
8059 | if (PTR_ERR(sqd) != -EPERM) | |
8060 | return sqd; | |
8061 | } | |
aa06165d | 8062 | |
534ca6d6 JA |
8063 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
8064 | if (!sqd) | |
8065 | return ERR_PTR(-ENOMEM); | |
8066 | ||
9e138a48 | 8067 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 8068 | refcount_set(&sqd->refs, 1); |
69fb2131 | 8069 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 8070 | mutex_init(&sqd->lock); |
534ca6d6 | 8071 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 8072 | init_completion(&sqd->exited); |
534ca6d6 JA |
8073 | return sqd; |
8074 | } | |
8075 | ||
6b06314c | 8076 | #if defined(CONFIG_UNIX) |
6b06314c JA |
8077 | /* |
8078 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
8079 | * the io_uring can be safely unregistered on process exit, even if we have | |
8080 | * loops in the file referencing. | |
8081 | */ | |
8082 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
8083 | { | |
8084 | struct sock *sk = ctx->ring_sock->sk; | |
8085 | struct scm_fp_list *fpl; | |
8086 | struct sk_buff *skb; | |
08a45173 | 8087 | int i, nr_files; |
6b06314c | 8088 | |
6b06314c JA |
8089 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
8090 | if (!fpl) | |
8091 | return -ENOMEM; | |
8092 | ||
8093 | skb = alloc_skb(0, GFP_KERNEL); | |
8094 | if (!skb) { | |
8095 | kfree(fpl); | |
8096 | return -ENOMEM; | |
8097 | } | |
8098 | ||
8099 | skb->sk = sk; | |
6b06314c | 8100 | |
08a45173 | 8101 | nr_files = 0; |
62e398be | 8102 | fpl->user = get_uid(current_user()); |
6b06314c | 8103 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
8104 | struct file *file = io_file_from_index(ctx, i + offset); |
8105 | ||
8106 | if (!file) | |
08a45173 | 8107 | continue; |
65e19f54 | 8108 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
8109 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
8110 | nr_files++; | |
6b06314c JA |
8111 | } |
8112 | ||
08a45173 JA |
8113 | if (nr_files) { |
8114 | fpl->max = SCM_MAX_FD; | |
8115 | fpl->count = nr_files; | |
8116 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 8117 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
8118 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
8119 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 8120 | |
40d8cb26 PB |
8121 | for (i = 0; i < nr; i++) { |
8122 | struct file *file = io_file_from_index(ctx, i + offset); | |
8123 | ||
8124 | if (file) | |
8125 | fput(file); | |
8126 | } | |
08a45173 JA |
8127 | } else { |
8128 | kfree_skb(skb); | |
54c4d35a | 8129 | free_uid(fpl->user); |
08a45173 JA |
8130 | kfree(fpl); |
8131 | } | |
6b06314c JA |
8132 | |
8133 | return 0; | |
8134 | } | |
8135 | ||
8136 | /* | |
8137 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
8138 | * causes regular reference counting to break down. We rely on the UNIX | |
8139 | * garbage collection to take care of this problem for us. | |
8140 | */ | |
8141 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8142 | { | |
8143 | unsigned left, total; | |
8144 | int ret = 0; | |
8145 | ||
8146 | total = 0; | |
8147 | left = ctx->nr_user_files; | |
8148 | while (left) { | |
8149 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
8150 | |
8151 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
8152 | if (ret) | |
8153 | break; | |
8154 | left -= this_files; | |
8155 | total += this_files; | |
8156 | } | |
8157 | ||
8158 | if (!ret) | |
8159 | return 0; | |
8160 | ||
8161 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
8162 | struct file *file = io_file_from_index(ctx, total); |
8163 | ||
8164 | if (file) | |
8165 | fput(file); | |
6b06314c JA |
8166 | total++; |
8167 | } | |
8168 | ||
8169 | return ret; | |
8170 | } | |
8171 | #else | |
8172 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8173 | { | |
8174 | return 0; | |
8175 | } | |
8176 | #endif | |
8177 | ||
47e90392 | 8178 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 8179 | { |
50238531 | 8180 | struct file *file = prsrc->file; |
05f3fb3c JA |
8181 | #if defined(CONFIG_UNIX) |
8182 | struct sock *sock = ctx->ring_sock->sk; | |
8183 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
8184 | struct sk_buff *skb; | |
8185 | int i; | |
8186 | ||
8187 | __skb_queue_head_init(&list); | |
8188 | ||
8189 | /* | |
8190 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
8191 | * remove this entry and rearrange the file array. | |
8192 | */ | |
8193 | skb = skb_dequeue(head); | |
8194 | while (skb) { | |
8195 | struct scm_fp_list *fp; | |
8196 | ||
8197 | fp = UNIXCB(skb).fp; | |
8198 | for (i = 0; i < fp->count; i++) { | |
8199 | int left; | |
8200 | ||
8201 | if (fp->fp[i] != file) | |
8202 | continue; | |
8203 | ||
8204 | unix_notinflight(fp->user, fp->fp[i]); | |
8205 | left = fp->count - 1 - i; | |
8206 | if (left) { | |
8207 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
8208 | left * sizeof(struct file *)); | |
8209 | } | |
8210 | fp->count--; | |
8211 | if (!fp->count) { | |
8212 | kfree_skb(skb); | |
8213 | skb = NULL; | |
8214 | } else { | |
8215 | __skb_queue_tail(&list, skb); | |
8216 | } | |
8217 | fput(file); | |
8218 | file = NULL; | |
8219 | break; | |
8220 | } | |
8221 | ||
8222 | if (!file) | |
8223 | break; | |
8224 | ||
8225 | __skb_queue_tail(&list, skb); | |
8226 | ||
8227 | skb = skb_dequeue(head); | |
8228 | } | |
8229 | ||
8230 | if (skb_peek(&list)) { | |
8231 | spin_lock_irq(&head->lock); | |
8232 | while ((skb = __skb_dequeue(&list)) != NULL) | |
8233 | __skb_queue_tail(head, skb); | |
8234 | spin_unlock_irq(&head->lock); | |
8235 | } | |
8236 | #else | |
8237 | fput(file); | |
8238 | #endif | |
8239 | } | |
8240 | ||
b895c9a6 | 8241 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 8242 | { |
b895c9a6 | 8243 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
8244 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
8245 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 8246 | |
269bbe5f BM |
8247 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
8248 | list_del(&prsrc->list); | |
b60c8dce PB |
8249 | |
8250 | if (prsrc->tag) { | |
8251 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
8252 | |
8253 | io_ring_submit_lock(ctx, lock_ring); | |
79ebeaee | 8254 | spin_lock(&ctx->completion_lock); |
b60c8dce | 8255 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 8256 | ctx->cq_extra++; |
b60c8dce | 8257 | io_commit_cqring(ctx); |
79ebeaee | 8258 | spin_unlock(&ctx->completion_lock); |
b60c8dce PB |
8259 | io_cqring_ev_posted(ctx); |
8260 | io_ring_submit_unlock(ctx, lock_ring); | |
8261 | } | |
8262 | ||
40ae0ff7 | 8263 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 8264 | kfree(prsrc); |
65e19f54 | 8265 | } |
05589553 | 8266 | |
28a9fe25 | 8267 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
8268 | if (atomic_dec_and_test(&rsrc_data->refs)) |
8269 | complete(&rsrc_data->done); | |
2faf852d | 8270 | } |
65e19f54 | 8271 | |
269bbe5f | 8272 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
8273 | { |
8274 | struct io_ring_ctx *ctx; | |
8275 | struct llist_node *node; | |
8276 | ||
269bbe5f BM |
8277 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
8278 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
8279 | |
8280 | while (node) { | |
b895c9a6 | 8281 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
8282 | struct llist_node *next = node->next; |
8283 | ||
b895c9a6 | 8284 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 8285 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
8286 | node = next; |
8287 | } | |
8288 | } | |
8289 | ||
6b06314c | 8290 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 8291 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
8292 | { |
8293 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 8294 | struct file *file; |
f3baed39 | 8295 | int fd, ret; |
846a4ef2 | 8296 | unsigned i; |
6b06314c | 8297 | |
05f3fb3c | 8298 | if (ctx->file_data) |
6b06314c JA |
8299 | return -EBUSY; |
8300 | if (!nr_args) | |
8301 | return -EINVAL; | |
8302 | if (nr_args > IORING_MAX_FIXED_FILES) | |
8303 | return -EMFILE; | |
3a1b8a4e PB |
8304 | if (nr_args > rlimit(RLIMIT_NOFILE)) |
8305 | return -EMFILE; | |
a7f0ed5a | 8306 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
8307 | if (ret) |
8308 | return ret; | |
d878c816 PB |
8309 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
8310 | &ctx->file_data); | |
8311 | if (ret) | |
8312 | return ret; | |
6b06314c | 8313 | |
f3baed39 | 8314 | ret = -ENOMEM; |
aeca241b | 8315 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 8316 | goto out_free; |
65e19f54 | 8317 | |
08a45173 | 8318 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 8319 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
8320 | ret = -EFAULT; |
8321 | goto out_fput; | |
8322 | } | |
08a45173 | 8323 | /* allow sparse sets */ |
792e3582 PB |
8324 | if (fd == -1) { |
8325 | ret = -EINVAL; | |
2d091d62 | 8326 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 8327 | goto out_fput; |
08a45173 | 8328 | continue; |
792e3582 | 8329 | } |
6b06314c | 8330 | |
05f3fb3c | 8331 | file = fget(fd); |
6b06314c | 8332 | ret = -EBADF; |
792e3582 | 8333 | if (unlikely(!file)) |
600cf3f8 | 8334 | goto out_fput; |
05f3fb3c | 8335 | |
6b06314c JA |
8336 | /* |
8337 | * Don't allow io_uring instances to be registered. If UNIX | |
8338 | * isn't enabled, then this causes a reference cycle and this | |
8339 | * instance can never get freed. If UNIX is enabled we'll | |
8340 | * handle it just fine, but there's still no point in allowing | |
8341 | * a ring fd as it doesn't support regular read/write anyway. | |
8342 | */ | |
05f3fb3c JA |
8343 | if (file->f_op == &io_uring_fops) { |
8344 | fput(file); | |
600cf3f8 | 8345 | goto out_fput; |
6b06314c | 8346 | } |
aeca241b | 8347 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
8348 | } |
8349 | ||
6b06314c | 8350 | ret = io_sqe_files_scm(ctx); |
05589553 | 8351 | if (ret) { |
08480400 | 8352 | __io_sqe_files_unregister(ctx); |
05589553 XW |
8353 | return ret; |
8354 | } | |
6b06314c | 8355 | |
a7f0ed5a | 8356 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 8357 | return ret; |
600cf3f8 PB |
8358 | out_fput: |
8359 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8360 | file = io_file_from_index(ctx, i); | |
8361 | if (file) | |
8362 | fput(file); | |
8363 | } | |
042b0d85 | 8364 | io_free_file_tables(&ctx->file_table); |
600cf3f8 | 8365 | ctx->nr_user_files = 0; |
600cf3f8 | 8366 | out_free: |
44b31f2f | 8367 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 8368 | ctx->file_data = NULL; |
6b06314c JA |
8369 | return ret; |
8370 | } | |
8371 | ||
c3a31e60 JA |
8372 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
8373 | int index) | |
8374 | { | |
8375 | #if defined(CONFIG_UNIX) | |
8376 | struct sock *sock = ctx->ring_sock->sk; | |
8377 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
8378 | struct sk_buff *skb; | |
8379 | ||
8380 | /* | |
8381 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
8382 | * file set. If there's no room, fall back to allocating a new skb | |
8383 | * and filling it in. | |
8384 | */ | |
8385 | spin_lock_irq(&head->lock); | |
8386 | skb = skb_peek(head); | |
8387 | if (skb) { | |
8388 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
8389 | ||
8390 | if (fpl->count < SCM_MAX_FD) { | |
8391 | __skb_unlink(skb, head); | |
8392 | spin_unlock_irq(&head->lock); | |
8393 | fpl->fp[fpl->count] = get_file(file); | |
8394 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
8395 | fpl->count++; | |
8396 | spin_lock_irq(&head->lock); | |
8397 | __skb_queue_head(head, skb); | |
8398 | } else { | |
8399 | skb = NULL; | |
8400 | } | |
8401 | } | |
8402 | spin_unlock_irq(&head->lock); | |
8403 | ||
8404 | if (skb) { | |
8405 | fput(file); | |
8406 | return 0; | |
8407 | } | |
8408 | ||
8409 | return __io_sqe_files_scm(ctx, 1, index); | |
8410 | #else | |
8411 | return 0; | |
8412 | #endif | |
8413 | } | |
8414 | ||
9c7b0ba8 PB |
8415 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
8416 | struct io_rsrc_node *node, void *rsrc) | |
8417 | { | |
c47dfb44 | 8418 | u64 *tag_slot = io_get_tag_slot(data, idx); |
9c7b0ba8 PB |
8419 | struct io_rsrc_put *prsrc; |
8420 | ||
8421 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); | |
8422 | if (!prsrc) | |
8423 | return -ENOMEM; | |
8424 | ||
c47dfb44 PB |
8425 | prsrc->tag = *tag_slot; |
8426 | *tag_slot = 0; | |
9c7b0ba8 PB |
8427 | prsrc->rsrc = rsrc; |
8428 | list_add(&prsrc->list, &node->rsrc_list); | |
8429 | return 0; | |
8430 | } | |
8431 | ||
b9445598 PB |
8432 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
8433 | unsigned int issue_flags, u32 slot_index) | |
8434 | { | |
8435 | struct io_ring_ctx *ctx = req->ctx; | |
8436 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; | |
9c7b0ba8 | 8437 | bool needs_switch = false; |
b9445598 PB |
8438 | struct io_fixed_file *file_slot; |
8439 | int ret = -EBADF; | |
8440 | ||
8441 | io_ring_submit_lock(ctx, !force_nonblock); | |
8442 | if (file->f_op == &io_uring_fops) | |
8443 | goto err; | |
8444 | ret = -ENXIO; | |
8445 | if (!ctx->file_data) | |
8446 | goto err; | |
8447 | ret = -EINVAL; | |
8448 | if (slot_index >= ctx->nr_user_files) | |
8449 | goto err; | |
8450 | ||
8451 | slot_index = array_index_nospec(slot_index, ctx->nr_user_files); | |
8452 | file_slot = io_fixed_file_slot(&ctx->file_table, slot_index); | |
9c7b0ba8 PB |
8453 | |
8454 | if (file_slot->file_ptr) { | |
8455 | struct file *old_file; | |
8456 | ||
8457 | ret = io_rsrc_node_switch_start(ctx); | |
8458 | if (ret) | |
8459 | goto err; | |
8460 | ||
8461 | old_file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
8462 | ret = io_queue_rsrc_removal(ctx->file_data, slot_index, | |
8463 | ctx->rsrc_node, old_file); | |
8464 | if (ret) | |
