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> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
92 | |
93 | /* | |
94 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
95 | */ | |
96 | #define IORING_FILE_TABLE_SHIFT 9 | |
97 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
98 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
99 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
100 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
101 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 102 | |
2d091d62 PB |
103 | #define IO_RSRC_TAG_TABLE_SHIFT 9 |
104 | #define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT) | |
105 | #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1) | |
106 | ||
489809e2 PB |
107 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
108 | ||
b16fed66 PB |
109 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
110 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
111 | IOSQE_BUFFER_SELECT) | |
112 | ||
09899b19 PB |
113 | #define IO_TCTX_REFS_CACHE_NR (1U << 10) |
114 | ||
2b188cc1 JA |
115 | struct io_uring { |
116 | u32 head ____cacheline_aligned_in_smp; | |
117 | u32 tail ____cacheline_aligned_in_smp; | |
118 | }; | |
119 | ||
1e84b97b | 120 | /* |
75b28aff HV |
121 | * This data is shared with the application through the mmap at offsets |
122 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
123 | * |
124 | * The offsets to the member fields are published through struct | |
125 | * io_sqring_offsets when calling io_uring_setup. | |
126 | */ | |
75b28aff | 127 | struct io_rings { |
1e84b97b SB |
128 | /* |
129 | * Head and tail offsets into the ring; the offsets need to be | |
130 | * masked to get valid indices. | |
131 | * | |
75b28aff HV |
132 | * The kernel controls head of the sq ring and the tail of the cq ring, |
133 | * and the application controls tail of the sq ring and the head of the | |
134 | * cq ring. | |
1e84b97b | 135 | */ |
75b28aff | 136 | struct io_uring sq, cq; |
1e84b97b | 137 | /* |
75b28aff | 138 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
139 | * ring_entries - 1) |
140 | */ | |
75b28aff HV |
141 | u32 sq_ring_mask, cq_ring_mask; |
142 | /* Ring sizes (constant, power of 2) */ | |
143 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
144 | /* |
145 | * Number of invalid entries dropped by the kernel due to | |
146 | * invalid index stored in array | |
147 | * | |
148 | * Written by the kernel, shouldn't be modified by the | |
149 | * application (i.e. get number of "new events" by comparing to | |
150 | * cached value). | |
151 | * | |
152 | * After a new SQ head value was read by the application this | |
153 | * counter includes all submissions that were dropped reaching | |
154 | * the new SQ head (and possibly more). | |
155 | */ | |
75b28aff | 156 | u32 sq_dropped; |
1e84b97b | 157 | /* |
0d9b5b3a | 158 | * Runtime SQ flags |
1e84b97b SB |
159 | * |
160 | * Written by the kernel, shouldn't be modified by the | |
161 | * application. | |
162 | * | |
163 | * The application needs a full memory barrier before checking | |
164 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
165 | */ | |
75b28aff | 166 | u32 sq_flags; |
0d9b5b3a SG |
167 | /* |
168 | * Runtime CQ flags | |
169 | * | |
170 | * Written by the application, shouldn't be modified by the | |
171 | * kernel. | |
172 | */ | |
173 | u32 cq_flags; | |
1e84b97b SB |
174 | /* |
175 | * Number of completion events lost because the queue was full; | |
176 | * this should be avoided by the application by making sure | |
0b4295b5 | 177 | * there are not more requests pending than there is space in |
1e84b97b SB |
178 | * the completion queue. |
179 | * | |
180 | * Written by the kernel, shouldn't be modified by the | |
181 | * application (i.e. get number of "new events" by comparing to | |
182 | * cached value). | |
183 | * | |
184 | * As completion events come in out of order this counter is not | |
185 | * ordered with any other data. | |
186 | */ | |
75b28aff | 187 | u32 cq_overflow; |
1e84b97b SB |
188 | /* |
189 | * Ring buffer of completion events. | |
190 | * | |
191 | * The kernel writes completion events fresh every time they are | |
192 | * produced, so the application is allowed to modify pending | |
193 | * entries. | |
194 | */ | |
75b28aff | 195 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
196 | }; |
197 | ||
45d189c6 PB |
198 | enum io_uring_cmd_flags { |
199 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 200 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
201 | }; |
202 | ||
edafccee JA |
203 | struct io_mapped_ubuf { |
204 | u64 ubuf; | |
4751f53d | 205 | u64 ubuf_end; |
edafccee | 206 | unsigned int nr_bvecs; |
de293938 | 207 | unsigned long acct_pages; |
41edf1a5 | 208 | struct bio_vec bvec[]; |
edafccee JA |
209 | }; |
210 | ||
50238531 BM |
211 | struct io_ring_ctx; |
212 | ||
6c2450ae PB |
213 | struct io_overflow_cqe { |
214 | struct io_uring_cqe cqe; | |
215 | struct list_head list; | |
216 | }; | |
217 | ||
a04b0ac0 PB |
218 | struct io_fixed_file { |
219 | /* file * with additional FFS_* flags */ | |
220 | unsigned long file_ptr; | |
221 | }; | |
222 | ||
269bbe5f BM |
223 | struct io_rsrc_put { |
224 | struct list_head list; | |
b60c8dce | 225 | u64 tag; |
50238531 BM |
226 | union { |
227 | void *rsrc; | |
228 | struct file *file; | |
bd54b6fe | 229 | struct io_mapped_ubuf *buf; |
50238531 | 230 | }; |
269bbe5f BM |
231 | }; |
232 | ||
aeca241b PB |
233 | struct io_file_table { |
234 | /* two level table */ | |
235 | struct io_fixed_file **files; | |
31b51510 JA |
236 | }; |
237 | ||
b895c9a6 | 238 | struct io_rsrc_node { |
05589553 XW |
239 | struct percpu_ref refs; |
240 | struct list_head node; | |
269bbe5f | 241 | struct list_head rsrc_list; |
b895c9a6 | 242 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 243 | struct llist_node llist; |
e297822b | 244 | bool done; |
05589553 XW |
245 | }; |
246 | ||
40ae0ff7 PB |
247 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
248 | ||
b895c9a6 | 249 | struct io_rsrc_data { |
05f3fb3c JA |
250 | struct io_ring_ctx *ctx; |
251 | ||
2d091d62 PB |
252 | u64 **tags; |
253 | unsigned int nr; | |
40ae0ff7 | 254 | rsrc_put_fn *do_put; |
3e942498 | 255 | atomic_t refs; |
05f3fb3c | 256 | struct completion done; |
8bad28d8 | 257 | bool quiesce; |
05f3fb3c JA |
258 | }; |
259 | ||
5a2e745d JA |
260 | struct io_buffer { |
261 | struct list_head list; | |
262 | __u64 addr; | |
d1f82808 | 263 | __u32 len; |
5a2e745d JA |
264 | __u16 bid; |
265 | }; | |
266 | ||
21b55dbc SG |
267 | struct io_restriction { |
268 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
269 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
270 | u8 sqe_flags_allowed; | |
271 | u8 sqe_flags_required; | |
7e84e1c7 | 272 | bool registered; |
21b55dbc SG |
273 | }; |
274 | ||
37d1e2e3 JA |
275 | enum { |
276 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
277 | IO_SQ_THREAD_SHOULD_PARK, | |
278 | }; | |
279 | ||
534ca6d6 JA |
280 | struct io_sq_data { |
281 | refcount_t refs; | |
9e138a48 | 282 | atomic_t park_pending; |
09a6f4ef | 283 | struct mutex lock; |
69fb2131 JA |
284 | |
285 | /* ctx's that are using this sqd */ | |
286 | struct list_head ctx_list; | |
69fb2131 | 287 | |
534ca6d6 JA |
288 | struct task_struct *thread; |
289 | struct wait_queue_head wait; | |
08369246 XW |
290 | |
291 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
292 | int sq_cpu; |
293 | pid_t task_pid; | |
5c2469e0 | 294 | pid_t task_tgid; |
37d1e2e3 JA |
295 | |
296 | unsigned long state; | |
37d1e2e3 | 297 | struct completion exited; |
534ca6d6 JA |
298 | }; |
299 | ||
258b29a9 | 300 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 301 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 302 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 303 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
304 | |
305 | struct io_comp_state { | |
6dd0be1e | 306 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 307 | unsigned int nr; |
c7dae4ba | 308 | /* inline/task_work completion list, under ->uring_lock */ |
1b4c351f | 309 | struct list_head free_list; |
258b29a9 PB |
310 | }; |
311 | ||
a1ab7b35 PB |
312 | struct io_submit_link { |
313 | struct io_kiocb *head; | |
314 | struct io_kiocb *last; | |
315 | }; | |
316 | ||
258b29a9 PB |
317 | struct io_submit_state { |
318 | struct blk_plug plug; | |
a1ab7b35 | 319 | struct io_submit_link link; |
258b29a9 PB |
320 | |
321 | /* | |
322 | * io_kiocb alloc cache | |
323 | */ | |
bf019da7 | 324 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
325 | unsigned int free_reqs; |
326 | ||
327 | bool plug_started; | |
328 | ||
329 | /* | |
330 | * Batch completion logic | |
331 | */ | |
332 | struct io_comp_state comp; | |
333 | ||
334 | /* | |
335 | * File reference cache | |
336 | */ | |
337 | struct file *file; | |
338 | unsigned int fd; | |
339 | unsigned int file_refs; | |
340 | unsigned int ios_left; | |
341 | }; | |
342 | ||
2b188cc1 | 343 | struct io_ring_ctx { |
b52ecf8c | 344 | /* const or read-mostly hot data */ |
2b188cc1 JA |
345 | struct { |
346 | struct percpu_ref refs; | |
2b188cc1 | 347 | |
b52ecf8c | 348 | struct io_rings *rings; |
2b188cc1 | 349 | unsigned int flags; |
e1d85334 | 350 | unsigned int compat: 1; |
e1d85334 RD |
351 | unsigned int drain_next: 1; |
352 | unsigned int eventfd_async: 1; | |
21b55dbc | 353 | unsigned int restricted: 1; |
f18ee4cf | 354 | unsigned int off_timeout_used: 1; |
441b8a78 | 355 | unsigned int drain_used: 1; |
b52ecf8c | 356 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 357 | |
7f1129d2 | 358 | /* submission data */ |
b52ecf8c | 359 | struct { |
0499e582 PB |
360 | struct mutex uring_lock; |
361 | ||
75b28aff HV |
362 | /* |
363 | * Ring buffer of indices into array of io_uring_sqe, which is | |
364 | * mmapped by the application using the IORING_OFF_SQES offset. | |
365 | * | |
366 | * This indirection could e.g. be used to assign fixed | |
367 | * io_uring_sqe entries to operations and only submit them to | |
368 | * the queue when needed. | |
369 | * | |
370 | * The kernel modifies neither the indices array nor the entries | |
371 | * array. | |
372 | */ | |
373 | u32 *sq_array; | |
c7af47cf | 374 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
375 | unsigned cached_sq_head; |
376 | unsigned sq_entries; | |
de0617e4 | 377 | struct list_head defer_list; |
7f1129d2 PB |
378 | |
379 | /* | |
380 | * Fixed resources fast path, should be accessed only under | |
381 | * uring_lock, and updated through io_uring_register(2) | |
382 | */ | |
383 | struct io_rsrc_node *rsrc_node; | |
384 | struct io_file_table file_table; | |
385 | unsigned nr_user_files; | |
386 | unsigned nr_user_bufs; | |
387 | struct io_mapped_ubuf **user_bufs; | |
388 | ||
389 | struct io_submit_state submit_state; | |
5262f567 | 390 | struct list_head timeout_list; |
1d7bb1d5 | 391 | struct list_head cq_overflow_list; |
7f1129d2 PB |
392 | struct xarray io_buffers; |
393 | struct xarray personalities; | |
394 | u32 pers_next; | |
395 | unsigned sq_thread_idle; | |
2b188cc1 JA |
396 | } ____cacheline_aligned_in_smp; |
397 | ||
d0acdee2 PB |
398 | /* IRQ completion list, under ->completion_lock */ |
399 | struct list_head locked_free_list; | |
400 | unsigned int locked_free_nr; | |
3c1a2ead | 401 | |
7c30f36a | 402 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
403 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
404 | ||
90554200 | 405 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 406 | struct list_head sqd_list; |
75b28aff | 407 | |
5ed7a37d PB |
408 | unsigned long check_cq_overflow; |
409 | ||
206aefde JA |
410 | struct { |
411 | unsigned cached_cq_tail; | |
412 | unsigned cq_entries; | |
0499e582 | 413 | struct eventfd_ctx *cq_ev_fd; |
311997b3 | 414 | struct wait_queue_head poll_wait; |
0499e582 PB |
415 | struct wait_queue_head cq_wait; |
416 | unsigned cq_extra; | |
417 | atomic_t cq_timeouts; | |
206aefde | 418 | struct fasync_struct *cq_fasync; |
0499e582 | 419 | unsigned cq_last_tm_flush; |
206aefde | 420 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 421 | |
2b188cc1 JA |
422 | struct { |
423 | spinlock_t completion_lock; | |
e94f141b | 424 | |
def596e9 | 425 | /* |
540e32a0 | 426 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
427 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
428 | * For SQPOLL, only the single threaded io_sq_thread() will | |
429 | * manipulate the list, hence no extra locking is needed there. | |
430 | */ | |
540e32a0 | 431 | struct list_head iopoll_list; |
78076bb6 JA |
432 | struct hlist_head *cancel_hash; |
433 | unsigned cancel_hash_bits; | |
e94f141b | 434 | bool poll_multi_file; |
2b188cc1 | 435 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 436 | |
21b55dbc | 437 | struct io_restriction restrictions; |
3c1a2ead | 438 | |
b13a8918 PB |
439 | /* slow path rsrc auxilary data, used by update/register */ |
440 | struct { | |
441 | struct io_rsrc_node *rsrc_backup_node; | |
442 | struct io_mapped_ubuf *dummy_ubuf; | |
443 | struct io_rsrc_data *file_data; | |
444 | struct io_rsrc_data *buf_data; | |
445 | ||
446 | struct delayed_work rsrc_put_work; | |
447 | struct llist_head rsrc_put_llist; | |
448 | struct list_head rsrc_ref_list; | |
449 | spinlock_t rsrc_ref_lock; | |
450 | }; | |
451 | ||
3c1a2ead | 452 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
453 | struct { |
454 | #if defined(CONFIG_UNIX) | |
455 | struct socket *ring_sock; | |
456 | #endif | |
457 | /* hashed buffered write serialization */ | |
458 | struct io_wq_hash *hash_map; | |
459 | ||
460 | /* Only used for accounting purposes */ | |
461 | struct user_struct *user; | |
462 | struct mm_struct *mm_account; | |
463 | ||
464 | /* ctx exit and cancelation */ | |
465 | struct callback_head *exit_task_work; | |
466 | struct work_struct exit_work; | |
467 | struct list_head tctx_list; | |
468 | struct completion ref_comp; | |
469 | }; | |
2b188cc1 JA |
470 | }; |
471 | ||
53e043b2 SM |
472 | struct io_uring_task { |
473 | /* submission side */ | |
09899b19 | 474 | int cached_refs; |
53e043b2 SM |
475 | struct xarray xa; |
476 | struct wait_queue_head wait; | |
ee53fb2b SM |
477 | const struct io_ring_ctx *last; |
478 | struct io_wq *io_wq; | |
53e043b2 | 479 | struct percpu_counter inflight; |
b303fe2e | 480 | atomic_t inflight_tracked; |
53e043b2 | 481 | atomic_t in_idle; |
53e043b2 SM |
482 | |
483 | spinlock_t task_lock; | |
484 | struct io_wq_work_list task_list; | |
485 | unsigned long task_state; | |
486 | struct callback_head task_work; | |
487 | }; | |
488 | ||
09bb8394 JA |
489 | /* |
490 | * First field must be the file pointer in all the | |
491 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
492 | */ | |
221c5eb2 JA |
493 | struct io_poll_iocb { |
494 | struct file *file; | |
018043be | 495 | struct wait_queue_head *head; |
221c5eb2 | 496 | __poll_t events; |
8c838788 | 497 | bool done; |
221c5eb2 | 498 | bool canceled; |
392edb45 | 499 | struct wait_queue_entry wait; |
221c5eb2 JA |
500 | }; |
501 | ||
9d805892 | 502 | struct io_poll_update { |
018043be | 503 | struct file *file; |
9d805892 PB |
504 | u64 old_user_data; |
505 | u64 new_user_data; | |
506 | __poll_t events; | |
b69de288 JA |
507 | bool update_events; |
508 | bool update_user_data; | |
018043be PB |
509 | }; |
510 | ||
b5dba59e JA |
511 | struct io_close { |
512 | struct file *file; | |
b5dba59e JA |
513 | int fd; |
514 | }; | |
515 | ||
ad8a48ac JA |
516 | struct io_timeout_data { |
517 | struct io_kiocb *req; | |
518 | struct hrtimer timer; | |
519 | struct timespec64 ts; | |
520 | enum hrtimer_mode mode; | |
521 | }; | |
522 | ||
8ed8d3c3 JA |
523 | struct io_accept { |
524 | struct file *file; | |
525 | struct sockaddr __user *addr; | |
526 | int __user *addr_len; | |
527 | int flags; | |
09952e3e | 528 | unsigned long nofile; |
8ed8d3c3 JA |
529 | }; |
530 | ||
531 | struct io_sync { | |
532 | struct file *file; | |
533 | loff_t len; | |
534 | loff_t off; | |
535 | int flags; | |
d63d1b5e | 536 | int mode; |
8ed8d3c3 JA |
537 | }; |
538 | ||
fbf23849 JA |
539 | struct io_cancel { |
540 | struct file *file; | |
541 | u64 addr; | |
542 | }; | |
543 | ||
b29472ee JA |
544 | struct io_timeout { |
545 | struct file *file; | |
bfe68a22 PB |
546 | u32 off; |
547 | u32 target_seq; | |
135fcde8 | 548 | struct list_head list; |
90cd7e42 PB |
549 | /* head of the link, used by linked timeouts only */ |
550 | struct io_kiocb *head; | |
b29472ee JA |
551 | }; |
552 | ||
0bdf7a2d PB |
553 | struct io_timeout_rem { |
554 | struct file *file; | |
555 | u64 addr; | |
9c8e11b3 PB |
556 | |
557 | /* timeout update */ | |
558 | struct timespec64 ts; | |
559 | u32 flags; | |
0bdf7a2d PB |
560 | }; |
561 | ||
9adbd45d JA |
562 | struct io_rw { |
563 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
564 | struct kiocb kiocb; | |
565 | u64 addr; | |
566 | u64 len; | |
567 | }; | |
568 | ||
3fbb51c1 JA |
569 | struct io_connect { |
570 | struct file *file; | |
571 | struct sockaddr __user *addr; | |
572 | int addr_len; | |
573 | }; | |
574 | ||
e47293fd JA |
575 | struct io_sr_msg { |
576 | struct file *file; | |
fddaface | 577 | union { |
4af3417a PB |
578 | struct compat_msghdr __user *umsg_compat; |
579 | struct user_msghdr __user *umsg; | |
580 | void __user *buf; | |
fddaface | 581 | }; |
e47293fd | 582 | int msg_flags; |
bcda7baa | 583 | int bgid; |
fddaface | 584 | size_t len; |
bcda7baa | 585 | struct io_buffer *kbuf; |
e47293fd JA |
586 | }; |
587 | ||
15b71abe JA |
588 | struct io_open { |
589 | struct file *file; | |
590 | int dfd; | |
15b71abe | 591 | struct filename *filename; |
c12cedf2 | 592 | struct open_how how; |
4022e7af | 593 | unsigned long nofile; |
15b71abe JA |
594 | }; |
595 | ||
269bbe5f | 596 | struct io_rsrc_update { |
05f3fb3c JA |
597 | struct file *file; |
598 | u64 arg; | |
599 | u32 nr_args; | |
600 | u32 offset; | |
601 | }; | |
602 | ||
4840e418 JA |
603 | struct io_fadvise { |
604 | struct file *file; | |
605 | u64 offset; | |
606 | u32 len; | |
607 | u32 advice; | |
608 | }; | |
609 | ||
c1ca757b JA |
610 | struct io_madvise { |
611 | struct file *file; | |
612 | u64 addr; | |
613 | u32 len; | |
614 | u32 advice; | |
615 | }; | |
616 | ||
3e4827b0 JA |
617 | struct io_epoll { |
618 | struct file *file; | |
619 | int epfd; | |
620 | int op; | |
621 | int fd; | |
622 | struct epoll_event event; | |
e47293fd JA |
623 | }; |
624 | ||
7d67af2c PB |
625 | struct io_splice { |
626 | struct file *file_out; | |
627 | struct file *file_in; | |
628 | loff_t off_out; | |
629 | loff_t off_in; | |
630 | u64 len; | |
631 | unsigned int flags; | |
632 | }; | |
633 | ||
ddf0322d JA |
634 | struct io_provide_buf { |
635 | struct file *file; | |
636 | __u64 addr; | |
38134ada | 637 | __u32 len; |
ddf0322d JA |
638 | __u32 bgid; |
639 | __u16 nbufs; | |
640 | __u16 bid; | |
641 | }; | |
642 | ||
1d9e1288 BM |
643 | struct io_statx { |
644 | struct file *file; | |
645 | int dfd; | |
646 | unsigned int mask; | |
647 | unsigned int flags; | |
e62753e4 | 648 | const char __user *filename; |
1d9e1288 BM |
649 | struct statx __user *buffer; |
650 | }; | |
651 | ||
36f4fa68 JA |
652 | struct io_shutdown { |
653 | struct file *file; | |
654 | int how; | |
655 | }; | |
656 | ||
80a261fd JA |
657 | struct io_rename { |
658 | struct file *file; | |
659 | int old_dfd; | |
660 | int new_dfd; | |
661 | struct filename *oldpath; | |
662 | struct filename *newpath; | |
663 | int flags; | |
664 | }; | |
665 | ||
14a1143b JA |
666 | struct io_unlink { |
667 | struct file *file; | |
668 | int dfd; | |
669 | int flags; | |
670 | struct filename *filename; | |
671 | }; | |
672 | ||
3ca405eb PB |
673 | struct io_completion { |
674 | struct file *file; | |
675 | struct list_head list; | |
8c3f9cd1 | 676 | u32 cflags; |
3ca405eb PB |
677 | }; |
678 | ||
f499a021 JA |
679 | struct io_async_connect { |
680 | struct sockaddr_storage address; | |
681 | }; | |
682 | ||
03b1230c JA |
683 | struct io_async_msghdr { |
684 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
685 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
686 | struct iovec *free_iov; | |
03b1230c JA |
687 | struct sockaddr __user *uaddr; |
688 | struct msghdr msg; | |
b537916c | 689 | struct sockaddr_storage addr; |
03b1230c JA |
690 | }; |
691 | ||
f67676d1 JA |
692 | struct io_async_rw { |
693 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
694 | const struct iovec *free_iovec; |
695 | struct iov_iter iter; | |
227c0c96 | 696 | size_t bytes_done; |
bcf5a063 | 697 | struct wait_page_queue wpq; |
f67676d1 JA |
698 | }; |
699 | ||
6b47ee6e PB |
700 | enum { |
701 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
702 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
703 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
704 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
705 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 706 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 707 | |
dddca226 | 708 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 709 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
710 | REQ_F_INFLIGHT_BIT, |
711 | REQ_F_CUR_POS_BIT, | |
712 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 713 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 714 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 715 | REQ_F_POLLED_BIT, |
bcda7baa | 716 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 717 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 718 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 719 | REQ_F_REISSUE_BIT, |
8c130827 | 720 | REQ_F_DONT_REISSUE_BIT, |
7b29f92d JA |
721 | /* keep async read/write and isreg together and in order */ |
722 | REQ_F_ASYNC_READ_BIT, | |
723 | REQ_F_ASYNC_WRITE_BIT, | |
724 | REQ_F_ISREG_BIT, | |
84557871 JA |
725 | |
726 | /* not a real bit, just to check we're not overflowing the space */ | |
727 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
728 | }; |
729 | ||
730 | enum { | |
731 | /* ctx owns file */ | |
732 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
733 | /* drain existing IO first */ | |
734 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
735 | /* linked sqes */ | |
736 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
737 | /* doesn't sever on completion < 0 */ | |
738 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
739 | /* IOSQE_ASYNC */ | |
740 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
741 | /* IOSQE_BUFFER_SELECT */ |
742 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 743 | |
6b47ee6e | 744 | /* fail rest of links */ |
93d2bcd2 | 745 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 746 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
747 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
748 | /* read/write uses file position */ | |
749 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
750 | /* must not punt to workers */ | |
751 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 752 | /* has or had linked timeout */ |
6b47ee6e | 753 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
754 | /* needs cleanup */ |
755 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
756 | /* already went through poll handler */ |
757 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
758 | /* buffer already selected */ |
759 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
760 | /* linked timeout is active, i.e. prepared by link's head */ |
761 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
762 | /* completion is deferred through io_comp_state */ |
763 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
764 | /* caller should reissue async */ |
765 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
766 | /* don't attempt request reissue, see io_rw_reissue() */ |
767 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
768 | /* supports async reads */ |
769 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
770 | /* supports async writes */ | |
771 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
772 | /* regular file */ | |
773 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
774 | }; |
775 | ||
776 | struct async_poll { | |
777 | struct io_poll_iocb poll; | |
807abcb0 | 778 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
779 | }; |
780 | ||
7cbf1722 JA |
781 | struct io_task_work { |
782 | struct io_wq_work_node node; | |
783 | task_work_func_t func; | |
784 | }; | |
785 | ||
992da01a PB |
786 | enum { |
787 | IORING_RSRC_FILE = 0, | |
788 | IORING_RSRC_BUFFER = 1, | |
789 | }; | |
790 | ||
09bb8394 JA |
791 | /* |
792 | * NOTE! Each of the iocb union members has the file pointer | |
793 | * as the first entry in their struct definition. So you can | |
794 | * access the file pointer through any of the sub-structs, | |
795 | * or directly as just 'ki_filp' in this struct. | |
796 | */ | |
2b188cc1 | 797 | struct io_kiocb { |
221c5eb2 | 798 | union { |
09bb8394 | 799 | struct file *file; |
9adbd45d | 800 | struct io_rw rw; |
221c5eb2 | 801 | struct io_poll_iocb poll; |
9d805892 | 802 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
803 | struct io_accept accept; |
804 | struct io_sync sync; | |
fbf23849 | 805 | struct io_cancel cancel; |
b29472ee | 806 | struct io_timeout timeout; |
0bdf7a2d | 807 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 808 | struct io_connect connect; |
e47293fd | 809 | struct io_sr_msg sr_msg; |
15b71abe | 810 | struct io_open open; |
b5dba59e | 811 | struct io_close close; |
269bbe5f | 812 | struct io_rsrc_update rsrc_update; |
4840e418 | 813 | struct io_fadvise fadvise; |
c1ca757b | 814 | struct io_madvise madvise; |
3e4827b0 | 815 | struct io_epoll epoll; |
7d67af2c | 816 | struct io_splice splice; |
ddf0322d | 817 | struct io_provide_buf pbuf; |
1d9e1288 | 818 | struct io_statx statx; |
36f4fa68 | 819 | struct io_shutdown shutdown; |
80a261fd | 820 | struct io_rename rename; |
14a1143b | 821 | struct io_unlink unlink; |
3ca405eb PB |
822 | /* use only after cleaning per-op data, see io_clean_op() */ |
823 | struct io_completion compl; | |
221c5eb2 | 824 | }; |
2b188cc1 | 825 | |
e8c2bc1f JA |
826 | /* opcode allocated if it needs to store data for async defer */ |
827 | void *async_data; | |
d625c6ee | 828 | u8 opcode; |
65a6543d XW |
829 | /* polled IO has completed */ |
830 | u8 iopoll_completed; | |
2b188cc1 | 831 | |
4f4eeba8 | 832 | u16 buf_index; |
9cf7c104 | 833 | u32 result; |
4f4eeba8 | 834 | |
010e8e6b PB |
835 | struct io_ring_ctx *ctx; |
836 | unsigned int flags; | |
abc54d63 | 837 | atomic_t refs; |
010e8e6b PB |
838 | struct task_struct *task; |
839 | u64 user_data; | |
d7718a9d | 840 | |
f2f87370 | 841 | struct io_kiocb *link; |
269bbe5f | 842 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 843 | |
b303fe2e | 844 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 845 | struct list_head inflight_entry; |
7cbf1722 JA |
846 | union { |
847 | struct io_task_work io_task_work; | |
848 | struct callback_head task_work; | |
849 | }; | |
010e8e6b PB |
850 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
851 | struct hlist_node hash_node; | |
852 | struct async_poll *apoll; | |
853 | struct io_wq_work work; | |
eae071c9 PB |
854 | /* store used ubuf, so we can prevent reloading */ |
855 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 856 | }; |
05589553 | 857 | |
13bf43f5 PB |
858 | struct io_tctx_node { |
859 | struct list_head ctx_node; | |
860 | struct task_struct *task; | |
13bf43f5 PB |
861 | struct io_ring_ctx *ctx; |
862 | }; | |
863 | ||
27dc8338 PB |
864 | struct io_defer_entry { |
865 | struct list_head list; | |
866 | struct io_kiocb *req; | |
9cf7c104 | 867 | u32 seq; |
2b188cc1 JA |
868 | }; |
869 | ||
d3656344 | 870 | struct io_op_def { |
d3656344 JA |
871 | /* needs req->file assigned */ |
872 | unsigned needs_file : 1; | |
d3656344 JA |
873 | /* hash wq insertion if file is a regular file */ |
874 | unsigned hash_reg_file : 1; | |
875 | /* unbound wq insertion if file is a non-regular file */ | |
876 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
877 | /* opcode is not supported by this kernel */ |
878 | unsigned not_supported : 1; | |
8a72758c JA |
879 | /* set if opcode supports polled "wait" */ |
880 | unsigned pollin : 1; | |
881 | unsigned pollout : 1; | |
bcda7baa JA |
882 | /* op supports buffer selection */ |
883 | unsigned buffer_select : 1; | |
26f0505a PB |
884 | /* do prep async if is going to be punted */ |
885 | unsigned needs_async_setup : 1; | |
27926b68 JA |
886 | /* should block plug */ |
887 | unsigned plug : 1; | |
e8c2bc1f JA |
888 | /* size of async data needed, if any */ |
889 | unsigned short async_size; | |
d3656344 JA |
890 | }; |
891 | ||
0918682b | 892 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
893 | [IORING_OP_NOP] = {}, |
894 | [IORING_OP_READV] = { | |
d3656344 JA |
895 | .needs_file = 1, |
896 | .unbound_nonreg_file = 1, | |
8a72758c | 897 | .pollin = 1, |
4d954c25 | 898 | .buffer_select = 1, |
26f0505a | 899 | .needs_async_setup = 1, |
27926b68 | 900 | .plug = 1, |
e8c2bc1f | 901 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 902 | }, |
0463b6c5 | 903 | [IORING_OP_WRITEV] = { |
d3656344 JA |
904 | .needs_file = 1, |
905 | .hash_reg_file = 1, | |
906 | .unbound_nonreg_file = 1, | |
8a72758c | 907 | .pollout = 1, |
26f0505a | 908 | .needs_async_setup = 1, |
27926b68 | 909 | .plug = 1, |
e8c2bc1f | 910 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 911 | }, |
0463b6c5 | 912 | [IORING_OP_FSYNC] = { |
d3656344 JA |
913 | .needs_file = 1, |
914 | }, | |
0463b6c5 | 915 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
916 | .needs_file = 1, |
917 | .unbound_nonreg_file = 1, | |
8a72758c | 918 | .pollin = 1, |
27926b68 | 919 | .plug = 1, |
e8c2bc1f | 920 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 921 | }, |
0463b6c5 | 922 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
923 | .needs_file = 1, |
924 | .hash_reg_file = 1, | |
925 | .unbound_nonreg_file = 1, | |
8a72758c | 926 | .pollout = 1, |
27926b68 | 927 | .plug = 1, |
e8c2bc1f | 928 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 929 | }, |
0463b6c5 | 930 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
931 | .needs_file = 1, |
932 | .unbound_nonreg_file = 1, | |
933 | }, | |
0463b6c5 PB |
934 | [IORING_OP_POLL_REMOVE] = {}, |
935 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
936 | .needs_file = 1, |
937 | }, | |
0463b6c5 | 938 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
939 | .needs_file = 1, |
940 | .unbound_nonreg_file = 1, | |
8a72758c | 941 | .pollout = 1, |
26f0505a | 942 | .needs_async_setup = 1, |
e8c2bc1f | 943 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 944 | }, |
0463b6c5 | 945 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
946 | .needs_file = 1, |
947 | .unbound_nonreg_file = 1, | |
8a72758c | 948 | .pollin = 1, |
52de1fe1 | 949 | .buffer_select = 1, |
26f0505a | 950 | .needs_async_setup = 1, |
e8c2bc1f | 951 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 952 | }, |
0463b6c5 | 953 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 954 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 955 | }, |
9c8e11b3 PB |
956 | [IORING_OP_TIMEOUT_REMOVE] = { |
957 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 958 | }, |
0463b6c5 | 959 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
960 | .needs_file = 1, |
961 | .unbound_nonreg_file = 1, | |
8a72758c | 962 | .pollin = 1, |
d3656344 | 963 | }, |
0463b6c5 PB |
964 | [IORING_OP_ASYNC_CANCEL] = {}, |
965 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 966 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 967 | }, |
0463b6c5 | 968 | [IORING_OP_CONNECT] = { |
d3656344 JA |
969 | .needs_file = 1, |
970 | .unbound_nonreg_file = 1, | |
8a72758c | 971 | .pollout = 1, |
26f0505a | 972 | .needs_async_setup = 1, |
e8c2bc1f | 973 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 974 | }, |
0463b6c5 | 975 | [IORING_OP_FALLOCATE] = { |
d3656344 | 976 | .needs_file = 1, |
d3656344 | 977 | }, |
44526bed JA |
978 | [IORING_OP_OPENAT] = {}, |
979 | [IORING_OP_CLOSE] = {}, | |
980 | [IORING_OP_FILES_UPDATE] = {}, | |
981 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 982 | [IORING_OP_READ] = { |
3a6820f2 JA |
983 | .needs_file = 1, |
984 | .unbound_nonreg_file = 1, | |
8a72758c | 985 | .pollin = 1, |
bcda7baa | 986 | .buffer_select = 1, |
27926b68 | 987 | .plug = 1, |
e8c2bc1f | 988 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 989 | }, |
0463b6c5 | 990 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
991 | .needs_file = 1, |
992 | .unbound_nonreg_file = 1, | |
8a72758c | 993 | .pollout = 1, |
27926b68 | 994 | .plug = 1, |
e8c2bc1f | 995 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 996 | }, |
0463b6c5 | 997 | [IORING_OP_FADVISE] = { |
4840e418 | 998 | .needs_file = 1, |
c1ca757b | 999 | }, |
44526bed | 1000 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1001 | [IORING_OP_SEND] = { |
fddaface JA |
1002 | .needs_file = 1, |
1003 | .unbound_nonreg_file = 1, | |
8a72758c | 1004 | .pollout = 1, |
fddaface | 1005 | }, |
0463b6c5 | 1006 | [IORING_OP_RECV] = { |
fddaface JA |
1007 | .needs_file = 1, |
1008 | .unbound_nonreg_file = 1, | |
8a72758c | 1009 | .pollin = 1, |
bcda7baa | 1010 | .buffer_select = 1, |
fddaface | 1011 | }, |
0463b6c5 | 1012 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1013 | }, |
3e4827b0 JA |
1014 | [IORING_OP_EPOLL_CTL] = { |
1015 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1016 | }, |
7d67af2c PB |
1017 | [IORING_OP_SPLICE] = { |
1018 | .needs_file = 1, | |
1019 | .hash_reg_file = 1, | |
1020 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1021 | }, |
1022 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1023 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1024 | [IORING_OP_TEE] = { |
1025 | .needs_file = 1, | |
1026 | .hash_reg_file = 1, | |
1027 | .unbound_nonreg_file = 1, | |
1028 | }, | |
36f4fa68 JA |
1029 | [IORING_OP_SHUTDOWN] = { |
1030 | .needs_file = 1, | |
1031 | }, | |
44526bed JA |
1032 | [IORING_OP_RENAMEAT] = {}, |
1033 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1034 | }; |
1035 | ||
7a612350 | 1036 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1037 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1038 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1039 | struct task_struct *task, | |
3dd0c97a | 1040 | bool cancel_all); |
78cc687b | 1041 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
b895c9a6 | 1042 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1043 | |
d4d19c19 PB |
1044 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1045 | long res, unsigned int cflags); | |
ec9c02ad | 1046 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1047 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1048 | static void io_dismantle_req(struct io_kiocb *req); |
1049 | static void io_put_task(struct task_struct *task, int nr); | |
94ae5e77 JA |
1050 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1051 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
fdecb662 | 1052 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1053 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1054 | unsigned nr_args); |
68fb8979 | 1055 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1056 | static struct file *io_file_get(struct io_submit_state *state, |
1057 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1058 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1059 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1060 | |
907d1df3 | 1061 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1062 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1063 | struct io_ring_ctx *ctx); | |
5082620f | 1064 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1065 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1066 | |
2b188cc1 JA |
1067 | static struct kmem_cache *req_cachep; |
1068 | ||
0918682b | 1069 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1070 | |
1071 | struct sock *io_uring_get_socket(struct file *file) | |
1072 | { | |
1073 | #if defined(CONFIG_UNIX) | |
1074 | if (file->f_op == &io_uring_fops) { | |
1075 | struct io_ring_ctx *ctx = file->private_data; | |
1076 | ||
1077 | return ctx->ring_sock->sk; | |
1078 | } | |
1079 | #endif | |
1080 | return NULL; | |
1081 | } | |
1082 | EXPORT_SYMBOL(io_uring_get_socket); | |
1083 | ||
f2f87370 PB |
1084 | #define io_for_each_link(pos, head) \ |
1085 | for (pos = (head); pos; pos = pos->link) | |
1086 | ||
b895c9a6 | 1087 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1088 | { |
1089 | struct io_ring_ctx *ctx = req->ctx; | |
1090 | ||
269bbe5f | 1091 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1092 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1093 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1094 | } |
1095 | } | |
1096 | ||
f70865db PB |
1097 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1098 | { | |
1099 | bool got = percpu_ref_tryget(ref); | |
1100 | ||
1101 | /* already at zero, wait for ->release() */ | |
1102 | if (!got) | |
1103 | wait_for_completion(compl); | |
1104 | percpu_ref_resurrect(ref); | |
1105 | if (got) | |
1106 | percpu_ref_put(ref); | |
1107 | } | |
1108 | ||
3dd0c97a PB |
1109 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1110 | bool cancel_all) | |
08d23634 PB |
1111 | { |
1112 | struct io_kiocb *req; | |
1113 | ||
68207680 | 1114 | if (task && head->task != task) |
08d23634 | 1115 | return false; |
3dd0c97a | 1116 | if (cancel_all) |
08d23634 PB |
1117 | return true; |
1118 | ||
1119 | io_for_each_link(req, head) { | |
b05a1bcd | 1120 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1121 | return true; |
08d23634 PB |
1122 | } |
1123 | return false; | |
1124 | } | |
1125 | ||
93d2bcd2 | 1126 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1127 | { |
93d2bcd2 | 1128 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1129 | } |
4a38aed2 | 1130 | |
2b188cc1 JA |
1131 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1132 | { | |
1133 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1134 | ||
0f158b4c | 1135 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1136 | } |
1137 | ||
8eb7e2d0 PB |
1138 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1139 | { | |
1140 | return !req->timeout.off; | |
1141 | } | |
1142 | ||
2b188cc1 JA |
1143 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1144 | { | |
1145 | struct io_ring_ctx *ctx; | |
78076bb6 | 1146 | int hash_bits; |
2b188cc1 JA |
1147 | |
1148 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1149 | if (!ctx) | |
1150 | return NULL; | |
1151 | ||
78076bb6 JA |
1152 | /* |
1153 | * Use 5 bits less than the max cq entries, that should give us around | |
1154 | * 32 entries per hash list if totally full and uniformly spread. | |
1155 | */ | |
1156 | hash_bits = ilog2(p->cq_entries); | |
1157 | hash_bits -= 5; | |
1158 | if (hash_bits <= 0) | |
1159 | hash_bits = 1; | |
1160 | ctx->cancel_hash_bits = hash_bits; | |
1161 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1162 | GFP_KERNEL); | |
1163 | if (!ctx->cancel_hash) | |
1164 | goto err; | |
1165 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1166 | ||
6224843d PB |
1167 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1168 | if (!ctx->dummy_ubuf) | |
1169 | goto err; | |
1170 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1171 | ctx->dummy_ubuf->ubuf = -1UL; | |
1172 | ||
21482896 | 1173 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1174 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1175 | goto err; | |
2b188cc1 JA |
1176 | |
1177 | ctx->flags = p->flags; | |
90554200 | 1178 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1179 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1180 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1181 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1182 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1183 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1184 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1185 | mutex_init(&ctx->uring_lock); |
311997b3 | 1186 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1187 | spin_lock_init(&ctx->completion_lock); |
540e32a0 | 1188 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1189 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1190 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1191 | spin_lock_init(&ctx->rsrc_ref_lock); |
1192 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1193 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1194 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1195 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1196 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
d0acdee2 | 1197 | INIT_LIST_HEAD(&ctx->locked_free_list); |
2b188cc1 | 1198 | return ctx; |
206aefde | 1199 | err: |
6224843d | 1200 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1201 | kfree(ctx->cancel_hash); |
206aefde JA |
1202 | kfree(ctx); |
1203 | return NULL; | |
2b188cc1 JA |
1204 | } |
1205 | ||
8f6ed49a PB |
1206 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1207 | { | |
1208 | struct io_rings *r = ctx->rings; | |
1209 | ||
1210 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1211 | ctx->cq_extra--; | |
1212 | } | |
1213 | ||
9cf7c104 | 1214 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1215 | { |
2bc9930e JA |
1216 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1217 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1218 | |
8f6ed49a | 1219 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1220 | } |
de0617e4 | 1221 | |
9d858b21 | 1222 | return false; |
de0617e4 JA |
1223 | } |
1224 | ||
ce3d5aae PB |
1225 | static void io_req_track_inflight(struct io_kiocb *req) |
1226 | { | |
ce3d5aae | 1227 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1228 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1229 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1230 | } |
1231 | } | |
1232 | ||
1e6fa521 JA |
1233 | static void io_prep_async_work(struct io_kiocb *req) |
1234 | { | |
1235 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1236 | struct io_ring_ctx *ctx = req->ctx; |
1237 | ||
003e8dcc JA |
1238 | if (!req->work.creds) |
1239 | req->work.creds = get_current_cred(); | |
1240 | ||
e1d675df PB |
1241 | req->work.list.next = NULL; |
1242 | req->work.flags = 0; | |
feaadc4f PB |
1243 | if (req->flags & REQ_F_FORCE_ASYNC) |
1244 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1245 | ||
1e6fa521 JA |
1246 | if (req->flags & REQ_F_ISREG) { |
1247 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1248 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1249 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1250 | if (def->unbound_nonreg_file) |
1251 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1252 | } | |
e1d675df PB |
1253 | |
1254 | switch (req->opcode) { | |
1255 | case IORING_OP_SPLICE: | |
1256 | case IORING_OP_TEE: | |
e1d675df PB |
1257 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1258 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1259 | break; | |
1260 | } | |
561fb04a | 1261 | } |
cccf0ee8 | 1262 | |
cbdcb435 | 1263 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1264 | { |
cbdcb435 | 1265 | struct io_kiocb *cur; |
54a91f3b | 1266 | |
f2f87370 PB |
1267 | io_for_each_link(cur, req) |
1268 | io_prep_async_work(cur); | |
561fb04a JA |
1269 | } |
1270 | ||
ebf93667 | 1271 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1272 | { |
a197f664 | 1273 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1274 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1275 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1276 | |
3bfe6106 JA |
1277 | BUG_ON(!tctx); |
1278 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1279 | |
cbdcb435 PB |
1280 | /* init ->work of the whole link before punting */ |
1281 | io_prep_async_link(req); | |
d07f1e8a PB |
1282 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1283 | &req->work, req->flags); | |
ebf93667 | 1284 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1285 | if (link) |
1286 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1287 | } |
1288 | ||
1ee4160c | 1289 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1290 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1291 | { |
e8c2bc1f | 1292 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1293 | |
fd9c7bc5 | 1294 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1295 | atomic_set(&req->ctx->cq_timeouts, |
1296 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1297 | list_del_init(&req->timeout.list); |
d4d19c19 | 1298 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
216578e5 | 1299 | io_put_req_deferred(req, 1); |
5262f567 JA |
1300 | } |
1301 | } | |
1302 | ||
441b8a78 | 1303 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1304 | { |
441b8a78 | 1305 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1306 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1307 | struct io_defer_entry, list); | |
de0617e4 | 1308 | |
9cf7c104 | 1309 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1310 | break; |
27dc8338 | 1311 | list_del_init(&de->list); |
907d1df3 | 1312 | io_req_task_queue(de->req); |
27dc8338 | 1313 | kfree(de); |
441b8a78 | 1314 | } |
04518945 PB |
1315 | } |
1316 | ||
360428f8 | 1317 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1318 | { |
441b8a78 | 1319 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1320 | |
f18ee4cf | 1321 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1322 | u32 events_needed, events_got; |
360428f8 | 1323 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1324 | struct io_kiocb, timeout.list); |
de0617e4 | 1325 | |
8eb7e2d0 | 1326 | if (io_is_timeout_noseq(req)) |
360428f8 | 1327 | break; |
f010505b MDG |
1328 | |
1329 | /* | |
1330 | * Since seq can easily wrap around over time, subtract | |
1331 | * the last seq at which timeouts were flushed before comparing. | |
1332 | * Assuming not more than 2^31-1 events have happened since, | |
1333 | * these subtractions won't have wrapped, so we can check if | |
1334 | * target is in [last_seq, current_seq] by comparing the two. | |
1335 | */ | |
1336 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1337 | events_got = seq - ctx->cq_last_tm_flush; | |
1338 | if (events_got < events_needed) | |
360428f8 | 1339 | break; |
bfe68a22 | 1340 | |
135fcde8 | 1341 | list_del_init(&req->timeout.list); |
1ee4160c | 1342 | io_kill_timeout(req, 0); |
f18ee4cf | 1343 | } |
f010505b | 1344 | ctx->cq_last_tm_flush = seq; |
360428f8 | 1345 | } |
5262f567 | 1346 | |
360428f8 PB |
1347 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1348 | { | |
441b8a78 PB |
1349 | if (unlikely(ctx->off_timeout_used || ctx->drain_used)) { |
1350 | if (ctx->off_timeout_used) | |
1351 | io_flush_timeouts(ctx); | |
1352 | if (ctx->drain_used) | |
1353 | io_queue_deferred(ctx); | |
1354 | } | |
ec30e04b PB |
1355 | /* order cqe stores with ring update */ |
1356 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1357 | } |
1358 | ||
90554200 JA |
1359 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1360 | { | |
1361 | struct io_rings *r = ctx->rings; | |
1362 | ||
a566c556 | 1363 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1364 | } |
1365 | ||
888aae2e PB |
1366 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1367 | { | |
1368 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1369 | } | |
1370 | ||
d068b506 | 1371 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1372 | { |
75b28aff | 1373 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1374 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1375 | |
115e12e5 SB |
1376 | /* |
1377 | * writes to the cq entry need to come after reading head; the | |
1378 | * control dependency is enough as we're using WRITE_ONCE to | |
1379 | * fill the cq entry | |
1380 | */ | |
a566c556 | 1381 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1382 | return NULL; |
1383 | ||
888aae2e | 1384 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1385 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1386 | } |
1387 | ||
f2842ab5 JA |
1388 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1389 | { | |
44c769de | 1390 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1391 | return false; |
7e55a19c SG |
1392 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1393 | return false; | |
44c769de | 1394 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1395 | } |
1396 | ||
b41e9852 | 1397 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1398 | { |
b1445e59 PB |
1399 | /* see waitqueue_active() comment */ |
1400 | smp_mb(); | |
1401 | ||
311997b3 PB |
1402 | if (waitqueue_active(&ctx->cq_wait)) |
1403 | wake_up(&ctx->cq_wait); | |
534ca6d6 JA |
1404 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1405 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1406 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1407 | eventfd_signal(ctx->cq_ev_fd, 1); |
311997b3 PB |
1408 | if (waitqueue_active(&ctx->poll_wait)) { |
1409 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1410 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1411 | } | |
1d7bb1d5 JA |
1412 | } |
1413 | ||
80c18e4a PB |
1414 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1415 | { | |
b1445e59 PB |
1416 | /* see waitqueue_active() comment */ |
1417 | smp_mb(); | |
1418 | ||
80c18e4a | 1419 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
311997b3 PB |
1420 | if (waitqueue_active(&ctx->cq_wait)) |
1421 | wake_up(&ctx->cq_wait); | |
80c18e4a PB |
1422 | } |
1423 | if (io_should_trigger_evfd(ctx)) | |
1424 | eventfd_signal(ctx->cq_ev_fd, 1); | |
311997b3 PB |
1425 | if (waitqueue_active(&ctx->poll_wait)) { |
1426 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1427 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1428 | } | |
80c18e4a PB |
1429 | } |
1430 | ||
c4a2ed72 | 1431 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1432 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1433 | { |
1d7bb1d5 | 1434 | unsigned long flags; |
b18032bb | 1435 | bool all_flushed, posted; |
1d7bb1d5 | 1436 | |
a566c556 | 1437 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1438 | return false; |
1d7bb1d5 | 1439 | |
b18032bb | 1440 | posted = false; |
1d7bb1d5 | 1441 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae | 1442 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1443 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1444 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1445 | |
1d7bb1d5 JA |
1446 | if (!cqe && !force) |
1447 | break; | |
6c2450ae PB |
1448 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1449 | struct io_overflow_cqe, list); | |
1450 | if (cqe) | |
1451 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1452 | else | |
8f6ed49a PB |
1453 | io_account_cq_overflow(ctx); |
1454 | ||
b18032bb | 1455 | posted = true; |
6c2450ae PB |
1456 | list_del(&ocqe->list); |
1457 | kfree(ocqe); | |
1d7bb1d5 JA |
1458 | } |
1459 | ||
09e88404 PB |
1460 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1461 | if (all_flushed) { | |
5ed7a37d | 1462 | clear_bit(0, &ctx->check_cq_overflow); |
09e88404 PB |
1463 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; |
1464 | } | |
46930143 | 1465 | |
b18032bb JA |
1466 | if (posted) |
1467 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1468 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1469 | if (posted) |
1470 | io_cqring_ev_posted(ctx); | |
09e88404 | 1471 | return all_flushed; |
1d7bb1d5 JA |
1472 | } |
1473 | ||
6c2450ae | 1474 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1475 | { |
ca0a2651 JA |
1476 | bool ret = true; |
1477 | ||
5ed7a37d | 1478 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1479 | /* iopoll syncs against uring_lock, not completion_lock */ |
1480 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1481 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1482 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1483 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1484 | mutex_unlock(&ctx->uring_lock); | |
1485 | } | |
ca0a2651 JA |
1486 | |
1487 | return ret; | |
6c503150 PB |
1488 | } |
1489 | ||
abc54d63 JA |
1490 | /* |
1491 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1492 | * see commit f958d7b528b1 for details. | |
1493 | */ | |
1494 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1495 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1496 | ||
de9b4cca JA |
1497 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1498 | { | |
abc54d63 | 1499 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1500 | } |
1501 | ||
1502 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1503 | { | |
abc54d63 JA |
1504 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1505 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1506 | } |
1507 | ||
1508 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1509 | { | |
abc54d63 JA |
1510 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1511 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1512 | } |
1513 | ||
1514 | static inline void req_ref_put(struct io_kiocb *req) | |
1515 | { | |
abc54d63 | 1516 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1517 | } |
1518 | ||
1519 | static inline void req_ref_get(struct io_kiocb *req) | |
1520 | { | |
abc54d63 JA |
1521 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1522 | atomic_inc(&req->refs); | |
de9b4cca JA |
1523 | } |
1524 | ||
d4d19c19 PB |
1525 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1526 | long res, unsigned int cflags) | |
2b188cc1 | 1527 | { |
cce4b8b0 | 1528 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1529 | |
cce4b8b0 PB |
1530 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1531 | if (!ocqe) { | |
1532 | /* | |
1533 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1534 | * or cannot allocate an overflow entry, then we need to drop it | |
1535 | * on the floor. | |
1536 | */ | |
8f6ed49a | 1537 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1538 | return false; |
2b188cc1 | 1539 | } |
cce4b8b0 | 1540 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1541 | set_bit(0, &ctx->check_cq_overflow); |
cce4b8b0 PB |
1542 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
1543 | } | |
d4d19c19 | 1544 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1545 | ocqe->cqe.res = res; |
1546 | ocqe->cqe.flags = cflags; | |
1547 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1548 | return true; | |
2b188cc1 JA |
1549 | } |
1550 | ||
d4d19c19 PB |
1551 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1552 | long res, unsigned int cflags) | |
2b188cc1 JA |
1553 | { |
1554 | struct io_uring_cqe *cqe; | |
1555 | ||
d4d19c19 | 1556 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1557 | |
2b188cc1 JA |
1558 | /* |
1559 | * If we can't get a cq entry, userspace overflowed the | |
1560 | * submission (by quite a lot). Increment the overflow count in | |
1561 | * the ring. | |
1562 | */ | |
d068b506 | 1563 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1564 | if (likely(cqe)) { |
d4d19c19 | 1565 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1566 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1567 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1568 | return true; |
2b188cc1 | 1569 | } |
d4d19c19 | 1570 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1571 | } |
1572 | ||
8d13326e | 1573 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1574 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1575 | long res, unsigned int cflags) | |
bcda7baa | 1576 | { |
d4d19c19 | 1577 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1578 | } |
1579 | ||
7a612350 PB |
1580 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1581 | unsigned int cflags) | |
2b188cc1 | 1582 | { |
78e19bbe | 1583 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1584 | unsigned long flags; |
1585 | ||
1586 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
d4d19c19 | 1587 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1588 | /* |
1589 | * If we're the last reference to this request, add to our locked | |
1590 | * free_list cache. | |
1591 | */ | |
de9b4cca | 1592 | if (req_ref_put_and_test(req)) { |
7a612350 | 1593 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
93d2bcd2 | 1594 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) |
7a612350 PB |
1595 | io_disarm_next(req); |
1596 | if (req->link) { | |
1597 | io_req_task_queue(req->link); | |
1598 | req->link = NULL; | |
1599 | } | |
1600 | } | |
c7dae4ba JA |
1601 | io_dismantle_req(req); |
1602 | io_put_task(req->task, 1); | |
d0acdee2 PB |
1603 | list_add(&req->compl.list, &ctx->locked_free_list); |
1604 | ctx->locked_free_nr++; | |
180f829f PB |
1605 | } else { |
1606 | if (!percpu_ref_tryget(&ctx->refs)) | |
1607 | req = NULL; | |
1608 | } | |
7a612350 | 1609 | io_commit_cqring(ctx); |
2b188cc1 | 1610 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1611 | |
180f829f PB |
1612 | if (req) { |
1613 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1614 | percpu_ref_put(&ctx->refs); |
180f829f | 1615 | } |
229a7b63 JA |
1616 | } |
1617 | ||
4e3d9ff9 JA |
1618 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1619 | { | |
75652a30 | 1620 | return req->flags & (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | |
3a0a6902 | 1621 | REQ_F_POLLED | REQ_F_INFLIGHT); |
4e3d9ff9 JA |
1622 | } |
1623 | ||
a38d68db | 1624 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1625 | unsigned int cflags) |
229a7b63 | 1626 | { |
4e3d9ff9 | 1627 | if (io_req_needs_clean(req)) |
68fb8979 | 1628 | io_clean_op(req); |
a38d68db PB |
1629 | req->result = res; |
1630 | req->compl.cflags = cflags; | |
e342c807 | 1631 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1632 | } |
1633 | ||
889fca73 PB |
1634 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1635 | long res, unsigned cflags) | |
bcda7baa | 1636 | { |
889fca73 PB |
1637 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1638 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1639 | else |
c7dae4ba | 1640 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1641 | } |
1642 | ||
a38d68db | 1643 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1644 | { |
889fca73 | 1645 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1646 | } |
1647 | ||
f41db273 PB |
1648 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1649 | { | |
93d2bcd2 | 1650 | req_set_fail(req); |
f41db273 PB |
1651 | io_put_req(req); |
1652 | io_req_complete_post(req, res, 0); | |
1653 | } | |
1654 | ||
dac7a098 PB |
1655 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1656 | struct io_comp_state *cs) | |
1657 | { | |
1658 | spin_lock_irq(&ctx->completion_lock); | |
d0acdee2 PB |
1659 | list_splice_init(&ctx->locked_free_list, &cs->free_list); |
1660 | ctx->locked_free_nr = 0; | |
dac7a098 PB |
1661 | spin_unlock_irq(&ctx->completion_lock); |
1662 | } | |
1663 | ||
dd78f492 | 1664 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1665 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1666 | { |
c7dae4ba JA |
1667 | struct io_submit_state *state = &ctx->submit_state; |
1668 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1669 | int nr; |
0ddf92e8 | 1670 | |
c7dae4ba JA |
1671 | /* |
1672 | * If we have more than a batch's worth of requests in our IRQ side | |
1673 | * locked cache, grab the lock and move them over to our submission | |
1674 | * side cache. | |
1675 | */ | |
d0acdee2 | 1676 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
dac7a098 | 1677 | io_flush_cached_locked_reqs(ctx, cs); |
0ddf92e8 | 1678 | |
dd78f492 | 1679 | nr = state->free_reqs; |
c7dae4ba | 1680 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1681 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1682 | struct io_kiocb, compl.list); | |
1683 | ||
1b4c351f | 1684 | list_del(&req->compl.list); |
dd78f492 PB |
1685 | state->reqs[nr++] = req; |
1686 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1687 | break; |
1b4c351f JA |
1688 | } |
1689 | ||
dd78f492 PB |
1690 | state->free_reqs = nr; |
1691 | return nr != 0; | |
0ddf92e8 JA |
1692 | } |
1693 | ||
e5d1bc0a | 1694 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1695 | { |
e5d1bc0a PB |
1696 | struct io_submit_state *state = &ctx->submit_state; |
1697 | ||
1698 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1699 | ||
f6b6c7d6 | 1700 | if (!state->free_reqs) { |
291b2821 | 1701 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1702 | int ret; |
1703 | ||
c7dae4ba | 1704 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1705 | goto got_req; |
1706 | ||
bf019da7 PB |
1707 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1708 | state->reqs); | |
fd6fab2c JA |
1709 | |
1710 | /* | |
1711 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1712 | * retry single alloc to be on the safe side. | |
1713 | */ | |
1714 | if (unlikely(ret <= 0)) { | |
1715 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1716 | if (!state->reqs[0]) | |
3893f39f | 1717 | return NULL; |
fd6fab2c JA |
1718 | ret = 1; |
1719 | } | |
291b2821 | 1720 | state->free_reqs = ret; |
2b188cc1 | 1721 | } |
e5d1bc0a | 1722 | got_req: |
291b2821 PB |
1723 | state->free_reqs--; |
1724 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1725 | } |
1726 | ||
e1d767f0 | 1727 | static inline void io_put_file(struct file *file) |
8da11c19 | 1728 | { |
e1d767f0 | 1729 | if (file) |
8da11c19 PB |
1730 | fput(file); |
1731 | } | |
1732 | ||
4edf20f9 | 1733 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1734 | { |
094bae49 | 1735 | unsigned int flags = req->flags; |
929a3af9 | 1736 | |
3a0a6902 PB |
1737 | if (io_req_needs_clean(req)) |
1738 | io_clean_op(req); | |
e1d767f0 PB |
1739 | if (!(flags & REQ_F_FIXED_FILE)) |
1740 | io_put_file(req->file); | |
269bbe5f BM |
1741 | if (req->fixed_rsrc_refs) |
1742 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1743 | if (req->async_data) |
1744 | kfree(req->async_data); | |
003e8dcc JA |
1745 | if (req->work.creds) { |
1746 | put_cred(req->work.creds); | |
1747 | req->work.creds = NULL; | |
1748 | } | |
e65ef56d JA |
1749 | } |
1750 | ||
b23fcf47 | 1751 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1752 | static inline void io_put_task(struct task_struct *task, int nr) |
1753 | { | |
1754 | struct io_uring_task *tctx = task->io_uring; | |
1755 | ||
1756 | percpu_counter_sub(&tctx->inflight, nr); | |
1757 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1758 | wake_up(&tctx->wait); | |
1759 | put_task_struct_many(task, nr); | |
1760 | } | |
1761 | ||
216578e5 | 1762 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1763 | { |
51a4cc11 | 1764 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1765 | |
216578e5 | 1766 | io_dismantle_req(req); |
7c660731 | 1767 | io_put_task(req->task, 1); |
c6ca97b3 | 1768 | |
3893f39f | 1769 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1770 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1771 | } |
1772 | ||
f2f87370 PB |
1773 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1774 | { | |
1775 | struct io_kiocb *nxt = req->link; | |
1776 | ||
1777 | req->link = nxt->link; | |
1778 | nxt->link = NULL; | |
1779 | } | |
1780 | ||
33cc89a9 PB |
1781 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1782 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1783 | { |
33cc89a9 | 1784 | struct io_kiocb *link = req->link; |
f2f87370 | 1785 | |
900fad45 PB |
1786 | /* |
1787 | * Can happen if a linked timeout fired and link had been like | |
1788 | * req -> link t-out -> link t-out [-> ...] | |
1789 | */ | |
c9abd7ad PB |
1790 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1791 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1792 | |
f2f87370 | 1793 | io_remove_next_linked(req); |
90cd7e42 | 1794 | link->timeout.head = NULL; |
fd9c7bc5 | 1795 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
d4d19c19 PB |
1796 | io_cqring_fill_event(link->ctx, link->user_data, |
1797 | -ECANCELED, 0); | |
33cc89a9 | 1798 | io_put_req_deferred(link, 1); |
d4729fbd | 1799 | return true; |
c9abd7ad PB |
1800 | } |
1801 | } | |
d4729fbd | 1802 | return false; |
7c86ffee PB |
1803 | } |
1804 | ||
d148ca4b | 1805 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1806 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1807 | { |
33cc89a9 | 1808 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1809 | |
f2f87370 | 1810 | req->link = NULL; |
f2f87370 PB |
1811 | while (link) { |
1812 | nxt = link->link; | |
1813 | link->link = NULL; | |
2665abfd | 1814 | |
f2f87370 | 1815 | trace_io_uring_fail_link(req, link); |
d4d19c19 | 1816 | io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0); |
1575f21a | 1817 | io_put_req_deferred(link, 2); |
f2f87370 | 1818 | link = nxt; |
9e645e11 | 1819 | } |
33cc89a9 | 1820 | } |
9e645e11 | 1821 | |
33cc89a9 PB |
1822 | static bool io_disarm_next(struct io_kiocb *req) |
1823 | __must_hold(&req->ctx->completion_lock) | |
1824 | { | |
1825 | bool posted = false; | |
1826 | ||
1827 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1828 | posted = io_kill_linked_timeout(req); | |
93d2bcd2 | 1829 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 1830 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
1831 | posted |= (req->link != NULL); |
1832 | io_fail_links(req); | |
1833 | } | |
1834 | return posted; | |
9e645e11 JA |
1835 | } |
1836 | ||
3fa5e0f3 | 1837 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1838 | { |
33cc89a9 | 1839 | struct io_kiocb *nxt; |
944e58bf | 1840 | |
9e645e11 JA |
1841 | /* |
1842 | * If LINK is set, we have dependent requests in this chain. If we | |
1843 | * didn't fail this request, queue the first one up, moving any other | |
1844 | * dependencies to the next request. In case of failure, fail the rest | |
1845 | * of the chain. | |
1846 | */ | |
93d2bcd2 | 1847 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) { |
33cc89a9 PB |
1848 | struct io_ring_ctx *ctx = req->ctx; |
1849 | unsigned long flags; | |
1850 | bool posted; | |
1851 | ||
1852 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1853 | posted = io_disarm_next(req); | |
1854 | if (posted) | |
1855 | io_commit_cqring(req->ctx); | |
1856 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1857 | if (posted) | |
1858 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1859 | } |
33cc89a9 PB |
1860 | nxt = req->link; |
1861 | req->link = NULL; | |
1862 | return nxt; | |
4d7dd462 | 1863 | } |
9e645e11 | 1864 | |
f2f87370 | 1865 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1866 | { |
cdbff982 | 1867 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1868 | return NULL; |
1869 | return __io_req_find_next(req); | |
1870 | } | |
1871 | ||
2c32395d PB |
1872 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1873 | { | |
1874 | if (!ctx) | |
1875 | return; | |
1876 | if (ctx->submit_state.comp.nr) { | |
1877 | mutex_lock(&ctx->uring_lock); | |
1878 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1879 | mutex_unlock(&ctx->uring_lock); | |
1880 | } | |
1881 | percpu_ref_put(&ctx->refs); | |
1882 | } | |
1883 | ||
7cbf1722 | 1884 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1885 | { |
65453d1e | 1886 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1887 | struct io_wq_work_list list; |
1888 | struct io_wq_work_node *node; | |
c2c4c83c | 1889 | |
7cbf1722 JA |
1890 | if (wq_list_empty(&tctx->task_list)) |
1891 | return false; | |
6200b0ae | 1892 | |
0b81e80c | 1893 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1894 | list = tctx->task_list; |
1895 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1896 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1897 | |
7cbf1722 JA |
1898 | node = list.first; |
1899 | while (node) { | |
1900 | struct io_wq_work_node *next = node->next; | |
1901 | struct io_kiocb *req; | |
0ba9c9ed | 1902 | |
7cbf1722 | 1903 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1904 | if (req->ctx != ctx) { |
1905 | ctx_flush_and_put(ctx); | |
1906 | ctx = req->ctx; | |
1907 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1908 | } |
65453d1e | 1909 | |
2c32395d PB |
1910 | req->task_work.func(&req->task_work); |
1911 | node = next; | |
7cbf1722 JA |
1912 | } |
1913 | ||
2c32395d | 1914 | ctx_flush_and_put(ctx); |
7cbf1722 | 1915 | return list.first != NULL; |
c2c4c83c JA |
1916 | } |
1917 | ||
7cbf1722 | 1918 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1919 | { |
7cbf1722 | 1920 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1921 | |
1d5f360d JA |
1922 | clear_bit(0, &tctx->task_state); |
1923 | ||
7cbf1722 JA |
1924 | while (__tctx_task_work(tctx)) |
1925 | cond_resched(); | |
7cbf1722 JA |
1926 | } |
1927 | ||
c15b79de | 1928 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1929 | { |
c15b79de | 1930 | struct task_struct *tsk = req->task; |
7cbf1722 | 1931 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1932 | enum task_work_notify_mode notify; |
7cbf1722 | 1933 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1934 | unsigned long flags; |
c15b79de PB |
1935 | int ret = 0; |
1936 | ||
1937 | if (unlikely(tsk->flags & PF_EXITING)) | |
1938 | return -ESRCH; | |
7cbf1722 JA |
1939 | |
1940 | WARN_ON_ONCE(!tctx); | |
1941 | ||
0b81e80c | 1942 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1943 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1944 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1945 | |
1946 | /* task_work already pending, we're done */ | |
1947 | if (test_bit(0, &tctx->task_state) || | |
1948 | test_and_set_bit(0, &tctx->task_state)) | |
1949 | return 0; | |
1950 | ||
c15b79de PB |
1951 | /* |
1952 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1953 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1954 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1955 | * will do the job. | |
1956 | */ | |
1957 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1958 | ||
1959 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1960 | wake_up_process(tsk); | |
7cbf1722 | 1961 | return 0; |
c15b79de | 1962 | } |
7cbf1722 JA |
1963 | |
1964 | /* | |
1965 | * Slow path - we failed, find and delete work. if the work is not | |
1966 | * in the list, it got run and we're fine. | |
1967 | */ | |
0b81e80c | 1968 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1969 | wq_list_for_each(node, prev, &tctx->task_list) { |
1970 | if (&req->io_task_work.node == node) { | |
1971 | wq_list_del(&tctx->task_list, node, prev); | |
1972 | ret = 1; | |
1973 | break; | |
1974 | } | |
1975 | } | |
0b81e80c | 1976 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1977 | clear_bit(0, &tctx->task_state); |
1978 | return ret; | |
1979 | } | |
1980 | ||
9b465711 PB |
1981 | static bool io_run_task_work_head(struct callback_head **work_head) |
1982 | { | |
1983 | struct callback_head *work, *next; | |
1984 | bool executed = false; | |
1985 | ||
1986 | do { | |
1987 | work = xchg(work_head, NULL); | |
1988 | if (!work) | |
1989 | break; | |
1990 | ||
1991 | do { | |
1992 | next = work->next; | |
1993 | work->func(work); | |
1994 | work = next; | |
1995 | cond_resched(); | |
1996 | } while (work); | |
1997 | executed = true; | |
1998 | } while (1); | |
1999 | ||
2000 | return executed; | |
2001 | } | |
2002 | ||
2003 | static void io_task_work_add_head(struct callback_head **work_head, | |
2004 | struct callback_head *task_work) | |
eab30c4d | 2005 | { |
7c25c0d1 | 2006 | struct callback_head *head; |
eab30c4d | 2007 | |
7c25c0d1 | 2008 | do { |
9b465711 PB |
2009 | head = READ_ONCE(*work_head); |
2010 | task_work->next = head; | |
2011 | } while (cmpxchg(work_head, head, task_work) != head); | |
2012 | } | |
2013 | ||
2014 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2015 | task_work_func_t cb) | |
2016 | { | |
2017 | init_task_work(&req->task_work, cb); | |
2018 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2019 | } |
2020 | ||
c40f6379 JA |
2021 | static void io_req_task_cancel(struct callback_head *cb) |
2022 | { | |
2023 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2024 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2025 | |
e83acd7d | 2026 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2027 | mutex_lock(&ctx->uring_lock); |
2593553a | 2028 | io_req_complete_failed(req, req->result); |