8465 | goto err; | |
8466 | file_slot->file_ptr = 0; | |
8467 | needs_switch = true; | |
8468 | } | |
b9445598 PB |
8469 | |
8470 | *io_get_tag_slot(ctx->file_data, slot_index) = 0; | |
8471 | io_fixed_file_set(file_slot, file); | |
8472 | ret = io_sqe_file_register(ctx, file, slot_index); | |
8473 | if (ret) { | |
8474 | file_slot->file_ptr = 0; | |
8475 | goto err; | |
8476 | } | |
8477 | ||
8478 | ret = 0; | |
8479 | err: | |
9c7b0ba8 PB |
8480 | if (needs_switch) |
8481 | io_rsrc_node_switch(ctx, ctx->file_data); | |
b9445598 PB |
8482 | io_ring_submit_unlock(ctx, !force_nonblock); |
8483 | if (ret) | |
8484 | fput(file); | |
8485 | return ret; | |
8486 | } | |
8487 | ||
7df778be PB |
8488 | static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) |
8489 | { | |
8490 | unsigned int offset = req->close.file_slot - 1; | |
8491 | struct io_ring_ctx *ctx = req->ctx; | |
8492 | struct io_fixed_file *file_slot; | |
8493 | struct file *file; | |
22a99dcb | 8494 | int ret; |
7df778be PB |
8495 | |
8496 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
8497 | ret = -ENXIO; | |
8498 | if (unlikely(!ctx->file_data)) | |
8499 | goto out; | |
8500 | ret = -EINVAL; | |
8501 | if (offset >= ctx->nr_user_files) | |
8502 | goto out; | |
8503 | ret = io_rsrc_node_switch_start(ctx); | |
8504 | if (ret) | |
8505 | goto out; | |
8506 | ||
22a99dcb PB |
8507 | offset = array_index_nospec(offset, ctx->nr_user_files); |
8508 | file_slot = io_fixed_file_slot(&ctx->file_table, offset); | |
7df778be PB |
8509 | ret = -EBADF; |
8510 | if (!file_slot->file_ptr) | |
8511 | goto out; | |
8512 | ||
8513 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
8514 | ret = io_queue_rsrc_removal(ctx->file_data, offset, ctx->rsrc_node, file); | |
8515 | if (ret) | |
8516 | goto out; | |
8517 | ||
8518 | file_slot->file_ptr = 0; | |
8519 | io_rsrc_node_switch(ctx, ctx->file_data); | |
8520 | ret = 0; | |
8521 | out: | |
8522 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
8523 | return ret; | |
8524 | } | |
8525 | ||
05f3fb3c | 8526 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
c3bdad02 | 8527 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
8528 | unsigned nr_args) |
8529 | { | |
c3bdad02 | 8530 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 8531 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 8532 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
8533 | struct io_fixed_file *file_slot; |
8534 | struct file *file; | |
98f0b3b4 PB |
8535 | int fd, i, err = 0; |
8536 | unsigned int done; | |
05589553 | 8537 | bool needs_switch = false; |
c3a31e60 | 8538 | |
98f0b3b4 PB |
8539 | if (!ctx->file_data) |
8540 | return -ENXIO; | |
8541 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
8542 | return -EINVAL; |
8543 | ||
67973b93 | 8544 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
8545 | u64 tag = 0; |
8546 | ||
8547 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
8548 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
8549 | err = -EFAULT; |
8550 | break; | |
8551 | } | |
c3bdad02 PB |
8552 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
8553 | err = -EINVAL; | |
8554 | break; | |
8555 | } | |
4e0377a1 | 8556 | if (fd == IORING_REGISTER_FILES_SKIP) |
8557 | continue; | |
8558 | ||
67973b93 | 8559 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 8560 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 8561 | |
a04b0ac0 PB |
8562 | if (file_slot->file_ptr) { |
8563 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
22a99dcb | 8564 | err = io_queue_rsrc_removal(data, i, ctx->rsrc_node, file); |
a5318d3c HD |
8565 | if (err) |
8566 | break; | |
a04b0ac0 | 8567 | file_slot->file_ptr = 0; |
05589553 | 8568 | needs_switch = true; |
c3a31e60 JA |
8569 | } |
8570 | if (fd != -1) { | |
c3a31e60 JA |
8571 | file = fget(fd); |
8572 | if (!file) { | |
8573 | err = -EBADF; | |
8574 | break; | |
8575 | } | |
8576 | /* | |
8577 | * Don't allow io_uring instances to be registered. If | |
8578 | * UNIX isn't enabled, then this causes a reference | |
8579 | * cycle and this instance can never get freed. If UNIX | |
8580 | * is enabled we'll handle it just fine, but there's | |
8581 | * still no point in allowing a ring fd as it doesn't | |
8582 | * support regular read/write anyway. | |
8583 | */ | |
8584 | if (file->f_op == &io_uring_fops) { | |
8585 | fput(file); | |
8586 | err = -EBADF; | |
8587 | break; | |
8588 | } | |
1dc25797 | 8589 | *io_get_tag_slot(data, i) = tag; |
9a321c98 | 8590 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 8591 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8592 | if (err) { |
a04b0ac0 | 8593 | file_slot->file_ptr = 0; |
f3bd9dae | 8594 | fput(file); |
c3a31e60 | 8595 | break; |
f3bd9dae | 8596 | } |
c3a31e60 | 8597 | } |
05f3fb3c JA |
8598 | } |
8599 | ||
a7f0ed5a PB |
8600 | if (needs_switch) |
8601 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
8602 | return done ? done : err; |
8603 | } | |
05589553 | 8604 | |
685fe7fe JA |
8605 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
8606 | struct task_struct *task) | |
24369c2e | 8607 | { |
e941894e | 8608 | struct io_wq_hash *hash; |
24369c2e | 8609 | struct io_wq_data data; |
24369c2e | 8610 | unsigned int concurrency; |
24369c2e | 8611 | |
362a9e65 | 8612 | mutex_lock(&ctx->uring_lock); |
e941894e JA |
8613 | hash = ctx->hash_map; |
8614 | if (!hash) { | |
8615 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
362a9e65 YY |
8616 | if (!hash) { |
8617 | mutex_unlock(&ctx->uring_lock); | |
e941894e | 8618 | return ERR_PTR(-ENOMEM); |
362a9e65 | 8619 | } |
e941894e JA |
8620 | refcount_set(&hash->refs, 1); |
8621 | init_waitqueue_head(&hash->wait); | |
8622 | ctx->hash_map = hash; | |
24369c2e | 8623 | } |
362a9e65 | 8624 | mutex_unlock(&ctx->uring_lock); |
24369c2e | 8625 | |
e941894e | 8626 | data.hash = hash; |
685fe7fe | 8627 | data.task = task; |
ebc11b6c | 8628 | data.free_work = io_wq_free_work; |
f5fa38c5 | 8629 | data.do_work = io_wq_submit_work; |
24369c2e | 8630 | |
d25e3a3d JA |
8631 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
8632 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 8633 | |
5aa75ed5 | 8634 | return io_wq_create(concurrency, &data); |
24369c2e PB |
8635 | } |
8636 | ||
5aa75ed5 JA |
8637 | static int io_uring_alloc_task_context(struct task_struct *task, |
8638 | struct io_ring_ctx *ctx) | |
0f212204 JA |
8639 | { |
8640 | struct io_uring_task *tctx; | |
d8a6df10 | 8641 | int ret; |
0f212204 | 8642 | |
09899b19 | 8643 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
8644 | if (unlikely(!tctx)) |
8645 | return -ENOMEM; | |
8646 | ||
d8a6df10 JA |
8647 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8648 | if (unlikely(ret)) { | |
8649 | kfree(tctx); | |
8650 | return ret; | |
8651 | } | |
8652 | ||
685fe7fe | 8653 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
8654 | if (IS_ERR(tctx->io_wq)) { |
8655 | ret = PTR_ERR(tctx->io_wq); | |
8656 | percpu_counter_destroy(&tctx->inflight); | |
8657 | kfree(tctx); | |
8658 | return ret; | |
8659 | } | |
8660 | ||
0f212204 JA |
8661 | xa_init(&tctx->xa); |
8662 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 8663 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 8664 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 8665 | task->io_uring = tctx; |
7cbf1722 JA |
8666 | spin_lock_init(&tctx->task_lock); |
8667 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 8668 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
8669 | return 0; |
8670 | } | |
8671 | ||
8672 | void __io_uring_free(struct task_struct *tsk) | |
8673 | { | |
8674 | struct io_uring_task *tctx = tsk->io_uring; | |
8675 | ||
8676 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 8677 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 8678 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 8679 | |
d8a6df10 | 8680 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8681 | kfree(tctx); |
8682 | tsk->io_uring = NULL; | |
8683 | } | |
8684 | ||
7e84e1c7 SG |
8685 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8686 | struct io_uring_params *p) | |
2b188cc1 JA |
8687 | { |
8688 | int ret; | |
8689 | ||
d25e3a3d JA |
8690 | /* Retain compatibility with failing for an invalid attach attempt */ |
8691 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
8692 | IORING_SETUP_ATTACH_WQ) { | |
8693 | struct fd f; | |
8694 | ||
8695 | f = fdget(p->wq_fd); | |
8696 | if (!f.file) | |
8697 | return -ENXIO; | |
0cc936f7 JA |
8698 | if (f.file->f_op != &io_uring_fops) { |
8699 | fdput(f); | |
f2a48dd0 | 8700 | return -EINVAL; |
0cc936f7 JA |
8701 | } |
8702 | fdput(f); | |
d25e3a3d | 8703 | } |
6c271ce2 | 8704 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 8705 | struct task_struct *tsk; |
534ca6d6 | 8706 | struct io_sq_data *sqd; |
26984fbf | 8707 | bool attached; |
534ca6d6 | 8708 | |
26984fbf | 8709 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8710 | if (IS_ERR(sqd)) { |
8711 | ret = PTR_ERR(sqd); | |
8712 | goto err; | |
8713 | } | |
69fb2131 | 8714 | |
7c30f36a | 8715 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8716 | ctx->sq_data = sqd; |
917257da JA |
8717 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8718 | if (!ctx->sq_thread_idle) | |
8719 | ctx->sq_thread_idle = HZ; | |
8720 | ||
78d7f6ba | 8721 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8722 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8723 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8724 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8725 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8726 | io_sq_thread_unpark(sqd); |
8727 | ||
de75a3d3 PB |
8728 | if (ret < 0) |
8729 | goto err; | |
8730 | if (attached) | |
5aa75ed5 | 8731 | return 0; |
aa06165d | 8732 | |
6c271ce2 | 8733 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8734 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8735 | |
917257da | 8736 | ret = -EINVAL; |
f2a48dd0 | 8737 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8738 | goto err_sqpoll; |
37d1e2e3 | 8739 | sqd->sq_cpu = cpu; |
6c271ce2 | 8740 | } else { |
37d1e2e3 | 8741 | sqd->sq_cpu = -1; |
6c271ce2 | 8742 | } |
37d1e2e3 JA |
8743 | |
8744 | sqd->task_pid = current->pid; | |
5c2469e0 | 8745 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8746 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8747 | if (IS_ERR(tsk)) { | |
8748 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8749 | goto err_sqpoll; |
6c271ce2 | 8750 | } |
97a73a0f | 8751 | |
46fe18b1 | 8752 | sqd->thread = tsk; |
97a73a0f | 8753 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8754 | wake_up_new_task(tsk); |
0f212204 JA |
8755 | if (ret) |
8756 | goto err; | |
6c271ce2 JA |
8757 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8758 | /* Can't have SQ_AFF without SQPOLL */ | |
8759 | ret = -EINVAL; | |
8760 | goto err; | |
8761 | } | |
8762 | ||
2b188cc1 | 8763 | return 0; |
f2a48dd0 PB |
8764 | err_sqpoll: |
8765 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8766 | err: |
37d1e2e3 | 8767 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8768 | return ret; |
8769 | } | |
8770 | ||
a087e2b5 BM |
8771 | static inline void __io_unaccount_mem(struct user_struct *user, |
8772 | unsigned long nr_pages) | |
2b188cc1 JA |
8773 | { |
8774 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8775 | } | |
8776 | ||
a087e2b5 BM |
8777 | static inline int __io_account_mem(struct user_struct *user, |
8778 | unsigned long nr_pages) | |
2b188cc1 JA |
8779 | { |
8780 | unsigned long page_limit, cur_pages, new_pages; | |
8781 | ||
8782 | /* Don't allow more pages than we can safely lock */ | |
8783 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8784 | ||
8785 | do { | |
8786 | cur_pages = atomic_long_read(&user->locked_vm); | |
8787 | new_pages = cur_pages + nr_pages; | |
8788 | if (new_pages > page_limit) | |
8789 | return -ENOMEM; | |
8790 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8791 | new_pages) != cur_pages); | |
8792 | ||
8793 | return 0; | |
8794 | } | |
8795 | ||
26bfa89e | 8796 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8797 | { |
62e398be | 8798 | if (ctx->user) |
a087e2b5 | 8799 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8800 | |
26bfa89e JA |
8801 | if (ctx->mm_account) |
8802 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8803 | } |
8804 | ||
26bfa89e | 8805 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8806 | { |
30975825 BM |
8807 | int ret; |
8808 | ||
62e398be | 8809 | if (ctx->user) { |
30975825 BM |
8810 | ret = __io_account_mem(ctx->user, nr_pages); |
8811 | if (ret) | |
8812 | return ret; | |
8813 | } | |
8814 | ||
26bfa89e JA |
8815 | if (ctx->mm_account) |
8816 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8817 | |
8818 | return 0; | |
8819 | } | |
8820 | ||
2b188cc1 JA |
8821 | static void io_mem_free(void *ptr) |
8822 | { | |
52e04ef4 MR |
8823 | struct page *page; |
8824 | ||
8825 | if (!ptr) | |
8826 | return; | |
2b188cc1 | 8827 | |
52e04ef4 | 8828 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8829 | if (put_page_testzero(page)) |
8830 | free_compound_page(page); | |
8831 | } | |
8832 | ||
8833 | static void *io_mem_alloc(size_t size) | |
8834 | { | |
44736143 | 8835 | gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP; |
2b188cc1 | 8836 | |
44736143 | 8837 | return (void *) __get_free_pages(gfp, get_order(size)); |
2b188cc1 JA |
8838 | } |
8839 | ||
75b28aff HV |
8840 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8841 | size_t *sq_offset) | |
8842 | { | |
8843 | struct io_rings *rings; | |
8844 | size_t off, sq_array_size; | |
8845 | ||
8846 | off = struct_size(rings, cqes, cq_entries); | |
8847 | if (off == SIZE_MAX) | |
8848 | return SIZE_MAX; | |
8849 | ||