792bb6eb | 2029 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2030 | } |
2031 | ||
2032 | static void __io_req_task_submit(struct io_kiocb *req) | |
2033 | { | |
2034 | struct io_ring_ctx *ctx = req->ctx; | |
2035 | ||
04fc6c80 | 2036 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2037 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2038 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2039 | __io_queue_sqe(req); |
81b6d05c | 2040 | else |
2593553a | 2041 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2042 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2043 | } |
2044 | ||
2045 | static void io_req_task_submit(struct callback_head *cb) | |
2046 | { | |
2047 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2048 | ||
2049 | __io_req_task_submit(req); | |
2050 | } | |
2051 | ||
2c4b8eb6 | 2052 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2053 | { |
2c4b8eb6 PB |
2054 | req->result = ret; |
2055 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2056 | |
2c4b8eb6 | 2057 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2058 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2059 | } |
2060 | ||
2c4b8eb6 | 2061 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2062 | { |
2c4b8eb6 | 2063 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2064 | |
2065 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2066 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2067 | } |
2068 | ||
f2f87370 | 2069 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2070 | { |
9b5f7bd9 | 2071 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2072 | |
2073 | if (nxt) | |
906a8c3f | 2074 | io_req_task_queue(nxt); |
c69f8dbe JL |
2075 | } |
2076 | ||
c3524383 | 2077 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2078 | { |
c3524383 PB |
2079 | io_queue_next(req); |
2080 | __io_free_req(req); | |
2081 | } | |
8766dd51 | 2082 | |
2d6500d4 | 2083 | struct req_batch { |
5af1d13e PB |
2084 | struct task_struct *task; |
2085 | int task_refs; | |
1b4c351f | 2086 | int ctx_refs; |
2d6500d4 PB |
2087 | }; |
2088 | ||
5af1d13e PB |
2089 | static inline void io_init_req_batch(struct req_batch *rb) |
2090 | { | |
5af1d13e | 2091 | rb->task_refs = 0; |
9ae72463 | 2092 | rb->ctx_refs = 0; |
5af1d13e PB |
2093 | rb->task = NULL; |
2094 | } | |
2095 | ||
2d6500d4 PB |
2096 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2097 | struct req_batch *rb) | |
2098 | { | |
6e833d53 | 2099 | if (rb->task) |
7c660731 | 2100 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2101 | if (rb->ctx_refs) |
2102 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2103 | } |
2104 | ||
6ff119a6 PB |
2105 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2106 | struct io_submit_state *state) | |
2d6500d4 | 2107 | { |
f2f87370 | 2108 | io_queue_next(req); |
96670657 | 2109 | io_dismantle_req(req); |
2d6500d4 | 2110 | |
e3bc8e9d | 2111 | if (req->task != rb->task) { |
7c660731 PB |
2112 | if (rb->task) |
2113 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2114 | rb->task = req->task; |
2115 | rb->task_refs = 0; | |
5af1d13e | 2116 | } |
e3bc8e9d | 2117 | rb->task_refs++; |
9ae72463 | 2118 | rb->ctx_refs++; |
5af1d13e | 2119 | |
bd759045 | 2120 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2121 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2122 | else |
2123 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2124 | } |
2125 | ||
905c172f PB |
2126 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2127 | struct io_ring_ctx *ctx) | |
2128 | { | |
2129 | int i, nr = cs->nr; | |
2130 | struct io_kiocb *req; | |
2131 | struct req_batch rb; | |
2132 | ||
2133 | io_init_req_batch(&rb); | |
2134 | spin_lock_irq(&ctx->completion_lock); | |
2135 | for (i = 0; i < nr; i++) { | |
2136 | req = cs->reqs[i]; | |
d4d19c19 PB |
2137 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2138 | req->compl.cflags); | |
905c172f PB |
2139 | } |
2140 | io_commit_cqring(ctx); | |
2141 | spin_unlock_irq(&ctx->completion_lock); | |
2142 | ||
2143 | io_cqring_ev_posted(ctx); | |
2144 | for (i = 0; i < nr; i++) { | |
2145 | req = cs->reqs[i]; | |
2146 | ||
2147 | /* submission and completion refs */ | |
de9b4cca | 2148 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2149 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2150 | } |
2151 | ||
2152 | io_req_free_batch_finish(ctx, &rb); | |
2153 | cs->nr = 0; | |
7a743e22 PB |
2154 | } |
2155 | ||
ba816ad6 JA |
2156 | /* |
2157 | * Drop reference to request, return next in chain (if there is one) if this | |
2158 | * was the last reference to this request. | |
2159 | */ | |
0d85035a | 2160 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2161 | { |
9b5f7bd9 PB |
2162 | struct io_kiocb *nxt = NULL; |
2163 | ||
de9b4cca | 2164 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2165 | nxt = io_req_find_next(req); |
4d7dd462 | 2166 | __io_free_req(req); |
2a44f467 | 2167 | } |
9b5f7bd9 | 2168 | return nxt; |
2b188cc1 JA |
2169 | } |
2170 | ||
0d85035a | 2171 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2172 | { |
de9b4cca | 2173 | if (req_ref_put_and_test(req)) |
e65ef56d | 2174 | io_free_req(req); |
2b188cc1 JA |
2175 | } |
2176 | ||
216578e5 PB |
2177 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2178 | { | |
2179 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2180 | ||
2181 | io_free_req(req); | |
2182 | } | |
2183 | ||
2184 | static void io_free_req_deferred(struct io_kiocb *req) | |
2185 | { | |
7cbf1722 | 2186 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2187 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2188 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2189 | } |
2190 | ||
2191 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2192 | { | |
de9b4cca | 2193 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2194 | io_free_req_deferred(req); |
2195 | } | |
2196 | ||
6c503150 | 2197 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2198 | { |
2199 | /* See comment at the top of this file */ | |
2200 | smp_rmb(); | |
e23de15f | 2201 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2202 | } |
2203 | ||
fb5ccc98 PB |
2204 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2205 | { | |
2206 | struct io_rings *rings = ctx->rings; | |
2207 | ||
2208 | /* make sure SQ entry isn't read before tail */ | |
2209 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2210 | } | |
2211 | ||
8ff069bf | 2212 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2213 | { |
8ff069bf | 2214 | unsigned int cflags; |
e94f141b | 2215 | |
bcda7baa JA |
2216 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2217 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2218 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2219 | kfree(kbuf); |
2220 | return cflags; | |
e94f141b JA |
2221 | } |
2222 | ||
8ff069bf | 2223 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2224 | { |
4d954c25 | 2225 | struct io_buffer *kbuf; |
bcda7baa | 2226 | |
4d954c25 | 2227 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2228 | return io_put_kbuf(req, kbuf); |
2229 | } | |
2230 | ||
4c6e277c JA |
2231 | static inline bool io_run_task_work(void) |
2232 | { | |
6200b0ae JA |
2233 | /* |
2234 | * Not safe to run on exiting task, and the task_work handling will | |
2235 | * not add work to such a task. | |
2236 | */ | |
2237 | if (unlikely(current->flags & PF_EXITING)) | |
2238 | return false; | |
4c6e277c JA |
2239 | if (current->task_works) { |
2240 | __set_current_state(TASK_RUNNING); | |
2241 | task_work_run(); | |
2242 | return true; | |
2243 | } | |
2244 | ||
2245 | return false; | |
bcda7baa JA |
2246 | } |
2247 | ||
def596e9 JA |
2248 | /* |
2249 | * Find and free completed poll iocbs | |
2250 | */ | |
2251 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2252 | struct list_head *done) | |
2253 | { | |
8237e045 | 2254 | struct req_batch rb; |
def596e9 | 2255 | struct io_kiocb *req; |
bbde017a XW |
2256 | |
2257 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2258 | smp_rmb(); | |
def596e9 | 2259 | |
5af1d13e | 2260 | io_init_req_batch(&rb); |
def596e9 | 2261 | while (!list_empty(done)) { |
bcda7baa JA |
2262 | int cflags = 0; |
2263 | ||
d21ffe7e | 2264 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2265 | list_del(&req->inflight_entry); |
2266 | ||
8c130827 PB |
2267 | if (READ_ONCE(req->result) == -EAGAIN && |
2268 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2269 | req->iopoll_completed = 0; |
8c130827 PB |
2270 | req_ref_get(req); |
2271 | io_queue_async_work(req); | |
2272 | continue; | |
bbde017a | 2273 | } |
def596e9 | 2274 | |
bcda7baa | 2275 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2276 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2277 | |
d4d19c19 | 2278 | __io_cqring_fill_event(ctx, req->user_data, req->result, cflags); |
def596e9 JA |
2279 | (*nr_events)++; |
2280 | ||
de9b4cca | 2281 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2282 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2283 | } |
def596e9 | 2284 | |
09bb8394 | 2285 | io_commit_cqring(ctx); |
80c18e4a | 2286 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2287 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2288 | } |
2289 | ||
def596e9 JA |
2290 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2291 | long min) | |
2292 | { | |
2293 | struct io_kiocb *req, *tmp; | |
2294 | LIST_HEAD(done); | |
2295 | bool spin; | |
2296 | int ret; | |
2297 | ||
2298 | /* | |
2299 | * Only spin for completions if we don't have multiple devices hanging | |
2300 | * off our complete list, and we're under the requested amount. | |
2301 | */ | |
2302 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2303 | ||
2304 | ret = 0; | |
d21ffe7e | 2305 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2306 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2307 | |
2308 | /* | |
581f9810 BM |
2309 | * Move completed and retryable entries to our local lists. |
2310 | * If we find a request that requires polling, break out | |
2311 | * and complete those lists first, if we have entries there. | |
def596e9 | 2312 | */ |
65a6543d | 2313 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2314 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2315 | continue; |
2316 | } | |
2317 | if (!list_empty(&done)) | |
2318 | break; | |
2319 | ||
2320 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2321 | if (ret < 0) | |
2322 | break; | |
2323 | ||
3aadc23e PB |
2324 | /* iopoll may have completed current req */ |
2325 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2326 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2327 | |
def596e9 JA |
2328 | if (ret && spin) |
2329 | spin = false; | |
2330 | ret = 0; | |
2331 | } | |
2332 | ||
2333 | if (!list_empty(&done)) | |
2334 | io_iopoll_complete(ctx, nr_events, &done); | |
2335 | ||
2336 | return ret; | |
2337 | } | |
2338 | ||
def596e9 JA |
2339 | /* |
2340 | * We can't just wait for polled events to come to us, we have to actively | |
2341 | * find and complete them. | |
2342 | */ | |
b2edc0a7 | 2343 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2344 | { |
2345 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2346 | return; | |
2347 | ||
2348 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2349 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2350 | unsigned int nr_events = 0; |
2351 | ||
b2edc0a7 | 2352 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2353 | |
b2edc0a7 PB |
2354 | /* let it sleep and repeat later if can't complete a request */ |
2355 | if (nr_events == 0) | |
2356 | break; | |
08f5439f JA |
2357 | /* |
2358 | * Ensure we allow local-to-the-cpu processing to take place, | |
2359 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2360 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2361 | */ |
3fcee5a6 PB |
2362 | if (need_resched()) { |
2363 | mutex_unlock(&ctx->uring_lock); | |
2364 | cond_resched(); | |
2365 | mutex_lock(&ctx->uring_lock); | |
2366 | } | |
def596e9 JA |
2367 | } |
2368 | mutex_unlock(&ctx->uring_lock); | |
2369 | } | |
2370 | ||
7668b92a | 2371 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2372 | { |
7668b92a | 2373 | unsigned int nr_events = 0; |
e9979b36 | 2374 | int ret = 0; |
500f9fba | 2375 | |
c7849be9 XW |
2376 | /* |
2377 | * We disallow the app entering submit/complete with polling, but we | |
2378 | * still need to lock the ring to prevent racing with polled issue | |
2379 | * that got punted to a workqueue. | |
2380 | */ | |
2381 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2382 | /* |
2383 | * Don't enter poll loop if we already have events pending. | |
2384 | * If we do, we can potentially be spinning for commands that | |
2385 | * already triggered a CQE (eg in error). | |
2386 | */ | |
5ed7a37d | 2387 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2388 | __io_cqring_overflow_flush(ctx, false); |
2389 | if (io_cqring_events(ctx)) | |
2390 | goto out; | |
def596e9 | 2391 | do { |
500f9fba JA |
2392 | /* |
2393 | * If a submit got punted to a workqueue, we can have the | |
2394 | * application entering polling for a command before it gets | |
2395 | * issued. That app will hold the uring_lock for the duration | |
2396 | * of the poll right here, so we need to take a breather every | |
2397 | * now and then to ensure that the issue has a chance to add | |
2398 | * the poll to the issued list. Otherwise we can spin here | |
2399 | * forever, while the workqueue is stuck trying to acquire the | |
2400 | * very same mutex. | |
2401 | */ | |
e9979b36 | 2402 | if (list_empty(&ctx->iopoll_list)) { |
500f9fba | 2403 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2404 | io_run_task_work(); |
500f9fba | 2405 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2406 | |
e9979b36 PB |
2407 | if (list_empty(&ctx->iopoll_list)) |
2408 | break; | |
500f9fba | 2409 | } |
f39c8a5b PB |
2410 | ret = io_do_iopoll(ctx, &nr_events, min); |
2411 | } while (!ret && nr_events < min && !need_resched()); | |
2412 | out: | |
500f9fba | 2413 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2414 | return ret; |
2415 | } | |
2416 | ||
491381ce | 2417 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2418 | { |
491381ce JA |
2419 | /* |
2420 | * Tell lockdep we inherited freeze protection from submission | |
2421 | * thread. | |
2422 | */ | |
2423 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2424 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2425 | |
1c98679d PB |
2426 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2427 | sb_end_write(sb); | |
2b188cc1 JA |
2428 | } |
2429 | } | |
2430 | ||
b63534c4 | 2431 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2432 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2433 | { |
ab454438 | 2434 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2435 | |
ab454438 PB |
2436 | if (!rw) |
2437 | return !io_req_prep_async(req); | |
2438 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2439 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2440 | return true; | |
b63534c4 | 2441 | } |
b63534c4 | 2442 | |
3e6a0d3c | 2443 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2444 | { |
355afaeb | 2445 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2446 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2447 | |
355afaeb JA |
2448 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2449 | return false; | |
3e6a0d3c JA |
2450 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2451 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2452 | return false; |
7c977a58 JA |
2453 | /* |
2454 | * If ref is dying, we might be running poll reap from the exit work. | |
2455 | * Don't attempt to reissue from that path, just let it fail with | |
2456 | * -EAGAIN. | |
2457 | */ | |
3e6a0d3c JA |
2458 | if (percpu_ref_is_dying(&ctx->refs)) |
2459 | return false; | |
2460 | return true; | |
2461 | } | |
e82ad485 | 2462 | #else |
a1ff1e3f | 2463 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2464 | { |
2465 | return false; | |
2466 | } | |
e82ad485 | 2467 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2468 | { |
b63534c4 JA |
2469 | return false; |
2470 | } | |
3e6a0d3c | 2471 | #endif |
b63534c4 | 2472 | |
a1d7c393 | 2473 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2474 | unsigned int issue_flags) |
a1d7c393 | 2475 | { |
2f8e45f1 PB |
2476 | int cflags = 0; |
2477 | ||
b65c128f PB |
2478 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2479 | kiocb_end_write(req); | |
9532b99b PB |
2480 | if (res != req->result) { |
2481 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2482 | io_rw_should_reissue(req)) { | |
2483 | req->flags |= REQ_F_REISSUE; | |
2484 | return; | |
2485 | } | |
93d2bcd2 | 2486 | req_set_fail(req); |
9532b99b | 2487 | } |
2f8e45f1 PB |
2488 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2489 | cflags = io_put_rw_kbuf(req); | |
2490 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2491 | } |
2492 | ||
2493 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2494 | { | |
9adbd45d | 2495 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2496 | |
889fca73 | 2497 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2498 | } |
2499 | ||
def596e9 JA |
2500 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2501 | { | |
9adbd45d | 2502 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2503 | |
491381ce JA |
2504 | if (kiocb->ki_flags & IOCB_WRITE) |
2505 | kiocb_end_write(req); | |
9532b99b | 2506 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2507 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2508 | io_resubmit_prep(req))) { | |
93d2bcd2 | 2509 | req_set_fail(req); |
9532b99b PB |
2510 | req->flags |= REQ_F_DONT_REISSUE; |
2511 | } | |
8c130827 | 2512 | } |
bbde017a XW |
2513 | |
2514 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2515 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2516 | smp_wmb(); |
2517 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2518 | } |
2519 | ||
2520 | /* | |
2521 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2522 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2523 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2524 | * accessing the kiocb cookie. |
2525 | */ | |
cb3d8972 | 2526 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2527 | { |
2528 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2529 | const bool in_async = io_wq_current_is_worker(); |
2530 | ||
2531 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2532 | if (unlikely(in_async)) | |
2533 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2534 | |
2535 | /* | |
2536 | * Track whether we have multiple files in our lists. This will impact | |
2537 | * how we do polling eventually, not spinning if we're on potentially | |
2538 | * different devices. | |
2539 | */ | |
540e32a0 | 2540 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2541 | ctx->poll_multi_file = false; |
2542 | } else if (!ctx->poll_multi_file) { | |
2543 | struct io_kiocb *list_req; | |
2544 | ||
540e32a0 | 2545 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2546 | inflight_entry); |
9adbd45d | 2547 | if (list_req->file != req->file) |
def596e9 JA |
2548 | ctx->poll_multi_file = true; |
2549 | } | |
2550 | ||
2551 | /* | |
2552 | * For fast devices, IO may have already completed. If it has, add | |
2553 | * it to the front so we find it first. | |
2554 | */ | |
65a6543d | 2555 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2556 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2557 | else |
d21ffe7e | 2558 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2559 | |
cb3d8972 PB |
2560 | if (unlikely(in_async)) { |
2561 | /* | |
2562 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2563 | * in sq thread task context or in io worker task context. If | |
2564 | * current task context is sq thread, we don't need to check | |
2565 | * whether should wake up sq thread. | |
2566 | */ | |
2567 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2568 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2569 | wake_up(&ctx->sq_data->wait); | |
2570 | ||
2571 | mutex_unlock(&ctx->uring_lock); | |
2572 | } | |
def596e9 JA |
2573 | } |
2574 | ||
9f13c35b PB |
2575 | static inline void io_state_file_put(struct io_submit_state *state) |
2576 | { | |
02b23a9a PB |
2577 | if (state->file_refs) { |
2578 | fput_many(state->file, state->file_refs); | |
2579 | state->file_refs = 0; | |
2580 | } | |
9a56a232 JA |
2581 | } |
2582 | ||
2583 | /* | |
2584 | * Get as many references to a file as we have IOs left in this submission, | |
2585 | * assuming most submissions are for one file, or at least that each file | |
2586 | * has more than one submission. | |
2587 | */ | |
8da11c19 | 2588 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2589 | { |
2590 | if (!state) | |
2591 | return fget(fd); | |
2592 | ||
6e1271e6 | 2593 | if (state->file_refs) { |
9a56a232 | 2594 | if (state->fd == fd) { |
6e1271e6 | 2595 | state->file_refs--; |
9a56a232 JA |
2596 | return state->file; |
2597 | } | |
02b23a9a | 2598 | io_state_file_put(state); |
9a56a232 JA |
2599 | } |
2600 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2601 | if (unlikely(!state->file)) |
9a56a232 JA |
2602 | return NULL; |
2603 | ||
2604 | state->fd = fd; | |
6e1271e6 | 2605 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2606 | return state->file; |
2607 | } | |
2608 | ||
4503b767 JA |
2609 | static bool io_bdev_nowait(struct block_device *bdev) |
2610 | { | |
9ba0d0c8 | 2611 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2612 | } |
2613 | ||
2b188cc1 JA |
2614 | /* |
2615 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2616 | * any file. For now, just ensure that anything potentially problematic is done | |
2617 | * inline. | |
2618 | */ | |
7b29f92d | 2619 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2620 | { |
2621 | umode_t mode = file_inode(file)->i_mode; | |
2622 | ||
4503b767 | 2623 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2624 | if (IS_ENABLED(CONFIG_BLOCK) && |
2625 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2626 | return true; |
2627 | return false; | |
2628 | } | |
976517f1 | 2629 | if (S_ISSOCK(mode)) |
2b188cc1 | 2630 | return true; |
4503b767 | 2631 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2632 | if (IS_ENABLED(CONFIG_BLOCK) && |
2633 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2634 | file->f_op != &io_uring_fops) |
2635 | return true; | |
2636 | return false; | |
2637 | } | |
2b188cc1 | 2638 | |
c5b85625 JA |
2639 | /* any ->read/write should understand O_NONBLOCK */ |
2640 | if (file->f_flags & O_NONBLOCK) | |
2641 | return true; | |
2642 | ||
af197f50 JA |
2643 | if (!(file->f_mode & FMODE_NOWAIT)) |
2644 | return false; | |
2645 | ||
2646 | if (rw == READ) | |
2647 | return file->f_op->read_iter != NULL; | |
2648 | ||
2649 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2650 | } |
2651 | ||
7b29f92d JA |
2652 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2653 | { | |
2654 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2655 | return true; | |
2656 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2657 | return true; | |
2658 | ||
2659 | return __io_file_supports_async(req->file, rw); | |
2660 | } | |
2661 | ||
a88fc400 | 2662 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2663 | { |
def596e9 | 2664 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2665 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2666 | struct file *file = req->file; |
09bb8394 JA |
2667 | unsigned ioprio; |
2668 | int ret; | |
2b188cc1 | 2669 | |
7b29f92d | 2670 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2671 | req->flags |= REQ_F_ISREG; |
2672 | ||
2b188cc1 | 2673 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2674 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2675 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2676 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2677 | } |
2b188cc1 | 2678 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2679 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2680 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2681 | if (unlikely(ret)) | |
2682 | return ret; | |
2b188cc1 | 2683 | |
75c668cd PB |
2684 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2685 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2686 | req->flags |= REQ_F_NOWAIT; | |
2687 | ||
2b188cc1 JA |
2688 | ioprio = READ_ONCE(sqe->ioprio); |
2689 | if (ioprio) { | |
2690 | ret = ioprio_check_cap(ioprio); | |
2691 | if (ret) | |
09bb8394 | 2692 | return ret; |
2b188cc1 JA |
2693 | |
2694 | kiocb->ki_ioprio = ioprio; | |
2695 | } else | |
2696 | kiocb->ki_ioprio = get_current_ioprio(); | |
2697 | ||
def596e9 | 2698 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2699 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2700 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2701 | return -EOPNOTSUPP; |
2b188cc1 | 2702 | |
def596e9 JA |
2703 | kiocb->ki_flags |= IOCB_HIPRI; |
2704 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2705 | req->iopoll_completed = 0; |
def596e9 | 2706 | } else { |
09bb8394 JA |
2707 | if (kiocb->ki_flags & IOCB_HIPRI) |
2708 | return -EINVAL; | |
def596e9 JA |
2709 | kiocb->ki_complete = io_complete_rw; |
2710 | } | |
9adbd45d | 2711 | |
eae071c9 PB |
2712 | if (req->opcode == IORING_OP_READ_FIXED || |
2713 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2714 | req->imu = NULL; | |
2715 | io_req_set_rsrc_node(req); | |
2716 | } | |
2717 | ||
3529d8c2 JA |
2718 | req->rw.addr = READ_ONCE(sqe->addr); |
2719 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2720 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2721 | return 0; |
2b188cc1 JA |
2722 | } |
2723 | ||
2724 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2725 | { | |
2726 | switch (ret) { | |
2727 | case -EIOCBQUEUED: | |
2728 | break; | |
2729 | case -ERESTARTSYS: | |
2730 | case -ERESTARTNOINTR: | |
2731 | case -ERESTARTNOHAND: | |
2732 | case -ERESTART_RESTARTBLOCK: | |
2733 | /* | |
2734 | * We can't just restart the syscall, since previously | |
2735 | * submitted sqes may already be in progress. Just fail this | |
2736 | * IO with EINTR. | |
2737 | */ | |
2738 | ret = -EINTR; | |
df561f66 | 2739 | fallthrough; |
2b188cc1 JA |
2740 | default: |
2741 | kiocb->ki_complete(kiocb, ret, 0); | |
2742 | } | |
2743 | } | |
2744 | ||
a1d7c393 | 2745 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2746 | unsigned int issue_flags) |
ba816ad6 | 2747 | { |
ba04291e | 2748 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2749 | struct io_async_rw *io = req->async_data; |
97284637 | 2750 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2751 | |
227c0c96 | 2752 | /* add previously done IO, if any */ |
e8c2bc1f | 2753 | if (io && io->bytes_done > 0) { |
227c0c96 | 2754 | if (ret < 0) |
e8c2bc1f | 2755 | ret = io->bytes_done; |
227c0c96 | 2756 | else |
e8c2bc1f | 2757 | ret += io->bytes_done; |
227c0c96 JA |
2758 | } |
2759 | ||
ba04291e JA |
2760 | if (req->flags & REQ_F_CUR_POS) |
2761 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2762 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2763 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2764 | else |
2765 | io_rw_done(kiocb, ret); | |
97284637 PB |
2766 | |
2767 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2768 | req->flags &= ~REQ_F_REISSUE; | |
a7be7c23 | 2769 | if (io_resubmit_prep(req)) { |
8c130827 PB |
2770 | req_ref_get(req); |
2771 | io_queue_async_work(req); | |
2772 | } else { | |
97284637 PB |
2773 | int cflags = 0; |
2774 | ||
93d2bcd2 | 2775 | req_set_fail(req); |
97284637 PB |
2776 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2777 | cflags = io_put_rw_kbuf(req); | |
2778 | __io_req_complete(req, issue_flags, ret, cflags); | |
2779 | } | |
2780 | } | |
ba816ad6 JA |
2781 | } |
2782 | ||
eae071c9 PB |
2783 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2784 | struct io_mapped_ubuf *imu) | |
edafccee | 2785 | { |
9adbd45d | 2786 | size_t len = req->rw.len; |
75769e3f | 2787 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2788 | size_t offset; |
edafccee | 2789 | |
75769e3f | 2790 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2791 | return -EFAULT; |
2792 | /* not inside the mapped region */ | |
4751f53d | 2793 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2794 | return -EFAULT; |
2795 | ||
2796 | /* | |
2797 | * May not be a start of buffer, set size appropriately | |
2798 | * and advance us to the beginning. | |
2799 | */ | |
2800 | offset = buf_addr - imu->ubuf; | |
2801 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2802 | |
2803 | if (offset) { | |
2804 | /* | |
2805 | * Don't use iov_iter_advance() here, as it's really slow for | |
2806 | * using the latter parts of a big fixed buffer - it iterates | |
2807 | * over each segment manually. We can cheat a bit here, because | |
2808 | * we know that: | |
2809 | * | |
2810 | * 1) it's a BVEC iter, we set it up | |
2811 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2812 | * first and last bvec | |
2813 | * | |
2814 | * So just find our index, and adjust the iterator afterwards. | |
2815 | * If the offset is within the first bvec (or the whole first | |
2816 | * bvec, just use iov_iter_advance(). This makes it easier | |
2817 | * since we can just skip the first segment, which may not | |
2818 | * be PAGE_SIZE aligned. | |
2819 | */ | |
2820 | const struct bio_vec *bvec = imu->bvec; | |
2821 | ||
2822 | if (offset <= bvec->bv_len) { | |
2823 | iov_iter_advance(iter, offset); | |
2824 | } else { | |
2825 | unsigned long seg_skip; | |
2826 | ||
2827 | /* skip first vec */ | |
2828 | offset -= bvec->bv_len; | |
2829 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2830 | ||
2831 | iter->bvec = bvec + seg_skip; | |
2832 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2833 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2834 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2835 | } |
2836 | } | |
2837 | ||
847595de | 2838 | return 0; |
edafccee JA |
2839 | } |
2840 | ||
eae071c9 PB |
2841 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
2842 | { | |
2843 | struct io_ring_ctx *ctx = req->ctx; | |
2844 | struct io_mapped_ubuf *imu = req->imu; | |
2845 | u16 index, buf_index = req->buf_index; | |
2846 | ||
2847 | if (likely(!imu)) { | |
2848 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
2849 | return -EFAULT; | |
2850 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2851 | imu = READ_ONCE(ctx->user_bufs[index]); | |
2852 | req->imu = imu; | |
2853 | } | |
2854 | return __io_import_fixed(req, rw, iter, imu); | |
2855 | } | |
2856 | ||
bcda7baa JA |
2857 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2858 | { | |
2859 | if (needs_lock) | |
2860 | mutex_unlock(&ctx->uring_lock); | |
2861 | } | |
2862 | ||
2863 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2864 | { | |
2865 | /* | |
2866 | * "Normal" inline submissions always hold the uring_lock, since we | |
2867 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2868 | * The only exception is when we've detached the request and issue it | |
2869 | * from an async worker thread, grab the lock for that case. | |
2870 | */ | |
2871 | if (needs_lock) | |
2872 | mutex_lock(&ctx->uring_lock); | |
2873 | } | |
2874 | ||
2875 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2876 | int bgid, struct io_buffer *kbuf, | |
2877 | bool needs_lock) | |
2878 | { | |
2879 | struct io_buffer *head; | |
2880 | ||
2881 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2882 | return kbuf; | |
2883 | ||
2884 | io_ring_submit_lock(req->ctx, needs_lock); | |
2885 | ||
2886 | lockdep_assert_held(&req->ctx->uring_lock); | |
2887 | ||
9e15c3a0 | 2888 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2889 | if (head) { |
2890 | if (!list_empty(&head->list)) { | |
2891 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2892 | list); | |
2893 | list_del(&kbuf->list); | |
2894 | } else { | |
2895 | kbuf = head; | |
9e15c3a0 | 2896 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2897 | } |
2898 | if (*len > kbuf->len) | |
2899 | *len = kbuf->len; | |
2900 | } else { | |
2901 | kbuf = ERR_PTR(-ENOBUFS); | |
2902 | } | |
2903 | ||
2904 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2905 | ||
2906 | return kbuf; | |
2907 | } | |
2908 | ||
4d954c25 JA |
2909 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2910 | bool needs_lock) | |
2911 | { | |
2912 | struct io_buffer *kbuf; | |
4f4eeba8 | 2913 | u16 bgid; |
4d954c25 JA |
2914 | |
2915 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2916 | bgid = req->buf_index; |
4d954c25 JA |
2917 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2918 | if (IS_ERR(kbuf)) | |
2919 | return kbuf; | |
2920 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2921 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2922 | return u64_to_user_ptr(kbuf->addr); | |
2923 | } | |
2924 | ||
2925 | #ifdef CONFIG_COMPAT | |
2926 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2927 | bool needs_lock) | |
2928 | { | |
2929 | struct compat_iovec __user *uiov; | |
2930 | compat_ssize_t clen; | |
2931 | void __user *buf; | |
2932 | ssize_t len; | |
2933 | ||
2934 | uiov = u64_to_user_ptr(req->rw.addr); | |
2935 | if (!access_ok(uiov, sizeof(*uiov))) | |
2936 | return -EFAULT; | |
2937 | if (__get_user(clen, &uiov->iov_len)) | |
2938 | return -EFAULT; | |
2939 | if (clen < 0) | |
2940 | return -EINVAL; | |
2941 | ||
2942 | len = clen; | |
2943 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2944 | if (IS_ERR(buf)) | |
2945 | return PTR_ERR(buf); | |
2946 | iov[0].iov_base = buf; | |
2947 | iov[0].iov_len = (compat_size_t) len; | |
2948 | return 0; | |
2949 | } | |
2950 | #endif | |
2951 | ||
2952 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2953 | bool needs_lock) | |
2954 | { | |
2955 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2956 | void __user *buf; | |
2957 | ssize_t len; | |
2958 | ||
2959 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2960 | return -EFAULT; | |
2961 | ||
2962 | len = iov[0].iov_len; | |
2963 | if (len < 0) | |
2964 | return -EINVAL; | |
2965 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2966 | if (IS_ERR(buf)) | |
2967 | return PTR_ERR(buf); | |
2968 | iov[0].iov_base = buf; | |
2969 | iov[0].iov_len = len; | |
2970 | return 0; | |
2971 | } | |
2972 | ||
2973 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2974 | bool needs_lock) | |
2975 | { | |
dddb3e26 JA |
2976 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2977 | struct io_buffer *kbuf; | |
2978 | ||
2979 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2980 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2981 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2982 | return 0; |
dddb3e26 | 2983 | } |
dd201662 | 2984 | if (req->rw.len != 1) |
4d954c25 JA |
2985 | return -EINVAL; |
2986 | ||
2987 | #ifdef CONFIG_COMPAT | |
2988 | if (req->ctx->compat) | |
2989 | return io_compat_import(req, iov, needs_lock); | |
2990 | #endif | |
2991 | ||
2992 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2993 | } | |
2994 | ||
847595de PB |
2995 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2996 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2997 | { |
9adbd45d JA |
2998 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2999 | size_t sqe_len = req->rw.len; | |
847595de | 3000 | u8 opcode = req->opcode; |
4d954c25 | 3001 | ssize_t ret; |
edafccee | 3002 | |
7d009165 | 3003 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3004 | *iovec = NULL; |
9adbd45d | 3005 | return io_import_fixed(req, rw, iter); |
edafccee | 3006 | } |
2b188cc1 | 3007 | |
bcda7baa | 3008 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3009 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3010 | return -EINVAL; |
3011 | ||
3a6820f2 | 3012 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3013 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3014 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3015 | if (IS_ERR(buf)) |
4d954c25 | 3016 | return PTR_ERR(buf); |
3f9d6441 | 3017 | req->rw.len = sqe_len; |
bcda7baa JA |
3018 | } |
3019 | ||
3a6820f2 JA |
3020 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3021 | *iovec = NULL; | |
10fc72e4 | 3022 | return ret; |
3a6820f2 JA |
3023 | } |
3024 | ||
4d954c25 JA |
3025 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3026 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3027 | if (!ret) |
3028 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3029 | *iovec = NULL; |
3030 | return ret; | |
3031 | } | |
3032 | ||
89cd35c5 CH |
3033 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3034 | req->ctx->compat); | |
2b188cc1 JA |
3035 | } |
3036 | ||
0fef9483 JA |
3037 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3038 | { | |
5b09e37e | 3039 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3040 | } |
3041 | ||
31b51510 | 3042 | /* |
32960613 JA |
3043 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3044 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3045 | */ |
4017eb91 | 3046 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3047 | { |
4017eb91 JA |
3048 | struct kiocb *kiocb = &req->rw.kiocb; |
3049 | struct file *file = req->file; | |
32960613 JA |
3050 | ssize_t ret = 0; |
3051 | ||
3052 | /* | |
3053 | * Don't support polled IO through this interface, and we can't | |
3054 | * support non-blocking either. For the latter, this just causes | |
3055 | * the kiocb to be handled from an async context. | |
3056 | */ | |
3057 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3058 | return -EOPNOTSUPP; | |
3059 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3060 | return -EAGAIN; | |
3061 | ||
3062 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3063 | struct iovec iovec; |
32960613 JA |
3064 | ssize_t nr; |
3065 | ||
311ae9e1 PB |
3066 | if (!iov_iter_is_bvec(iter)) { |
3067 | iovec = iov_iter_iovec(iter); | |
3068 | } else { | |
4017eb91 JA |
3069 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3070 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3071 | } |
3072 | ||
32960613 JA |
3073 | if (rw == READ) { |
3074 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3075 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3076 | } else { |
3077 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3078 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3079 | } |
3080 | ||
3081 | if (nr < 0) { | |
3082 | if (!ret) | |
3083 | ret = nr; | |
3084 | break; | |
3085 | } | |
3086 | ret += nr; | |
3087 | if (nr != iovec.iov_len) | |
3088 | break; | |
4017eb91 JA |
3089 | req->rw.len -= nr; |
3090 | req->rw.addr += nr; | |
32960613 JA |
3091 | iov_iter_advance(iter, nr); |
3092 | } | |
3093 | ||
3094 | return ret; | |
3095 | } | |
3096 | ||
ff6165b2 JA |
3097 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3098 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3099 | { |
e8c2bc1f | 3100 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3101 | |
ff6165b2 | 3102 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3103 | rw->free_iovec = iovec; |
227c0c96 | 3104 | rw->bytes_done = 0; |
ff6165b2 | 3105 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3106 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3107 | return; |
b64e3444 | 3108 | if (!iovec) { |
ff6165b2 JA |
3109 | unsigned iov_off = 0; |
3110 | ||
3111 | rw->iter.iov = rw->fast_iov; | |
3112 | if (iter->iov != fast_iov) { | |
3113 | iov_off = iter->iov - fast_iov; | |
3114 | rw->iter.iov += iov_off; | |
3115 | } | |
3116 | if (rw->fast_iov != fast_iov) | |
3117 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3118 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3119 | } else { |
3120 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3121 | } |
3122 | } | |
3123 | ||
6cb78689 | 3124 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3125 | { |
e8c2bc1f JA |
3126 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3127 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3128 | return req->async_data == NULL; | |
3d9932a8 XW |
3129 | } |
3130 | ||
ff6165b2 JA |
3131 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3132 | const struct iovec *fast_iov, | |
227c0c96 | 3133 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3134 | { |
26f0505a | 3135 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3136 | return 0; |
e8c2bc1f | 3137 | if (!req->async_data) { |
6cb78689 | 3138 | if (io_alloc_async_data(req)) { |
6bf985dc | 3139 | kfree(iovec); |
5d204bcf | 3140 | return -ENOMEM; |
6bf985dc | 3141 | } |
b7bb4f7d | 3142 | |
ff6165b2 | 3143 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3144 | } |
b7bb4f7d | 3145 | return 0; |
f67676d1 JA |
3146 | } |
3147 | ||
73debe68 | 3148 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3149 | { |
e8c2bc1f | 3150 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3151 | struct iovec *iov = iorw->fast_iov; |
847595de | 3152 | int ret; |
c3e330a4 | 3153 | |
2846c481 | 3154 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3155 | if (unlikely(ret < 0)) |
3156 | return ret; | |
3157 | ||
ab0b196c PB |
3158 | iorw->bytes_done = 0; |
3159 | iorw->free_iovec = iov; | |
3160 | if (iov) | |
3161 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3162 | return 0; |
3163 | } | |
3164 | ||
73debe68 | 3165 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3166 | { |
3529d8c2 JA |
3167 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3168 | return -EBADF; | |
93642ef8 | 3169 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3170 | } |
3171 | ||
c1dd91d1 JA |
3172 | /* |
3173 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3174 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3175 | * This gets called when the page is unlocked, and we generally expect that to | |
3176 | * happen when the page IO is completed and the page is now uptodate. This will | |
3177 | * queue a task_work based retry of the operation, attempting to copy the data | |
3178 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3179 | * do a thread based blocking retry of the operation. That's the unexpected | |
3180 | * slow path. | |
3181 | */ | |
bcf5a063 JA |
3182 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3183 | int sync, void *arg) | |
3184 | { | |
3185 | struct wait_page_queue *wpq; | |
3186 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3187 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3188 | |
3189 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3190 | ||
cdc8fcb4 LT |
3191 | if (!wake_page_match(wpq, key)) |
3192 | return 0; | |
3193 | ||
c8d317aa | 3194 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3195 | list_del_init(&wait->entry); |
3196 | ||
bcf5a063 | 3197 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3198 | req_ref_get(req); |
921b9054 | 3199 | io_req_task_queue(req); |
bcf5a063 JA |
3200 | return 1; |
3201 | } | |
3202 | ||
c1dd91d1 JA |
3203 | /* |
3204 | * This controls whether a given IO request should be armed for async page | |
3205 | * based retry. If we return false here, the request is handed to the async | |
3206 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3207 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3208 | * will either succeed because the page is now uptodate and unlocked, or it | |
3209 | * will register a callback when the page is unlocked at IO completion. Through | |
3210 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3211 | * That retry will attempt the buffered read again. The retry will generally | |
3212 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3213 | * async worker threads for a blocking retry. | |
3214 | */ | |
227c0c96 | 3215 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3216 | { |
e8c2bc1f JA |
3217 | struct io_async_rw *rw = req->async_data; |
3218 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3219 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3220 | |
bcf5a063 JA |
3221 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3222 | if (req->flags & REQ_F_NOWAIT) | |
3223 | return false; | |
f67676d1 | 3224 | |
227c0c96 | 3225 | /* Only for buffered IO */ |
3b2a4439 | 3226 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3227 | return false; |
3b2a4439 | 3228 | |
bcf5a063 JA |
3229 | /* |
3230 | * just use poll if we can, and don't attempt if the fs doesn't | |
3231 | * support callback based unlocks | |
3232 | */ | |
3233 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3234 | return false; | |
f67676d1 | 3235 | |
3b2a4439 JA |
3236 | wait->wait.func = io_async_buf_func; |
3237 | wait->wait.private = req; | |
3238 | wait->wait.flags = 0; | |
3239 | INIT_LIST_HEAD(&wait->wait.entry); | |
3240 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3241 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3242 | kiocb->ki_waitq = wait; |
3b2a4439 | 3243 | return true; |
bcf5a063 JA |
3244 | } |
3245 | ||
aeab9506 | 3246 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3247 | { |
3248 | if (req->file->f_op->read_iter) | |
3249 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3250 | else if (req->file->f_op->read) |
4017eb91 | 3251 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3252 | else |
3253 | return -EINVAL; | |
f67676d1 JA |
3254 | } |
3255 | ||
889fca73 | 3256 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3257 | { |
3258 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3259 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3260 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3261 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3262 | ssize_t io_size, ret, ret2; |
45d189c6 | 3263 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3264 | |
2846c481 | 3265 | if (rw) { |
e8c2bc1f | 3266 | iter = &rw->iter; |
2846c481 PB |
3267 | iovec = NULL; |
3268 | } else { | |
3269 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3270 | if (ret < 0) | |
3271 | return ret; | |
3272 | } | |
632546c4 | 3273 | io_size = iov_iter_count(iter); |
fa15bafb | 3274 | req->result = io_size; |
2b188cc1 | 3275 | |
fd6c2e4c JA |
3276 | /* Ensure we clear previously set non-block flag */ |
3277 | if (!force_nonblock) | |
29de5f6a | 3278 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3279 | else |
3280 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3281 | ||
24c74678 | 3282 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3283 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3284 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3285 | return ret ?: -EAGAIN; |
6713e7a6 | 3286 | } |
9e645e11 | 3287 | |
632546c4 | 3288 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3289 | if (unlikely(ret)) { |
3290 | kfree(iovec); | |
3291 | return ret; | |
3292 | } | |
2b188cc1 | 3293 | |
227c0c96 | 3294 | ret = io_iter_do_read(req, iter); |
32960613 | 3295 | |
230d50d4 | 3296 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3297 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3298 | /* IOPOLL retry should happen for io-wq threads */ |
3299 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3300 | goto done; |
75c668cd PB |
3301 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3302 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3303 | goto done; |
84216315 | 3304 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3305 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3306 | ret = 0; |
230d50d4 JA |
3307 | } else if (ret == -EIOCBQUEUED) { |
3308 | goto out_free; | |
7335e3bf | 3309 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3310 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3311 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3312 | goto done; |
227c0c96 JA |
3313 | } |
3314 | ||
227c0c96 | 3315 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3316 | if (ret2) |
3317 | return ret2; | |
3318 | ||
fe1cdd55 | 3319 | iovec = NULL; |
e8c2bc1f | 3320 | rw = req->async_data; |
227c0c96 | 3321 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3322 | iter = &rw->iter; |
227c0c96 | 3323 | |
b23df91b PB |
3324 | do { |
3325 | io_size -= ret; | |
3326 | rw->bytes_done += ret; | |
3327 | /* if we can retry, do so with the callbacks armed */ | |
3328 | if (!io_rw_should_retry(req)) { | |
3329 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3330 | return -EAGAIN; | |
3331 | } | |
3332 | ||
3333 | /* | |
3334 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3335 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3336 | * desired page gets unlocked. We can also get a partial read | |
3337 | * here, and if we do, then just retry at the new offset. | |
3338 | */ | |
3339 | ret = io_iter_do_read(req, iter); | |
3340 | if (ret == -EIOCBQUEUED) | |
3341 | return 0; | |
227c0c96 | 3342 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3343 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3344 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3345 | done: |
889fca73 | 3346 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3347 | out_free: |
3348 | /* it's faster to check here then delegate to kfree */ | |
3349 | if (iovec) | |
3350 | kfree(iovec); | |
5ea5dd45 | 3351 | return 0; |
2b188cc1 JA |
3352 | } |
3353 | ||
73debe68 | 3354 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3355 | { |
3529d8c2 JA |
3356 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3357 | return -EBADF; | |
93642ef8 | 3358 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3359 | } |
3360 | ||
889fca73 | 3361 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3362 | { |
3363 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3364 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3365 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3366 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3367 | ssize_t ret, ret2, io_size; |
45d189c6 | 3368 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3369 | |
2846c481 | 3370 | if (rw) { |
e8c2bc1f | 3371 | iter = &rw->iter; |
2846c481 PB |
3372 | iovec = NULL; |
3373 | } else { | |
3374 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3375 | if (ret < 0) | |
3376 | return ret; | |
3377 | } | |
632546c4 | 3378 | io_size = iov_iter_count(iter); |
fa15bafb | 3379 | req->result = io_size; |
2b188cc1 | 3380 | |
fd6c2e4c JA |
3381 | /* Ensure we clear previously set non-block flag */ |
3382 | if (!force_nonblock) | |
a88fc400 PB |
3383 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3384 | else | |
3385 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3386 | |
24c74678 | 3387 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3388 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3389 | goto copy_iov; |
31b51510 | 3390 | |
10d59345 JA |
3391 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3392 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3393 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3394 | goto copy_iov; |
31b51510 | 3395 | |
632546c4 | 3396 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3397 | if (unlikely(ret)) |
3398 | goto out_free; | |
4ed734b0 | 3399 | |
fa15bafb PB |
3400 | /* |
3401 | * Open-code file_start_write here to grab freeze protection, | |
3402 | * which will be released by another thread in | |
3403 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3404 | * released so that it doesn't complain about the held lock when | |
3405 | * we return to userspace. | |
3406 | */ | |
3407 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3408 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3409 | __sb_writers_release(file_inode(req->file)->i_sb, |
3410 | SB_FREEZE_WRITE); | |
3411 | } | |
3412 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3413 | |
fa15bafb | 3414 | if (req->file->f_op->write_iter) |
ff6165b2 | 3415 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3416 | else if (req->file->f_op->write) |
4017eb91 | 3417 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3418 | else |
3419 | ret2 = -EINVAL; | |
4ed734b0 | 3420 | |
6ad7f233 PB |
3421 | if (req->flags & REQ_F_REISSUE) { |
3422 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3423 | ret2 = -EAGAIN; |
6ad7f233 | 3424 | } |
230d50d4 | 3425 | |
fa15bafb PB |
3426 | /* |
3427 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3428 | * retry them without IOCB_NOWAIT. | |
3429 | */ | |
3430 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3431 | ret2 = -EAGAIN; | |
75c668cd PB |
3432 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3433 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3434 | goto done; |
fa15bafb | 3435 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3436 | /* IOPOLL retry should happen for io-wq threads */ |
3437 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3438 | goto copy_iov; | |
355afaeb | 3439 | done: |
889fca73 | 3440 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3441 | } else { |
f67676d1 | 3442 | copy_iov: |
84216315 | 3443 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3444 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3445 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3446 | return ret ?: -EAGAIN; |
2b188cc1 | 3447 | } |
31b51510 | 3448 | out_free: |
f261c168 | 3449 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3450 | if (iovec) |
6f2cc166 | 3451 | kfree(iovec); |
2b188cc1 JA |
3452 | return ret; |
3453 | } | |
3454 | ||
80a261fd JA |
3455 | static int io_renameat_prep(struct io_kiocb *req, |
3456 | const struct io_uring_sqe *sqe) | |
3457 | { | |
3458 | struct io_rename *ren = &req->rename; | |
3459 | const char __user *oldf, *newf; | |
3460 | ||
3461 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3462 | return -EBADF; | |
3463 | ||
3464 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3465 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3466 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3467 | ren->new_dfd = READ_ONCE(sqe->len); | |
3468 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3469 | ||
3470 | ren->oldpath = getname(oldf); | |
3471 | if (IS_ERR(ren->oldpath)) | |
3472 | return PTR_ERR(ren->oldpath); | |
3473 | ||
3474 | ren->newpath = getname(newf); | |
3475 | if (IS_ERR(ren->newpath)) { | |
3476 | putname(ren->oldpath); | |
3477 | return PTR_ERR(ren->newpath); | |
3478 | } | |
3479 | ||
3480 | req->flags |= REQ_F_NEED_CLEANUP; | |
3481 | return 0; | |
3482 | } | |
3483 | ||
45d189c6 | 3484 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3485 | { |
3486 | struct io_rename *ren = &req->rename; | |
3487 | int ret; | |
3488 | ||
45d189c6 | 3489 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3490 | return -EAGAIN; |
3491 | ||
3492 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3493 | ren->newpath, ren->flags); | |
3494 | ||
3495 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3496 | if (ret < 0) | |
93d2bcd2 | 3497 | req_set_fail(req); |
80a261fd JA |
3498 | io_req_complete(req, ret); |
3499 | return 0; | |
3500 | } | |
3501 | ||
14a1143b JA |
3502 | static int io_unlinkat_prep(struct io_kiocb *req, |
3503 | const struct io_uring_sqe *sqe) | |
3504 | { | |
3505 | struct io_unlink *un = &req->unlink; | |
3506 | const char __user *fname; | |
3507 | ||
3508 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3509 | return -EBADF; | |
3510 | ||
3511 | un->dfd = READ_ONCE(sqe->fd); | |
3512 | ||
3513 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3514 | if (un->flags & ~AT_REMOVEDIR) | |
3515 | return -EINVAL; | |
3516 | ||
3517 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3518 | un->filename = getname(fname); | |
3519 | if (IS_ERR(un->filename)) | |
3520 | return PTR_ERR(un->filename); | |
3521 | ||
3522 | req->flags |= REQ_F_NEED_CLEANUP; | |
3523 | return 0; | |
3524 | } | |
3525 | ||
45d189c6 | 3526 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3527 | { |
3528 | struct io_unlink *un = &req->unlink; | |
3529 | int ret; | |
3530 | ||
45d189c6 | 3531 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3532 | return -EAGAIN; |
3533 | ||
3534 | if (un->flags & AT_REMOVEDIR) | |
3535 | ret = do_rmdir(un->dfd, un->filename); | |
3536 | else | |
3537 | ret = do_unlinkat(un->dfd, un->filename); | |
3538 | ||
3539 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3540 | if (ret < 0) | |
93d2bcd2 | 3541 | req_set_fail(req); |
14a1143b JA |
3542 | io_req_complete(req, ret); |
3543 | return 0; | |
3544 | } | |
3545 | ||
36f4fa68 JA |
3546 | static int io_shutdown_prep(struct io_kiocb *req, |
3547 | const struct io_uring_sqe *sqe) | |
3548 | { | |
3549 | #if defined(CONFIG_NET) | |
3550 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3551 | return -EINVAL; | |
3552 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3553 | sqe->buf_index) | |
3554 | return -EINVAL; | |
3555 | ||
3556 | req->shutdown.how = READ_ONCE(sqe->len); | |
3557 | return 0; | |
3558 | #else | |
3559 | return -EOPNOTSUPP; | |
3560 | #endif | |
3561 | } | |
3562 | ||
45d189c6 | 3563 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3564 | { |
3565 | #if defined(CONFIG_NET) | |
3566 | struct socket *sock; | |
3567 | int ret; | |
3568 | ||
45d189c6 | 3569 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3570 | return -EAGAIN; |
3571 | ||
48aba79b | 3572 | sock = sock_from_file(req->file); |
36f4fa68 | 3573 | if (unlikely(!sock)) |
48aba79b | 3574 | return -ENOTSOCK; |
36f4fa68 JA |
3575 | |
3576 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3577 | if (ret < 0) |
93d2bcd2 | 3578 | req_set_fail(req); |
36f4fa68 JA |
3579 | io_req_complete(req, ret); |
3580 | return 0; | |
3581 | #else | |
3582 | return -EOPNOTSUPP; | |
3583 | #endif | |
3584 | } | |
3585 | ||
f2a8d5c7 PB |
3586 | static int __io_splice_prep(struct io_kiocb *req, |
3587 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3588 | { |
3589 | struct io_splice* sp = &req->splice; | |
3590 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3591 | |
3232dd02 PB |
3592 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3593 | return -EINVAL; | |
7d67af2c PB |
3594 | |
3595 | sp->file_in = NULL; | |
7d67af2c PB |
3596 | sp->len = READ_ONCE(sqe->len); |
3597 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3598 | ||
3599 | if (unlikely(sp->flags & ~valid_flags)) | |
3600 | return -EINVAL; | |
3601 | ||
8371adf5 PB |
3602 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3603 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3604 | if (!sp->file_in) | |
3605 | return -EBADF; | |
7d67af2c | 3606 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3607 | return 0; |
3608 | } | |
3609 | ||
f2a8d5c7 PB |
3610 | static int io_tee_prep(struct io_kiocb *req, |
3611 | const struct io_uring_sqe *sqe) | |
3612 | { | |
3613 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3614 | return -EINVAL; | |
3615 | return __io_splice_prep(req, sqe); | |
3616 | } | |
3617 | ||
45d189c6 | 3618 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3619 | { |
3620 | struct io_splice *sp = &req->splice; | |
3621 | struct file *in = sp->file_in; | |
3622 | struct file *out = sp->file_out; | |
3623 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3624 | long ret = 0; | |
3625 | ||
45d189c6 | 3626 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3627 | return -EAGAIN; |
3628 | if (sp->len) | |
3629 | ret = do_tee(in, out, sp->len, flags); | |
3630 | ||
e1d767f0 PB |
3631 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3632 | io_put_file(in); | |
f2a8d5c7 PB |
3633 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3634 | ||
f2a8d5c7 | 3635 | if (ret != sp->len) |
93d2bcd2 | 3636 | req_set_fail(req); |
e1e16097 | 3637 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3638 | return 0; |
3639 | } | |
3640 | ||
3641 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3642 | { | |
3643 | struct io_splice* sp = &req->splice; | |
3644 | ||
3645 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3646 | sp->off_out = READ_ONCE(sqe->off); | |
3647 | return __io_splice_prep(req, sqe); | |
3648 | } | |
3649 | ||
45d189c6 | 3650 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3651 | { |
3652 | struct io_splice *sp = &req->splice; | |
3653 | struct file *in = sp->file_in; | |
3654 | struct file *out = sp->file_out; | |
3655 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3656 | loff_t *poff_in, *poff_out; | |
c9687426 | 3657 | long ret = 0; |
7d67af2c | 3658 | |
45d189c6 | 3659 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3660 | return -EAGAIN; |
7d67af2c PB |
3661 | |
3662 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3663 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3664 | |
948a7749 | 3665 | if (sp->len) |
c9687426 | 3666 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3667 | |
e1d767f0 PB |
3668 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3669 | io_put_file(in); | |
7d67af2c PB |
3670 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3671 | ||
7d67af2c | 3672 | if (ret != sp->len) |
93d2bcd2 | 3673 | req_set_fail(req); |
e1e16097 | 3674 | io_req_complete(req, ret); |
7d67af2c PB |
3675 | return 0; |
3676 | } | |
3677 | ||
2b188cc1 JA |
3678 | /* |
3679 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3680 | */ | |
889fca73 | 3681 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3682 | { |
3683 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3684 | |
def596e9 JA |
3685 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3686 | return -EINVAL; | |
3687 | ||
889fca73 | 3688 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3689 | return 0; |
3690 | } | |
3691 | ||
1155c76a | 3692 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3693 | { |
6b06314c | 3694 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3695 | |
09bb8394 JA |
3696 | if (!req->file) |
3697 | return -EBADF; | |
c992fe29 | 3698 | |
6b06314c | 3699 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3700 | return -EINVAL; |
edafccee | 3701 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3702 | return -EINVAL; |
3703 | ||
8ed8d3c3 JA |
3704 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3705 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3706 | return -EINVAL; | |
3707 | ||
3708 | req->sync.off = READ_ONCE(sqe->off); | |
3709 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3710 | return 0; |
3711 | } | |
3712 | ||
45d189c6 | 3713 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3714 | { |
8ed8d3c3 | 3715 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3716 | int ret; |
3717 | ||
ac45abc0 | 3718 | /* fsync always requires a blocking context */ |
45d189c6 | 3719 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3720 | return -EAGAIN; |
3721 | ||
9adbd45d | 3722 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3723 | end > 0 ? end : LLONG_MAX, |
3724 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3725 | if (ret < 0) | |
93d2bcd2 | 3726 | req_set_fail(req); |
e1e16097 | 3727 | io_req_complete(req, ret); |
c992fe29 CH |
3728 | return 0; |
3729 | } | |
3730 | ||
d63d1b5e JA |
3731 | static int io_fallocate_prep(struct io_kiocb *req, |
3732 | const struct io_uring_sqe *sqe) | |
3733 | { | |
3734 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3735 | return -EINVAL; | |
3232dd02 PB |
3736 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3737 | return -EINVAL; | |
d63d1b5e JA |
3738 | |
3739 | req->sync.off = READ_ONCE(sqe->off); | |
3740 | req->sync.len = READ_ONCE(sqe->addr); | |
3741 | req->sync.mode = READ_ONCE(sqe->len); | |
3742 | return 0; | |
3743 | } | |
3744 | ||
45d189c6 | 3745 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3746 | { |
ac45abc0 PB |
3747 | int ret; |
3748 | ||
d63d1b5e | 3749 | /* fallocate always requiring blocking context */ |
45d189c6 | 3750 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3751 | return -EAGAIN; |
ac45abc0 PB |
3752 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3753 | req->sync.len); | |
ac45abc0 | 3754 | if (ret < 0) |
93d2bcd2 | 3755 | req_set_fail(req); |
e1e16097 | 3756 | io_req_complete(req, ret); |
5d17b4a4 JA |
3757 | return 0; |
3758 | } | |
3759 | ||
ec65fea5 | 3760 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3761 | { |
f8748881 | 3762 | const char __user *fname; |
15b71abe | 3763 | int ret; |
b7bb4f7d | 3764 | |
ec65fea5 | 3765 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3766 | return -EINVAL; |
ec65fea5 | 3767 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3768 | return -EBADF; |
03b1230c | 3769 | |
ec65fea5 PB |
3770 | /* open.how should be already initialised */ |
3771 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3772 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3773 | |
25e72d10 PB |
3774 | req->open.dfd = READ_ONCE(sqe->fd); |
3775 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3776 | req->open.filename = getname(fname); |
15b71abe JA |
3777 | if (IS_ERR(req->open.filename)) { |
3778 | ret = PTR_ERR(req->open.filename); | |
3779 | req->open.filename = NULL; | |
3780 | return ret; | |
3781 | } | |
4022e7af | 3782 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3783 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3784 | return 0; |
03b1230c JA |
3785 | } |
3786 | ||
ec65fea5 PB |
3787 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3788 | { | |
3789 | u64 flags, mode; | |
3790 | ||
14587a46 | 3791 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3792 | return -EINVAL; |
ec65fea5 PB |
3793 | mode = READ_ONCE(sqe->len); |
3794 | flags = READ_ONCE(sqe->open_flags); | |
3795 | req->open.how = build_open_how(flags, mode); | |
3796 | return __io_openat_prep(req, sqe); | |
3797 | } | |
3798 | ||
cebdb986 | 3799 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3800 | { |
cebdb986 | 3801 | struct open_how __user *how; |
cebdb986 | 3802 | size_t len; |
0fa03c62 JA |
3803 | int ret; |
3804 | ||
14587a46 | 3805 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3806 | return -EINVAL; |
cebdb986 JA |
3807 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3808 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3809 | if (len < OPEN_HOW_SIZE_VER0) |
3810 | return -EINVAL; | |
3529d8c2 | 3811 | |
cebdb986 JA |
3812 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3813 | len); | |
3814 | if (ret) | |
3815 | return ret; | |
3529d8c2 | 3816 | |
ec65fea5 | 3817 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3818 | } |
3819 | ||
45d189c6 | 3820 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3821 | { |
3822 | struct open_flags op; | |
15b71abe | 3823 | struct file *file; |
3a81fd02 JA |
3824 | bool nonblock_set; |
3825 | bool resolve_nonblock; | |
15b71abe JA |
3826 | int ret; |
3827 | ||
cebdb986 | 3828 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3829 | if (ret) |
3830 | goto err; | |
3a81fd02 JA |
3831 | nonblock_set = op.open_flag & O_NONBLOCK; |
3832 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3833 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3834 | /* |
3835 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3836 | * it'll always -EAGAIN | |
3837 | */ | |
3838 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3839 | return -EAGAIN; | |
3840 | op.lookup_flags |= LOOKUP_CACHED; | |
3841 | op.open_flag |= O_NONBLOCK; | |
3842 | } | |
15b71abe | 3843 | |
4022e7af | 3844 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3845 | if (ret < 0) |
3846 | goto err; | |
3847 | ||
3848 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3849 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3850 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3851 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3852 | /* |
3a81fd02 JA |
3853 | * We could hang on to this 'fd', but seems like marginal |
3854 | * gain for something that is now known to be a slower path. | |
3855 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3856 | */ |
3a81fd02 JA |
3857 | put_unused_fd(ret); |
3858 | return -EAGAIN; | |
3859 | } | |
3860 | ||
15b71abe JA |
3861 | if (IS_ERR(file)) { |
3862 | put_unused_fd(ret); | |
3863 | ret = PTR_ERR(file); | |
3864 | } else { | |
45d189c6 | 3865 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3866 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3867 | fsnotify_open(file); |
3868 | fd_install(ret, file); | |
3869 | } | |
3870 | err: | |
3871 | putname(req->open.filename); | |
8fef80bf | 3872 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 3873 | if (ret < 0) |
93d2bcd2 | 3874 | req_set_fail(req); |
0bdf3398 | 3875 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3876 | return 0; |
3877 | } | |
3878 | ||
45d189c6 | 3879 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3880 | { |
e45cff58 | 3881 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3882 | } |
3883 | ||
067524e9 JA |
3884 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3885 | const struct io_uring_sqe *sqe) | |
3886 | { | |
3887 | struct io_provide_buf *p = &req->pbuf; | |
3888 | u64 tmp; | |
3889 | ||
3890 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3891 | return -EINVAL; | |
3892 | ||
3893 | tmp = READ_ONCE(sqe->fd); | |
3894 | if (!tmp || tmp > USHRT_MAX) | |
3895 | return -EINVAL; | |
3896 | ||
3897 | memset(p, 0, sizeof(*p)); | |
3898 | p->nbufs = tmp; | |
3899 | p->bgid = READ_ONCE(sqe->buf_group); | |
3900 | return 0; | |
3901 | } | |
3902 | ||
3903 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3904 | int bgid, unsigned nbufs) | |
3905 | { | |
3906 | unsigned i = 0; | |
3907 | ||
3908 | /* shouldn't happen */ | |
3909 | if (!nbufs) | |
3910 | return 0; | |
3911 | ||
3912 | /* the head kbuf is the list itself */ | |
3913 | while (!list_empty(&buf->list)) { | |
3914 | struct io_buffer *nxt; | |
3915 | ||
3916 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3917 | list_del(&nxt->list); | |
3918 | kfree(nxt); | |
3919 | if (++i == nbufs) | |
3920 | return i; | |
3921 | } | |
3922 | i++; | |
3923 | kfree(buf); | |
9e15c3a0 | 3924 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3925 | |
3926 | return i; | |
3927 | } | |
3928 | ||
889fca73 | 3929 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3930 | { |
3931 | struct io_provide_buf *p = &req->pbuf; | |
3932 | struct io_ring_ctx *ctx = req->ctx; | |
3933 | struct io_buffer *head; | |
3934 | int ret = 0; | |
45d189c6 | 3935 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3936 | |
3937 | io_ring_submit_lock(ctx, !force_nonblock); | |
3938 | ||
3939 | lockdep_assert_held(&ctx->uring_lock); | |
3940 | ||
3941 | ret = -ENOENT; | |
9e15c3a0 | 3942 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3943 | if (head) |
3944 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 3945 | if (ret < 0) |
93d2bcd2 | 3946 | req_set_fail(req); |
067524e9 | 3947 | |
9fb8cb49 PB |
3948 | /* complete before unlock, IOPOLL may need the lock */ |
3949 | __io_req_complete(req, issue_flags, ret, 0); | |
3950 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3951 | return 0; |
3952 | } | |
3953 | ||
ddf0322d JA |
3954 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3955 | const struct io_uring_sqe *sqe) | |
3956 | { | |
38134ada | 3957 | unsigned long size, tmp_check; |
ddf0322d JA |
3958 | struct io_provide_buf *p = &req->pbuf; |
3959 | u64 tmp; | |
3960 | ||
3961 | if (sqe->ioprio || sqe->rw_flags) | |
3962 | return -EINVAL; | |
3963 | ||
3964 | tmp = READ_ONCE(sqe->fd); | |
3965 | if (!tmp || tmp > USHRT_MAX) | |
3966 | return -E2BIG; | |
3967 | p->nbufs = tmp; | |
3968 | p->addr = READ_ONCE(sqe->addr); | |
3969 | p->len = READ_ONCE(sqe->len); | |
3970 | ||
38134ada PB |
3971 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
3972 | &size)) | |
3973 | return -EOVERFLOW; | |
3974 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
3975 | return -EOVERFLOW; | |
3976 | ||
d81269fe PB |
3977 | size = (unsigned long)p->len * p->nbufs; |
3978 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3979 | return -EFAULT; |
3980 | ||
3981 | p->bgid = READ_ONCE(sqe->buf_group); | |
3982 | tmp = READ_ONCE(sqe->off); | |
3983 | if (tmp > USHRT_MAX) | |
3984 | return -E2BIG; | |
3985 | p->bid = tmp; | |
3986 | return 0; | |
3987 | } | |
3988 | ||
3989 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3990 | { | |
3991 | struct io_buffer *buf; | |
3992 | u64 addr = pbuf->addr; | |
3993 | int i, bid = pbuf->bid; | |
3994 | ||
3995 | for (i = 0; i < pbuf->nbufs; i++) { | |