8850 | #ifdef CONFIG_SMP | |
8851 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8852 | if (off == 0) | |
8853 | return SIZE_MAX; | |
8854 | #endif | |
8855 | ||
b36200f5 DV |
8856 | if (sq_offset) |
8857 | *sq_offset = off; | |
8858 | ||
75b28aff HV |
8859 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8860 | if (sq_array_size == SIZE_MAX) | |
8861 | return SIZE_MAX; | |
8862 | ||
8863 | if (check_add_overflow(off, sq_array_size, &off)) | |
8864 | return SIZE_MAX; | |
8865 | ||
75b28aff HV |
8866 | return off; |
8867 | } | |
8868 | ||
41edf1a5 | 8869 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8870 | { |
41edf1a5 | 8871 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8872 | unsigned int i; |
8873 | ||
6224843d PB |
8874 | if (imu != ctx->dummy_ubuf) { |
8875 | for (i = 0; i < imu->nr_bvecs; i++) | |
8876 | unpin_user_page(imu->bvec[i].bv_page); | |
8877 | if (imu->acct_pages) | |
8878 | io_unaccount_mem(ctx, imu->acct_pages); | |
8879 | kvfree(imu); | |
8880 | } | |
41edf1a5 | 8881 | *slot = NULL; |
7f61a1e9 PB |
8882 | } |
8883 | ||
bd54b6fe | 8884 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8885 | { |
634d00df PB |
8886 | io_buffer_unmap(ctx, &prsrc->buf); |
8887 | prsrc->buf = NULL; | |
bd54b6fe | 8888 | } |
edafccee | 8889 | |
bd54b6fe BM |
8890 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8891 | { | |
8892 | unsigned int i; | |
edafccee | 8893 | |
7f61a1e9 PB |
8894 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8895 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8896 | kfree(ctx->user_bufs); |
bb6659cc | 8897 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8898 | ctx->user_bufs = NULL; |
bd54b6fe | 8899 | ctx->buf_data = NULL; |
edafccee | 8900 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8901 | } |
8902 | ||
0a96bbe4 | 8903 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8904 | { |
bd54b6fe | 8905 | int ret; |
edafccee | 8906 | |
bd54b6fe | 8907 | if (!ctx->buf_data) |
edafccee JA |
8908 | return -ENXIO; |
8909 | ||
bd54b6fe BM |
8910 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8911 | if (!ret) | |
8912 | __io_sqe_buffers_unregister(ctx); | |
8913 | return ret; | |
edafccee JA |
8914 | } |
8915 | ||
8916 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8917 | void __user *arg, unsigned index) | |
8918 | { | |
8919 | struct iovec __user *src; | |
8920 | ||
8921 | #ifdef CONFIG_COMPAT | |
8922 | if (ctx->compat) { | |
8923 | struct compat_iovec __user *ciovs; | |
8924 | struct compat_iovec ciov; | |
8925 | ||
8926 | ciovs = (struct compat_iovec __user *) arg; | |
8927 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8928 | return -EFAULT; | |
8929 | ||
d55e5f5b | 8930 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8931 | dst->iov_len = ciov.iov_len; |
8932 | return 0; | |
8933 | } | |
8934 | #endif | |
8935 | src = (struct iovec __user *) arg; | |
8936 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8937 | return -EFAULT; | |
8938 | return 0; | |
8939 | } | |
8940 | ||
de293938 JA |
8941 | /* |
8942 | * Not super efficient, but this is just a registration time. And we do cache | |
8943 | * the last compound head, so generally we'll only do a full search if we don't | |
8944 | * match that one. | |
8945 | * | |
8946 | * We check if the given compound head page has already been accounted, to | |
8947 | * avoid double accounting it. This allows us to account the full size of the | |
8948 | * page, not just the constituent pages of a huge page. | |
8949 | */ | |
8950 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8951 | int nr_pages, struct page *hpage) | |
8952 | { | |
8953 | int i, j; | |
8954 | ||
8955 | /* check current page array */ | |
8956 | for (i = 0; i < nr_pages; i++) { | |
8957 | if (!PageCompound(pages[i])) | |
8958 | continue; | |
8959 | if (compound_head(pages[i]) == hpage) | |
8960 | return true; | |
8961 | } | |
8962 | ||
8963 | /* check previously registered pages */ | |
8964 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8965 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8966 | |
8967 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8968 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8969 | continue; | |
8970 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8971 | return true; | |
8972 | } | |
8973 | } | |
8974 | ||
8975 | return false; | |
8976 | } | |
8977 | ||
8978 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8979 | int nr_pages, struct io_mapped_ubuf *imu, | |
8980 | struct page **last_hpage) | |
8981 | { | |
8982 | int i, ret; | |
8983 | ||
216e5835 | 8984 | imu->acct_pages = 0; |
de293938 JA |
8985 | for (i = 0; i < nr_pages; i++) { |
8986 | if (!PageCompound(pages[i])) { | |
8987 | imu->acct_pages++; | |
8988 | } else { | |
8989 | struct page *hpage; | |
8990 | ||
8991 | hpage = compound_head(pages[i]); | |
8992 | if (hpage == *last_hpage) | |
8993 | continue; | |
8994 | *last_hpage = hpage; | |
8995 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8996 | continue; | |
8997 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8998 | } | |
8999 | } | |
9000 | ||
9001 | if (!imu->acct_pages) | |
9002 | return 0; | |
9003 | ||
26bfa89e | 9004 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
9005 | if (ret) |
9006 | imu->acct_pages = 0; | |
9007 | return ret; | |
9008 | } | |
9009 | ||
0a96bbe4 | 9010 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 9011 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 9012 | struct page **last_hpage) |
edafccee | 9013 | { |
41edf1a5 | 9014 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
9015 | struct vm_area_struct **vmas = NULL; |
9016 | struct page **pages = NULL; | |
0a96bbe4 BM |
9017 | unsigned long off, start, end, ubuf; |
9018 | size_t size; | |
9019 | int ret, pret, nr_pages, i; | |
9020 | ||
6224843d PB |
9021 | if (!iov->iov_base) { |
9022 | *pimu = ctx->dummy_ubuf; | |
9023 | return 0; | |
9024 | } | |
9025 | ||
0a96bbe4 BM |
9026 | ubuf = (unsigned long) iov->iov_base; |
9027 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
9028 | start = ubuf >> PAGE_SHIFT; | |
9029 | nr_pages = end - start; | |
9030 | ||
41edf1a5 | 9031 | *pimu = NULL; |
0a96bbe4 BM |
9032 | ret = -ENOMEM; |
9033 | ||
9034 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
9035 | if (!pages) | |
9036 | goto done; | |
9037 | ||
9038 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
9039 | GFP_KERNEL); | |
9040 | if (!vmas) | |
9041 | goto done; | |
edafccee | 9042 | |
41edf1a5 | 9043 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 9044 | if (!imu) |
0a96bbe4 BM |
9045 | goto done; |
9046 | ||
9047 | ret = 0; | |
9048 | mmap_read_lock(current->mm); | |
9049 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
9050 | pages, vmas); | |
9051 | if (pret == nr_pages) { | |
9052 | /* don't support file backed memory */ | |
9053 | for (i = 0; i < nr_pages; i++) { | |
9054 | struct vm_area_struct *vma = vmas[i]; | |
9055 | ||
40dad765 PB |
9056 | if (vma_is_shmem(vma)) |
9057 | continue; | |
0a96bbe4 BM |
9058 | if (vma->vm_file && |
9059 | !is_file_hugepages(vma->vm_file)) { | |
9060 | ret = -EOPNOTSUPP; | |
9061 | break; | |
9062 | } | |
9063 | } | |
9064 | } else { | |
9065 | ret = pret < 0 ? pret : -EFAULT; | |
9066 | } | |
9067 | mmap_read_unlock(current->mm); | |
9068 | if (ret) { | |
9069 | /* | |
9070 | * if we did partial map, or found file backed vmas, | |
9071 | * release any pages we did get | |
9072 | */ | |
9073 | if (pret > 0) | |
9074 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
9075 | goto done; |
9076 | } | |
9077 | ||
9078 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
9079 | if (ret) { | |
9080 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
9081 | goto done; |
9082 | } | |
9083 | ||
9084 | off = ubuf & ~PAGE_MASK; | |
9085 | size = iov->iov_len; | |
9086 | for (i = 0; i < nr_pages; i++) { | |
9087 | size_t vec_len; | |
9088 | ||
9089 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
9090 | imu->bvec[i].bv_page = pages[i]; | |
9091 | imu->bvec[i].bv_len = vec_len; | |
9092 | imu->bvec[i].bv_offset = off; | |
9093 | off = 0; | |
9094 | size -= vec_len; | |
9095 | } | |
9096 | /* store original address for later verification */ | |
9097 | imu->ubuf = ubuf; | |
4751f53d | 9098 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 9099 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 9100 | *pimu = imu; |
0a96bbe4 BM |
9101 | ret = 0; |
9102 | done: | |
41edf1a5 PB |
9103 | if (ret) |
9104 | kvfree(imu); | |
0a96bbe4 BM |
9105 | kvfree(pages); |
9106 | kvfree(vmas); | |
9107 | return ret; | |
9108 | } | |
9109 | ||
2b358604 | 9110 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 9111 | { |
87094465 PB |
9112 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
9113 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 9114 | } |
edafccee | 9115 | |
2b358604 BM |
9116 | static int io_buffer_validate(struct iovec *iov) |
9117 | { | |
50e96989 PB |
9118 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
9119 | ||
2b358604 BM |
9120 | /* |
9121 | * Don't impose further limits on the size and buffer | |
9122 | * constraints here, we'll -EINVAL later when IO is | |
9123 | * submitted if they are wrong. | |
9124 | */ | |
6224843d PB |
9125 | if (!iov->iov_base) |
9126 | return iov->iov_len ? -EFAULT : 0; | |
9127 | if (!iov->iov_len) | |
2b358604 | 9128 | return -EFAULT; |
edafccee | 9129 | |
2b358604 BM |
9130 | /* arbitrary limit, but we need something */ |
9131 | if (iov->iov_len > SZ_1G) | |
9132 | return -EFAULT; | |
edafccee | 9133 | |
50e96989 PB |
9134 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
9135 | return -EOVERFLOW; | |
9136 | ||
2b358604 BM |
9137 | return 0; |
9138 | } | |
edafccee | 9139 | |
2b358604 | 9140 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 9141 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 9142 | { |
bd54b6fe BM |
9143 | struct page *last_hpage = NULL; |
9144 | struct io_rsrc_data *data; | |
2b358604 BM |
9145 | int i, ret; |
9146 | struct iovec iov; | |
edafccee | 9147 | |
87094465 PB |
9148 | if (ctx->user_bufs) |
9149 | return -EBUSY; | |
489809e2 | 9150 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 9151 | return -EINVAL; |
bd54b6fe | 9152 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
9153 | if (ret) |
9154 | return ret; | |
d878c816 PB |
9155 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
9156 | if (ret) | |
9157 | return ret; | |
bd54b6fe BM |
9158 | ret = io_buffers_map_alloc(ctx, nr_args); |
9159 | if (ret) { | |
bb6659cc | 9160 | io_rsrc_data_free(data); |
bd54b6fe BM |
9161 | return ret; |
9162 | } | |
edafccee | 9163 | |
87094465 | 9164 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
9165 | ret = io_copy_iov(ctx, &iov, arg, i); |
9166 | if (ret) | |
0a96bbe4 | 9167 | break; |
2b358604 BM |
9168 | ret = io_buffer_validate(&iov); |
9169 | if (ret) | |
0a96bbe4 | 9170 | break; |
2d091d62 | 9171 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
9172 | ret = -EINVAL; |
9173 | break; | |
9174 | } | |
edafccee | 9175 | |
41edf1a5 PB |
9176 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
9177 | &last_hpage); | |
0a96bbe4 BM |
9178 | if (ret) |
9179 | break; | |
edafccee | 9180 | } |
0a96bbe4 | 9181 | |
bd54b6fe | 9182 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 9183 | |
bd54b6fe BM |
9184 | ctx->buf_data = data; |
9185 | if (ret) | |
9186 | __io_sqe_buffers_unregister(ctx); | |
9187 | else | |
9188 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
9189 | return ret; |
9190 | } | |
9191 | ||
634d00df PB |
9192 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
9193 | struct io_uring_rsrc_update2 *up, | |
9194 | unsigned int nr_args) | |
9195 | { | |
9196 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
9197 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
9198 | struct page *last_hpage = NULL; |
9199 | bool needs_switch = false; | |
9200 | __u32 done; | |
9201 | int i, err; | |
9202 | ||
9203 | if (!ctx->buf_data) | |
9204 | return -ENXIO; | |
9205 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
9206 | return -EINVAL; | |
9207 | ||
9208 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
9209 | struct io_mapped_ubuf *imu; |
9210 | int offset = up->offset + done; | |
634d00df PB |
9211 | u64 tag = 0; |
9212 | ||
9213 | err = io_copy_iov(ctx, &iov, iovs, done); | |
9214 | if (err) | |
9215 | break; | |
9216 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
9217 | err = -EFAULT; | |
9218 | break; | |
9219 | } | |
0b8c0e7c PB |
9220 | err = io_buffer_validate(&iov); |
9221 | if (err) | |
9222 | break; | |
cf3770e7 CIK |
9223 | if (!iov.iov_base && tag) { |
9224 | err = -EINVAL; | |
9225 | break; | |
9226 | } | |
0b8c0e7c PB |
9227 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
9228 | if (err) | |
9229 | break; | |
634d00df | 9230 | |
0b8c0e7c | 9231 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 9232 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
22a99dcb | 9233 | err = io_queue_rsrc_removal(ctx->buf_data, i, |
0b8c0e7c PB |
9234 | ctx->rsrc_node, ctx->user_bufs[i]); |
9235 | if (unlikely(err)) { | |
9236 | io_buffer_unmap(ctx, &imu); | |
634d00df | 9237 | break; |
0b8c0e7c | 9238 | } |
634d00df PB |
9239 | ctx->user_bufs[i] = NULL; |
9240 | needs_switch = true; | |
9241 | } | |
9242 | ||
0b8c0e7c | 9243 | ctx->user_bufs[i] = imu; |
2d091d62 | 9244 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
9245 | } |
9246 | ||
9247 | if (needs_switch) | |
9248 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
9249 | return done ? done : err; | |
9250 | } | |
9251 | ||
9b402849 JA |
9252 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
9253 | { | |
9254 | __s32 __user *fds = arg; | |
9255 | int fd; | |