3996 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3997 | if (!buf) | |
3998 | break; | |
3999 | ||
4000 | buf->addr = addr; | |
d1f82808 | 4001 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
4002 | buf->bid = bid; |
4003 | addr += pbuf->len; | |
4004 | bid++; | |
4005 | if (!*head) { | |
4006 | INIT_LIST_HEAD(&buf->list); | |
4007 | *head = buf; | |
4008 | } else { | |
4009 | list_add_tail(&buf->list, &(*head)->list); | |
4010 | } | |
4011 | } | |
4012 | ||
4013 | return i ? i : -ENOMEM; | |
4014 | } | |
4015 | ||
889fca73 | 4016 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4017 | { |
4018 | struct io_provide_buf *p = &req->pbuf; | |
4019 | struct io_ring_ctx *ctx = req->ctx; | |
4020 | struct io_buffer *head, *list; | |
4021 | int ret = 0; | |
45d189c6 | 4022 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4023 | |
4024 | io_ring_submit_lock(ctx, !force_nonblock); | |
4025 | ||
4026 | lockdep_assert_held(&ctx->uring_lock); | |
4027 | ||
9e15c3a0 | 4028 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4029 | |
4030 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4031 | if (ret >= 0 && !list) { |
4032 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4033 | if (ret < 0) | |
067524e9 | 4034 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4035 | } |
ddf0322d | 4036 | if (ret < 0) |
93d2bcd2 | 4037 | req_set_fail(req); |
9fb8cb49 PB |
4038 | /* complete before unlock, IOPOLL may need the lock */ |
4039 | __io_req_complete(req, issue_flags, ret, 0); | |
4040 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4041 | return 0; |
cebdb986 JA |
4042 | } |
4043 | ||
3e4827b0 JA |
4044 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4045 | const struct io_uring_sqe *sqe) | |
4046 | { | |
4047 | #if defined(CONFIG_EPOLL) | |
4048 | if (sqe->ioprio || sqe->buf_index) | |
4049 | return -EINVAL; | |
2d74d042 | 4050 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4051 | return -EINVAL; |
3e4827b0 JA |
4052 | |
4053 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4054 | req->epoll.op = READ_ONCE(sqe->len); | |
4055 | req->epoll.fd = READ_ONCE(sqe->off); | |
4056 | ||
4057 | if (ep_op_has_event(req->epoll.op)) { | |
4058 | struct epoll_event __user *ev; | |
4059 | ||
4060 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4061 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4062 | return -EFAULT; | |
4063 | } | |
4064 | ||
4065 | return 0; | |
4066 | #else | |
4067 | return -EOPNOTSUPP; | |
4068 | #endif | |
4069 | } | |
4070 | ||
889fca73 | 4071 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4072 | { |
4073 | #if defined(CONFIG_EPOLL) | |
4074 | struct io_epoll *ie = &req->epoll; | |
4075 | int ret; | |
45d189c6 | 4076 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4077 | |
4078 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4079 | if (force_nonblock && ret == -EAGAIN) | |
4080 | return -EAGAIN; | |
4081 | ||
4082 | if (ret < 0) | |
93d2bcd2 | 4083 | req_set_fail(req); |
889fca73 | 4084 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4085 | return 0; |
4086 | #else | |
4087 | return -EOPNOTSUPP; | |
4088 | #endif | |
4089 | } | |
4090 | ||
c1ca757b JA |
4091 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4092 | { | |
4093 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4094 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4095 | return -EINVAL; | |
3232dd02 PB |
4096 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4097 | return -EINVAL; | |
c1ca757b JA |
4098 | |
4099 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4100 | req->madvise.len = READ_ONCE(sqe->len); | |
4101 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4102 | return 0; | |
4103 | #else | |
4104 | return -EOPNOTSUPP; | |
4105 | #endif | |
4106 | } | |
4107 | ||
45d189c6 | 4108 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4109 | { |
4110 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4111 | struct io_madvise *ma = &req->madvise; | |
4112 | int ret; | |
4113 | ||
45d189c6 | 4114 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4115 | return -EAGAIN; |
4116 | ||
0726b01e | 4117 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4118 | if (ret < 0) |
93d2bcd2 | 4119 | req_set_fail(req); |
e1e16097 | 4120 | io_req_complete(req, ret); |
c1ca757b JA |
4121 | return 0; |
4122 | #else | |
4123 | return -EOPNOTSUPP; | |
4124 | #endif | |
4125 | } | |
4126 | ||
4840e418 JA |
4127 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4128 | { | |
4129 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4130 | return -EINVAL; | |
3232dd02 PB |
4131 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4132 | return -EINVAL; | |
4840e418 JA |
4133 | |
4134 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4135 | req->fadvise.len = READ_ONCE(sqe->len); | |
4136 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4137 | return 0; | |
4138 | } | |
4139 | ||
45d189c6 | 4140 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4141 | { |
4142 | struct io_fadvise *fa = &req->fadvise; | |
4143 | int ret; | |
4144 | ||
45d189c6 | 4145 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4146 | switch (fa->advice) { |
4147 | case POSIX_FADV_NORMAL: | |
4148 | case POSIX_FADV_RANDOM: | |
4149 | case POSIX_FADV_SEQUENTIAL: | |
4150 | break; | |
4151 | default: | |
4152 | return -EAGAIN; | |
4153 | } | |
4154 | } | |
4840e418 JA |
4155 | |
4156 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4157 | if (ret < 0) | |
93d2bcd2 | 4158 | req_set_fail(req); |
0bdf3398 | 4159 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4160 | return 0; |
4161 | } | |
4162 | ||
eddc7ef5 JA |
4163 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4164 | { | |
2d74d042 | 4165 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4166 | return -EINVAL; |
eddc7ef5 JA |
4167 | if (sqe->ioprio || sqe->buf_index) |
4168 | return -EINVAL; | |
9c280f90 | 4169 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4170 | return -EBADF; |
eddc7ef5 | 4171 | |
1d9e1288 BM |
4172 | req->statx.dfd = READ_ONCE(sqe->fd); |
4173 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4174 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4175 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4176 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4177 | |
4178 | return 0; | |
4179 | } | |
4180 | ||
45d189c6 | 4181 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4182 | { |
1d9e1288 | 4183 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4184 | int ret; |
4185 | ||
59d70013 | 4186 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4187 | return -EAGAIN; |
4188 | ||
e62753e4 BM |
4189 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4190 | ctx->buffer); | |
eddc7ef5 | 4191 | |
eddc7ef5 | 4192 | if (ret < 0) |
93d2bcd2 | 4193 | req_set_fail(req); |
e1e16097 | 4194 | io_req_complete(req, ret); |
eddc7ef5 JA |
4195 | return 0; |
4196 | } | |
4197 | ||
b5dba59e JA |
4198 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4199 | { | |
14587a46 | 4200 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4201 | return -EINVAL; |
b5dba59e JA |
4202 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4203 | sqe->rw_flags || sqe->buf_index) | |
4204 | return -EINVAL; | |
9c280f90 | 4205 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4206 | return -EBADF; |
b5dba59e JA |
4207 | |
4208 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4209 | return 0; |
b5dba59e JA |
4210 | } |
4211 | ||
889fca73 | 4212 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4213 | { |
9eac1904 | 4214 | struct files_struct *files = current->files; |
3af73b28 | 4215 | struct io_close *close = &req->close; |
9eac1904 | 4216 | struct fdtable *fdt; |
a1fde923 PB |
4217 | struct file *file = NULL; |
4218 | int ret = -EBADF; | |
b5dba59e | 4219 | |
9eac1904 JA |
4220 | spin_lock(&files->file_lock); |
4221 | fdt = files_fdtable(files); | |
4222 | if (close->fd >= fdt->max_fds) { | |
4223 | spin_unlock(&files->file_lock); | |
4224 | goto err; | |
4225 | } | |
4226 | file = fdt->fd[close->fd]; | |
a1fde923 | 4227 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4228 | spin_unlock(&files->file_lock); |
4229 | file = NULL; | |
4230 | goto err; | |
3af73b28 | 4231 | } |
b5dba59e JA |
4232 | |
4233 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4234 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4235 | spin_unlock(&files->file_lock); |
0bf0eefd | 4236 | return -EAGAIN; |
a2100672 | 4237 | } |
b5dba59e | 4238 | |
9eac1904 JA |
4239 | ret = __close_fd_get_file(close->fd, &file); |
4240 | spin_unlock(&files->file_lock); | |
4241 | if (ret < 0) { | |
4242 | if (ret == -ENOENT) | |
4243 | ret = -EBADF; | |
4244 | goto err; | |
4245 | } | |
4246 | ||
3af73b28 | 4247 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4248 | ret = filp_close(file, current->files); |
4249 | err: | |
3af73b28 | 4250 | if (ret < 0) |
93d2bcd2 | 4251 | req_set_fail(req); |
9eac1904 JA |
4252 | if (file) |
4253 | fput(file); | |
889fca73 | 4254 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4255 | return 0; |
b5dba59e JA |
4256 | } |
4257 | ||
1155c76a | 4258 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4259 | { |
4260 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4261 | |
5d17b4a4 JA |
4262 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4263 | return -EINVAL; | |
4264 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4265 | return -EINVAL; | |
4266 | ||
8ed8d3c3 JA |
4267 | req->sync.off = READ_ONCE(sqe->off); |
4268 | req->sync.len = READ_ONCE(sqe->len); | |
4269 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4270 | return 0; |
4271 | } | |
4272 | ||
45d189c6 | 4273 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4274 | { |
8ed8d3c3 JA |
4275 | int ret; |
4276 | ||
ac45abc0 | 4277 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4278 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4279 | return -EAGAIN; |
4280 | ||
9adbd45d | 4281 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4282 | req->sync.flags); |
4283 | if (ret < 0) | |
93d2bcd2 | 4284 | req_set_fail(req); |
e1e16097 | 4285 | io_req_complete(req, ret); |
5d17b4a4 JA |
4286 | return 0; |
4287 | } | |
4288 | ||
469956e8 | 4289 | #if defined(CONFIG_NET) |
02d27d89 PB |
4290 | static int io_setup_async_msg(struct io_kiocb *req, |
4291 | struct io_async_msghdr *kmsg) | |
4292 | { | |
e8c2bc1f JA |
4293 | struct io_async_msghdr *async_msg = req->async_data; |
4294 | ||
4295 | if (async_msg) | |
02d27d89 | 4296 | return -EAGAIN; |
e8c2bc1f | 4297 | if (io_alloc_async_data(req)) { |
257e84a5 | 4298 | kfree(kmsg->free_iov); |
02d27d89 PB |
4299 | return -ENOMEM; |
4300 | } | |
e8c2bc1f | 4301 | async_msg = req->async_data; |
02d27d89 | 4302 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4303 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4304 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4305 | /* if were using fast_iov, set it to the new one */ |
4306 | if (!async_msg->free_iov) | |
4307 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4308 | ||
02d27d89 PB |
4309 | return -EAGAIN; |
4310 | } | |
4311 | ||
2ae523ed PB |
4312 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4313 | struct io_async_msghdr *iomsg) | |
4314 | { | |
2ae523ed | 4315 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4316 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4317 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4318 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4319 | } |
4320 | ||
93642ef8 PB |
4321 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4322 | { | |
4323 | int ret; | |
4324 | ||
93642ef8 PB |
4325 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4326 | if (!ret) | |
4327 | req->flags |= REQ_F_NEED_CLEANUP; | |
4328 | return ret; | |
4329 | } | |
4330 | ||
3529d8c2 | 4331 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4332 | { |
e47293fd | 4333 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4334 | |
d2b6f48b PB |
4335 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4336 | return -EINVAL; | |
4337 | ||
270a5940 | 4338 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4339 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4340 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4341 | if (sr->msg_flags & MSG_DONTWAIT) | |
4342 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4343 | |
d8768362 JA |
4344 | #ifdef CONFIG_COMPAT |
4345 | if (req->ctx->compat) | |
4346 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4347 | #endif | |
93642ef8 | 4348 | return 0; |
03b1230c JA |
4349 | } |
4350 | ||
889fca73 | 4351 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4352 | { |
6b754c8b | 4353 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4354 | struct socket *sock; |
7a7cacba | 4355 | unsigned flags; |
0031275d | 4356 | int min_ret = 0; |
0fa03c62 JA |
4357 | int ret; |
4358 | ||
dba4a925 | 4359 | sock = sock_from_file(req->file); |
7a7cacba | 4360 | if (unlikely(!sock)) |
dba4a925 | 4361 | return -ENOTSOCK; |
3529d8c2 | 4362 | |
257e84a5 PB |
4363 | kmsg = req->async_data; |
4364 | if (!kmsg) { | |
7a7cacba PB |
4365 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4366 | if (ret) | |
4367 | return ret; | |
4368 | kmsg = &iomsg; | |
0fa03c62 | 4369 | } |
0fa03c62 | 4370 | |
04411806 PB |
4371 | flags = req->sr_msg.msg_flags; |
4372 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4373 | flags |= MSG_DONTWAIT; |
0031275d SM |
4374 | if (flags & MSG_WAITALL) |
4375 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4376 | ||
7a7cacba | 4377 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4378 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4379 | return io_setup_async_msg(req, kmsg); |
4380 | if (ret == -ERESTARTSYS) | |
4381 | ret = -EINTR; | |
0fa03c62 | 4382 | |
257e84a5 PB |
4383 | /* fast path, check for non-NULL to avoid function call */ |
4384 | if (kmsg->free_iov) | |
4385 | kfree(kmsg->free_iov); | |
99bc4c38 | 4386 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4387 | if (ret < min_ret) |
93d2bcd2 | 4388 | req_set_fail(req); |
889fca73 | 4389 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4390 | return 0; |
03b1230c | 4391 | } |
aa1fa28f | 4392 | |
889fca73 | 4393 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4394 | { |
7a7cacba PB |
4395 | struct io_sr_msg *sr = &req->sr_msg; |
4396 | struct msghdr msg; | |
4397 | struct iovec iov; | |
fddaface | 4398 | struct socket *sock; |
7a7cacba | 4399 | unsigned flags; |
0031275d | 4400 | int min_ret = 0; |
fddaface JA |
4401 | int ret; |
4402 | ||
dba4a925 | 4403 | sock = sock_from_file(req->file); |
7a7cacba | 4404 | if (unlikely(!sock)) |
dba4a925 | 4405 | return -ENOTSOCK; |
fddaface | 4406 | |
7a7cacba PB |
4407 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4408 | if (unlikely(ret)) | |
14db8411 | 4409 | return ret; |
fddaface | 4410 | |
7a7cacba PB |
4411 | msg.msg_name = NULL; |
4412 | msg.msg_control = NULL; | |
4413 | msg.msg_controllen = 0; | |
4414 | msg.msg_namelen = 0; | |
fddaface | 4415 | |
04411806 PB |
4416 | flags = req->sr_msg.msg_flags; |
4417 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4418 | flags |= MSG_DONTWAIT; |
0031275d SM |
4419 | if (flags & MSG_WAITALL) |
4420 | min_ret = iov_iter_count(&msg.msg_iter); | |
4421 | ||
7a7cacba PB |
4422 | msg.msg_flags = flags; |
4423 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4424 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4425 | return -EAGAIN; |
4426 | if (ret == -ERESTARTSYS) | |
4427 | ret = -EINTR; | |
fddaface | 4428 | |
0031275d | 4429 | if (ret < min_ret) |
93d2bcd2 | 4430 | req_set_fail(req); |
889fca73 | 4431 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4432 | return 0; |
fddaface JA |
4433 | } |
4434 | ||
1400e697 PB |
4435 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4436 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4437 | { |
4438 | struct io_sr_msg *sr = &req->sr_msg; | |
4439 | struct iovec __user *uiov; | |
4440 | size_t iov_len; | |
4441 | int ret; | |
4442 | ||
1400e697 PB |
4443 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4444 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4445 | if (ret) |
4446 | return ret; | |
4447 | ||
4448 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4449 | if (iov_len > 1) | |
4450 | return -EINVAL; | |
5476dfed | 4451 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4452 | return -EFAULT; |
5476dfed | 4453 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4454 | iomsg->free_iov = NULL; |
52de1fe1 | 4455 | } else { |
257e84a5 | 4456 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4457 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4458 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4459 | false); |
52de1fe1 JA |
4460 | if (ret > 0) |
4461 | ret = 0; | |
4462 | } | |
4463 | ||
4464 | return ret; | |
4465 | } | |
4466 | ||
4467 | #ifdef CONFIG_COMPAT | |
4468 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4469 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4470 | { |
52de1fe1 JA |
4471 | struct io_sr_msg *sr = &req->sr_msg; |
4472 | struct compat_iovec __user *uiov; | |
4473 | compat_uptr_t ptr; | |
4474 | compat_size_t len; | |
4475 | int ret; | |
4476 | ||
4af3417a PB |
4477 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4478 | &ptr, &len); | |
52de1fe1 JA |
4479 | if (ret) |
4480 | return ret; | |
4481 | ||
4482 | uiov = compat_ptr(ptr); | |
4483 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4484 | compat_ssize_t clen; | |
4485 | ||
4486 | if (len > 1) | |
4487 | return -EINVAL; | |
4488 | if (!access_ok(uiov, sizeof(*uiov))) | |
4489 | return -EFAULT; | |
4490 | if (__get_user(clen, &uiov->iov_len)) | |
4491 | return -EFAULT; | |
4492 | if (clen < 0) | |
4493 | return -EINVAL; | |
2d280bc8 | 4494 | sr->len = clen; |
257e84a5 | 4495 | iomsg->free_iov = NULL; |
52de1fe1 | 4496 | } else { |
257e84a5 | 4497 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4498 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4499 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4500 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4501 | if (ret < 0) |
4502 | return ret; | |
4503 | } | |
4504 | ||
4505 | return 0; | |
4506 | } | |
4507 | #endif | |
4508 | ||
1400e697 PB |
4509 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4510 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4511 | { |
1400e697 | 4512 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4513 | |
4514 | #ifdef CONFIG_COMPAT | |
4515 | if (req->ctx->compat) | |
1400e697 | 4516 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4517 | #endif |
52de1fe1 | 4518 | |
1400e697 | 4519 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4520 | } |
4521 | ||
bcda7baa | 4522 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4523 | bool needs_lock) |
bcda7baa JA |
4524 | { |
4525 | struct io_sr_msg *sr = &req->sr_msg; | |
4526 | struct io_buffer *kbuf; | |
4527 | ||
bcda7baa JA |
4528 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4529 | if (IS_ERR(kbuf)) | |
4530 | return kbuf; | |
4531 | ||
4532 | sr->kbuf = kbuf; | |
4533 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4534 | return kbuf; |
fddaface JA |
4535 | } |
4536 | ||
7fbb1b54 PB |
4537 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4538 | { | |
4539 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4540 | } | |
4541 | ||
93642ef8 | 4542 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4543 | { |
99bc4c38 | 4544 | int ret; |
3529d8c2 | 4545 | |
93642ef8 PB |
4546 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4547 | if (!ret) | |
4548 | req->flags |= REQ_F_NEED_CLEANUP; | |
4549 | return ret; | |
4550 | } | |
4551 | ||
4552 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4553 | { | |
4554 | struct io_sr_msg *sr = &req->sr_msg; | |
4555 | ||
d2b6f48b PB |
4556 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4557 | return -EINVAL; | |
4558 | ||
270a5940 | 4559 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4560 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4561 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4562 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4563 | if (sr->msg_flags & MSG_DONTWAIT) | |
4564 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4565 | |
d8768362 JA |
4566 | #ifdef CONFIG_COMPAT |
4567 | if (req->ctx->compat) | |
4568 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4569 | #endif | |
93642ef8 | 4570 | return 0; |
aa1fa28f JA |
4571 | } |
4572 | ||
889fca73 | 4573 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4574 | { |
6b754c8b | 4575 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4576 | struct socket *sock; |
7fbb1b54 | 4577 | struct io_buffer *kbuf; |
7a7cacba | 4578 | unsigned flags; |
0031275d | 4579 | int min_ret = 0; |
52de1fe1 | 4580 | int ret, cflags = 0; |
45d189c6 | 4581 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4582 | |
dba4a925 | 4583 | sock = sock_from_file(req->file); |
7a7cacba | 4584 | if (unlikely(!sock)) |
dba4a925 | 4585 | return -ENOTSOCK; |
3529d8c2 | 4586 | |
257e84a5 PB |
4587 | kmsg = req->async_data; |
4588 | if (!kmsg) { | |
7a7cacba PB |
4589 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4590 | if (ret) | |
681fda8d | 4591 | return ret; |
7a7cacba PB |
4592 | kmsg = &iomsg; |
4593 | } | |
03b1230c | 4594 | |
bc02ef33 | 4595 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4596 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4597 | if (IS_ERR(kbuf)) |
52de1fe1 | 4598 | return PTR_ERR(kbuf); |
7a7cacba | 4599 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4600 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4601 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4602 | 1, req->sr_msg.len); |
4603 | } | |
52de1fe1 | 4604 | |
04411806 PB |
4605 | flags = req->sr_msg.msg_flags; |
4606 | if (force_nonblock) | |
7a7cacba | 4607 | flags |= MSG_DONTWAIT; |
0031275d SM |
4608 | if (flags & MSG_WAITALL) |
4609 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4610 | ||
7a7cacba PB |
4611 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4612 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4613 | if (force_nonblock && ret == -EAGAIN) |
4614 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4615 | if (ret == -ERESTARTSYS) |
4616 | ret = -EINTR; | |
03b1230c | 4617 | |
7fbb1b54 PB |
4618 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4619 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4620 | /* fast path, check for non-NULL to avoid function call */ |
4621 | if (kmsg->free_iov) | |
4622 | kfree(kmsg->free_iov); | |
99bc4c38 | 4623 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4624 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4625 | req_set_fail(req); |
889fca73 | 4626 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4627 | return 0; |
0fa03c62 | 4628 | } |
5d17b4a4 | 4629 | |
889fca73 | 4630 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4631 | { |
6b754c8b | 4632 | struct io_buffer *kbuf; |
7a7cacba PB |
4633 | struct io_sr_msg *sr = &req->sr_msg; |
4634 | struct msghdr msg; | |
4635 | void __user *buf = sr->buf; | |
fddaface | 4636 | struct socket *sock; |
7a7cacba PB |
4637 | struct iovec iov; |
4638 | unsigned flags; | |
0031275d | 4639 | int min_ret = 0; |
bcda7baa | 4640 | int ret, cflags = 0; |
45d189c6 | 4641 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4642 | |
dba4a925 | 4643 | sock = sock_from_file(req->file); |
7a7cacba | 4644 | if (unlikely(!sock)) |
dba4a925 | 4645 | return -ENOTSOCK; |
fddaface | 4646 | |
bc02ef33 | 4647 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4648 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4649 | if (IS_ERR(kbuf)) |
4650 | return PTR_ERR(kbuf); | |
7a7cacba | 4651 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4652 | } |
bcda7baa | 4653 | |
7a7cacba | 4654 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4655 | if (unlikely(ret)) |
4656 | goto out_free; | |
fddaface | 4657 | |
7a7cacba PB |
4658 | msg.msg_name = NULL; |
4659 | msg.msg_control = NULL; | |
4660 | msg.msg_controllen = 0; | |
4661 | msg.msg_namelen = 0; | |
4662 | msg.msg_iocb = NULL; | |
4663 | msg.msg_flags = 0; | |
fddaface | 4664 | |
04411806 PB |
4665 | flags = req->sr_msg.msg_flags; |
4666 | if (force_nonblock) | |
7a7cacba | 4667 | flags |= MSG_DONTWAIT; |
0031275d SM |
4668 | if (flags & MSG_WAITALL) |
4669 | min_ret = iov_iter_count(&msg.msg_iter); | |
4670 | ||
7a7cacba PB |
4671 | ret = sock_recvmsg(sock, &msg, flags); |
4672 | if (force_nonblock && ret == -EAGAIN) | |
4673 | return -EAGAIN; | |
4674 | if (ret == -ERESTARTSYS) | |
4675 | ret = -EINTR; | |
14c32eee | 4676 | out_free: |
7fbb1b54 PB |
4677 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4678 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4679 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4680 | req_set_fail(req); |
889fca73 | 4681 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4682 | return 0; |
fddaface JA |
4683 | } |
4684 | ||
3529d8c2 | 4685 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4686 | { |
8ed8d3c3 JA |
4687 | struct io_accept *accept = &req->accept; |
4688 | ||
14587a46 | 4689 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4690 | return -EINVAL; |
8042d6ce | 4691 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4692 | return -EINVAL; |
4693 | ||
d55e5f5b JA |
4694 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4695 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4696 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4697 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4698 | return 0; |
8ed8d3c3 | 4699 | } |
17f2fe35 | 4700 | |
889fca73 | 4701 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4702 | { |
4703 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4704 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4705 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4706 | int ret; |
4707 | ||
e697deed JX |
4708 | if (req->file->f_flags & O_NONBLOCK) |
4709 | req->flags |= REQ_F_NOWAIT; | |
4710 | ||
8ed8d3c3 | 4711 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4712 | accept->addr_len, accept->flags, |
4713 | accept->nofile); | |
8ed8d3c3 | 4714 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4715 | return -EAGAIN; |
ac45abc0 PB |
4716 | if (ret < 0) { |
4717 | if (ret == -ERESTARTSYS) | |
4718 | ret = -EINTR; | |
93d2bcd2 | 4719 | req_set_fail(req); |
ac45abc0 | 4720 | } |
889fca73 | 4721 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4722 | return 0; |
8ed8d3c3 JA |
4723 | } |
4724 | ||
93642ef8 PB |
4725 | static int io_connect_prep_async(struct io_kiocb *req) |
4726 | { | |
4727 | struct io_async_connect *io = req->async_data; | |
4728 | struct io_connect *conn = &req->connect; | |
4729 | ||
4730 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4731 | } | |
4732 | ||
3529d8c2 | 4733 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4734 | { |
3529d8c2 | 4735 | struct io_connect *conn = &req->connect; |
f499a021 | 4736 | |
14587a46 | 4737 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4738 | return -EINVAL; |
4739 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4740 | return -EINVAL; | |
4741 | ||
3529d8c2 JA |
4742 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4743 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4744 | return 0; |
f499a021 JA |
4745 | } |
4746 | ||
889fca73 | 4747 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4748 | { |
e8c2bc1f | 4749 | struct io_async_connect __io, *io; |
f8e85cf2 | 4750 | unsigned file_flags; |
3fbb51c1 | 4751 | int ret; |
45d189c6 | 4752 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4753 | |
e8c2bc1f JA |
4754 | if (req->async_data) { |
4755 | io = req->async_data; | |
f499a021 | 4756 | } else { |
3529d8c2 JA |
4757 | ret = move_addr_to_kernel(req->connect.addr, |
4758 | req->connect.addr_len, | |
e8c2bc1f | 4759 | &__io.address); |
f499a021 JA |
4760 | if (ret) |
4761 | goto out; | |
4762 | io = &__io; | |
4763 | } | |
4764 | ||
3fbb51c1 JA |
4765 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4766 | ||
e8c2bc1f | 4767 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4768 | req->connect.addr_len, file_flags); |
87f80d62 | 4769 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4770 | if (req->async_data) |
b7bb4f7d | 4771 | return -EAGAIN; |
e8c2bc1f | 4772 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4773 | ret = -ENOMEM; |
4774 | goto out; | |
4775 | } | |
e8c2bc1f | 4776 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4777 | return -EAGAIN; |
f499a021 | 4778 | } |
f8e85cf2 JA |
4779 | if (ret == -ERESTARTSYS) |
4780 | ret = -EINTR; | |
f499a021 | 4781 | out: |
4e88d6e7 | 4782 | if (ret < 0) |
93d2bcd2 | 4783 | req_set_fail(req); |
889fca73 | 4784 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4785 | return 0; |
469956e8 Y |
4786 | } |
4787 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4788 | #define IO_NETOP_FN(op) \ |
4789 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4790 | { \ | |
4791 | return -EOPNOTSUPP; \ | |
4792 | } | |
4793 | ||
4794 | #define IO_NETOP_PREP(op) \ | |
4795 | IO_NETOP_FN(op) \ | |
4796 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4797 | { \ | |
4798 | return -EOPNOTSUPP; \ | |
4799 | } \ | |
4800 | ||
4801 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4802 | IO_NETOP_PREP(op) \ | |
4803 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4804 | { \ | |
4805 | return -EOPNOTSUPP; \ | |
4806 | } | |
4807 | ||
4808 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4809 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4810 | IO_NETOP_PREP_ASYNC(connect); | |
4811 | IO_NETOP_PREP(accept); | |
4812 | IO_NETOP_FN(send); | |
4813 | IO_NETOP_FN(recv); | |
469956e8 | 4814 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4815 | |
d7718a9d JA |
4816 | struct io_poll_table { |
4817 | struct poll_table_struct pt; | |
4818 | struct io_kiocb *req; | |
4819 | int error; | |
4820 | }; | |
ce593a6c | 4821 | |
d7718a9d JA |
4822 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4823 | __poll_t mask, task_work_func_t func) | |
4824 | { | |
aa96bf8a | 4825 | int ret; |
d7718a9d JA |
4826 | |
4827 | /* for instances that support it check for an event match first: */ | |
4828 | if (mask && !(mask & poll->events)) | |
4829 | return 0; | |
4830 | ||
4831 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4832 | ||
4833 | list_del_init(&poll->wait.entry); | |
4834 | ||
d7718a9d | 4835 | req->result = mask; |
7cbf1722 | 4836 | req->task_work.func = func; |
6d816e08 | 4837 | |
d7718a9d | 4838 | /* |
e3aabf95 JA |
4839 | * If this fails, then the task is exiting. When a task exits, the |
4840 | * work gets canceled, so just cancel this request as well instead | |
4841 | * of executing it. We can't safely execute it anyway, as we may not | |
4842 | * have the needed state needed for it anyway. | |
d7718a9d | 4843 | */ |
355fb9e2 | 4844 | ret = io_req_task_work_add(req); |
aa96bf8a | 4845 | if (unlikely(ret)) { |
e3aabf95 | 4846 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4847 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4848 | } |
d7718a9d JA |
4849 | return 1; |
4850 | } | |
4851 | ||
74ce6ce4 JA |
4852 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4853 | __acquires(&req->ctx->completion_lock) | |
4854 | { | |
4855 | struct io_ring_ctx *ctx = req->ctx; | |
4856 | ||
4857 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4858 | struct poll_table_struct pt = { ._key = poll->events }; | |
4859 | ||
4860 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4861 | } | |
4862 | ||
4863 | spin_lock_irq(&ctx->completion_lock); | |
4864 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4865 | add_wait_queue(poll->head, &poll->wait); | |
4866 | return true; | |
4867 | } | |
4868 | ||
4869 | return false; | |
4870 | } | |
4871 | ||
d4e7cd36 | 4872 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4873 | { |
e8c2bc1f | 4874 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4875 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4876 | return req->async_data; |
d4e7cd36 JA |
4877 | return req->apoll->double_poll; |
4878 | } | |
4879 | ||
4880 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4881 | { | |
4882 | if (req->opcode == IORING_OP_POLL_ADD) | |
4883 | return &req->poll; | |
4884 | return &req->apoll->poll; | |
4885 | } | |
4886 | ||
4887 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4888 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4889 | { |
4890 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4891 | |
4892 | lockdep_assert_held(&req->ctx->completion_lock); | |
4893 | ||
4894 | if (poll && poll->head) { | |
4895 | struct wait_queue_head *head = poll->head; | |
4896 | ||
4897 | spin_lock(&head->lock); | |
4898 | list_del_init(&poll->wait.entry); | |
4899 | if (poll->wait.private) | |
de9b4cca | 4900 | req_ref_put(req); |
18bceab1 JA |
4901 | poll->head = NULL; |
4902 | spin_unlock(&head->lock); | |
4903 | } | |
4904 | } | |
4905 | ||
e27414be | 4906 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4907 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4908 | { |
4909 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4910 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4911 | int error; |
18bceab1 | 4912 | |
e27414be | 4913 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4914 | error = -ECANCELED; |
88e41cf9 | 4915 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4916 | } else { |
5082620f | 4917 | error = mangle_poll(mask); |
e27414be | 4918 | } |
b69de288 JA |
4919 | if (req->poll.events & EPOLLONESHOT) |
4920 | flags = 0; | |
d4d19c19 | 4921 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5082620f | 4922 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4923 | req->poll.done = true; |
4924 | flags = 0; | |
4925 | } | |
7b289c38 HX |
4926 | if (flags & IORING_CQE_F_MORE) |
4927 | ctx->cq_extra++; | |
18bceab1 | 4928 | |
18bceab1 | 4929 | io_commit_cqring(ctx); |
88e41cf9 | 4930 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4931 | } |
4932 | ||
dd221f46 | 4933 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4934 | { |
dd221f46 | 4935 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4936 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4937 | struct io_kiocb *nxt; |
18bceab1 JA |
4938 | |
4939 | if (io_poll_rewait(req, &req->poll)) { | |
4940 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4941 | } else { |
f40b964a | 4942 | bool done; |
18bceab1 | 4943 | |
e27414be | 4944 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4945 | if (done) { |
4946 | hash_del(&req->hash_node); | |
f40b964a | 4947 | } else { |
88e41cf9 JA |
4948 | req->result = 0; |
4949 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4950 | } | |
dd221f46 | 4951 | spin_unlock_irq(&ctx->completion_lock); |
dd221f46 | 4952 | io_cqring_ev_posted(ctx); |
18bceab1 | 4953 | |
88e41cf9 JA |
4954 | if (done) { |
4955 | nxt = io_put_req_find_next(req); | |
4956 | if (nxt) | |
4957 | __io_req_task_submit(nxt); | |
4958 | } | |
dd221f46 | 4959 | } |
18bceab1 JA |
4960 | } |
4961 | ||
4962 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4963 | int sync, void *key) | |
4964 | { | |
4965 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4966 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4967 | __poll_t mask = key_to_poll(key); |
4968 | ||
4969 | /* for instances that support it check for an event match first: */ | |
4970 | if (mask && !(mask & poll->events)) | |
4971 | return 0; | |
88e41cf9 JA |
4972 | if (!(poll->events & EPOLLONESHOT)) |
4973 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4974 | |
8706e04e JA |
4975 | list_del_init(&wait->entry); |
4976 | ||
807abcb0 | 4977 | if (poll && poll->head) { |
18bceab1 JA |
4978 | bool done; |
4979 | ||
807abcb0 JA |
4980 | spin_lock(&poll->head->lock); |
4981 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4982 | if (!done) |
807abcb0 | 4983 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4984 | /* make sure double remove sees this as being gone */ |
4985 | wait->private = NULL; | |
807abcb0 | 4986 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4987 | if (!done) { |
4988 | /* use wait func handler, so it matches the rq type */ | |
4989 | poll->wait.func(&poll->wait, mode, sync, key); | |
4990 | } | |
18bceab1 | 4991 | } |
de9b4cca | 4992 | req_ref_put(req); |
18bceab1 JA |
4993 | return 1; |
4994 | } | |
4995 | ||
4996 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4997 | wait_queue_func_t wake_func) | |
4998 | { | |
4999 | poll->head = NULL; | |
5000 | poll->done = false; | |
5001 | poll->canceled = false; | |
464dca61 JA |
5002 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5003 | /* mask in events that we always want/need */ | |
5004 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5005 | INIT_LIST_HEAD(&poll->wait.entry); |
5006 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5007 | } | |
5008 | ||
5009 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5010 | struct wait_queue_head *head, |
5011 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5012 | { |
5013 | struct io_kiocb *req = pt->req; | |
5014 | ||
5015 | /* | |
5016 | * If poll->head is already set, it's because the file being polled | |
5017 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5018 | * for write). Setup a separate io_poll_iocb if this happens. | |
5019 | */ | |
5020 | if (unlikely(poll->head)) { | |
58852d4d PB |
5021 | struct io_poll_iocb *poll_one = poll; |
5022 | ||
18bceab1 | 5023 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5024 | if (*poll_ptr) { |
18bceab1 JA |
5025 | pt->error = -EINVAL; |
5026 | return; | |
5027 | } | |
ea6a693d JA |
5028 | /* |
5029 | * Can't handle multishot for double wait for now, turn it | |
5030 | * into one-shot mode. | |
5031 | */ | |
7a274727 PB |
5032 | if (!(poll_one->events & EPOLLONESHOT)) |
5033 | poll_one->events |= EPOLLONESHOT; | |
1c3b3e65 | 5034 | /* double add on the same waitqueue head, ignore */ |
7a274727 | 5035 | if (poll_one->head == head) |
1c3b3e65 | 5036 | return; |
18bceab1 JA |
5037 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5038 | if (!poll) { | |
5039 | pt->error = -ENOMEM; | |
5040 | return; | |
5041 | } | |
58852d4d | 5042 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5043 | req_ref_get(req); |
18bceab1 | 5044 | poll->wait.private = req; |
807abcb0 | 5045 | *poll_ptr = poll; |
18bceab1 JA |
5046 | } |
5047 | ||
5048 | pt->error = 0; | |
5049 | poll->head = head; | |
a31eb4a2 JX |
5050 | |
5051 | if (poll->events & EPOLLEXCLUSIVE) | |
5052 | add_wait_queue_exclusive(head, &poll->wait); | |
5053 | else | |
5054 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5055 | } |
5056 | ||
5057 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5058 | struct poll_table_struct *p) | |
5059 | { | |
5060 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5061 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5062 | |
807abcb0 | 5063 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5064 | } |
5065 | ||
d7718a9d JA |
5066 | static void io_async_task_func(struct callback_head *cb) |
5067 | { | |
5068 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5069 | struct async_poll *apoll = req->apoll; | |
5070 | struct io_ring_ctx *ctx = req->ctx; | |
5071 | ||
5072 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5073 | ||
74ce6ce4 | 5074 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5075 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5076 | return; |
d7718a9d JA |
5077 | } |
5078 | ||
0ea13b44 | 5079 | hash_del(&req->hash_node); |
d4e7cd36 | 5080 | io_poll_remove_double(req); |
74ce6ce4 JA |
5081 | spin_unlock_irq(&ctx->completion_lock); |
5082 | ||
0be0b0e3 PB |
5083 | if (!READ_ONCE(apoll->poll.canceled)) |
5084 | __io_req_task_submit(req); | |
5085 | else | |
2593553a | 5086 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5087 | } |
5088 | ||
5089 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5090 | void *key) | |
5091 | { | |
5092 | struct io_kiocb *req = wait->private; | |
5093 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5094 | ||
5095 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5096 | key_to_poll(key)); | |
5097 | ||
5098 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5099 | } | |
5100 | ||
5101 | static void io_poll_req_insert(struct io_kiocb *req) | |
5102 | { | |
5103 | struct io_ring_ctx *ctx = req->ctx; | |
5104 | struct hlist_head *list; | |
5105 | ||
5106 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5107 | hlist_add_head(&req->hash_node, list); | |
5108 | } | |
5109 | ||
5110 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5111 | struct io_poll_iocb *poll, | |
5112 | struct io_poll_table *ipt, __poll_t mask, | |
5113 | wait_queue_func_t wake_func) | |
5114 | __acquires(&ctx->completion_lock) | |
5115 | { | |
5116 | struct io_ring_ctx *ctx = req->ctx; | |
5117 | bool cancel = false; | |
5118 | ||
4d52f338 | 5119 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5120 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5121 | poll->file = req->file; |
18bceab1 | 5122 | poll->wait.private = req; |
d7718a9d JA |
5123 | |
5124 | ipt->pt._key = mask; | |
5125 | ipt->req = req; | |
5126 | ipt->error = -EINVAL; | |
5127 | ||
d7718a9d JA |
5128 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5129 | ||
5130 | spin_lock_irq(&ctx->completion_lock); | |
5131 | if (likely(poll->head)) { | |
5132 | spin_lock(&poll->head->lock); | |
5133 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5134 | if (ipt->error) | |
5135 | cancel = true; | |
5136 | ipt->error = 0; | |
5137 | mask = 0; | |
5138 | } | |
88e41cf9 | 5139 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5140 | list_del_init(&poll->wait.entry); |
5141 | else if (cancel) | |
5142 | WRITE_ONCE(poll->canceled, true); | |
5143 | else if (!poll->done) /* actually waiting for an event */ | |
5144 | io_poll_req_insert(req); | |
5145 | spin_unlock(&poll->head->lock); | |
5146 | } | |
5147 | ||
5148 | return mask; | |
5149 | } | |
5150 | ||
5151 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5152 | { | |
5153 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5154 | struct io_ring_ctx *ctx = req->ctx; | |
5155 | struct async_poll *apoll; | |
5156 | struct io_poll_table ipt; | |
5157 | __poll_t mask, ret; | |
9dab14b8 | 5158 | int rw; |
d7718a9d JA |
5159 | |
5160 | if (!req->file || !file_can_poll(req->file)) | |
5161 | return false; | |
24c74678 | 5162 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5163 | return false; |
9dab14b8 JA |
5164 | if (def->pollin) |
5165 | rw = READ; | |
5166 | else if (def->pollout) | |
5167 | rw = WRITE; | |
5168 | else | |
5169 | return false; | |
5170 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5171 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5172 | return false; |
5173 | ||
5174 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5175 | if (unlikely(!apoll)) | |
5176 | return false; | |
807abcb0 | 5177 | apoll->double_poll = NULL; |
d7718a9d JA |
5178 | |
5179 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5180 | req->apoll = apoll; |
d7718a9d | 5181 | |
88e41cf9 | 5182 | mask = EPOLLONESHOT; |
d7718a9d | 5183 | if (def->pollin) |
8755d97a | 5184 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5185 | if (def->pollout) |
5186 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5187 | |
5188 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5189 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5190 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5191 | mask &= ~POLLIN; | |
5192 | ||
d7718a9d JA |
5193 | mask |= POLLERR | POLLPRI; |
5194 | ||
5195 | ipt.pt._qproc = io_async_queue_proc; | |
5196 | ||
5197 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5198 | io_async_wake); | |
a36da65c | 5199 | if (ret || ipt.error) { |
d4e7cd36 | 5200 | io_poll_remove_double(req); |
d7718a9d | 5201 | spin_unlock_irq(&ctx->completion_lock); |
d7718a9d JA |
5202 | return false; |
5203 | } | |
5204 | spin_unlock_irq(&ctx->completion_lock); | |
5205 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5206 | apoll->poll.events); | |
5207 | return true; | |
5208 | } | |
5209 | ||
5210 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5211 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5212 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5213 | { |
b41e9852 | 5214 | bool do_complete = false; |
221c5eb2 | 5215 | |
5082620f JA |
5216 | if (!poll->head) |
5217 | return false; | |
221c5eb2 | 5218 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5219 | if (do_cancel) |
5220 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5221 | if (!list_empty(&poll->wait.entry)) { |
5222 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5223 | do_complete = true; |
221c5eb2 JA |
5224 | } |
5225 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5226 | hash_del(&req->hash_node); |
d7718a9d JA |
5227 | return do_complete; |
5228 | } | |
5229 | ||
b2c3f7e1 | 5230 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5231 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5232 | { |
5233 | bool do_complete; | |
5234 | ||
d4e7cd36 | 5235 | io_poll_remove_double(req); |
e31001a3 | 5236 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5237 | |
e31001a3 | 5238 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
d7718a9d | 5239 | /* non-poll requests have submit ref still */ |
e31001a3 | 5240 | req_ref_put(req); |
b1f573bd | 5241 | } |
b2c3f7e1 JA |
5242 | return do_complete; |
5243 | } | |
5244 | ||
5245 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5246 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5247 | { |
5248 | bool do_complete; | |
b1f573bd | 5249 | |
b2c3f7e1 | 5250 | do_complete = io_poll_remove_waitqs(req); |
b41e9852 | 5251 | if (do_complete) { |
d4d19c19 | 5252 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5253 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5254 | req_set_fail(req); |
216578e5 | 5255 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5256 | } |
5257 | ||
5258 | return do_complete; | |
221c5eb2 JA |
5259 | } |
5260 | ||
76e1b642 JA |
5261 | /* |
5262 | * Returns true if we found and killed one or more poll requests | |
5263 | */ | |
6b81928d | 5264 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5265 | bool cancel_all) |
221c5eb2 | 5266 | { |
78076bb6 | 5267 | struct hlist_node *tmp; |
221c5eb2 | 5268 | struct io_kiocb *req; |
8e2e1faf | 5269 | int posted = 0, i; |
221c5eb2 JA |
5270 | |
5271 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5272 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5273 | struct hlist_head *list; | |
5274 | ||
5275 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5276 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5277 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5278 | posted += io_poll_remove_one(req); |
5279 | } | |
221c5eb2 JA |
5280 | } |
5281 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5282 | |
8e2e1faf JA |
5283 | if (posted) |
5284 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5285 | |
5286 | return posted != 0; | |
221c5eb2 JA |
5287 | } |
5288 | ||
9ba5fac8 PB |
5289 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5290 | bool poll_only) | |
e07785b0 | 5291 | __must_hold(&ctx->completion_lock) |
47f46768 | 5292 | { |
78076bb6 | 5293 | struct hlist_head *list; |
47f46768 JA |
5294 | struct io_kiocb *req; |
5295 | ||
78076bb6 JA |
5296 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5297 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5298 | if (sqe_addr != req->user_data) |
5299 | continue; | |
9ba5fac8 PB |
5300 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5301 | continue; | |
b2cb805f | 5302 | return req; |
47f46768 | 5303 | } |
b2cb805f JA |
5304 | return NULL; |
5305 | } | |
5306 | ||
9ba5fac8 PB |
5307 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5308 | bool poll_only) | |
e07785b0 | 5309 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5310 | { |
5311 | struct io_kiocb *req; | |
5312 | ||
9ba5fac8 | 5313 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5314 | if (!req) |
5315 | return -ENOENT; | |
5316 | if (io_poll_remove_one(req)) | |
5317 | return 0; | |
5318 | ||
5319 | return -EALREADY; | |
47f46768 JA |
5320 | } |
5321 | ||
9096af3e PB |
5322 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5323 | unsigned int flags) | |
5324 | { | |
5325 | u32 events; | |
47f46768 | 5326 | |
9096af3e PB |
5327 | events = READ_ONCE(sqe->poll32_events); |
5328 | #ifdef __BIG_ENDIAN | |
5329 | events = swahw32(events); | |
5330 | #endif | |
5331 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5332 | events |= EPOLLONESHOT; | |
5333 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5334 | } |
5335 | ||
c5de0036 | 5336 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5337 | const struct io_uring_sqe *sqe) |
0969e783 | 5338 | { |
c5de0036 PB |
5339 | struct io_poll_update *upd = &req->poll_update; |
5340 | u32 flags; | |
5341 | ||
0969e783 JA |
5342 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5343 | return -EINVAL; | |
c5de0036 PB |
5344 | if (sqe->ioprio || sqe->buf_index) |
5345 | return -EINVAL; | |
5346 | flags = READ_ONCE(sqe->len); | |
5347 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5348 | IORING_POLL_ADD_MULTI)) | |
5349 | return -EINVAL; | |
5350 | /* meaningless without update */ | |
5351 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5352 | return -EINVAL; |
5353 | ||
c5de0036 PB |
5354 | upd->old_user_data = READ_ONCE(sqe->addr); |
5355 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5356 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5357 | |
c5de0036 PB |
5358 | upd->new_user_data = READ_ONCE(sqe->off); |
5359 | if (!upd->update_user_data && upd->new_user_data) | |
5360 | return -EINVAL; | |
5361 | if (upd->update_events) | |
5362 | upd->events = io_poll_parse_events(sqe, flags); | |
5363 | else if (sqe->poll32_events) | |
5364 | return -EINVAL; | |
221c5eb2 | 5365 | |
221c5eb2 JA |
5366 | return 0; |
5367 | } | |
5368 | ||
221c5eb2 JA |
5369 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5370 | void *key) | |
5371 | { | |
c2f2eb7d JA |
5372 | struct io_kiocb *req = wait->private; |
5373 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5374 | |
d7718a9d | 5375 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5376 | } |
5377 | ||
221c5eb2 JA |
5378 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5379 | struct poll_table_struct *p) | |
5380 | { | |
5381 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5382 | ||
e8c2bc1f | 5383 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5384 | } |
5385 | ||
3529d8c2 | 5386 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5387 | { |
5388 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5389 | u32 flags; |
221c5eb2 JA |
5390 | |
5391 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5392 | return -EINVAL; | |
c5de0036 | 5393 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5394 | return -EINVAL; |
5395 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5396 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5397 | return -EINVAL; |
5398 | ||
c5de0036 | 5399 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5400 | return 0; |
5401 | } | |
5402 | ||
61e98203 | 5403 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5404 | { |
5405 | struct io_poll_iocb *poll = &req->poll; | |
5406 | struct io_ring_ctx *ctx = req->ctx; | |
5407 | struct io_poll_table ipt; | |
0969e783 | 5408 | __poll_t mask; |
0969e783 | 5409 | |
d7718a9d | 5410 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5411 | |
d7718a9d JA |
5412 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5413 | io_poll_wake); | |
221c5eb2 | 5414 | |
8c838788 | 5415 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5416 | ipt.error = 0; |
e27414be | 5417 | io_poll_complete(req, mask); |
221c5eb2 | 5418 | } |
221c5eb2 JA |
5419 | spin_unlock_irq(&ctx->completion_lock); |
5420 | ||
8c838788 JA |
5421 | if (mask) { |
5422 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5423 | if (poll->events & EPOLLONESHOT) |
5424 | io_put_req(req); | |
221c5eb2 | 5425 | } |
8c838788 | 5426 | return ipt.error; |
221c5eb2 JA |
5427 | } |
5428 | ||
c5de0036 | 5429 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5430 | { |
5431 | struct io_ring_ctx *ctx = req->ctx; | |
5432 | struct io_kiocb *preq; | |
cb3b200e | 5433 | bool completing; |
b69de288 JA |
5434 | int ret; |
5435 | ||
5436 | spin_lock_irq(&ctx->completion_lock); | |
9ba5fac8 | 5437 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5438 | if (!preq) { |
5439 | ret = -ENOENT; | |
5440 | goto err; | |
b69de288 | 5441 | } |
cb3b200e | 5442 | |
c5de0036 PB |
5443 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5444 | completing = true; | |
5445 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5446 | goto err; | |
5447 | } | |
5448 | ||
cb3b200e JA |
5449 | /* |
5450 | * Don't allow racy completion with singleshot, as we cannot safely | |
5451 | * update those. For multishot, if we're racing with completion, just | |
5452 | * let completion re-add it. | |
5453 | */ | |
5454 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5455 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5456 | ret = -EALREADY; | |
5457 | goto err; | |
b69de288 JA |
5458 | } |
5459 | /* we now have a detached poll request. reissue. */ | |
5460 | ret = 0; | |
5461 | err: | |
b69de288 | 5462 | if (ret < 0) { |
cb3b200e | 5463 | spin_unlock_irq(&ctx->completion_lock); |
93d2bcd2 | 5464 | req_set_fail(req); |
b69de288 JA |
5465 | io_req_complete(req, ret); |
5466 | return 0; | |
5467 | } | |
5468 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5469 | if (req->poll_update.update_events) { |
b69de288 | 5470 | preq->poll.events &= ~0xffff; |
9d805892 | 5471 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5472 | preq->poll.events |= IO_POLL_UNMASK; |
5473 | } | |
9d805892 PB |
5474 | if (req->poll_update.update_user_data) |
5475 | preq->user_data = req->poll_update.new_user_data; | |
cb3b200e JA |
5476 | spin_unlock_irq(&ctx->completion_lock); |
5477 | ||
b69de288 JA |
5478 | /* complete update request, we're done with it */ |
5479 | io_req_complete(req, ret); | |
5480 | ||
cb3b200e | 5481 | if (!completing) { |
c5de0036 | 5482 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5483 | if (ret < 0) { |
93d2bcd2 | 5484 | req_set_fail(preq); |
cb3b200e JA |
5485 | io_req_complete(preq, ret); |
5486 | } | |
b69de288 JA |
5487 | } |
5488 | return 0; | |
5489 | } | |
5490 | ||
5262f567 JA |
5491 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5492 | { | |
ad8a48ac JA |
5493 | struct io_timeout_data *data = container_of(timer, |
5494 | struct io_timeout_data, timer); | |
5495 | struct io_kiocb *req = data->req; | |
5496 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5497 | unsigned long flags; |
5498 | ||
5262f567 | 5499 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5500 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5501 | atomic_set(&req->ctx->cq_timeouts, |
5502 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5503 | ||
d4d19c19 | 5504 | io_cqring_fill_event(ctx, req->user_data, -ETIME, 0); |
5262f567 JA |
5505 | io_commit_cqring(ctx); |
5506 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5507 | ||
5508 | io_cqring_ev_posted(ctx); | |
93d2bcd2 | 5509 | req_set_fail(req); |
5262f567 JA |
5510 | io_put_req(req); |
5511 | return HRTIMER_NORESTART; | |
5512 | } | |
5513 | ||
fbd15848 PB |
5514 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5515 | __u64 user_data) | |
e07785b0 | 5516 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5517 | { |
fbd15848 | 5518 | struct io_timeout_data *io; |
47f46768 | 5519 | struct io_kiocb *req; |
fd9c7bc5 | 5520 | bool found = false; |
f254ac04 | 5521 | |
135fcde8 | 5522 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5523 | found = user_data == req->user_data; |
5524 | if (found) | |
47f46768 | 5525 | break; |
47f46768 | 5526 | } |
fd9c7bc5 PB |
5527 | if (!found) |
5528 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5529 | |
5530 | io = req->async_data; | |
fd9c7bc5 | 5531 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5532 | return ERR_PTR(-EALREADY); |
a71976f3 | 5533 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5534 | return req; |
5535 | } | |
47f46768 | 5536 | |
fbd15848 | 5537 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5538 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5539 | { |
5540 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5541 | ||
5542 | if (IS_ERR(req)) | |
5543 | return PTR_ERR(req); | |
f254ac04 | 5544 | |
93d2bcd2 | 5545 | req_set_fail(req); |
d4d19c19 | 5546 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
216578e5 | 5547 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5548 | return 0; |
5549 | } | |
5550 | ||
9c8e11b3 PB |
5551 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5552 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5553 | __must_hold(&ctx->completion_lock) |
47f46768 | 5554 | { |
9c8e11b3 PB |
5555 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5556 | struct io_timeout_data *data; | |
47f46768 | 5557 | |
9c8e11b3 PB |
5558 | if (IS_ERR(req)) |
5559 | return PTR_ERR(req); | |
47f46768 | 5560 | |
9c8e11b3 PB |
5561 | req->timeout.off = 0; /* noseq */ |
5562 | data = req->async_data; | |
5563 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5564 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5565 | data->timer.function = io_timeout_fn; | |
5566 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5567 | return 0; | |
47f46768 JA |
5568 | } |
5569 | ||
3529d8c2 JA |
5570 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5571 | const struct io_uring_sqe *sqe) | |
b29472ee | 5572 | { |
9c8e11b3 PB |
5573 | struct io_timeout_rem *tr = &req->timeout_rem; |
5574 | ||
b29472ee JA |
5575 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5576 | return -EINVAL; | |
61710e43 DA |
5577 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5578 | return -EINVAL; | |
9c8e11b3 | 5579 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5580 | return -EINVAL; |
5581 | ||
9c8e11b3 PB |
5582 | tr->addr = READ_ONCE(sqe->addr); |
5583 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5584 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5585 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5586 | return -EINVAL; | |
5587 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5588 | return -EFAULT; | |
5589 | } else if (tr->flags) { | |
5590 | /* timeout removal doesn't support flags */ | |
b29472ee | 5591 | return -EINVAL; |
9c8e11b3 | 5592 | } |
b29472ee | 5593 | |
b29472ee JA |
5594 | return 0; |
5595 | } | |
5596 | ||
8662daec PB |
5597 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5598 | { | |
5599 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5600 | : HRTIMER_MODE_REL; | |
5601 | } | |
5602 | ||
11365043 JA |
5603 | /* |
5604 | * Remove or update an existing timeout command | |
5605 | */ | |
61e98203 | 5606 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5607 | { |
9c8e11b3 | 5608 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5609 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5610 | int ret; |
11365043 | 5611 | |
11365043 | 5612 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5613 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5614 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5615 | else |
5616 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5617 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5618 | |
d4d19c19 | 5619 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
11365043 JA |
5620 | io_commit_cqring(ctx); |
5621 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5622 | io_cqring_ev_posted(ctx); |
4e88d6e7 | 5623 | if (ret < 0) |
93d2bcd2 | 5624 | req_set_fail(req); |
ec9c02ad | 5625 | io_put_req(req); |
11365043 | 5626 | return 0; |
5262f567 JA |
5627 | } |
5628 | ||
3529d8c2 | 5629 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5630 | bool is_timeout_link) |
5262f567 | 5631 | { |
ad8a48ac | 5632 | struct io_timeout_data *data; |
a41525ab | 5633 | unsigned flags; |
56080b02 | 5634 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5635 | |
ad8a48ac | 5636 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5637 | return -EINVAL; |
ad8a48ac | 5638 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5639 | return -EINVAL; |
56080b02 | 5640 | if (off && is_timeout_link) |
2d28390a | 5641 | return -EINVAL; |
a41525ab JA |
5642 | flags = READ_ONCE(sqe->timeout_flags); |
5643 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5644 | return -EINVAL; |
bdf20073 | 5645 | |
bfe68a22 | 5646 | req->timeout.off = off; |
f18ee4cf PB |
5647 | if (unlikely(off && !req->ctx->off_timeout_used)) |
5648 | req->ctx->off_timeout_used = true; | |
26a61679 | 5649 | |
e8c2bc1f | 5650 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5651 | return -ENOMEM; |
5652 | ||
e8c2bc1f | 5653 | data = req->async_data; |
ad8a48ac | 5654 | data->req = req; |
ad8a48ac JA |
5655 | |
5656 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5657 | return -EFAULT; |
5658 | ||
8662daec | 5659 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5660 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5661 | if (is_timeout_link) |
5662 | io_req_track_inflight(req); | |
ad8a48ac JA |
5663 | return 0; |
5664 | } | |
5665 | ||
61e98203 | 5666 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5667 | { |
ad8a48ac | 5668 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5669 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5670 | struct list_head *entry; |
bfe68a22 | 5671 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5672 | |
733f5c95 | 5673 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5674 | |
5262f567 JA |
5675 | /* |
5676 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5677 | * timeout event to be satisfied. If it isn't set, then this is |
5678 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5679 | */ |
8eb7e2d0 | 5680 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5681 | entry = ctx->timeout_list.prev; |
5682 | goto add; | |
5683 | } | |
5262f567 | 5684 | |
bfe68a22 PB |
5685 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5686 | req->timeout.target_seq = tail + off; | |
5262f567 | 5687 | |
f010505b MDG |
5688 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5689 | * This is safe because ->completion_lock is held, and submissions | |
5690 | * and completions are never mixed in the same ->completion_lock section. | |
5691 | */ | |
5692 | ctx->cq_last_tm_flush = tail; | |
5693 | ||
5262f567 JA |
5694 | /* |
5695 | * Insertion sort, ensuring the first entry in the list is always | |
5696 | * the one we need first. | |
5697 | */ | |
5262f567 | 5698 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5699 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5700 | timeout.list); | |
5262f567 | 5701 | |
8eb7e2d0 | 5702 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5703 | continue; |
bfe68a22 PB |
5704 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5705 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5706 | break; |
5707 | } | |
93bd25bb | 5708 | add: |
135fcde8 | 5709 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5710 | data->timer.function = io_timeout_fn; |
5711 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5712 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5713 | return 0; |
5714 | } | |
5262f567 | 5715 | |
f458dd84 PB |
5716 | struct io_cancel_data { |
5717 | struct io_ring_ctx *ctx; | |
5718 | u64 user_data; | |
5719 | }; | |
5720 | ||
62755e35 JA |
5721 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5722 | { | |
5723 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5724 | struct io_cancel_data *cd = data; |
62755e35 | 5725 | |
f458dd84 | 5726 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5727 | } |
5728 | ||
f458dd84 PB |
5729 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5730 | struct io_ring_ctx *ctx) | |
62755e35 | 5731 | { |
f458dd84 | 5732 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5733 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5734 | int ret = 0; |
5735 | ||
f458dd84 | 5736 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5737 | return -ENOENT; |
5738 | ||
f458dd84 | 5739 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5740 | switch (cancel_ret) { |
5741 | case IO_WQ_CANCEL_OK: | |
5742 | ret = 0; | |
5743 | break; | |
5744 | case IO_WQ_CANCEL_RUNNING: | |
5745 | ret = -EALREADY; | |
5746 | break; | |
5747 | case IO_WQ_CANCEL_NOTFOUND: | |
5748 | ret = -ENOENT; | |
5749 | break; | |
5750 | } | |
5751 | ||
e977d6d3 JA |
5752 | return ret; |
5753 | } | |
5754 | ||
47f46768 JA |
5755 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5756 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5757 | int success_ret) |
47f46768 JA |
5758 | { |
5759 | unsigned long flags; | |
5760 | int ret; | |
5761 | ||
f458dd84 | 5762 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5763 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5764 | if (ret != -ENOENT) |
5765 | goto done; | |
47f46768 JA |
5766 | ret = io_timeout_cancel(ctx, sqe_addr); |
5767 | if (ret != -ENOENT) | |
5768 | goto done; | |
9ba5fac8 | 5769 | ret = io_poll_cancel(ctx, sqe_addr, false); |
47f46768 | 5770 | done: |
b0dd8a41 JA |
5771 | if (!ret) |
5772 | ret = success_ret; | |
d4d19c19 | 5773 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
47f46768 JA |
5774 | io_commit_cqring(ctx); |
5775 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5776 | io_cqring_ev_posted(ctx); | |
5777 | ||
4e88d6e7 | 5778 | if (ret < 0) |
93d2bcd2 | 5779 | req_set_fail(req); |
47f46768 JA |
5780 | } |
5781 | ||
3529d8c2 JA |
5782 | static int io_async_cancel_prep(struct io_kiocb *req, |
5783 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5784 | { |
fbf23849 | 5785 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5786 | return -EINVAL; |
61710e43 DA |
5787 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5788 | return -EINVAL; | |
5789 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5790 | return -EINVAL; |
5791 | ||
fbf23849 JA |
5792 | req->cancel.addr = READ_ONCE(sqe->addr); |
5793 | return 0; | |
5794 | } | |
5795 | ||
61e98203 | 5796 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5797 | { |
5798 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5799 | u64 sqe_addr = req->cancel.addr; |
5800 | struct io_tctx_node *node; | |
5801 | int ret; | |
5802 | ||
5803 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5804 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5805 | spin_lock_irq(&ctx->completion_lock); | |
5806 | if (ret != -ENOENT) | |
5807 | goto done; | |
5808 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5809 | if (ret != -ENOENT) | |
5810 | goto done; | |
9ba5fac8 | 5811 | ret = io_poll_cancel(ctx, sqe_addr, false); |
58f99373 PB |
5812 | if (ret != -ENOENT) |
5813 | goto done; | |
5814 | spin_unlock_irq(&ctx->completion_lock); | |
5815 | ||
5816 | /* slow path, try all io-wq's */ | |
5817 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5818 | ret = -ENOENT; | |
5819 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5820 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5821 | |
58f99373 PB |
5822 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5823 | if (ret != -ENOENT) | |
5824 | break; | |
5825 | } | |
5826 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5827 | ||
5828 | spin_lock_irq(&ctx->completion_lock); | |
5829 | done: | |
d4d19c19 | 5830 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
58f99373 PB |
5831 | io_commit_cqring(ctx); |
5832 | spin_unlock_irq(&ctx->completion_lock); | |
5833 | io_cqring_ev_posted(ctx); | |
5834 | ||
5835 | if (ret < 0) | |
93d2bcd2 | 5836 | req_set_fail(req); |
58f99373 | 5837 | io_put_req(req); |
5262f567 JA |
5838 | return 0; |
5839 | } | |
5840 | ||
269bbe5f | 5841 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5842 | const struct io_uring_sqe *sqe) |
5843 | { | |
61710e43 DA |
5844 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5845 | return -EINVAL; | |
5846 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5847 | return -EINVAL; |
5848 | ||
269bbe5f BM |
5849 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5850 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5851 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5852 | return -EINVAL; |
269bbe5f | 5853 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5854 | return 0; |
5855 | } | |
5856 | ||
889fca73 | 5857 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5858 | { |
5859 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 5860 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 5861 | int ret; |
fbf23849 | 5862 | |
45d189c6 | 5863 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5864 | return -EAGAIN; |
05f3fb3c | 5865 | |
269bbe5f BM |
5866 | up.offset = req->rsrc_update.offset; |
5867 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
5868 | up.nr = 0; |
5869 | up.tags = 0; | |
615cee49 | 5870 | up.resv = 0; |
05f3fb3c JA |
5871 | |
5872 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 5873 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 5874 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5875 | mutex_unlock(&ctx->uring_lock); |
5876 | ||
5877 | if (ret < 0) | |
93d2bcd2 | 5878 | req_set_fail(req); |
889fca73 | 5879 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5880 | return 0; |
5881 | } | |
5882 | ||
bfe76559 | 5883 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5884 | { |
d625c6ee | 5885 | switch (req->opcode) { |
e781573e | 5886 | case IORING_OP_NOP: |
bfe76559 | 5887 | return 0; |
f67676d1 JA |
5888 | case IORING_OP_READV: |
5889 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5890 | case IORING_OP_READ: |
bfe76559 | 5891 | return io_read_prep(req, sqe); |
f67676d1 JA |
5892 | case IORING_OP_WRITEV: |
5893 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5894 | case IORING_OP_WRITE: |
bfe76559 | 5895 | return io_write_prep(req, sqe); |
0969e783 | 5896 | case IORING_OP_POLL_ADD: |
bfe76559 | 5897 | return io_poll_add_prep(req, sqe); |
0969e783 | 5898 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5899 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5900 | case IORING_OP_FSYNC: |
1155c76a | 5901 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5902 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5903 | return io_sfr_prep(req, sqe); |
03b1230c | 5904 | case IORING_OP_SENDMSG: |
fddaface | 5905 | case IORING_OP_SEND: |
bfe76559 | 5906 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5907 | case IORING_OP_RECVMSG: |
fddaface | 5908 | case IORING_OP_RECV: |
bfe76559 | 5909 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5910 | case IORING_OP_CONNECT: |
bfe76559 | 5911 | return io_connect_prep(req, sqe); |
2d28390a | 5912 | case IORING_OP_TIMEOUT: |
bfe76559 | 5913 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5914 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5915 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5916 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5917 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5918 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5919 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5920 | case IORING_OP_ACCEPT: |
bfe76559 | 5921 | return io_accept_prep(req, sqe); |
d63d1b5e | 5922 | case IORING_OP_FALLOCATE: |
bfe76559 | 5923 | return io_fallocate_prep(req, sqe); |
15b71abe | 5924 | case IORING_OP_OPENAT: |
bfe76559 | 5925 | return io_openat_prep(req, sqe); |
b5dba59e | 5926 | case IORING_OP_CLOSE: |
bfe76559 | 5927 | return io_close_prep(req, sqe); |
05f3fb3c | 5928 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5929 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5930 | case IORING_OP_STATX: |
bfe76559 | 5931 | return io_statx_prep(req, sqe); |
4840e418 | 5932 | case IORING_OP_FADVISE: |
bfe76559 | 5933 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5934 | case IORING_OP_MADVISE: |