9256 | ||
9257 | if (ctx->cq_ev_fd) | |
9258 | return -EBUSY; | |
9259 | ||
9260 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
9261 | return -EFAULT; | |
9262 | ||
9263 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
9264 | if (IS_ERR(ctx->cq_ev_fd)) { | |
9265 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
fe7e3257 | 9266 | |
9b402849 JA |
9267 | ctx->cq_ev_fd = NULL; |
9268 | return ret; | |
9269 | } | |
9270 | ||
9271 | return 0; | |
9272 | } | |
9273 | ||
9274 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
9275 | { | |
9276 | if (ctx->cq_ev_fd) { | |
9277 | eventfd_ctx_put(ctx->cq_ev_fd); | |
9278 | ctx->cq_ev_fd = NULL; | |
9279 | return 0; | |
9280 | } | |
9281 | ||
9282 | return -ENXIO; | |
9283 | } | |
9284 | ||
5a2e745d JA |
9285 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
9286 | { | |
9e15c3a0 JA |
9287 | struct io_buffer *buf; |
9288 | unsigned long index; | |
9289 | ||
07edfd19 | 9290 | xa_for_each(&ctx->io_buffers, index, buf) |
9e15c3a0 | 9291 | __io_remove_buffers(ctx, buf, index, -1U); |
5a2e745d JA |
9292 | } |
9293 | ||
7255834e | 9294 | static void io_req_cache_free(struct list_head *list) |
1b4c351f | 9295 | { |
68e68ee6 | 9296 | struct io_kiocb *req, *nxt; |
1b4c351f | 9297 | |
bb943b82 PB |
9298 | list_for_each_entry_safe(req, nxt, list, inflight_entry) { |
9299 | list_del(&req->inflight_entry); | |
1b4c351f JA |
9300 | kmem_cache_free(req_cachep, req); |
9301 | } | |
9302 | } | |
9303 | ||
4010fec4 | 9304 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 9305 | { |
cd0ca2e0 | 9306 | struct io_submit_state *state = &ctx->submit_state; |
bf019da7 | 9307 | |
9a4fdbd8 JA |
9308 | mutex_lock(&ctx->uring_lock); |
9309 | ||
cd0ca2e0 PB |
9310 | if (state->free_reqs) { |
9311 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); | |
9312 | state->free_reqs = 0; | |
8e5c66c4 | 9313 | } |
9a4fdbd8 | 9314 | |
cd0ca2e0 PB |
9315 | io_flush_cached_locked_reqs(ctx, state); |
9316 | io_req_cache_free(&state->free_list); | |
9a4fdbd8 JA |
9317 | mutex_unlock(&ctx->uring_lock); |
9318 | } | |
9319 | ||
43597aac | 9320 | static void io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 9321 | { |
43597aac | 9322 | if (data && !atomic_dec_and_test(&data->refs)) |
bd54b6fe | 9323 | wait_for_completion(&data->done); |
bd54b6fe | 9324 | } |
04fc6c80 | 9325 | |
2b188cc1 JA |
9326 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
9327 | { | |
37d1e2e3 | 9328 | io_sq_thread_finish(ctx); |
2aede0e4 | 9329 | |
37d1e2e3 | 9330 | if (ctx->mm_account) { |
2aede0e4 JA |
9331 | mmdrop(ctx->mm_account); |
9332 | ctx->mm_account = NULL; | |
30975825 | 9333 | } |
def596e9 | 9334 | |
43597aac PB |
9335 | /* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */ |
9336 | io_wait_rsrc_data(ctx->buf_data); | |
9337 | io_wait_rsrc_data(ctx->file_data); | |
9338 | ||
8bad28d8 | 9339 | mutex_lock(&ctx->uring_lock); |
43597aac | 9340 | if (ctx->buf_data) |
bd54b6fe | 9341 | __io_sqe_buffers_unregister(ctx); |
43597aac | 9342 | if (ctx->file_data) |
08480400 | 9343 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
9344 | if (ctx->rings) |
9345 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 9346 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 9347 | io_eventfd_unregister(ctx); |
5a2e745d | 9348 | io_destroy_buffers(ctx); |
07db298a PB |
9349 | if (ctx->sq_creds) |
9350 | put_cred(ctx->sq_creds); | |
def596e9 | 9351 | |
a7f0ed5a PB |
9352 | /* there are no registered resources left, nobody uses it */ |
9353 | if (ctx->rsrc_node) | |
9354 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 9355 | if (ctx->rsrc_backup_node) |
b895c9a6 | 9356 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
9357 | flush_delayed_work(&ctx->rsrc_put_work); |
9358 | ||
9359 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
9360 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 9361 | |
2b188cc1 | 9362 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
9363 | if (ctx->ring_sock) { |
9364 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 9365 | sock_release(ctx->ring_sock); |
355e8d26 | 9366 | } |
2b188cc1 | 9367 | #endif |
ef9dd637 | 9368 | WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list)); |
2b188cc1 | 9369 | |
75b28aff | 9370 | io_mem_free(ctx->rings); |
2b188cc1 | 9371 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
9372 | |
9373 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 9374 | free_uid(ctx->user); |
4010fec4 | 9375 | io_req_caches_free(ctx); |
e941894e JA |
9376 | if (ctx->hash_map) |
9377 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 9378 | kfree(ctx->cancel_hash); |
6224843d | 9379 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
9380 | kfree(ctx); |
9381 | } | |
9382 | ||
9383 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
9384 | { | |
9385 | struct io_ring_ctx *ctx = file->private_data; | |
9386 | __poll_t mask = 0; | |
9387 | ||
311997b3 | 9388 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
9389 | /* |
9390 | * synchronizes with barrier from wq_has_sleeper call in | |
9391 | * io_commit_cqring | |
9392 | */ | |
2b188cc1 | 9393 | smp_rmb(); |
90554200 | 9394 | if (!io_sqring_full(ctx)) |
2b188cc1 | 9395 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
9396 | |
9397 | /* | |
9398 | * Don't flush cqring overflow list here, just do a simple check. | |
9399 | * Otherwise there could possible be ABBA deadlock: | |
9400 | * CPU0 CPU1 | |
9401 | * ---- ---- | |
9402 | * lock(&ctx->uring_lock); | |
9403 | * lock(&ep->mtx); | |
9404 | * lock(&ctx->uring_lock); | |
9405 | * lock(&ep->mtx); | |
9406 | * | |
9407 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
9408 | * pushs them to do the flush. | |
9409 | */ | |
5ed7a37d | 9410 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
9411 | mask |= EPOLLIN | EPOLLRDNORM; |
9412 | ||
9413 | return mask; | |
9414 | } | |
9415 | ||
0bead8cd | 9416 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 9417 | { |
4379bf8b | 9418 | const struct cred *creds; |
071698e1 | 9419 | |
61cf9370 | 9420 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
9421 | if (creds) { |
9422 | put_cred(creds); | |
0bead8cd | 9423 | return 0; |
1e6fa521 | 9424 | } |
0bead8cd YD |
9425 | |
9426 | return -EINVAL; | |
9427 | } | |
9428 | ||
d56d938b PB |
9429 | struct io_tctx_exit { |
9430 | struct callback_head task_work; | |
9431 | struct completion completion; | |
baf186c4 | 9432 | struct io_ring_ctx *ctx; |
d56d938b PB |
9433 | }; |
9434 | ||
9435 | static void io_tctx_exit_cb(struct callback_head *cb) | |
9436 | { | |
9437 | struct io_uring_task *tctx = current->io_uring; | |
9438 | struct io_tctx_exit *work; | |
9439 | ||
9440 | work = container_of(cb, struct io_tctx_exit, task_work); | |
9441 | /* | |
9442 | * When @in_idle, we're in cancellation and it's racy to remove the | |
9443 | * node. It'll be removed by the end of cancellation, just ignore it. | |
9444 | */ | |
9445 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 9446 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
9447 | complete(&work->completion); |
9448 | } | |
9449 | ||
28090c13 PB |
9450 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
9451 | { | |
9452 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
9453 | ||
9454 | return req->ctx == data; | |
9455 | } | |
9456 | ||
85faa7b8 JA |
9457 | static void io_ring_exit_work(struct work_struct *work) |
9458 | { | |
d56d938b | 9459 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 9460 | unsigned long timeout = jiffies + HZ * 60 * 5; |
58d3be2c | 9461 | unsigned long interval = HZ / 20; |
d56d938b PB |
9462 | struct io_tctx_exit exit; |
9463 | struct io_tctx_node *node; | |
9464 | int ret; | |
85faa7b8 | 9465 | |
56952e91 JA |
9466 | /* |
9467 | * If we're doing polled IO and end up having requests being | |
9468 | * submitted async (out-of-line), then completions can come in while | |
9469 | * we're waiting for refs to drop. We need to reap these manually, | |
9470 | * as nobody else will be looking for them. | |
9471 | */ | |
b2edc0a7 | 9472 | do { |
3dd0c97a | 9473 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
9474 | if (ctx->sq_data) { |
9475 | struct io_sq_data *sqd = ctx->sq_data; | |
9476 | struct task_struct *tsk; | |
9477 | ||
9478 | io_sq_thread_park(sqd); | |
9479 | tsk = sqd->thread; | |
9480 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
9481 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
9482 | io_cancel_ctx_cb, ctx, true); | |
9483 | io_sq_thread_unpark(sqd); | |
9484 | } | |
b5bb3a24 | 9485 | |
58d3be2c PB |
9486 | if (WARN_ON_ONCE(time_after(jiffies, timeout))) { |
9487 | /* there is little hope left, don't run it too often */ | |
9488 | interval = HZ * 60; | |
9489 | } | |
9490 | } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); | |
d56d938b | 9491 | |
7f00651a PB |
9492 | init_completion(&exit.completion); |
9493 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
9494 | exit.ctx = ctx; | |
89b5066e PB |
9495 | /* |
9496 | * Some may use context even when all refs and requests have been put, | |
9497 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 9498 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
9499 | * this lock/unlock section also waits them to finish. |
9500 | */ | |
d56d938b PB |
9501 | mutex_lock(&ctx->uring_lock); |
9502 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
9503 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
9504 | ||
d56d938b PB |
9505 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
9506 | ctx_node); | |
7f00651a PB |
9507 | /* don't spin on a single task if cancellation failed */ |
9508 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
9509 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
9510 | if (WARN_ON_ONCE(ret)) | |
9511 | continue; | |
9512 | wake_up_process(node->task); | |
9513 | ||
9514 | mutex_unlock(&ctx->uring_lock); | |
9515 | wait_for_completion(&exit.completion); | |
d56d938b PB |
9516 | mutex_lock(&ctx->uring_lock); |
9517 | } | |
9518 | mutex_unlock(&ctx->uring_lock); | |
79ebeaee JA |
9519 | spin_lock(&ctx->completion_lock); |
9520 | spin_unlock(&ctx->completion_lock); | |
d56d938b | 9521 | |
85faa7b8 JA |
9522 | io_ring_ctx_free(ctx); |
9523 | } | |
9524 | ||
80c4cbdb PB |
9525 | /* Returns true if we found and killed one or more timeouts */ |
9526 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 9527 | bool cancel_all) |
80c4cbdb PB |
9528 | { |
9529 | struct io_kiocb *req, *tmp; | |
9530 | int canceled = 0; | |
9531 | ||
79ebeaee JA |
9532 | spin_lock(&ctx->completion_lock); |
9533 | spin_lock_irq(&ctx->timeout_lock); | |
80c4cbdb | 9534 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
3dd0c97a | 9535 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
9536 | io_kill_timeout(req, -ECANCELED); |
9537 | canceled++; | |
9538 | } | |
9539 | } | |
79ebeaee | 9540 | spin_unlock_irq(&ctx->timeout_lock); |
51520426 PB |
9541 | if (canceled != 0) |
9542 | io_commit_cqring(ctx); | |
79ebeaee | 9543 | spin_unlock(&ctx->completion_lock); |
80c4cbdb PB |
9544 | if (canceled != 0) |
9545 | io_cqring_ev_posted(ctx); | |
9546 | return canceled != 0; | |
9547 | } | |
9548 | ||
2b188cc1 JA |
9549 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
9550 | { | |
61cf9370 MWO |
9551 | unsigned long index; |
9552 | struct creds *creds; | |
9553 | ||
2b188cc1 JA |
9554 | mutex_lock(&ctx->uring_lock); |
9555 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 9556 | if (ctx->rings) |
6c2450ae | 9557 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
9558 | xa_for_each(&ctx->personalities, index, creds) |
9559 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
9560 | mutex_unlock(&ctx->uring_lock); |
9561 | ||
3dd0c97a PB |
9562 | io_kill_timeouts(ctx, NULL, true); |
9563 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 9564 | |
15dff286 | 9565 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 9566 | io_iopoll_try_reap_events(ctx); |
309fc03a | 9567 | |
85faa7b8 | 9568 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
9569 | /* |
9570 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
9571 | * if we're exiting a ton of rings at the same time. It just adds | |
9572 | * noise and overhead, there's no discernable change in runtime | |
9573 | * over using system_wq. | |
9574 | */ | |
9575 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
9576 | } |
9577 | ||
9578 | static int io_uring_release(struct inode *inode, struct file *file) | |
9579 | { | |
9580 | struct io_ring_ctx *ctx = file->private_data; | |
9581 | ||
9582 | file->private_data = NULL; | |
9583 | io_ring_ctx_wait_and_kill(ctx); | |
9584 | return 0; | |
9585 | } | |
9586 | ||
f6edbabb PB |
9587 | struct io_task_cancel { |
9588 | struct task_struct *task; | |
3dd0c97a | 9589 | bool all; |
f6edbabb | 9590 | }; |
f254ac04 | 9591 | |
f6edbabb | 9592 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 9593 | { |
9a472ef7 | 9594 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 9595 | struct io_task_cancel *cancel = data; |
9a472ef7 | 9596 | |
f0baed8e | 9597 | return io_match_task_safe(req, cancel->task, cancel->all); |
b711d4ea JA |
9598 | } |
9599 | ||
e1915f76 | 9600 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 9601 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 9602 | { |
e1915f76 | 9603 | struct io_defer_entry *de; |
b7ddce3c PB |
9604 | LIST_HEAD(list); |
9605 | ||
79ebeaee | 9606 | spin_lock(&ctx->completion_lock); |
b7ddce3c | 9607 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { |
f0baed8e | 9608 | if (io_match_task_safe(de->req, task, cancel_all)) { |
b7ddce3c PB |
9609 | list_cut_position(&list, &ctx->defer_list, &de->list); |
9610 | break; | |
9611 | } | |
9612 | } | |
79ebeaee | 9613 | spin_unlock(&ctx->completion_lock); |
e1915f76 PB |
9614 | if (list_empty(&list)) |
9615 | return false; | |
b7ddce3c PB |
9616 | |
9617 | while (!list_empty(&list)) { | |
9618 | de = list_first_entry(&list, struct io_defer_entry, list); | |
9619 | list_del_init(&de->list); | |
f41db273 | 9620 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
9621 | kfree(de); |
9622 | } | |
e1915f76 | 9623 | return true; |
b7ddce3c PB |
9624 | } |
9625 | ||
1b00764f PB |
9626 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
9627 | { | |
9628 | struct io_tctx_node *node; | |
9629 | enum io_wq_cancel cret; | |
9630 | bool ret = false; | |
9631 | ||
9632 | mutex_lock(&ctx->uring_lock); | |
9633 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
9634 | struct io_uring_task *tctx = node->task->io_uring; | |
9635 | ||
9636 | /* | |
9637 | * io_wq will stay alive while we hold uring_lock, because it's | |
9638 | * killed after ctx nodes, which requires to take the lock. | |
9639 | */ | |
9640 | if (!tctx || !tctx->io_wq) | |
9641 | continue; | |
9642 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
9643 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9644 | } | |
9645 | mutex_unlock(&ctx->uring_lock); | |
9646 | ||
9647 | return ret; | |
9648 | } | |
9649 | ||
9936c7c2 PB |
9650 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
9651 | struct task_struct *task, | |
3dd0c97a | 9652 | bool cancel_all) |
9936c7c2 | 9653 | { |
3dd0c97a | 9654 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 9655 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
9656 | |
9657 | while (1) { | |
9658 | enum io_wq_cancel cret; | |
9659 | bool ret = false; | |
9660 | ||
1b00764f PB |
9661 | if (!task) { |
9662 | ret |= io_uring_try_cancel_iowq(ctx); | |
9663 | } else if (tctx && tctx->io_wq) { | |
9664 | /* | |
9665 | * Cancels requests of all rings, not only @ctx, but | |
9666 | * it's fine as the task is in exit/exec. | |
9667 | */ | |
5aa75ed5 | 9668 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
9669 | &cancel, true); |
9670 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9671 | } | |
9672 | ||
9673 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 9674 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 9675 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
9676 | while (!list_empty_careful(&ctx->iopoll_list)) { |
9677 | io_iopoll_try_reap_events(ctx); | |
9678 | ret = true; | |
9679 | } | |
9680 | } | |
9681 | ||
3dd0c97a PB |
9682 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
9683 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
9684 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
9685 | if (task) |
9686 | ret |= io_run_task_work(); | |
9936c7c2 PB |
9687 | if (!ret) |
9688 | break; | |
9689 | cond_resched(); | |
9690 | } | |
9691 | } | |
9692 | ||
eef51daa | 9693 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 9694 | { |
236434c3 | 9695 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 9696 | struct io_tctx_node *node; |
a528b04e | 9697 | int ret; |
236434c3 MWO |
9698 | |
9699 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9700 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9701 | if (unlikely(ret)) |
9702 | return ret; | |
e139a1ec | 9703 | |
236434c3 | 9704 | tctx = current->io_uring; |
e139a1ec PB |
9705 | if (ctx->iowq_limits_set) { |
9706 | unsigned int limits[2] = { ctx->iowq_limits[0], | |
9707 | ctx->iowq_limits[1], }; | |
9708 | ||
9709 | ret = io_wq_max_workers(tctx->io_wq, limits); | |
9710 | if (ret) | |
9711 | return ret; | |
9712 | } | |
0f212204 | 9713 | } |
cf27f3b1 PB |
9714 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9715 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9716 | if (!node) | |
9717 | return -ENOMEM; | |
9718 | node->ctx = ctx; | |
9719 | node->task = current; | |
13bf43f5 | 9720 | |
cf27f3b1 PB |
9721 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9722 | node, GFP_KERNEL)); | |
9723 | if (ret) { | |
9724 | kfree(node); | |
9725 | return ret; | |
0f212204 | 9726 | } |
cf27f3b1 PB |
9727 | |
9728 | mutex_lock(&ctx->uring_lock); | |
9729 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9730 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9731 | } |
cf27f3b1 | 9732 | tctx->last = ctx; |
0f212204 JA |
9733 | return 0; |
9734 | } | |
9735 | ||
cf27f3b1 PB |
9736 | /* |
9737 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9738 | */ | |
eef51daa | 9739 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9740 | { |
9741 | struct io_uring_task *tctx = current->io_uring; | |
9742 | ||
9743 | if (likely(tctx && tctx->last == ctx)) | |
9744 | return 0; | |
eef51daa | 9745 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9746 | } |
9747 | ||
0f212204 JA |
9748 | /* |
9749 | * Remove this io_uring_file -> task mapping. | |
9750 | */ | |
eef51daa | 9751 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9752 | { |
9753 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9754 | struct io_tctx_node *node; |
2941267b | 9755 | |
eebd2e37 PB |
9756 | if (!tctx) |
9757 | return; | |
13bf43f5 PB |
9758 | node = xa_erase(&tctx->xa, index); |
9759 | if (!node) | |
2941267b | 9760 | return; |
0f212204 | 9761 | |
13bf43f5 PB |
9762 | WARN_ON_ONCE(current != node->task); |
9763 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9764 | ||
9765 | mutex_lock(&node->ctx->uring_lock); | |
9766 | list_del(&node->ctx_node); | |
9767 | mutex_unlock(&node->ctx->uring_lock); | |
9768 | ||
baf186c4 | 9769 | if (tctx->last == node->ctx) |
0f212204 | 9770 | tctx->last = NULL; |
13bf43f5 | 9771 | kfree(node); |
0f212204 JA |
9772 | } |
9773 | ||
8452d4a6 | 9774 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9775 | { |
ba5ef6dc | 9776 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9777 | struct io_tctx_node *node; |
de7f1d9e PB |
9778 | unsigned long index; |
9779 | ||
8bab4c09 | 9780 | xa_for_each(&tctx->xa, index, node) { |
eef51daa | 9781 | io_uring_del_tctx_node(index); |
8bab4c09 JA |
9782 | cond_resched(); |
9783 | } | |
b16ef427 ME |
9784 | if (wq) { |
9785 | /* | |
9786 | * Must be after io_uring_del_task_file() (removes nodes under | |
9787 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9788 | */ | |
ba5ef6dc | 9789 | io_wq_put_and_exit(wq); |
dadebc35 | 9790 | tctx->io_wq = NULL; |
b16ef427 | 9791 | } |
de7f1d9e PB |
9792 | } |
9793 | ||
3f48cf18 | 9794 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9795 | { |
3f48cf18 PB |
9796 | if (tracked) |
9797 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9798 | return percpu_counter_sum(&tctx->inflight); |
9799 | } | |
9800 | ||
78cc687b PB |
9801 | /* |
9802 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
b16590fd | 9803 | * requests. @sqd should be not-null IFF it's an SQPOLL thread cancellation. |
78cc687b PB |
9804 | */ |
9805 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9806 | { |
521d6a73 | 9807 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9808 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9809 | s64 inflight; |
9810 | DEFINE_WAIT(wait); | |
fdaf083c | 9811 | |
78cc687b PB |
9812 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9813 | ||
6d042ffb PO |
9814 | if (!current->io_uring) |
9815 | return; | |
17a91051 PB |
9816 | if (tctx->io_wq) |
9817 | io_wq_exit_start(tctx->io_wq); | |
9818 | ||
0e9ddb39 PB |
9819 | atomic_inc(&tctx->in_idle); |
9820 | do { | |
e9dbe221 | 9821 | io_uring_drop_tctx_refs(current); |
0e9ddb39 | 9822 | /* read completions before cancelations */ |
78cc687b | 9823 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9824 | if (!inflight) |
9825 | break; | |
fdaf083c | 9826 | |
78cc687b PB |
9827 | if (!sqd) { |
9828 | struct io_tctx_node *node; | |
9829 | unsigned long index; | |
0f212204 | 9830 | |
78cc687b PB |
9831 | xa_for_each(&tctx->xa, index, node) { |
9832 | /* sqpoll task will cancel all its requests */ | |
9833 | if (node->ctx->sq_data) | |
9834 | continue; | |
9835 | io_uring_try_cancel_requests(node->ctx, current, | |
9836 | cancel_all); | |
9837 | } | |
9838 | } else { | |
9839 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9840 | io_uring_try_cancel_requests(ctx, current, | |
9841 | cancel_all); | |
9842 | } | |
17a91051 | 9843 | |
b16590fd JA |
9844 | prepare_to_wait(&tctx->wait, &wait, TASK_INTERRUPTIBLE); |
9845 | io_run_task_work(); | |
e9dbe221 | 9846 | io_uring_drop_tctx_refs(current); |
b16590fd | 9847 | |
0f212204 | 9848 | /* |
a1bb3cd5 PB |
9849 | * If we've seen completions, retry without waiting. This |
9850 | * avoids a race where a completion comes in before we did | |
9851 | * prepare_to_wait(). | |
0f212204 | 9852 | */ |
3dd0c97a | 9853 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9854 | schedule(); |
f57555ed | 9855 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9856 | } while (1); |
de7f1d9e | 9857 | |
8452d4a6 | 9858 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9859 | if (cancel_all) { |
abdebba9 PB |
9860 | /* |
9861 | * We shouldn't run task_works after cancel, so just leave | |
9862 | * ->in_idle set for normal exit. | |
9863 | */ | |
9864 | atomic_dec(&tctx->in_idle); | |
3f48cf18 PB |
9865 | /* for exec all current's requests should be gone, kill tctx */ |
9866 | __io_uring_free(current); | |
9867 | } | |
44e728b8 PB |
9868 | } |
9869 | ||
f552a27a | 9870 | void __io_uring_cancel(bool cancel_all) |
78cc687b | 9871 | { |
f552a27a | 9872 | io_uring_cancel_generic(cancel_all, NULL); |
78cc687b PB |
9873 | } |
9874 | ||
6c5c240e RP |
9875 | static void *io_uring_validate_mmap_request(struct file *file, |
9876 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9877 | { |
2b188cc1 | 9878 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9879 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9880 | struct page *page; |
9881 | void *ptr; | |
9882 | ||
9883 | switch (offset) { | |
9884 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9885 | case IORING_OFF_CQ_RING: |
9886 | ptr = ctx->rings; | |
2b188cc1 JA |
9887 | break; |
9888 | case IORING_OFF_SQES: | |
9889 | ptr = ctx->sq_sqes; | |
9890 | break; | |
2b188cc1 | 9891 | default: |
6c5c240e | 9892 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9893 | } |
9894 | ||
9895 | page = virt_to_head_page(ptr); | |
a50b854e | 9896 | if (sz > page_size(page)) |
6c5c240e RP |
9897 | return ERR_PTR(-EINVAL); |
9898 | ||
9899 | return ptr; | |
9900 | } | |
9901 | ||
9902 | #ifdef CONFIG_MMU | |
9903 | ||
9904 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9905 | { | |
9906 | size_t sz = vma->vm_end - vma->vm_start; | |
9907 | unsigned long pfn; | |
9908 | void *ptr; | |
9909 | ||
9910 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9911 | if (IS_ERR(ptr)) | |
9912 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9913 | |
9914 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9915 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9916 | } | |
9917 | ||
6c5c240e RP |
9918 | #else /* !CONFIG_MMU */ |
9919 | ||
9920 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9921 | { | |
9922 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9923 | } | |
9924 | ||
9925 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9926 | { | |
9927 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9928 | } | |
9929 | ||
9930 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9931 | unsigned long addr, unsigned long len, | |
9932 | unsigned long pgoff, unsigned long flags) | |
9933 | { | |
9934 | void *ptr; | |
9935 | ||
9936 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9937 | if (IS_ERR(ptr)) | |
9938 | return PTR_ERR(ptr); | |
9939 | ||
9940 | return (unsigned long) ptr; | |
9941 | } | |
9942 | ||
9943 | #endif /* !CONFIG_MMU */ | |
9944 | ||
d9d05217 | 9945 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9946 | { |
9947 | DEFINE_WAIT(wait); | |
9948 | ||
9949 | do { | |
9950 | if (!io_sqring_full(ctx)) | |
9951 | break; | |
90554200 JA |
9952 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9953 | ||
9954 | if (!io_sqring_full(ctx)) | |
9955 | break; | |
90554200 JA |
9956 | schedule(); |
9957 | } while (!signal_pending(current)); | |
9958 | ||
9959 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9960 | return 0; |
90554200 JA |
9961 | } |
9962 | ||
c73ebb68 HX |
9963 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9964 | struct __kernel_timespec __user **ts, | |
9965 | const sigset_t __user **sig) | |
9966 | { | |
9967 | struct io_uring_getevents_arg arg; | |
9968 | ||
9969 | /* | |
9970 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9971 | * is just a pointer to the sigset_t. | |
9972 | */ | |
9973 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9974 | *sig = (const sigset_t __user *) argp; | |
9975 | *ts = NULL; | |
9976 | return 0; | |
9977 | } | |
9978 | ||
9979 | /* | |
9980 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9981 | * timespec and sigset_t pointers if good. | |
9982 | */ | |
9983 | if (*argsz != sizeof(arg)) | |
9984 | return -EINVAL; | |
9985 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9986 | return -EFAULT; | |
02a84aaf DY |
9987 | if (arg.pad) |
9988 | return -EINVAL; | |
c73ebb68 HX |
9989 | *sig = u64_to_user_ptr(arg.sigmask); |
9990 | *argsz = arg.sigmask_sz; | |
9991 | *ts = u64_to_user_ptr(arg.ts); | |
9992 | return 0; | |
9993 | } | |
9994 | ||
2b188cc1 | 9995 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9996 | u32, min_complete, u32, flags, const void __user *, argp, |
9997 | size_t, argsz) | |
2b188cc1 JA |
9998 | { |
9999 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
10000 | int submitted = 0; |
10001 | struct fd f; | |
33f993da | 10002 | long ret; |
2b188cc1 | 10003 | |
4c6e277c | 10004 | io_run_task_work(); |
b41e9852 | 10005 | |
33f993da PB |