bfe76559 | 5935 | return io_madvise_prep(req, sqe); |
cebdb986 | 5936 | case IORING_OP_OPENAT2: |
bfe76559 | 5937 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5938 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5939 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5940 | case IORING_OP_SPLICE: |
bfe76559 | 5941 | return io_splice_prep(req, sqe); |
ddf0322d | 5942 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5943 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5944 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5945 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5946 | case IORING_OP_TEE: |
bfe76559 | 5947 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5948 | case IORING_OP_SHUTDOWN: |
5949 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5950 | case IORING_OP_RENAMEAT: |
5951 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5952 | case IORING_OP_UNLINKAT: |
5953 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5954 | } |
5955 | ||
bfe76559 PB |
5956 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5957 | req->opcode); | |
bd54b6fe | 5958 | return -EINVAL; |
bfe76559 PB |
5959 | } |
5960 | ||
93642ef8 | 5961 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5962 | { |
b7e298d2 PB |
5963 | if (!io_op_defs[req->opcode].needs_async_setup) |
5964 | return 0; | |
5965 | if (WARN_ON_ONCE(req->async_data)) | |
5966 | return -EFAULT; | |
5967 | if (io_alloc_async_data(req)) | |
5968 | return -EAGAIN; | |
5969 | ||
93642ef8 PB |
5970 | switch (req->opcode) { |
5971 | case IORING_OP_READV: | |
93642ef8 PB |
5972 | return io_rw_prep_async(req, READ); |
5973 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5974 | return io_rw_prep_async(req, WRITE); |
5975 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5976 | return io_sendmsg_prep_async(req); |
5977 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5978 | return io_recvmsg_prep_async(req); |
5979 | case IORING_OP_CONNECT: | |
5980 | return io_connect_prep_async(req); | |
5981 | } | |
b7e298d2 PB |
5982 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5983 | req->opcode); | |
5984 | return -EFAULT; | |
f67676d1 JA |
5985 | } |
5986 | ||
9cf7c104 PB |
5987 | static u32 io_get_sequence(struct io_kiocb *req) |
5988 | { | |
5989 | struct io_kiocb *pos; | |
5990 | struct io_ring_ctx *ctx = req->ctx; | |
15641e42 | 5991 | u32 nr_reqs = 0; |
9cf7c104 | 5992 | |
f2f87370 PB |
5993 | io_for_each_link(pos, req) |
5994 | nr_reqs++; | |
15641e42 | 5995 | return ctx->cached_sq_head - nr_reqs; |
9cf7c104 PB |
5996 | } |
5997 | ||
76cc33d7 | 5998 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 5999 | { |
a197f664 | 6000 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6001 | struct io_defer_entry *de; |
f67676d1 | 6002 | int ret; |
9cf7c104 | 6003 | u32 seq; |
de0617e4 | 6004 | |
9d858b21 | 6005 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6006 | if (likely(list_empty_careful(&ctx->defer_list) && |
6007 | !(req->flags & REQ_F_IO_DRAIN))) | |
76cc33d7 | 6008 | return false; |
9cf7c104 PB |
6009 | |
6010 | seq = io_get_sequence(req); | |
6011 | /* Still a chance to pass the sequence check */ | |
6012 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6013 | return false; |
de0617e4 | 6014 | |
b7e298d2 | 6015 | ret = io_req_prep_async(req); |
be7053b7 PB |
6016 | if (ret) |
6017 | return ret; | |
cbdcb435 | 6018 | io_prep_async_link(req); |
27dc8338 | 6019 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 PB |
6020 | if (!de) { |
6021 | io_req_complete_failed(req, ret); | |
6022 | return true; | |
6023 | } | |
2d28390a | 6024 | |
de0617e4 | 6025 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6026 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6027 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6028 | kfree(de); |
ae34817b | 6029 | io_queue_async_work(req); |
76cc33d7 | 6030 | return true; |
de0617e4 JA |
6031 | } |
6032 | ||
915967f6 | 6033 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6034 | de->req = req; |
9cf7c104 | 6035 | de->seq = seq; |
27dc8338 | 6036 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 | 6037 | spin_unlock_irq(&ctx->completion_lock); |
76cc33d7 | 6038 | return true; |
de0617e4 JA |
6039 | } |
6040 | ||
68fb8979 | 6041 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6042 | { |
0e1b6fe3 PB |
6043 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6044 | switch (req->opcode) { | |
6045 | case IORING_OP_READV: | |
6046 | case IORING_OP_READ_FIXED: | |
6047 | case IORING_OP_READ: | |
bcda7baa | 6048 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6049 | break; |
6050 | case IORING_OP_RECVMSG: | |
6051 | case IORING_OP_RECV: | |
bcda7baa | 6052 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6053 | break; |
6054 | } | |
6055 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6056 | } |
6057 | ||
0e1b6fe3 PB |
6058 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6059 | switch (req->opcode) { | |
6060 | case IORING_OP_READV: | |
6061 | case IORING_OP_READ_FIXED: | |
6062 | case IORING_OP_READ: | |
6063 | case IORING_OP_WRITEV: | |
6064 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6065 | case IORING_OP_WRITE: { |
6066 | struct io_async_rw *io = req->async_data; | |
6067 | if (io->free_iovec) | |
6068 | kfree(io->free_iovec); | |
0e1b6fe3 | 6069 | break; |
e8c2bc1f | 6070 | } |
0e1b6fe3 | 6071 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6072 | case IORING_OP_SENDMSG: { |
6073 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6074 | |
6075 | kfree(io->free_iov); | |
0e1b6fe3 | 6076 | break; |
e8c2bc1f | 6077 | } |
0e1b6fe3 PB |
6078 | case IORING_OP_SPLICE: |
6079 | case IORING_OP_TEE: | |
e1d767f0 PB |
6080 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6081 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6082 | break; |
f3cd4850 JA |
6083 | case IORING_OP_OPENAT: |
6084 | case IORING_OP_OPENAT2: | |
6085 | if (req->open.filename) | |
6086 | putname(req->open.filename); | |
6087 | break; | |
80a261fd JA |
6088 | case IORING_OP_RENAMEAT: |
6089 | putname(req->rename.oldpath); | |
6090 | putname(req->rename.newpath); | |
6091 | break; | |
14a1143b JA |
6092 | case IORING_OP_UNLINKAT: |
6093 | putname(req->unlink.filename); | |
6094 | break; | |
0e1b6fe3 PB |
6095 | } |
6096 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6097 | } |
75652a30 JA |
6098 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6099 | kfree(req->apoll->double_poll); | |
6100 | kfree(req->apoll); | |
6101 | req->apoll = NULL; | |
6102 | } | |
3a0a6902 PB |
6103 | if (req->flags & REQ_F_INFLIGHT) { |
6104 | struct io_uring_task *tctx = req->task->io_uring; | |
6105 | ||
6106 | atomic_dec(&tctx->inflight_tracked); | |
6107 | req->flags &= ~REQ_F_INFLIGHT; | |
6108 | } | |
99bc4c38 PB |
6109 | } |
6110 | ||
889fca73 | 6111 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6112 | { |
a197f664 | 6113 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6114 | const struct cred *creds = NULL; |
d625c6ee | 6115 | int ret; |
2b188cc1 | 6116 | |
003e8dcc JA |
6117 | if (req->work.creds && req->work.creds != current_cred()) |
6118 | creds = override_creds(req->work.creds); | |
5730b27e | 6119 | |
d625c6ee | 6120 | switch (req->opcode) { |
2b188cc1 | 6121 | case IORING_OP_NOP: |
889fca73 | 6122 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6123 | break; |
6124 | case IORING_OP_READV: | |
edafccee | 6125 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6126 | case IORING_OP_READ: |
889fca73 | 6127 | ret = io_read(req, issue_flags); |
edafccee | 6128 | break; |
3529d8c2 | 6129 | case IORING_OP_WRITEV: |
edafccee | 6130 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6131 | case IORING_OP_WRITE: |
889fca73 | 6132 | ret = io_write(req, issue_flags); |
2b188cc1 | 6133 | break; |
c992fe29 | 6134 | case IORING_OP_FSYNC: |
45d189c6 | 6135 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6136 | break; |
221c5eb2 | 6137 | case IORING_OP_POLL_ADD: |
61e98203 | 6138 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6139 | break; |
6140 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6141 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6142 | break; |
5d17b4a4 | 6143 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6144 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6145 | break; |
0fa03c62 | 6146 | case IORING_OP_SENDMSG: |
889fca73 | 6147 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6148 | break; |
fddaface | 6149 | case IORING_OP_SEND: |
889fca73 | 6150 | ret = io_send(req, issue_flags); |
0fa03c62 | 6151 | break; |
aa1fa28f | 6152 | case IORING_OP_RECVMSG: |
889fca73 | 6153 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6154 | break; |
fddaface | 6155 | case IORING_OP_RECV: |
889fca73 | 6156 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6157 | break; |
5262f567 | 6158 | case IORING_OP_TIMEOUT: |
61e98203 | 6159 | ret = io_timeout(req, issue_flags); |
5262f567 | 6160 | break; |
11365043 | 6161 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6162 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6163 | break; |
17f2fe35 | 6164 | case IORING_OP_ACCEPT: |
889fca73 | 6165 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6166 | break; |
f8e85cf2 | 6167 | case IORING_OP_CONNECT: |
889fca73 | 6168 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6169 | break; |
62755e35 | 6170 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6171 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6172 | break; |
d63d1b5e | 6173 | case IORING_OP_FALLOCATE: |
45d189c6 | 6174 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6175 | break; |
15b71abe | 6176 | case IORING_OP_OPENAT: |
45d189c6 | 6177 | ret = io_openat(req, issue_flags); |
15b71abe | 6178 | break; |
b5dba59e | 6179 | case IORING_OP_CLOSE: |
889fca73 | 6180 | ret = io_close(req, issue_flags); |
b5dba59e | 6181 | break; |
05f3fb3c | 6182 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6183 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6184 | break; |
eddc7ef5 | 6185 | case IORING_OP_STATX: |
45d189c6 | 6186 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6187 | break; |
4840e418 | 6188 | case IORING_OP_FADVISE: |
45d189c6 | 6189 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6190 | break; |
c1ca757b | 6191 | case IORING_OP_MADVISE: |
45d189c6 | 6192 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6193 | break; |
cebdb986 | 6194 | case IORING_OP_OPENAT2: |
45d189c6 | 6195 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6196 | break; |
3e4827b0 | 6197 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6198 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6199 | break; |
7d67af2c | 6200 | case IORING_OP_SPLICE: |
45d189c6 | 6201 | ret = io_splice(req, issue_flags); |
7d67af2c | 6202 | break; |
ddf0322d | 6203 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6204 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6205 | break; |
067524e9 | 6206 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6207 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6208 | break; |
f2a8d5c7 | 6209 | case IORING_OP_TEE: |
45d189c6 | 6210 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6211 | break; |
36f4fa68 | 6212 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6213 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6214 | break; |
80a261fd | 6215 | case IORING_OP_RENAMEAT: |
45d189c6 | 6216 | ret = io_renameat(req, issue_flags); |
80a261fd | 6217 | break; |
14a1143b | 6218 | case IORING_OP_UNLINKAT: |
45d189c6 | 6219 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6220 | break; |
2b188cc1 JA |
6221 | default: |
6222 | ret = -EINVAL; | |
6223 | break; | |
6224 | } | |
6225 | ||
5730b27e JA |
6226 | if (creds) |
6227 | revert_creds(creds); | |
def596e9 JA |
6228 | if (ret) |
6229 | return ret; | |
b532576e | 6230 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6231 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6232 | io_iopoll_req_issued(req); | |
def596e9 JA |
6233 | |
6234 | return 0; | |
2b188cc1 JA |
6235 | } |
6236 | ||
5280f7e5 | 6237 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6238 | { |
6239 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6240 | struct io_kiocb *timeout; |
561fb04a | 6241 | int ret = 0; |
2b188cc1 | 6242 | |
6df1db6b PB |
6243 | timeout = io_prep_linked_timeout(req); |
6244 | if (timeout) | |
6245 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6246 | |
4014d943 | 6247 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6248 | ret = -ECANCELED; |
31b51510 | 6249 | |
561fb04a | 6250 | if (!ret) { |
561fb04a | 6251 | do { |
889fca73 | 6252 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6253 | /* |
6254 | * We can get EAGAIN for polled IO even though we're | |
6255 | * forcing a sync submission from here, since we can't | |
6256 | * wait for request slots on the block side. | |
6257 | */ | |
6258 | if (ret != -EAGAIN) | |
6259 | break; | |
6260 | cond_resched(); | |
6261 | } while (1); | |
6262 | } | |
31b51510 | 6263 | |
a3df7698 | 6264 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6265 | if (ret) { |
a3df7698 | 6266 | /* io-wq is going to take one down */ |
de9b4cca | 6267 | req_ref_get(req); |
a3df7698 | 6268 | io_req_task_queue_fail(req, ret); |
edafccee | 6269 | } |
2b188cc1 JA |
6270 | } |
6271 | ||
7b29f92d JA |
6272 | #define FFS_ASYNC_READ 0x1UL |
6273 | #define FFS_ASYNC_WRITE 0x2UL | |
6274 | #ifdef CONFIG_64BIT | |
6275 | #define FFS_ISREG 0x4UL | |
6276 | #else | |
6277 | #define FFS_ISREG 0x0UL | |
6278 | #endif | |
6279 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6280 | ||
aeca241b | 6281 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6282 | unsigned i) |
65e19f54 | 6283 | { |
aeca241b | 6284 | struct io_fixed_file *table_l2; |
65e19f54 | 6285 | |
aeca241b PB |
6286 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6287 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6288 | } |
6289 | ||
65e19f54 JA |
6290 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6291 | int index) | |
6292 | { | |
aeca241b | 6293 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6294 | |
a04b0ac0 | 6295 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6296 | } |
6297 | ||
a04b0ac0 | 6298 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6299 | { |
6300 | unsigned long file_ptr = (unsigned long) file; | |
6301 | ||
6302 | if (__io_file_supports_async(file, READ)) | |
6303 | file_ptr |= FFS_ASYNC_READ; | |
6304 | if (__io_file_supports_async(file, WRITE)) | |
6305 | file_ptr |= FFS_ASYNC_WRITE; | |
6306 | if (S_ISREG(file_inode(file)->i_mode)) | |
6307 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6308 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6309 | } |
6310 | ||
8371adf5 PB |
6311 | static struct file *io_file_get(struct io_submit_state *state, |
6312 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6313 | { |
a197f664 | 6314 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6315 | struct file *file; |
09bb8394 | 6316 | |
8da11c19 | 6317 | if (fixed) { |
7b29f92d JA |
6318 | unsigned long file_ptr; |
6319 | ||
479f517b | 6320 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6321 | return NULL; |
b7620121 | 6322 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6323 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6324 | file = (struct file *) (file_ptr & FFS_MASK); |
6325 | file_ptr &= ~FFS_MASK; | |
6326 | /* mask in overlapping REQ_F and FFS bits */ | |
6327 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6328 | io_req_set_rsrc_node(req); |
09bb8394 | 6329 | } else { |
c826bd7a | 6330 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6331 | file = __io_file_get(state, fd); |
d44f554e JA |
6332 | |
6333 | /* we don't allow fixed io_uring files */ | |
6334 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6335 | io_req_track_inflight(req); | |
09bb8394 JA |
6336 | } |
6337 | ||
8371adf5 | 6338 | return file; |
09bb8394 JA |
6339 | } |
6340 | ||
2665abfd | 6341 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6342 | { |
ad8a48ac JA |
6343 | struct io_timeout_data *data = container_of(timer, |
6344 | struct io_timeout_data, timer); | |
90cd7e42 | 6345 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6346 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6347 | unsigned long flags; |
2665abfd JA |
6348 | |
6349 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6350 | prev = req->timeout.head; |
6351 | req->timeout.head = NULL; | |
2665abfd JA |
6352 | |
6353 | /* | |
6354 | * We don't expect the list to be empty, that will only happen if we | |
6355 | * race with the completion of the linked work. | |
6356 | */ | |
447c19f3 | 6357 | if (prev) { |
f2f87370 | 6358 | io_remove_next_linked(prev); |
447c19f3 PB |
6359 | if (!req_ref_inc_not_zero(prev)) |
6360 | prev = NULL; | |
6361 | } | |
2665abfd JA |
6362 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6363 | ||
6364 | if (prev) { | |
014db007 | 6365 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6366 | io_put_req_deferred(prev, 1); |
a298232e | 6367 | io_put_req_deferred(req, 1); |
47f46768 | 6368 | } else { |
9ae1f8dd | 6369 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6370 | } |
2665abfd JA |
6371 | return HRTIMER_NORESTART; |
6372 | } | |
6373 | ||
de968c18 | 6374 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6375 | { |
de968c18 PB |
6376 | struct io_ring_ctx *ctx = req->ctx; |
6377 | ||
6378 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6379 | /* |
f2f87370 PB |
6380 | * If the back reference is NULL, then our linked request finished |
6381 | * before we got a chance to setup the timer | |
76a46e06 | 6382 | */ |
90cd7e42 | 6383 | if (req->timeout.head) { |
e8c2bc1f | 6384 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6385 | |
ad8a48ac JA |
6386 | data->timer.function = io_link_timeout_fn; |
6387 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6388 | data->mode); | |
2665abfd | 6389 | } |
76a46e06 | 6390 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6391 | /* drop submission reference */ |
76a46e06 JA |
6392 | io_put_req(req); |
6393 | } | |
2665abfd | 6394 | |
ad8a48ac | 6395 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6396 | { |
f2f87370 | 6397 | struct io_kiocb *nxt = req->link; |
2665abfd | 6398 | |
f2f87370 PB |
6399 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6400 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6401 | return NULL; |
2665abfd | 6402 | |
90cd7e42 | 6403 | nxt->timeout.head = req; |
900fad45 | 6404 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6405 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6406 | return nxt; |
2665abfd JA |
6407 | } |
6408 | ||
c5eef2b9 | 6409 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6410 | { |
d3d7298d | 6411 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6412 | int ret; |
2b188cc1 | 6413 | |
c5eef2b9 | 6414 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6415 | |
491381ce JA |
6416 | /* |
6417 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6418 | * doesn't support non-blocking read/write attempts | |
6419 | */ | |
1840038e | 6420 | if (likely(!ret)) { |
0d63c148 | 6421 | /* drop submission reference */ |
e342c807 | 6422 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6423 | struct io_ring_ctx *ctx = req->ctx; |
6424 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6425 | |
6dd0be1e | 6426 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6427 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6428 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6429 | } else { |
d3d7298d | 6430 | io_put_req(req); |
0d63c148 | 6431 | } |
1840038e PB |
6432 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6433 | if (!io_arm_poll_handler(req)) { | |
6434 | /* | |
6435 | * Queued up for async execution, worker will release | |
6436 | * submit reference when the iocb is actually submitted. | |
6437 | */ | |
6438 | io_queue_async_work(req); | |
6439 | } | |
0d63c148 | 6440 | } else { |
f41db273 | 6441 | io_req_complete_failed(req, ret); |
9e645e11 | 6442 | } |
d3d7298d PB |
6443 | if (linked_timeout) |
6444 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6445 | } |
6446 | ||
441b8a78 | 6447 | static inline void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 | 6448 | { |
441b8a78 | 6449 | if (unlikely(req->ctx->drain_used) && io_drain_req(req)) |
76cc33d7 | 6450 | return; |
4fe2c963 | 6451 | |
76cc33d7 | 6452 | if (likely(!(req->flags & REQ_F_FORCE_ASYNC))) { |
c5eef2b9 | 6453 | __io_queue_sqe(req); |
76cc33d7 PB |
6454 | } else { |
6455 | int ret = io_req_prep_async(req); | |
6456 | ||
6457 | if (unlikely(ret)) | |
6458 | io_req_complete_failed(req, ret); | |
6459 | else | |
6460 | io_queue_async_work(req); | |
ce35a47a | 6461 | } |
4fe2c963 JL |
6462 | } |
6463 | ||
b16fed66 PB |
6464 | /* |
6465 | * Check SQE restrictions (opcode and flags). | |
6466 | * | |
6467 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6468 | */ | |
6469 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6470 | struct io_kiocb *req, | |
6471 | unsigned int sqe_flags) | |
4fe2c963 | 6472 | { |
b16fed66 PB |
6473 | if (!ctx->restricted) |
6474 | return true; | |
6475 | ||
6476 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6477 | return false; | |
6478 | ||
6479 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6480 | ctx->restrictions.sqe_flags_required) | |
6481 | return false; | |
6482 | ||
6483 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6484 | ctx->restrictions.sqe_flags_required)) | |
6485 | return false; | |
6486 | ||
6487 | return true; | |
4fe2c963 JL |
6488 | } |
6489 | ||
b16fed66 PB |
6490 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6491 | const struct io_uring_sqe *sqe) | |
6492 | { | |
6493 | struct io_submit_state *state; | |
6494 | unsigned int sqe_flags; | |
003e8dcc | 6495 | int personality, ret = 0; |
b16fed66 PB |
6496 | |
6497 | req->opcode = READ_ONCE(sqe->opcode); | |
6498 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6499 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6500 | req->user_data = READ_ONCE(sqe->user_data); | |
6501 | req->async_data = NULL; | |
6502 | req->file = NULL; | |
6503 | req->ctx = ctx; | |
6504 | req->link = NULL; | |
6505 | req->fixed_rsrc_refs = NULL; | |
6506 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6507 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6508 | req->task = current; |
6509 | req->result = 0; | |
93e68e03 | 6510 | req->work.creds = NULL; |
b16fed66 PB |
6511 | |
6512 | /* enforce forwards compatibility on users */ | |
dddca226 | 6513 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 6514 | return -EINVAL; |
b16fed66 PB |
6515 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
6516 | return -EINVAL; | |
b16fed66 PB |
6517 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6518 | return -EACCES; | |
6519 | ||
6520 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6521 | !io_op_defs[req->opcode].buffer_select) | |
6522 | return -EOPNOTSUPP; | |
863e0560 | 6523 | |
003e8dcc JA |
6524 | personality = READ_ONCE(sqe->personality); |
6525 | if (personality) { | |
61cf9370 | 6526 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6527 | if (!req->work.creds) |
6528 | return -EINVAL; | |
6529 | get_cred(req->work.creds); | |
003e8dcc | 6530 | } |
b16fed66 PB |
6531 | state = &ctx->submit_state; |
6532 | ||
6533 | /* | |
6534 | * Plug now if we have more than 1 IO left after this, and the target | |
6535 | * is potentially a read/write to block based storage. | |
6536 | */ | |
6537 | if (!state->plug_started && state->ios_left > 1 && | |
6538 | io_op_defs[req->opcode].plug) { | |
6539 | blk_start_plug(&state->plug); | |
6540 | state->plug_started = true; | |
6541 | } | |
6542 | ||
6543 | if (io_op_defs[req->opcode].needs_file) { | |
6544 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6545 | ||
6546 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6547 | if (unlikely(!req->file)) | |
6548 | ret = -EBADF; | |
6549 | } | |
6550 | ||
6551 | state->ios_left--; | |
6552 | return ret; | |
6553 | } | |
6554 | ||
a6b8cadc | 6555 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6556 | const struct io_uring_sqe *sqe) |
9e645e11 | 6557 | { |
a1ab7b35 | 6558 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6559 | int ret; |
9e645e11 | 6560 | |
a6b8cadc PB |
6561 | ret = io_init_req(ctx, req, sqe); |
6562 | if (unlikely(ret)) { | |
6563 | fail_req: | |
de59bc10 PB |
6564 | if (link->head) { |
6565 | /* fail even hard links since we don't submit */ | |
93d2bcd2 | 6566 | req_set_fail(link->head); |
f41db273 | 6567 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6568 | link->head = NULL; |
6569 | } | |
f41db273 | 6570 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6571 | return ret; |
6572 | } | |
441b8a78 PB |
6573 | |
6574 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { | |
6575 | ctx->drain_used = true; | |
6576 | ||
6577 | /* | |
6578 | * Taking sequential execution of a link, draining both sides | |
6579 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6580 | * requests in the link. So, it drains the head and the | |
6581 | * next after the link request. The last one is done via | |
6582 | * drain_next flag to persist the effect across calls. | |
6583 | */ | |
6584 | if (link->head) { | |
6585 | link->head->flags |= REQ_F_IO_DRAIN; | |
6586 | ctx->drain_next = 1; | |
6587 | } | |
6588 | } | |
6589 | ||
be7053b7 PB |
6590 | ret = io_req_prep(req, sqe); |
6591 | if (unlikely(ret)) | |
6592 | goto fail_req; | |
a6b8cadc | 6593 | |
be7053b7 | 6594 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6595 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6596 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6597 | ||
9e645e11 JA |
6598 | /* |
6599 | * If we already have a head request, queue this one for async | |
6600 | * submittal once the head completes. If we don't have a head but | |
6601 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6602 | * submitted sync once the chain is complete. If none of those | |
6603 | * conditions are true (normal request), then just queue it. | |
6604 | */ | |
863e0560 PB |
6605 | if (link->head) { |
6606 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6607 | |
b7e298d2 | 6608 | ret = io_req_prep_async(req); |
cf109604 | 6609 | if (unlikely(ret)) |
a6b8cadc | 6610 | goto fail_req; |
9d76377f | 6611 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6612 | link->last->link = req; |
863e0560 | 6613 | link->last = req; |
32fe525b PB |
6614 | |
6615 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6616 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 6617 | link->head = NULL; |
5e159204 | 6618 | io_queue_sqe(head); |
32fe525b | 6619 | } |
9e645e11 | 6620 | } else { |
711be031 PB |
6621 | if (unlikely(ctx->drain_next)) { |
6622 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6623 | ctx->drain_next = 0; |
711be031 | 6624 | } |
ef4ff581 | 6625 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6626 | link->head = req; |
6627 | link->last = req; | |
711be031 | 6628 | } else { |
be7053b7 | 6629 | io_queue_sqe(req); |
711be031 | 6630 | } |
9e645e11 | 6631 | } |
2e6e1fde | 6632 | |
1d4240cc | 6633 | return 0; |
9e645e11 JA |
6634 | } |
6635 | ||
9a56a232 JA |
6636 | /* |
6637 | * Batched submission is done, ensure local IO is flushed out. | |
6638 | */ | |
ba88ff11 PB |
6639 | static void io_submit_state_end(struct io_submit_state *state, |
6640 | struct io_ring_ctx *ctx) | |
9a56a232 | 6641 | { |
a1ab7b35 | 6642 | if (state->link.head) |
de59bc10 | 6643 | io_queue_sqe(state->link.head); |
6dd0be1e | 6644 | if (state->comp.nr) |
ba88ff11 | 6645 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6646 | if (state->plug_started) |
6647 | blk_finish_plug(&state->plug); | |
9f13c35b | 6648 | io_state_file_put(state); |
9a56a232 JA |
6649 | } |
6650 | ||
6651 | /* | |
6652 | * Start submission side cache. | |
6653 | */ | |
6654 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6655 | unsigned int max_ios) |
9a56a232 | 6656 | { |
27926b68 | 6657 | state->plug_started = false; |
9a56a232 | 6658 | state->ios_left = max_ios; |
a1ab7b35 PB |
6659 | /* set only head, no need to init link_last in advance */ |
6660 | state->link.head = NULL; | |
9a56a232 JA |
6661 | } |
6662 | ||
2b188cc1 JA |
6663 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6664 | { | |
75b28aff | 6665 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6666 | |
caf582c6 PB |
6667 | /* |
6668 | * Ensure any loads from the SQEs are done at this point, | |
6669 | * since once we write the new head, the application could | |
6670 | * write new data to them. | |
6671 | */ | |
6672 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6673 | } |
6674 | ||
2b188cc1 | 6675 | /* |
3529d8c2 | 6676 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6677 | * that is mapped by userspace. This means that care needs to be taken to |
6678 | * ensure that reads are stable, as we cannot rely on userspace always | |
6679 | * being a good citizen. If members of the sqe are validated and then later | |
6680 | * used, it's important that those reads are done through READ_ONCE() to | |
6681 | * prevent a re-load down the line. | |
6682 | */ | |
709b302f | 6683 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6684 | { |
ea5ab3b5 | 6685 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 6686 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
6687 | |
6688 | /* | |
6689 | * The cached sq head (or cq tail) serves two purposes: | |
6690 | * | |
6691 | * 1) allows us to batch the cost of updating the user visible | |
6692 | * head updates. | |
6693 | * 2) allows the kernel side to track the head on its own, even | |
6694 | * though the application is the one updating it. | |
6695 | */ | |
17d3aeb3 | 6696 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
6697 | if (likely(head < ctx->sq_entries)) |
6698 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6699 | |
6700 | /* drop invalid entries */ | |
15641e42 PB |
6701 | ctx->cq_extra--; |
6702 | WRITE_ONCE(ctx->rings->sq_dropped, | |
6703 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
6704 | return NULL; |
6705 | } | |
6706 | ||
0f212204 | 6707 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6708 | { |
09899b19 | 6709 | struct io_uring_task *tctx; |
46c4e16a | 6710 | int submitted = 0; |
6c271ce2 | 6711 | |
ee7d46d9 PB |
6712 | /* make sure SQ entry isn't read before tail */ |
6713 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
6714 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6715 | return -EAGAIN; | |
6c271ce2 | 6716 | |
09899b19 PB |
6717 | tctx = current->io_uring; |
6718 | tctx->cached_refs -= nr; | |
6719 | if (unlikely(tctx->cached_refs < 0)) { | |
6720 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
6721 | ||
6722 | percpu_counter_add(&tctx->inflight, refill); | |
6723 | refcount_add(refill, ¤t->usage); | |
6724 | tctx->cached_refs += refill; | |
6725 | } | |
ba88ff11 | 6726 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6727 | |
46c4e16a | 6728 | while (submitted < nr) { |
3529d8c2 | 6729 | const struct io_uring_sqe *sqe; |
196be95c | 6730 | struct io_kiocb *req; |
fb5ccc98 | 6731 | |
258b29a9 | 6732 | req = io_alloc_req(ctx); |
196be95c PB |
6733 | if (unlikely(!req)) { |
6734 | if (!submitted) | |
6735 | submitted = -EAGAIN; | |
fb5ccc98 | 6736 | break; |
196be95c | 6737 | } |
4fccfcbb PB |
6738 | sqe = io_get_sqe(ctx); |
6739 | if (unlikely(!sqe)) { | |
6740 | kmem_cache_free(req_cachep, req); | |
6741 | break; | |
6742 | } | |
d3656344 JA |
6743 | /* will complete beyond this point, count as submitted */ |
6744 | submitted++; | |
a1ab7b35 | 6745 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6746 | break; |
6c271ce2 JA |
6747 | } |
6748 | ||
9466f437 PB |
6749 | if (unlikely(submitted != nr)) { |
6750 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 6751 | int unused = nr - ref_used; |
9466f437 | 6752 | |
09899b19 | 6753 | current->io_uring->cached_refs += unused; |
d8a6df10 | 6754 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 6755 | } |
6c271ce2 | 6756 | |
a1ab7b35 | 6757 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6758 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6759 | io_commit_sqring(ctx); | |
6760 | ||
6c271ce2 JA |
6761 | return submitted; |
6762 | } | |
6763 | ||
e4b6d902 PB |
6764 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
6765 | { | |
6766 | return READ_ONCE(sqd->state); | |
6767 | } | |
6768 | ||
23b3628e XW |
6769 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6770 | { | |
6771 | /* Tell userspace we may need a wakeup call */ | |
6772 | spin_lock_irq(&ctx->completion_lock); | |
6773 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6774 | spin_unlock_irq(&ctx->completion_lock); | |
6775 | } | |
6776 | ||
6777 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6778 | { | |
6779 | spin_lock_irq(&ctx->completion_lock); | |
6780 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6781 | spin_unlock_irq(&ctx->completion_lock); | |
6782 | } | |
6783 | ||
08369246 | 6784 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6785 | { |
c8d1ba58 | 6786 | unsigned int to_submit; |
bdcd3eab | 6787 | int ret = 0; |
6c271ce2 | 6788 | |
c8d1ba58 | 6789 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6790 | /* if we're handling multiple rings, cap submit size for fairness */ |
6791 | if (cap_entries && to_submit > 8) | |
6792 | to_submit = 8; | |
6793 | ||
906a3c6f | 6794 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6795 | unsigned nr_events = 0; |
a4c0b3de | 6796 | |
c8d1ba58 | 6797 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6798 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6799 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6800 | |
3b763ba1 PB |
6801 | /* |
6802 | * Don't submit if refs are dying, good for io_uring_register(), | |
6803 | * but also it is relied upon by io_ring_exit_work() | |
6804 | */ | |
0298ef96 PB |
6805 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6806 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6807 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 6808 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 6809 | |
acfb381d PB |
6810 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6811 | wake_up(&ctx->sqo_sq_wait); | |
6812 | } | |
6c271ce2 | 6813 | |
08369246 XW |
6814 | return ret; |
6815 | } | |
6c271ce2 | 6816 | |
08369246 XW |
6817 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6818 | { | |
6819 | struct io_ring_ctx *ctx; | |
6820 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6821 | |
c9dca27d PB |
6822 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6823 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6824 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6825 | } |
6c271ce2 | 6826 | |
e4b6d902 PB |
6827 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
6828 | { | |
6829 | bool did_sig = false; | |
6830 | struct ksignal ksig; | |
6831 | ||
6832 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
6833 | signal_pending(current)) { | |
6834 | mutex_unlock(&sqd->lock); | |
6835 | if (signal_pending(current)) | |
6836 | did_sig = get_signal(&ksig); | |
6837 | cond_resched(); | |
6838 | mutex_lock(&sqd->lock); | |
6839 | } | |
6840 | io_run_task_work(); | |
e4b6d902 PB |
6841 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
6842 | } | |
6843 | ||
c8d1ba58 JA |
6844 | static int io_sq_thread(void *data) |
6845 | { | |
69fb2131 JA |
6846 | struct io_sq_data *sqd = data; |
6847 | struct io_ring_ctx *ctx; | |
a0d9205f | 6848 | unsigned long timeout = 0; |
37d1e2e3 | 6849 | char buf[TASK_COMM_LEN]; |
08369246 | 6850 | DEFINE_WAIT(wait); |
6c271ce2 | 6851 | |
696ee88a | 6852 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6853 | set_task_comm(current, buf); |
37d1e2e3 JA |
6854 | |
6855 | if (sqd->sq_cpu != -1) | |
6856 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6857 | else | |
6858 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6859 | current->flags |= PF_NO_SETAFFINITY; | |