10006 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
10007 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
10008 | return -EINVAL; |
10009 | ||
10010 | f = fdget(fd); | |
33f993da | 10011 | if (unlikely(!f.file)) |
2b188cc1 JA |
10012 | return -EBADF; |
10013 | ||
10014 | ret = -EOPNOTSUPP; | |
33f993da | 10015 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
10016 | goto out_fput; |
10017 | ||
10018 | ret = -ENXIO; | |
10019 | ctx = f.file->private_data; | |
33f993da | 10020 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
10021 | goto out_fput; |
10022 | ||
7e84e1c7 | 10023 | ret = -EBADFD; |
33f993da | 10024 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
10025 | goto out; |
10026 | ||
6c271ce2 JA |
10027 | /* |
10028 | * For SQ polling, the thread will do all submissions and completions. | |
10029 | * Just return the requested submit count, and wake the thread if | |
10030 | * we were asked to. | |
10031 | */ | |
b2a9eada | 10032 | ret = 0; |
6c271ce2 | 10033 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
90f67366 | 10034 | io_cqring_overflow_flush(ctx); |
89448c47 | 10035 | |
21f96522 JA |
10036 | if (unlikely(ctx->sq_data->thread == NULL)) { |
10037 | ret = -EOWNERDEAD; | |
04147488 | 10038 | goto out; |
21f96522 | 10039 | } |
6c271ce2 | 10040 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 10041 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
10042 | if (flags & IORING_ENTER_SQ_WAIT) { |
10043 | ret = io_sqpoll_wait_sq(ctx); | |
10044 | if (ret) | |
10045 | goto out; | |
10046 | } | |
6c271ce2 | 10047 | submitted = to_submit; |
b2a9eada | 10048 | } else if (to_submit) { |
eef51daa | 10049 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
10050 | if (unlikely(ret)) |
10051 | goto out; | |
2b188cc1 | 10052 | mutex_lock(&ctx->uring_lock); |
0f212204 | 10053 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 10054 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
10055 | |
10056 | if (submitted != to_submit) | |
10057 | goto out; | |
2b188cc1 JA |
10058 | } |
10059 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
10060 | const sigset_t __user *sig; |
10061 | struct __kernel_timespec __user *ts; | |
10062 | ||
10063 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
10064 | if (unlikely(ret)) | |
10065 | goto out; | |
10066 | ||
2b188cc1 JA |
10067 | min_complete = min(min_complete, ctx->cq_entries); |
10068 | ||
32b2244a XW |
10069 | /* |
10070 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
10071 | * space applications don't need to do io completion events | |
10072 | * polling again, they can rely on io_sq_thread to do polling | |
10073 | * work, which can reduce cpu usage and uring_lock contention. | |
10074 | */ | |
10075 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
10076 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 10077 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 10078 | } else { |
c73ebb68 | 10079 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 10080 | } |
2b188cc1 JA |
10081 | } |
10082 | ||
7c504e65 | 10083 | out: |
6805b32e | 10084 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
10085 | out_fput: |
10086 | fdput(f); | |
10087 | return submitted ? submitted : ret; | |
10088 | } | |
10089 | ||
bebdb65e | 10090 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
10091 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
10092 | const struct cred *cred) | |
87ce955b | 10093 | { |
87ce955b JA |
10094 | struct user_namespace *uns = seq_user_ns(m); |
10095 | struct group_info *gi; | |
10096 | kernel_cap_t cap; | |
10097 | unsigned __capi; | |
10098 | int g; | |
10099 | ||
10100 | seq_printf(m, "%5d\n", id); | |
10101 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
10102 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
10103 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
10104 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
10105 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
10106 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
10107 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
10108 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
10109 | seq_puts(m, "\n\tGroups:\t"); | |
10110 | gi = cred->group_info; | |
10111 | for (g = 0; g < gi->ngroups; g++) { | |
10112 | seq_put_decimal_ull(m, g ? " " : "", | |
10113 | from_kgid_munged(uns, gi->gid[g])); | |
10114 | } | |
10115 | seq_puts(m, "\n\tCapEff:\t"); | |
10116 | cap = cred->cap_effective; | |
10117 | CAP_FOR_EACH_U32(__capi) | |
10118 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
10119 | seq_putc(m, '\n'); | |
10120 | return 0; | |
10121 | } | |
10122 | ||
10123 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
10124 | { | |
dbbe9c64 | 10125 | struct io_sq_data *sq = NULL; |
fad8e0de | 10126 | bool has_lock; |
87ce955b JA |
10127 | int i; |
10128 | ||
fad8e0de JA |
10129 | /* |
10130 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
10131 | * since fdinfo case grabs it in the opposite direction of normal use | |
10132 | * cases. If we fail to get the lock, we just don't iterate any | |
10133 | * structures that could be going away outside the io_uring mutex. | |
10134 | */ | |
10135 | has_lock = mutex_trylock(&ctx->uring_lock); | |
10136 | ||
5f3f26f9 | 10137 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 10138 | sq = ctx->sq_data; |
5f3f26f9 JA |
10139 | if (!sq->thread) |
10140 | sq = NULL; | |
10141 | } | |
dbbe9c64 JQ |
10142 | |
10143 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
10144 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 10145 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 10146 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 10147 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 10148 | |
87ce955b JA |
10149 | if (f) |
10150 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
10151 | else | |
10152 | seq_printf(m, "%5u: <none>\n", i); | |
10153 | } | |
10154 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 10155 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 10156 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 10157 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 10158 | |
4751f53d | 10159 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 10160 | } |
61cf9370 MWO |
10161 | if (has_lock && !xa_empty(&ctx->personalities)) { |
10162 | unsigned long index; | |
10163 | const struct cred *cred; | |
10164 | ||
87ce955b | 10165 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
10166 | xa_for_each(&ctx->personalities, index, cred) |
10167 | io_uring_show_cred(m, index, cred); | |
87ce955b | 10168 | } |
d7718a9d | 10169 | seq_printf(m, "PollList:\n"); |
79ebeaee | 10170 | spin_lock(&ctx->completion_lock); |
d7718a9d JA |
10171 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
10172 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
10173 | struct io_kiocb *req; | |
10174 | ||
10175 | hlist_for_each_entry(req, list, hash_node) | |
10176 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
10177 | req->task->task_works != NULL); | |
10178 | } | |
79ebeaee | 10179 | spin_unlock(&ctx->completion_lock); |
fad8e0de JA |
10180 | if (has_lock) |
10181 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
10182 | } |
10183 | ||
10184 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
10185 | { | |
10186 | struct io_ring_ctx *ctx = f->private_data; | |
10187 | ||
10188 | if (percpu_ref_tryget(&ctx->refs)) { | |
10189 | __io_uring_show_fdinfo(ctx, m); | |
10190 | percpu_ref_put(&ctx->refs); | |
10191 | } | |
10192 | } | |
bebdb65e | 10193 | #endif |
87ce955b | 10194 | |
2b188cc1 JA |
10195 | static const struct file_operations io_uring_fops = { |
10196 | .release = io_uring_release, | |
10197 | .mmap = io_uring_mmap, | |
6c5c240e RP |
10198 | #ifndef CONFIG_MMU |
10199 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
10200 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
10201 | #endif | |
2b188cc1 | 10202 | .poll = io_uring_poll, |
bebdb65e | 10203 | #ifdef CONFIG_PROC_FS |
87ce955b | 10204 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 10205 | #endif |
2b188cc1 JA |
10206 | }; |
10207 | ||
10208 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
10209 | struct io_uring_params *p) | |
10210 | { | |
75b28aff HV |
10211 | struct io_rings *rings; |
10212 | size_t size, sq_array_offset; | |
2b188cc1 | 10213 | |
bd740481 JA |
10214 | /* make sure these are sane, as we already accounted them */ |
10215 | ctx->sq_entries = p->sq_entries; | |
10216 | ctx->cq_entries = p->cq_entries; | |
10217 | ||
75b28aff HV |
10218 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
10219 | if (size == SIZE_MAX) | |
10220 | return -EOVERFLOW; | |
10221 | ||
10222 | rings = io_mem_alloc(size); | |
10223 | if (!rings) | |
2b188cc1 JA |
10224 | return -ENOMEM; |
10225 | ||
75b28aff HV |
10226 | ctx->rings = rings; |
10227 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
10228 | rings->sq_ring_mask = p->sq_entries - 1; | |
10229 | rings->cq_ring_mask = p->cq_entries - 1; | |
10230 | rings->sq_ring_entries = p->sq_entries; | |
10231 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
10232 | |
10233 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
10234 | if (size == SIZE_MAX) { |
10235 | io_mem_free(ctx->rings); | |
10236 | ctx->rings = NULL; | |
2b188cc1 | 10237 | return -EOVERFLOW; |
eb065d30 | 10238 | } |
2b188cc1 JA |
10239 | |
10240 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
10241 | if (!ctx->sq_sqes) { |
10242 | io_mem_free(ctx->rings); | |
10243 | ctx->rings = NULL; | |
2b188cc1 | 10244 | return -ENOMEM; |
eb065d30 | 10245 | } |
2b188cc1 | 10246 | |
2b188cc1 JA |
10247 | return 0; |
10248 | } | |
10249 | ||
9faadcc8 PB |
10250 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
10251 | { | |
10252 | int ret, fd; | |
10253 | ||
10254 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
10255 | if (fd < 0) | |
10256 | return fd; | |
10257 | ||
eef51daa | 10258 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
10259 | if (ret) { |
10260 | put_unused_fd(fd); | |
10261 | return ret; | |
10262 | } | |
10263 | fd_install(fd, file); | |
10264 | return fd; | |
10265 | } | |
10266 | ||
2b188cc1 JA |
10267 | /* |
10268 | * Allocate an anonymous fd, this is what constitutes the application | |
10269 | * visible backing of an io_uring instance. The application mmaps this | |
10270 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
10271 | * we have to tie this fd to a socket for file garbage collection purposes. | |
10272 | */ | |
9faadcc8 | 10273 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
10274 | { |
10275 | struct file *file; | |
9faadcc8 | 10276 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
10277 | int ret; |
10278 | ||
2b188cc1 JA |
10279 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
10280 | &ctx->ring_sock); | |
10281 | if (ret) | |
9faadcc8 | 10282 | return ERR_PTR(ret); |
2b188cc1 JA |
10283 | #endif |
10284 | ||
2b188cc1 JA |
10285 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
10286 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 10287 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
10288 | if (IS_ERR(file)) { |
10289 | sock_release(ctx->ring_sock); | |
10290 | ctx->ring_sock = NULL; | |
10291 | } else { | |
10292 | ctx->ring_sock->file = file; | |
0f212204 | 10293 | } |
2b188cc1 | 10294 | #endif |
9faadcc8 | 10295 | return file; |
2b188cc1 JA |
10296 | } |
10297 | ||
7f13657d XW |
10298 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
10299 | struct io_uring_params __user *params) | |
2b188cc1 | 10300 | { |
2b188cc1 | 10301 | struct io_ring_ctx *ctx; |
9faadcc8 | 10302 | struct file *file; |
2b188cc1 JA |
10303 | int ret; |
10304 | ||
8110c1a6 | 10305 | if (!entries) |
2b188cc1 | 10306 | return -EINVAL; |
8110c1a6 JA |
10307 | if (entries > IORING_MAX_ENTRIES) { |
10308 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
10309 | return -EINVAL; | |
10310 | entries = IORING_MAX_ENTRIES; | |
10311 | } | |
2b188cc1 JA |
10312 | |
10313 | /* | |
10314 | * Use twice as many entries for the CQ ring. It's possible for the | |
10315 | * application to drive a higher depth than the size of the SQ ring, | |
10316 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
10317 | * some flexibility in overcommitting a bit. If the application has |
10318 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
10319 | * of CQ ring entries manually. | |
2b188cc1 JA |
10320 | */ |
10321 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
10322 | if (p->flags & IORING_SETUP_CQSIZE) { |
10323 | /* | |
10324 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
10325 | * to a power-of-two, if it isn't already. We do NOT impose | |
10326 | * any cq vs sq ring sizing. | |
10327 | */ | |
eb2667b3 | 10328 | if (!p->cq_entries) |
33a107f0 | 10329 | return -EINVAL; |
8110c1a6 JA |
10330 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
10331 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
10332 | return -EINVAL; | |
10333 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
10334 | } | |
eb2667b3 JQ |
10335 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
10336 | if (p->cq_entries < p->sq_entries) | |
10337 | return -EINVAL; | |
33a107f0 JA |
10338 | } else { |
10339 | p->cq_entries = 2 * p->sq_entries; | |
10340 | } | |
2b188cc1 | 10341 | |
2b188cc1 | 10342 | ctx = io_ring_ctx_alloc(p); |
62e398be | 10343 | if (!ctx) |
2b188cc1 | 10344 | return -ENOMEM; |
2b188cc1 | 10345 | ctx->compat = in_compat_syscall(); |
62e398be JA |
10346 | if (!capable(CAP_IPC_LOCK)) |
10347 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
10348 | |
10349 | /* | |
10350 | * This is just grabbed for accounting purposes. When a process exits, | |
10351 | * the mm is exited and dropped before the files, hence we need to hang | |
10352 | * on to this mm purely for the purposes of being able to unaccount | |
10353 | * memory (locked/pinned vm). It's not used for anything else. | |
10354 | */ | |
6b7898eb | 10355 | mmgrab(current->mm); |
2aede0e4 | 10356 | ctx->mm_account = current->mm; |
6b7898eb | 10357 | |
2b188cc1 JA |
10358 | ret = io_allocate_scq_urings(ctx, p); |
10359 | if (ret) | |
10360 | goto err; | |
10361 | ||
7e84e1c7 | 10362 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
10363 | if (ret) |
10364 | goto err; | |
eae071c9 | 10365 | /* always set a rsrc node */ |
47b228ce PB |
10366 | ret = io_rsrc_node_switch_start(ctx); |
10367 | if (ret) | |
10368 | goto err; | |
eae071c9 | 10369 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 10370 | |
2b188cc1 | 10371 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
10372 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
10373 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
10374 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
10375 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
10376 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
10377 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
10378 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
10379 | |
10380 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
10381 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
10382 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
10383 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
10384 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
10385 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
10386 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 10387 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 10388 | |
7f13657d XW |
10389 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
10390 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 10391 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 10392 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
10393 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
10394 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
10395 | |
10396 | if (copy_to_user(params, p, sizeof(*p))) { | |
10397 | ret = -EFAULT; | |
10398 | goto err; | |
10399 | } | |
d1719f70 | 10400 | |
9faadcc8 PB |
10401 | file = io_uring_get_file(ctx); |
10402 | if (IS_ERR(file)) { | |
10403 | ret = PTR_ERR(file); | |
10404 | goto err; | |
10405 | } | |
10406 | ||
044c1ab3 JA |
10407 | /* |
10408 | * Install ring fd as the very last thing, so we don't risk someone | |
10409 | * having closed it before we finish setup | |
10410 | */ | |
9faadcc8 PB |
10411 | ret = io_uring_install_fd(ctx, file); |
10412 | if (ret < 0) { | |
10413 | /* fput will clean it up */ | |
10414 | fput(file); | |
10415 | return ret; | |
10416 | } | |
044c1ab3 | 10417 | |
c826bd7a | 10418 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
10419 | return ret; |
10420 | err: | |
10421 | io_ring_ctx_wait_and_kill(ctx); | |
10422 | return ret; | |
10423 | } | |
10424 | ||
10425 | /* | |
10426 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
10427 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
10428 | * params structure passed in. | |
10429 | */ | |
10430 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
10431 | { | |
10432 | struct io_uring_params p; | |
2b188cc1 JA |
10433 | int i; |
10434 | ||
10435 | if (copy_from_user(&p, params, sizeof(p))) | |
10436 | return -EFAULT; | |
10437 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
10438 | if (p.resv[i]) | |
10439 | return -EINVAL; | |
10440 | } | |
10441 | ||
6c271ce2 | 10442 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 10443 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
10444 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
10445 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
10446 | return -EINVAL; |
10447 | ||
7f13657d | 10448 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
10449 | } |
10450 | ||
10451 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
10452 | struct io_uring_params __user *, params) | |
10453 | { | |
10454 | return io_uring_setup(entries, params); | |
10455 | } | |
10456 | ||
66f4af93 JA |
10457 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
10458 | { | |
10459 | struct io_uring_probe *p; | |
10460 | size_t size; | |
10461 | int i, ret; | |
10462 | ||
10463 | size = struct_size(p, ops, nr_args); | |
10464 | if (size == SIZE_MAX) | |
10465 | return -EOVERFLOW; | |
10466 | p = kzalloc(size, GFP_KERNEL); | |
10467 | if (!p) | |
10468 | return -ENOMEM; | |
10469 | ||
10470 | ret = -EFAULT; | |
10471 | if (copy_from_user(p, arg, size)) | |
10472 | goto out; | |
10473 | ret = -EINVAL; | |
10474 | if (memchr_inv(p, 0, size)) | |
10475 | goto out; | |
10476 | ||
10477 | p->last_op = IORING_OP_LAST - 1; | |
10478 | if (nr_args > IORING_OP_LAST) | |
10479 | nr_args = IORING_OP_LAST; | |
10480 | ||
10481 | for (i = 0; i < nr_args; i++) { | |
10482 | p->ops[i].op = i; | |
10483 | if (!io_op_defs[i].not_supported) | |
10484 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
10485 | } | |
10486 | p->ops_len = i; | |
10487 | ||
10488 | ret = 0; | |
10489 | if (copy_to_user(arg, p, size)) | |
10490 | ret = -EFAULT; | |
10491 | out: | |
10492 | kfree(p); | |
10493 | return ret; | |
10494 | } | |
10495 | ||
071698e1 JA |
10496 | static int io_register_personality(struct io_ring_ctx *ctx) |
10497 | { | |
4379bf8b | 10498 | const struct cred *creds; |
61cf9370 | 10499 | u32 id; |
1e6fa521 | 10500 | int ret; |
071698e1 | 10501 | |
4379bf8b | 10502 | creds = get_current_cred(); |
1e6fa521 | 10503 | |
61cf9370 MWO |
10504 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
10505 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
a30f895a JA |
10506 | if (ret < 0) { |
10507 | put_cred(creds); | |
10508 | return ret; | |
10509 | } | |
10510 | return id; | |
071698e1 JA |
10511 | } |
10512 | ||
21b55dbc SG |
10513 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
10514 | unsigned int nr_args) | |
10515 | { | |
10516 | struct io_uring_restriction *res; | |
10517 | size_t size; | |
10518 | int i, ret; | |
10519 | ||
7e84e1c7 SG |
10520 | /* Restrictions allowed only if rings started disabled */ |
10521 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10522 | return -EBADFD; | |
10523 | ||
21b55dbc | 10524 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 10525 | if (ctx->restrictions.registered) |
21b55dbc SG |
10526 | return -EBUSY; |
10527 | ||
10528 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
10529 | return -EINVAL; | |
10530 | ||
10531 | size = array_size(nr_args, sizeof(*res)); | |
10532 | if (size == SIZE_MAX) | |
10533 | return -EOVERFLOW; | |
10534 | ||
10535 | res = memdup_user(arg, size); | |
10536 | if (IS_ERR(res)) | |
10537 | return PTR_ERR(res); | |
10538 | ||
10539 | ret = 0; | |
10540 | ||
10541 | for (i = 0; i < nr_args; i++) { | |
10542 | switch (res[i].opcode) { | |
10543 | case IORING_RESTRICTION_REGISTER_OP: | |
10544 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
10545 | ret = -EINVAL; | |
10546 | goto out; | |
10547 | } | |
10548 | ||
10549 | __set_bit(res[i].register_op, | |
10550 | ctx->restrictions.register_op); | |
10551 | break; | |
10552 | case IORING_RESTRICTION_SQE_OP: | |
10553 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
10554 | ret = -EINVAL; | |
10555 | goto out; | |
10556 | } | |
10557 | ||
10558 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
10559 | break; | |
10560 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
10561 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
10562 | break; | |
10563 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
10564 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
10565 | break; | |
10566 | default: | |
10567 | ret = -EINVAL; | |
10568 | goto out; | |
10569 | } | |
10570 | } | |
10571 | ||
10572 | out: | |
10573 | /* Reset all restrictions if an error happened */ | |
10574 | if (ret != 0) | |
10575 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
10576 | else | |
7e84e1c7 | 10577 | ctx->restrictions.registered = true; |
21b55dbc SG |
10578 | |
10579 | kfree(res); | |
10580 | return ret; | |
10581 | } | |
10582 | ||
7e84e1c7 SG |
10583 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10584 | { | |
10585 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10586 | return -EBADFD; | |
10587 | ||
10588 | if (ctx->restrictions.registered) | |
10589 | ctx->restricted = 1; | |
10590 | ||
0298ef96 PB |
10591 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
10592 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
10593 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
10594 | return 0; |
10595 | } | |
10596 | ||
fdecb662 | 10597 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 10598 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
10599 | unsigned nr_args) |
10600 | { | |
10601 | __u32 tmp; | |
10602 | int err; | |
10603 | ||
10604 | if (check_add_overflow(up->offset, nr_args, &tmp)) | |
10605 | return -EOVERFLOW; | |
10606 | err = io_rsrc_node_switch_start(ctx); | |
10607 | if (err) | |
10608 | return err; | |
10609 | ||
fdecb662 PB |
10610 | switch (type) { |
10611 | case IORING_RSRC_FILE: | |
98f0b3b4 | 10612 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
10613 | case IORING_RSRC_BUFFER: |
10614 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
10615 | } |
10616 | return -EINVAL; | |
10617 | } | |
10618 | ||
c3bdad02 PB |
10619 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
10620 | unsigned nr_args) | |
98f0b3b4 | 10621 | { |
c3bdad02 | 10622 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
10623 | |
10624 | if (!nr_args) | |
10625 | return -EINVAL; | |
c3bdad02 PB |
10626 | memset(&up, 0, sizeof(up)); |
10627 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
10628 | return -EFAULT; | |
c9747fa5 | 10629 | if (up.resv || up.resv2) |
4dd8f4c2 | 10630 | return -EINVAL; |
c3bdad02 PB |
10631 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); |
10632 | } | |
10633 | ||
10634 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 10635 | unsigned size, unsigned type) |
c3bdad02 PB |
10636 | { |
10637 | struct io_uring_rsrc_update2 up; | |
10638 | ||
10639 | if (size != sizeof(up)) | |
10640 | return -EINVAL; | |
98f0b3b4 PB |
10641 | if (copy_from_user(&up, arg, sizeof(up))) |
10642 | return -EFAULT; | |
c9747fa5 | 10643 | if (!up.nr || up.resv || up.resv2) |
98f0b3b4 | 10644 | return -EINVAL; |
992da01a | 10645 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
10646 | } |
10647 | ||
792e3582 | 10648 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 10649 | unsigned int size, unsigned int type) |
792e3582 PB |
10650 | { |
10651 | struct io_uring_rsrc_register rr; | |
10652 | ||
10653 | /* keep it extendible */ | |
10654 | if (size != sizeof(rr)) | |
10655 | return -EINVAL; | |
10656 | ||
10657 | memset(&rr, 0, sizeof(rr)); | |
10658 | if (copy_from_user(&rr, arg, size)) | |
10659 | return -EFAULT; | |
992da01a | 10660 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
10661 | return -EINVAL; |
10662 | ||
992da01a | 10663 | switch (type) { |
792e3582 PB |
10664 | case IORING_RSRC_FILE: |
10665 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
10666 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
10667 | case IORING_RSRC_BUFFER: |
10668 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
10669 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
10670 | } |
10671 | return -EINVAL; | |
10672 | } | |
10673 | ||
fe76421d JA |
10674 | static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg, |
10675 | unsigned len) | |
10676 | { | |
10677 | struct io_uring_task *tctx = current->io_uring; | |
10678 | cpumask_var_t new_mask; | |
10679 | int ret; | |
10680 | ||
10681 | if (!tctx || !tctx->io_wq) | |
10682 | return -EINVAL; | |
10683 | ||
10684 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
10685 | return -ENOMEM; | |
10686 | ||
10687 | cpumask_clear(new_mask); | |
10688 | if (len > cpumask_size()) | |
10689 | len = cpumask_size(); | |
10690 | ||
fd213661 ES |
10691 | if (in_compat_syscall()) { |
10692 | ret = compat_get_bitmap(cpumask_bits(new_mask), | |
10693 | (const compat_ulong_t __user *)arg, | |
10694 | len * 8 /* CHAR_BIT */); | |
10695 | } else { | |
10696 | ret = copy_from_user(new_mask, arg, len); | |
10697 | } | |
10698 | ||
10699 | if (ret) { | |
fe76421d JA |
10700 | free_cpumask_var(new_mask); |
10701 | return -EFAULT; | |
10702 | } | |
10703 | ||
10704 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
10705 | free_cpumask_var(new_mask); | |
10706 | return ret; | |
10707 | } | |
10708 | ||
10709 | static int io_unregister_iowq_aff(struct io_ring_ctx *ctx) | |
10710 | { | |
10711 | struct io_uring_task *tctx = current->io_uring; | |
10712 | ||
10713 | if (!tctx || !tctx->io_wq) | |
10714 | return -EINVAL; | |
10715 | ||
10716 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
10717 | } | |
10718 | ||
2e480058 JA |
10719 | static int io_register_iowq_max_workers(struct io_ring_ctx *ctx, |
10720 | void __user *arg) | |
b22fa62a | 10721 | __must_hold(&ctx->uring_lock) |
2e480058 | 10722 | { |
b22fa62a | 10723 | struct io_tctx_node *node; |
fa84693b JA |
10724 | struct io_uring_task *tctx = NULL; |
10725 | struct io_sq_data *sqd = NULL; | |
2e480058 JA |
10726 | __u32 new_count[2]; |
10727 | int i, ret; | |
10728 | ||
2e480058 JA |
10729 | if (copy_from_user(new_count, arg, sizeof(new_count))) |
10730 | return -EFAULT; | |
10731 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
10732 | if (new_count[i] > INT_MAX) | |
10733 | return -EINVAL; | |
10734 | ||
fa84693b JA |
10735 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
10736 | sqd = ctx->sq_data; | |
10737 | if (sqd) { | |
009ad9f0 JA |
10738 | /* |
10739 | * Observe the correct sqd->lock -> ctx->uring_lock | |
10740 | * ordering. Fine to drop uring_lock here, we hold | |
10741 | * a ref to the ctx. | |
10742 | */ | |
41d3a6bd | 10743 | refcount_inc(&sqd->refs); |
009ad9f0 | 10744 | mutex_unlock(&ctx->uring_lock); |
fa84693b | 10745 | mutex_lock(&sqd->lock); |
009ad9f0 | 10746 | mutex_lock(&ctx->uring_lock); |
41d3a6bd JA |
10747 | if (sqd->thread) |
10748 | tctx = sqd->thread->io_uring; | |
fa84693b JA |
10749 | } |
10750 | } else { | |
10751 | tctx = current->io_uring; | |
10752 | } | |
10753 | ||
e139a1ec | 10754 | BUILD_BUG_ON(sizeof(new_count) != sizeof(ctx->iowq_limits)); |
fa84693b | 10755 | |
c29d963b PB |
10756 | for (i = 0; i < ARRAY_SIZE(new_count); i++) |
10757 | if (new_count[i]) | |
10758 | ctx->iowq_limits[i] = new_count[i]; | |
e139a1ec PB |
10759 | ctx->iowq_limits_set = true; |
10760 | ||
10761 | ret = -EINVAL; | |
10762 | if (tctx && tctx->io_wq) { | |
10763 | ret = io_wq_max_workers(tctx->io_wq, new_count); | |
10764 | if (ret) | |
10765 | goto err; | |
10766 | } else { | |
10767 | memset(new_count, 0, sizeof(new_count)); | |
10768 | } | |
fa84693b | 10769 | |
41d3a6bd | 10770 | if (sqd) { |
fa84693b | 10771 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
10772 | io_put_sq_data(sqd); |
10773 | } | |
2e480058 JA |
10774 | |
10775 | if (copy_to_user(arg, new_count, sizeof(new_count))) | |
10776 | return -EFAULT; | |
10777 | ||
b22fa62a PB |
10778 | /* that's it for SQPOLL, only the SQPOLL task creates requests */ |
10779 | if (sqd) | |
10780 | return 0; | |
10781 | ||
10782 | /* now propagate the restriction to all registered users */ | |
10783 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
10784 | struct io_uring_task *tctx = node->task->io_uring; | |
10785 | ||
10786 | if (WARN_ON_ONCE(!tctx->io_wq)) | |
10787 | continue; | |
10788 | ||
10789 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
10790 | new_count[i] = ctx->iowq_limits[i]; | |
10791 | /* ignore errors, it always returns zero anyway */ | |
10792 | (void)io_wq_max_workers(tctx->io_wq, new_count); | |
10793 | } | |
2e480058 | 10794 | return 0; |
fa84693b | 10795 | err: |
41d3a6bd | 10796 | if (sqd) { |
fa84693b | 10797 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
10798 | io_put_sq_data(sqd); |
10799 | } | |
fa84693b | 10800 | return ret; |
2e480058 JA |
10801 | } |
10802 | ||
071698e1 JA |
10803 | static bool io_register_op_must_quiesce(int op) |
10804 | { | |
10805 | switch (op) { | |
bd54b6fe BM |
10806 | case IORING_REGISTER_BUFFERS: |
10807 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 10808 | case IORING_REGISTER_FILES: |
071698e1 JA |
10809 | case IORING_UNREGISTER_FILES: |
10810 | case IORING_REGISTER_FILES_UPDATE: | |
10811 | case IORING_REGISTER_PROBE: | |
10812 | case IORING_REGISTER_PERSONALITY: | |
10813 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
10814 | case IORING_REGISTER_FILES2: |
10815 | case IORING_REGISTER_FILES_UPDATE2: | |
10816 | case IORING_REGISTER_BUFFERS2: | |
10817 | case IORING_REGISTER_BUFFERS_UPDATE: | |
fe76421d JA |
10818 | case IORING_REGISTER_IOWQ_AFF: |
10819 | case IORING_UNREGISTER_IOWQ_AFF: | |
2e480058 | 10820 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
071698e1 JA |
10821 | return false; |
10822 | default: | |
10823 | return true; | |
10824 | } | |
10825 | } | |
10826 | ||
e73c5c7c PB |
10827 | static int io_ctx_quiesce(struct io_ring_ctx *ctx) |
10828 | { | |
10829 | long ret; | |
10830 | ||
10831 | percpu_ref_kill(&ctx->refs); | |
10832 | ||
10833 | /* | |
10834 | * Drop uring mutex before waiting for references to exit. If another | |
10835 | * thread is currently inside io_uring_enter() it might need to grab the | |
10836 | * uring_lock to make progress. If we hold it here across the drain | |
10837 | * wait, then we can deadlock. It's safe to drop the mutex here, since | |
10838 | * no new references will come in after we've killed the percpu ref. | |
10839 | */ | |
10840 | mutex_unlock(&ctx->uring_lock); | |
10841 | do { | |
10842 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10843 | if (!ret) | |
10844 | break; | |
10845 | ret = io_run_task_work_sig(); | |
10846 | } while (ret >= 0); | |
10847 | mutex_lock(&ctx->uring_lock); | |
10848 | ||
10849 | if (ret) | |
10850 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); | |
10851 | return ret; | |
10852 | } | |
10853 | ||
edafccee JA |
10854 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10855 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10856 | __releases(ctx->uring_lock) |
10857 | __acquires(ctx->uring_lock) | |
edafccee JA |
10858 | { |
10859 | int ret; | |
10860 | ||
35fa71a0 JA |
10861 | /* |
10862 | * We're inside the ring mutex, if the ref is already dying, then | |
10863 | * someone else killed the ctx or is already going through | |
10864 | * io_uring_register(). | |
10865 | */ | |
10866 | if (percpu_ref_is_dying(&ctx->refs)) | |
10867 | return -ENXIO; | |
10868 | ||
75c4021a PB |
10869 | if (ctx->restricted) { |
10870 | if (opcode >= IORING_REGISTER_LAST) | |
10871 | return -EINVAL; | |
10872 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10873 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10874 | return -EACCES; | |
10875 | } | |
10876 | ||
071698e1 | 10877 | if (io_register_op_must_quiesce(opcode)) { |
e73c5c7c PB |
10878 | ret = io_ctx_quiesce(ctx); |
10879 | if (ret) | |
f70865db | 10880 | return ret; |
05f3fb3c | 10881 | } |
edafccee JA |
10882 | |
10883 | switch (opcode) { | |
10884 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10885 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10886 | break; |
10887 | case IORING_UNREGISTER_BUFFERS: | |
10888 | ret = -EINVAL; | |
10889 | if (arg || nr_args) | |
10890 | break; | |
0a96bbe4 | 10891 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10892 | break; |
6b06314c | 10893 | case IORING_REGISTER_FILES: |
792e3582 | 10894 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10895 | break; |
10896 | case IORING_UNREGISTER_FILES: | |
10897 | ret = -EINVAL; | |
10898 | if (arg || nr_args) | |
10899 | break; | |
10900 | ret = io_sqe_files_unregister(ctx); | |
10901 | break; | |
c3a31e60 | 10902 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10903 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10904 | break; |
9b402849 | 10905 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10906 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10907 | ret = -EINVAL; |
10908 | if (nr_args != 1) | |
10909 | break; | |
10910 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10911 | if (ret) |
10912 | break; | |
10913 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10914 | ctx->eventfd_async = 1; | |
10915 | else | |
10916 | ctx->eventfd_async = 0; | |
9b402849 JA |
10917 | break; |
10918 | case IORING_UNREGISTER_EVENTFD: | |
10919 | ret = -EINVAL; | |
10920 | if (arg || nr_args) | |
10921 | break; | |
10922 | ret = io_eventfd_unregister(ctx); | |
10923 | break; | |
66f4af93 JA |
10924 | case IORING_REGISTER_PROBE: |
10925 | ret = -EINVAL; | |
10926 | if (!arg || nr_args > 256) | |
10927 | break; | |
10928 | ret = io_probe(ctx, arg, nr_args); | |
10929 | break; | |
071698e1 JA |
10930 | case IORING_REGISTER_PERSONALITY: |
10931 | ret = -EINVAL; | |
10932 | if (arg || nr_args) | |
10933 | break; | |
10934 | ret = io_register_personality(ctx); | |
10935 | break; | |
10936 | case IORING_UNREGISTER_PERSONALITY: | |
10937 | ret = -EINVAL; | |
10938 | if (arg) | |
10939 | break; | |
10940 | ret = io_unregister_personality(ctx, nr_args); | |
10941 | break; | |
7e84e1c7 SG |
10942 | case IORING_REGISTER_ENABLE_RINGS: |
10943 | ret = -EINVAL; | |
10944 | if (arg || nr_args) | |
10945 | break; | |
10946 | ret = io_register_enable_rings(ctx); | |
10947 | break; | |
21b55dbc SG |
10948 | case IORING_REGISTER_RESTRICTIONS: |
10949 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10950 | break; | |
992da01a PB |
10951 | case IORING_REGISTER_FILES2: |
10952 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10953 | break; | |
10954 | case IORING_REGISTER_FILES_UPDATE2: | |
10955 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10956 | IORING_RSRC_FILE); | |
10957 | break; | |
10958 | case IORING_REGISTER_BUFFERS2: | |
10959 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10960 | break; |
992da01a PB |
10961 | case IORING_REGISTER_BUFFERS_UPDATE: |
10962 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10963 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10964 | break; |
fe76421d JA |
10965 | case IORING_REGISTER_IOWQ_AFF: |
10966 | ret = -EINVAL; | |
10967 | if (!arg || !nr_args) | |
10968 | break; | |
10969 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
10970 | break; | |
10971 | case IORING_UNREGISTER_IOWQ_AFF: | |
10972 | ret = -EINVAL; | |
10973 | if (arg || nr_args) | |
10974 | break; | |
10975 | ret = io_unregister_iowq_aff(ctx); | |
10976 | break; | |
2e480058 JA |
10977 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
10978 | ret = -EINVAL; | |
10979 | if (!arg || nr_args != 2) | |
10980 | break; | |
10981 | ret = io_register_iowq_max_workers(ctx, arg); | |
10982 | break; | |
edafccee JA |
10983 | default: |
10984 | ret = -EINVAL; | |
10985 | break; | |
10986 | } | |
10987 | ||
071698e1 | 10988 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10989 | /* bring the ctx back to life */ |
05f3fb3c | 10990 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10991 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10992 | } |
edafccee JA |
10993 | return ret; |
10994 | } | |
10995 | ||
10996 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10997 | void __user *, arg, unsigned int, nr_args) | |
10998 | { | |
10999 | struct io_ring_ctx *ctx; | |
11000 | long ret = -EBADF; | |
11001 | struct fd f; | |
11002 | ||
11003 | f = fdget(fd); | |
11004 | if (!f.file) | |
11005 | return -EBADF; | |
11006 | ||
11007 | ret = -EOPNOTSUPP; | |
11008 | if (f.file->f_op != &io_uring_fops) | |
11009 | goto out_fput; | |
11010 | ||
11011 | ctx = f.file->private_data; | |
11012 | ||
b6c23dd5 PB |
11013 | io_run_task_work(); |
11014 | ||
edafccee JA |
11015 | mutex_lock(&ctx->uring_lock); |
11016 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
11017 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
11018 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
11019 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
11020 | out_fput: |
11021 | fdput(f); | |
11022 | return ret; | |
11023 | } | |
11024 | ||
2b188cc1 JA |
11025 | static int __init io_uring_init(void) |
11026 | { | |
d7f62e82 SM |
11027 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
11028 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
11029 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
11030 | } while (0) | |
11031 | ||
11032 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
11033 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
11034 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
11035 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
11036 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
11037 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
11038 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
11039 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
11040 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
11041 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 11042 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
11043 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
11044 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
11045 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
11046 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
11047 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
11048 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
11049 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
11050 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
11051 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
11052 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
11053 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
11054 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
11055 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
11056 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
11057 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 11058 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
11059 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
11060 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 11061 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 11062 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 11063 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
b9445598 | 11064 | BUILD_BUG_SQE_ELEM(44, __u32, file_index); |
d7f62e82 | 11065 | |
b0d658ec PB |
11066 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
11067 | sizeof(struct io_uring_rsrc_update)); | |
11068 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
11069 | sizeof(struct io_uring_rsrc_update2)); | |
90499ad0 PB |
11070 | |
11071 | /* ->buf_index is u16 */ | |
11072 | BUILD_BUG_ON(IORING_MAX_REG_BUFFERS >= (1u << 16)); | |
11073 | ||
b0d658ec PB |
11074 | /* should fit into one byte */ |
11075 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
11076 | ||
d3656344 | 11077 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
32c2d33e | 11078 | BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int)); |
16340eab | 11079 | |
91f245d5 JA |
11080 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
11081 | SLAB_ACCOUNT); | |
2b188cc1 JA |
11082 | return 0; |
11083 | }; | |
11084 | __initcall(io_uring_init); |