6860 | ||
09a6f4ef | 6861 | mutex_lock(&sqd->lock); |
e4b6d902 | 6862 | while (1) { |
08369246 XW |
6863 | int ret; |
6864 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6865 | |
e4b6d902 PB |
6866 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
6867 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 6868 | break; |
08369246 | 6869 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6870 | continue; |
08369246 | 6871 | } |
e4b6d902 | 6872 | |
08369246 | 6873 | sqt_spin = false; |
e95eee2d | 6874 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6875 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6876 | const struct cred *creds = NULL; |
6877 | ||
6878 | if (ctx->sq_creds != current_cred()) | |
6879 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6880 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6881 | if (creds) |
6882 | revert_creds(creds); | |
08369246 XW |
6883 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6884 | sqt_spin = true; | |
69fb2131 | 6885 | } |
6c271ce2 | 6886 | |
08369246 | 6887 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6888 | io_run_task_work(); |
6889 | cond_resched(); | |
08369246 XW |
6890 | if (sqt_spin) |
6891 | timeout = jiffies + sqd->sq_thread_idle; | |
6892 | continue; | |
6893 | } | |
6894 | ||
08369246 | 6895 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
e4b6d902 | 6896 | if (!io_sqd_events_pending(sqd)) { |
724cb4f9 HX |
6897 | needs_sched = true; |
6898 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
aaa9f0f4 PB |
6899 | io_ring_set_wakeup_flag(ctx); |
6900 | ||
724cb4f9 HX |
6901 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
6902 | !list_empty_careful(&ctx->iopoll_list)) { | |
6903 | needs_sched = false; | |
6904 | break; | |
6905 | } | |
6906 | if (io_sqring_entries(ctx)) { | |
6907 | needs_sched = false; | |
6908 | break; | |
6909 | } | |
6910 | } | |
6911 | ||
6912 | if (needs_sched) { | |
6913 | mutex_unlock(&sqd->lock); | |
6914 | schedule(); | |
6915 | mutex_lock(&sqd->lock); | |
6916 | } | |
69fb2131 JA |
6917 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6918 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6919 | } |
08369246 XW |
6920 | |
6921 | finish_wait(&sqd->wait, &wait); | |
6922 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6923 | } |
28cea78a | 6924 | |
78cc687b | 6925 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 6926 | sqd->thread = NULL; |
05962f95 | 6927 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6928 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 6929 | io_run_task_work(); |
734551df PB |
6930 | mutex_unlock(&sqd->lock); |
6931 | ||
37d1e2e3 JA |
6932 | complete(&sqd->exited); |
6933 | do_exit(0); | |
6c271ce2 JA |
6934 | } |
6935 | ||
bda52162 JA |
6936 | struct io_wait_queue { |
6937 | struct wait_queue_entry wq; | |
6938 | struct io_ring_ctx *ctx; | |
6939 | unsigned to_wait; | |
6940 | unsigned nr_timeouts; | |
6941 | }; | |
6942 | ||
6c503150 | 6943 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6944 | { |
6945 | struct io_ring_ctx *ctx = iowq->ctx; | |
6946 | ||
6947 | /* | |
d195a66e | 6948 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6949 | * started waiting. For timeouts, we always want to return to userspace, |
6950 | * regardless of event count. | |
6951 | */ | |
6c503150 | 6952 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6953 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6954 | } | |
6955 | ||
6956 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6957 | int wake_flags, void *key) | |
6958 | { | |
6959 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6960 | wq); | |
6961 | ||
6c503150 PB |
6962 | /* |
6963 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6964 | * the task, and the next invocation will do it. | |
6965 | */ | |
5ed7a37d | 6966 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
6967 | return autoremove_wake_function(curr, mode, wake_flags, key); |
6968 | return -1; | |
bda52162 JA |
6969 | } |
6970 | ||
af9c1a44 JA |
6971 | static int io_run_task_work_sig(void) |
6972 | { | |
6973 | if (io_run_task_work()) | |
6974 | return 1; | |
6975 | if (!signal_pending(current)) | |
6976 | return 0; | |
0b8cfa97 | 6977 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6978 | return -ERESTARTSYS; |
af9c1a44 JA |
6979 | return -EINTR; |
6980 | } | |
6981 | ||
eeb60b9a PB |
6982 | /* when returns >0, the caller should retry */ |
6983 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6984 | struct io_wait_queue *iowq, | |
6985 | signed long *timeout) | |
6986 | { | |
6987 | int ret; | |
6988 | ||
6989 | /* make sure we run task_work before checking for signals */ | |
6990 | ret = io_run_task_work_sig(); | |
6991 | if (ret || io_should_wake(iowq)) | |
6992 | return ret; | |
6993 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 6994 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
6995 | return 1; |
6996 | ||
6997 | *timeout = schedule_timeout(*timeout); | |
6998 | return !*timeout ? -ETIME : 1; | |
6999 | } | |
7000 | ||
2b188cc1 JA |
7001 | /* |
7002 | * Wait until events become available, if we don't already have some. The | |
7003 | * application must reap them itself, as they reside on the shared cq ring. | |
7004 | */ | |
7005 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7006 | const sigset_t __user *sig, size_t sigsz, |
7007 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7008 | { |
bda52162 JA |
7009 | struct io_wait_queue iowq = { |
7010 | .wq = { | |
7011 | .private = current, | |
7012 | .func = io_wake_function, | |
7013 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7014 | }, | |
7015 | .ctx = ctx, | |
7016 | .to_wait = min_events, | |
7017 | }; | |
75b28aff | 7018 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7019 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7020 | int ret; | |
2b188cc1 | 7021 | |
b41e9852 | 7022 | do { |
6c2450ae | 7023 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 7024 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7025 | return 0; |
4c6e277c | 7026 | if (!io_run_task_work()) |
b41e9852 | 7027 | break; |
b41e9852 | 7028 | } while (1); |
2b188cc1 JA |
7029 | |
7030 | if (sig) { | |
9e75ad5d AB |
7031 | #ifdef CONFIG_COMPAT |
7032 | if (in_compat_syscall()) | |
7033 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7034 | sigsz); |
9e75ad5d AB |
7035 | else |
7036 | #endif | |
b772434b | 7037 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7038 | |
2b188cc1 JA |
7039 | if (ret) |
7040 | return ret; | |
7041 | } | |
7042 | ||
c73ebb68 | 7043 | if (uts) { |
c1d5a224 PB |
7044 | struct timespec64 ts; |
7045 | ||
c73ebb68 HX |
7046 | if (get_timespec64(&ts, uts)) |
7047 | return -EFAULT; | |
7048 | timeout = timespec64_to_jiffies(&ts); | |
7049 | } | |
7050 | ||
bda52162 | 7051 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7052 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7053 | do { |
ca0a2651 | 7054 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7055 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7056 | ret = -EBUSY; |
7057 | break; | |
7058 | } | |
311997b3 | 7059 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7060 | TASK_INTERRUPTIBLE); |
eeb60b9a | 7061 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
311997b3 | 7062 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7063 | cond_resched(); |
eeb60b9a | 7064 | } while (ret > 0); |
bda52162 | 7065 | |
b7db41c9 | 7066 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7067 | |
75b28aff | 7068 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7069 | } |
7070 | ||
9123c8ff | 7071 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7072 | { |
9123c8ff | 7073 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7074 | |
846a4ef2 | 7075 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7076 | kfree(table[i]); |
7077 | kfree(table); | |
7078 | } | |
7079 | ||
7080 | static void **io_alloc_page_table(size_t size) | |
7081 | { | |
7082 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7083 | size_t init_size = size; | |
7084 | void **table; | |
7085 | ||
7086 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL); | |
7087 | if (!table) | |
7088 | return NULL; | |
7089 | ||
7090 | for (i = 0; i < nr_tables; i++) { | |
7091 | unsigned int this_size = min(size, PAGE_SIZE); | |
7092 | ||
7093 | table[i] = kzalloc(this_size, GFP_KERNEL); | |
7094 | if (!table[i]) { | |
7095 | io_free_page_table(table, init_size); | |
7096 | return NULL; | |
7097 | } | |
7098 | size -= this_size; | |
7099 | } | |
7100 | return table; | |
05f3fb3c JA |
7101 | } |
7102 | ||
2a63b2d9 | 7103 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7104 | { |
2a63b2d9 | 7105 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7106 | } |
7107 | ||
2a63b2d9 | 7108 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7109 | { |
2a63b2d9 BM |
7110 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7111 | } | |
65e19f54 | 7112 | |
28a9fe25 | 7113 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7114 | { |
28a9fe25 PB |
7115 | percpu_ref_exit(&ref_node->refs); |
7116 | kfree(ref_node); | |
1642b445 PB |
7117 | } |
7118 | ||
a7f0ed5a PB |
7119 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7120 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7121 | { |
a7f0ed5a PB |
7122 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7123 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7124 | |
a7f0ed5a PB |
7125 | if (data_to_kill) { |
7126 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7127 | |
a7f0ed5a PB |
7128 | rsrc_node->rsrc_data = data_to_kill; |
7129 | io_rsrc_ref_lock(ctx); | |
7130 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7131 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7132 | |
3e942498 | 7133 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7134 | percpu_ref_kill(&rsrc_node->refs); |
7135 | ctx->rsrc_node = NULL; | |
7136 | } | |
6b06314c | 7137 | |
a7f0ed5a PB |
7138 | if (!ctx->rsrc_node) { |
7139 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7140 | ctx->rsrc_backup_node = NULL; | |
7141 | } | |
8bad28d8 HX |
7142 | } |
7143 | ||
a7f0ed5a | 7144 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7145 | { |
7146 | if (ctx->rsrc_backup_node) | |
7147 | return 0; | |
b895c9a6 | 7148 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7149 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7150 | } |
7151 | ||
40ae0ff7 | 7152 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7153 | { |
7154 | int ret; | |
05589553 | 7155 | |
215c3902 | 7156 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7157 | if (data->quiesce) |
7158 | return -ENXIO; | |
05589553 | 7159 | |
8bad28d8 | 7160 | data->quiesce = true; |
1ffc5422 | 7161 | do { |
a7f0ed5a | 7162 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7163 | if (ret) |
f2303b1f | 7164 | break; |
a7f0ed5a | 7165 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7166 | |
3e942498 PB |
7167 | /* kill initial ref, already quiesced if zero */ |
7168 | if (atomic_dec_and_test(&data->refs)) | |
7169 | break; | |
8bad28d8 | 7170 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7171 | ret = wait_for_completion_interruptible(&data->done); |
7172 | if (!ret) | |
7173 | break; | |
8bad28d8 | 7174 | |
3e942498 PB |
7175 | atomic_inc(&data->refs); |
7176 | /* wait for all works potentially completing data->done */ | |
7177 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7178 | reinit_completion(&data->done); |
8dd03afe | 7179 | |
8bad28d8 | 7180 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7181 | ret = io_run_task_work_sig(); |
8bad28d8 | 7182 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7183 | } while (ret >= 0); |
8bad28d8 | 7184 | data->quiesce = false; |
05f3fb3c | 7185 | |
8bad28d8 | 7186 | return ret; |
d7954b2b BM |
7187 | } |
7188 | ||
2d091d62 PB |
7189 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7190 | { | |
7191 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7192 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7193 | ||
7194 | return &data->tags[table_idx][off]; | |
7195 | } | |
7196 | ||
44b31f2f | 7197 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7198 | { |
2d091d62 PB |
7199 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7200 | ||
7201 | if (data->tags) | |
7202 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7203 | kfree(data); |
7204 | } | |
7205 | ||
d878c816 PB |
7206 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7207 | u64 __user *utags, unsigned nr, | |
7208 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7209 | { |
b895c9a6 | 7210 | struct io_rsrc_data *data; |
2d091d62 | 7211 | int ret = -ENOMEM; |
d878c816 | 7212 | unsigned i; |
1ad555c6 BM |
7213 | |
7214 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7215 | if (!data) | |
d878c816 | 7216 | return -ENOMEM; |
2d091d62 | 7217 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7218 | if (!data->tags) { |
1ad555c6 | 7219 | kfree(data); |
d878c816 PB |
7220 | return -ENOMEM; |
7221 | } | |
2d091d62 PB |
7222 | |
7223 | data->nr = nr; | |
7224 | data->ctx = ctx; | |
7225 | data->do_put = do_put; | |
d878c816 | 7226 | if (utags) { |
2d091d62 | 7227 | ret = -EFAULT; |
d878c816 | 7228 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7229 | u64 *tag_slot = io_get_tag_slot(data, i); |
7230 | ||
7231 | if (copy_from_user(tag_slot, &utags[i], | |
7232 | sizeof(*tag_slot))) | |
2d091d62 | 7233 | goto fail; |
d878c816 | 7234 | } |
1ad555c6 | 7235 | } |
b60c8dce | 7236 | |
3e942498 | 7237 | atomic_set(&data->refs, 1); |
1ad555c6 | 7238 | init_completion(&data->done); |
d878c816 PB |
7239 | *pdata = data; |
7240 | return 0; | |
2d091d62 PB |
7241 | fail: |
7242 | io_rsrc_data_free(data); | |
7243 | return ret; | |
1ad555c6 BM |
7244 | } |
7245 | ||
9123c8ff PB |
7246 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7247 | { | |
7248 | size_t size = nr_files * sizeof(struct io_fixed_file); | |
7249 | ||
7250 | table->files = (struct io_fixed_file **)io_alloc_page_table(size); | |
7251 | return !!table->files; | |
7252 | } | |
7253 | ||
7254 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) | |
7255 | { | |
7256 | size_t size = nr_files * sizeof(struct io_fixed_file); | |
7257 | ||
7258 | io_free_page_table((void **)table->files, size); | |
7259 | table->files = NULL; | |
7260 | } | |
7261 | ||
fff4db76 | 7262 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7263 | { |
fff4db76 PB |
7264 | #if defined(CONFIG_UNIX) |
7265 | if (ctx->ring_sock) { | |
7266 | struct sock *sock = ctx->ring_sock->sk; | |
7267 | struct sk_buff *skb; | |
7268 | ||
7269 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7270 | kfree_skb(skb); | |
7271 | } | |
7272 | #else | |
7273 | int i; | |
7274 | ||
7275 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7276 | struct file *file; | |
7277 | ||
7278 | file = io_file_from_index(ctx, i); | |
7279 | if (file) | |
7280 | fput(file); | |
7281 | } | |
7282 | #endif | |
7283 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); | |
44b31f2f | 7284 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7285 | ctx->file_data = NULL; |
7286 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7287 | } |
7288 | ||
d7954b2b BM |
7289 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7290 | { | |
d7954b2b BM |
7291 | int ret; |
7292 | ||
08480400 | 7293 | if (!ctx->file_data) |
d7954b2b | 7294 | return -ENXIO; |
08480400 PB |
7295 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7296 | if (!ret) | |
7297 | __io_sqe_files_unregister(ctx); | |
7298 | return ret; | |
6b06314c JA |
7299 | } |
7300 | ||
37d1e2e3 | 7301 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7302 | __releases(&sqd->lock) |
37d1e2e3 | 7303 | { |
521d6a73 PB |
7304 | WARN_ON_ONCE(sqd->thread == current); |
7305 | ||
9e138a48 PB |
7306 | /* |
7307 | * Do the dance but not conditional clear_bit() because it'd race with | |
7308 | * other threads incrementing park_pending and setting the bit. | |
7309 | */ | |
37d1e2e3 | 7310 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7311 | if (atomic_dec_return(&sqd->park_pending)) |
7312 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7313 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7314 | } |
7315 | ||
86e0d676 | 7316 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7317 | __acquires(&sqd->lock) |
37d1e2e3 | 7318 | { |
521d6a73 PB |
7319 | WARN_ON_ONCE(sqd->thread == current); |
7320 | ||
9e138a48 | 7321 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7322 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7323 | mutex_lock(&sqd->lock); |
05962f95 | 7324 | if (sqd->thread) |
86e0d676 | 7325 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7326 | } |
7327 | ||
7328 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7329 | { | |
521d6a73 | 7330 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7331 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7332 | |
05962f95 | 7333 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7334 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7335 | if (sqd->thread) |
7336 | wake_up_process(sqd->thread); | |
09a6f4ef | 7337 | mutex_unlock(&sqd->lock); |
05962f95 | 7338 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7339 | } |
7340 | ||
534ca6d6 | 7341 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7342 | { |
534ca6d6 | 7343 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7344 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7345 | ||
37d1e2e3 JA |
7346 | io_sq_thread_stop(sqd); |
7347 | kfree(sqd); | |
7348 | } | |
7349 | } | |
7350 | ||
7351 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7352 | { | |
7353 | struct io_sq_data *sqd = ctx->sq_data; | |
7354 | ||
7355 | if (sqd) { | |
05962f95 | 7356 | io_sq_thread_park(sqd); |
521d6a73 | 7357 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7358 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7359 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7360 | |
7361 | io_put_sq_data(sqd); | |
7362 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7363 | } |
7364 | } | |
7365 | ||
aa06165d JA |
7366 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7367 | { | |
7368 | struct io_ring_ctx *ctx_attach; | |
7369 | struct io_sq_data *sqd; | |
7370 | struct fd f; | |
7371 | ||
7372 | f = fdget(p->wq_fd); | |
7373 | if (!f.file) | |
7374 | return ERR_PTR(-ENXIO); | |
7375 | if (f.file->f_op != &io_uring_fops) { | |
7376 | fdput(f); | |
7377 | return ERR_PTR(-EINVAL); | |
7378 | } | |
7379 | ||
7380 | ctx_attach = f.file->private_data; | |
7381 | sqd = ctx_attach->sq_data; | |
7382 | if (!sqd) { | |
7383 | fdput(f); | |
7384 | return ERR_PTR(-EINVAL); | |
7385 | } | |
5c2469e0 JA |
7386 | if (sqd->task_tgid != current->tgid) { |
7387 | fdput(f); | |
7388 | return ERR_PTR(-EPERM); | |
7389 | } | |
aa06165d JA |
7390 | |
7391 | refcount_inc(&sqd->refs); | |
7392 | fdput(f); | |
7393 | return sqd; | |
7394 | } | |
7395 | ||
26984fbf PB |
7396 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7397 | bool *attached) | |
534ca6d6 JA |
7398 | { |
7399 | struct io_sq_data *sqd; | |
7400 | ||
26984fbf | 7401 | *attached = false; |
5c2469e0 JA |
7402 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7403 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7404 | if (!IS_ERR(sqd)) { |
7405 | *attached = true; | |
5c2469e0 | 7406 | return sqd; |
26984fbf | 7407 | } |
5c2469e0 JA |
7408 | /* fall through for EPERM case, setup new sqd/task */ |
7409 | if (PTR_ERR(sqd) != -EPERM) | |
7410 | return sqd; | |
7411 | } | |
aa06165d | 7412 | |
534ca6d6 JA |
7413 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7414 | if (!sqd) | |
7415 | return ERR_PTR(-ENOMEM); | |
7416 | ||
9e138a48 | 7417 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7418 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7419 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7420 | mutex_init(&sqd->lock); |
534ca6d6 | 7421 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7422 | init_completion(&sqd->exited); |
534ca6d6 JA |
7423 | return sqd; |
7424 | } | |
7425 | ||
6b06314c | 7426 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7427 | /* |
7428 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7429 | * the io_uring can be safely unregistered on process exit, even if we have | |
7430 | * loops in the file referencing. | |
7431 | */ | |
7432 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7433 | { | |
7434 | struct sock *sk = ctx->ring_sock->sk; | |
7435 | struct scm_fp_list *fpl; | |
7436 | struct sk_buff *skb; | |
08a45173 | 7437 | int i, nr_files; |
6b06314c | 7438 | |
6b06314c JA |
7439 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7440 | if (!fpl) | |
7441 | return -ENOMEM; | |
7442 | ||
7443 | skb = alloc_skb(0, GFP_KERNEL); | |
7444 | if (!skb) { | |
7445 | kfree(fpl); | |
7446 | return -ENOMEM; | |
7447 | } | |
7448 | ||
7449 | skb->sk = sk; | |
6b06314c | 7450 | |
08a45173 | 7451 | nr_files = 0; |
62e398be | 7452 | fpl->user = get_uid(current_user()); |
6b06314c | 7453 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7454 | struct file *file = io_file_from_index(ctx, i + offset); |
7455 | ||
7456 | if (!file) | |
08a45173 | 7457 | continue; |
65e19f54 | 7458 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7459 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7460 | nr_files++; | |
6b06314c JA |
7461 | } |
7462 | ||
08a45173 JA |
7463 | if (nr_files) { |
7464 | fpl->max = SCM_MAX_FD; | |
7465 | fpl->count = nr_files; | |
7466 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7467 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7468 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7469 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7470 | |
08a45173 JA |
7471 | for (i = 0; i < nr_files; i++) |
7472 | fput(fpl->fp[i]); | |
7473 | } else { | |
7474 | kfree_skb(skb); | |
7475 | kfree(fpl); | |
7476 | } | |
6b06314c JA |
7477 | |
7478 | return 0; | |
7479 | } | |
7480 | ||
7481 | /* | |
7482 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7483 | * causes regular reference counting to break down. We rely on the UNIX | |
7484 | * garbage collection to take care of this problem for us. | |
7485 | */ | |
7486 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7487 | { | |
7488 | unsigned left, total; | |
7489 | int ret = 0; | |
7490 | ||
7491 | total = 0; | |
7492 | left = ctx->nr_user_files; | |
7493 | while (left) { | |
7494 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7495 | |
7496 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7497 | if (ret) | |
7498 | break; | |
7499 | left -= this_files; | |
7500 | total += this_files; | |
7501 | } | |
7502 | ||
7503 | if (!ret) | |
7504 | return 0; | |
7505 | ||
7506 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7507 | struct file *file = io_file_from_index(ctx, total); |
7508 | ||
7509 | if (file) | |
7510 | fput(file); | |
6b06314c JA |
7511 | total++; |
7512 | } | |
7513 | ||
7514 | return ret; | |
7515 | } | |
7516 | #else | |
7517 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7518 | { | |
7519 | return 0; | |
7520 | } | |
7521 | #endif | |
7522 | ||
47e90392 | 7523 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7524 | { |
50238531 | 7525 | struct file *file = prsrc->file; |
05f3fb3c JA |
7526 | #if defined(CONFIG_UNIX) |
7527 | struct sock *sock = ctx->ring_sock->sk; | |
7528 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7529 | struct sk_buff *skb; | |
7530 | int i; | |
7531 | ||
7532 | __skb_queue_head_init(&list); | |
7533 | ||
7534 | /* | |
7535 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7536 | * remove this entry and rearrange the file array. | |
7537 | */ | |
7538 | skb = skb_dequeue(head); | |
7539 | while (skb) { | |
7540 | struct scm_fp_list *fp; | |
7541 | ||
7542 | fp = UNIXCB(skb).fp; | |
7543 | for (i = 0; i < fp->count; i++) { | |
7544 | int left; | |
7545 | ||
7546 | if (fp->fp[i] != file) | |
7547 | continue; | |
7548 | ||
7549 | unix_notinflight(fp->user, fp->fp[i]); | |
7550 | left = fp->count - 1 - i; | |
7551 | if (left) { | |
7552 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7553 | left * sizeof(struct file *)); | |
7554 | } | |
7555 | fp->count--; | |
7556 | if (!fp->count) { | |
7557 | kfree_skb(skb); | |
7558 | skb = NULL; | |
7559 | } else { | |
7560 | __skb_queue_tail(&list, skb); | |
7561 | } | |
7562 | fput(file); | |
7563 | file = NULL; | |
7564 | break; | |
7565 | } | |
7566 | ||
7567 | if (!file) | |
7568 | break; | |
7569 | ||
7570 | __skb_queue_tail(&list, skb); | |
7571 | ||
7572 | skb = skb_dequeue(head); | |
7573 | } | |
7574 | ||
7575 | if (skb_peek(&list)) { | |
7576 | spin_lock_irq(&head->lock); | |
7577 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7578 | __skb_queue_tail(head, skb); | |
7579 | spin_unlock_irq(&head->lock); | |
7580 | } | |
7581 | #else | |
7582 | fput(file); | |
7583 | #endif | |
7584 | } | |
7585 | ||
b895c9a6 | 7586 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7587 | { |
b895c9a6 | 7588 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7589 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7590 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7591 | |
269bbe5f BM |
7592 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7593 | list_del(&prsrc->list); | |
b60c8dce PB |
7594 | |
7595 | if (prsrc->tag) { | |
7596 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
7597 | |
7598 | io_ring_submit_lock(ctx, lock_ring); | |
157d257f | 7599 | spin_lock_irq(&ctx->completion_lock); |
b60c8dce | 7600 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 7601 | ctx->cq_extra++; |
b60c8dce | 7602 | io_commit_cqring(ctx); |
157d257f | 7603 | spin_unlock_irq(&ctx->completion_lock); |
b60c8dce PB |
7604 | io_cqring_ev_posted(ctx); |
7605 | io_ring_submit_unlock(ctx, lock_ring); | |
7606 | } | |
7607 | ||
40ae0ff7 | 7608 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7609 | kfree(prsrc); |
65e19f54 | 7610 | } |
05589553 | 7611 | |
28a9fe25 | 7612 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7613 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7614 | complete(&rsrc_data->done); | |
2faf852d | 7615 | } |
65e19f54 | 7616 | |
269bbe5f | 7617 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7618 | { |
7619 | struct io_ring_ctx *ctx; | |
7620 | struct llist_node *node; | |
7621 | ||
269bbe5f BM |
7622 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7623 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7624 | |
7625 | while (node) { | |
b895c9a6 | 7626 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7627 | struct llist_node *next = node->next; |
7628 | ||
b895c9a6 | 7629 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7630 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7631 | node = next; |
7632 | } | |
7633 | } | |
7634 | ||
00835dce | 7635 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7636 | { |
b895c9a6 | 7637 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7638 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7639 | bool first_add = false; |
e297822b | 7640 | |
2a63b2d9 | 7641 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7642 | node->done = true; |
e297822b | 7643 | |
d67d2263 | 7644 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7645 | node = list_first_entry(&ctx->rsrc_ref_list, |
7646 | struct io_rsrc_node, node); | |
e297822b | 7647 | /* recycle ref nodes in order */ |
b895c9a6 | 7648 | if (!node->done) |
e297822b | 7649 | break; |
b895c9a6 PB |
7650 | list_del(&node->node); |
7651 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7652 | } |
2a63b2d9 | 7653 | io_rsrc_ref_unlock(ctx); |
05589553 | 7654 | |
3e942498 PB |
7655 | if (first_add) |
7656 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7657 | } |
65e19f54 | 7658 | |
b895c9a6 | 7659 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7660 | { |
b895c9a6 | 7661 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7662 | |
05589553 XW |
7663 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7664 | if (!ref_node) | |
3e2224c5 | 7665 | return NULL; |
05f3fb3c | 7666 | |
00835dce | 7667 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7668 | 0, GFP_KERNEL)) { |
7669 | kfree(ref_node); | |
3e2224c5 | 7670 | return NULL; |
05589553 XW |
7671 | } |
7672 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7673 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7674 | ref_node->done = false; |
05589553 | 7675 | return ref_node; |
05589553 XW |
7676 | } |
7677 | ||
6b06314c | 7678 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 7679 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
7680 | { |
7681 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7682 | struct file *file; |
f3baed39 | 7683 | int fd, ret; |
846a4ef2 | 7684 | unsigned i; |
6b06314c | 7685 | |
05f3fb3c | 7686 | if (ctx->file_data) |
6b06314c JA |
7687 | return -EBUSY; |
7688 | if (!nr_args) | |
7689 | return -EINVAL; | |
7690 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7691 | return -EMFILE; | |
a7f0ed5a | 7692 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7693 | if (ret) |
7694 | return ret; | |
d878c816 PB |
7695 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
7696 | &ctx->file_data); | |
7697 | if (ret) | |
7698 | return ret; | |
6b06314c | 7699 | |
f3baed39 | 7700 | ret = -ENOMEM; |
aeca241b | 7701 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7702 | goto out_free; |
65e19f54 | 7703 | |
08a45173 | 7704 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 7705 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
7706 | ret = -EFAULT; |
7707 | goto out_fput; | |
7708 | } | |
08a45173 | 7709 | /* allow sparse sets */ |
792e3582 PB |
7710 | if (fd == -1) { |
7711 | ret = -EINVAL; | |
2d091d62 | 7712 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 7713 | goto out_fput; |
08a45173 | 7714 | continue; |
792e3582 | 7715 | } |
6b06314c | 7716 | |
05f3fb3c | 7717 | file = fget(fd); |
6b06314c | 7718 | ret = -EBADF; |
792e3582 | 7719 | if (unlikely(!file)) |
600cf3f8 | 7720 | goto out_fput; |
05f3fb3c | 7721 | |
6b06314c JA |
7722 | /* |
7723 | * Don't allow io_uring instances to be registered. If UNIX | |
7724 | * isn't enabled, then this causes a reference cycle and this | |
7725 | * instance can never get freed. If UNIX is enabled we'll | |
7726 | * handle it just fine, but there's still no point in allowing | |
7727 | * a ring fd as it doesn't support regular read/write anyway. | |
7728 | */ | |
05f3fb3c JA |
7729 | if (file->f_op == &io_uring_fops) { |
7730 | fput(file); | |
600cf3f8 | 7731 | goto out_fput; |
6b06314c | 7732 | } |
aeca241b | 7733 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7734 | } |
7735 | ||
6b06314c | 7736 | ret = io_sqe_files_scm(ctx); |
05589553 | 7737 | if (ret) { |
08480400 | 7738 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7739 | return ret; |
7740 | } | |
6b06314c | 7741 | |
a7f0ed5a | 7742 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7743 | return ret; |
600cf3f8 PB |
7744 | out_fput: |
7745 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7746 | file = io_file_from_index(ctx, i); | |
7747 | if (file) | |
7748 | fput(file); | |
7749 | } | |
aeca241b | 7750 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7751 | ctx->nr_user_files = 0; |
600cf3f8 | 7752 | out_free: |
44b31f2f | 7753 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7754 | ctx->file_data = NULL; |
6b06314c JA |
7755 | return ret; |
7756 | } | |
7757 | ||
c3a31e60 JA |
7758 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7759 | int index) | |
7760 | { | |
7761 | #if defined(CONFIG_UNIX) | |
7762 | struct sock *sock = ctx->ring_sock->sk; | |
7763 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7764 | struct sk_buff *skb; | |
7765 | ||
7766 | /* | |
7767 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7768 | * file set. If there's no room, fall back to allocating a new skb | |
7769 | * and filling it in. | |
7770 | */ | |
7771 | spin_lock_irq(&head->lock); | |
7772 | skb = skb_peek(head); | |
7773 | if (skb) { | |
7774 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7775 | ||
7776 | if (fpl->count < SCM_MAX_FD) { | |
7777 | __skb_unlink(skb, head); | |
7778 | spin_unlock_irq(&head->lock); | |
7779 | fpl->fp[fpl->count] = get_file(file); | |
7780 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7781 | fpl->count++; | |
7782 | spin_lock_irq(&head->lock); | |
7783 | __skb_queue_head(head, skb); | |
7784 | } else { | |
7785 | skb = NULL; | |
7786 | } | |
7787 | } | |
7788 | spin_unlock_irq(&head->lock); | |
7789 | ||
7790 | if (skb) { | |
7791 | fput(file); | |
7792 | return 0; | |
7793 | } | |
7794 | ||
7795 | return __io_sqe_files_scm(ctx, 1, index); | |
7796 | #else | |
7797 | return 0; | |
7798 | #endif | |
7799 | } | |
7800 | ||
b60c8dce | 7801 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 7802 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 7803 | { |
269bbe5f | 7804 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7805 | |
269bbe5f BM |
7806 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7807 | if (!prsrc) | |
a5318d3c | 7808 | return -ENOMEM; |
05f3fb3c | 7809 | |
2d091d62 | 7810 | prsrc->tag = *io_get_tag_slot(data, idx); |
50238531 | 7811 | prsrc->rsrc = rsrc; |
e7c78371 | 7812 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7813 | return 0; |
05f3fb3c JA |
7814 | } |
7815 | ||
7816 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 7817 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
7818 | unsigned nr_args) |
7819 | { | |
c3bdad02 | 7820 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 7821 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 7822 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7823 | struct io_fixed_file *file_slot; |
7824 | struct file *file; | |
98f0b3b4 PB |
7825 | int fd, i, err = 0; |
7826 | unsigned int done; | |
05589553 | 7827 | bool needs_switch = false; |
c3a31e60 | 7828 | |
98f0b3b4 PB |
7829 | if (!ctx->file_data) |
7830 | return -ENXIO; | |
7831 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
7832 | return -EINVAL; |
7833 | ||
67973b93 | 7834 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
7835 | u64 tag = 0; |
7836 | ||
7837 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
7838 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
7839 | err = -EFAULT; |
7840 | break; | |
7841 | } | |
c3bdad02 PB |
7842 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
7843 | err = -EINVAL; | |
7844 | break; | |
7845 | } | |
4e0377a1 | 7846 | if (fd == IORING_REGISTER_FILES_SKIP) |
7847 | continue; | |
7848 | ||
67973b93 | 7849 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7850 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7851 | |
a04b0ac0 PB |
7852 | if (file_slot->file_ptr) { |
7853 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
7854 | err = io_queue_rsrc_removal(data, up->offset + done, |
7855 | ctx->rsrc_node, file); | |
a5318d3c HD |
7856 | if (err) |
7857 | break; | |
a04b0ac0 | 7858 | file_slot->file_ptr = 0; |
05589553 | 7859 | needs_switch = true; |
c3a31e60 JA |
7860 | } |
7861 | if (fd != -1) { | |
c3a31e60 JA |
7862 | file = fget(fd); |
7863 | if (!file) { | |
7864 | err = -EBADF; | |
7865 | break; | |
7866 | } | |
7867 | /* | |
7868 | * Don't allow io_uring instances to be registered. If | |
7869 | * UNIX isn't enabled, then this causes a reference | |
7870 | * cycle and this instance can never get freed. If UNIX | |
7871 | * is enabled we'll handle it just fine, but there's | |
7872 | * still no point in allowing a ring fd as it doesn't | |
7873 | * support regular read/write anyway. | |
7874 | */ | |
7875 | if (file->f_op == &io_uring_fops) { | |
7876 | fput(file); | |
7877 | err = -EBADF; | |
7878 | break; | |
7879 | } | |
2d091d62 | 7880 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 7881 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7882 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7883 | if (err) { |
a04b0ac0 | 7884 | file_slot->file_ptr = 0; |
f3bd9dae | 7885 | fput(file); |
c3a31e60 | 7886 | break; |
f3bd9dae | 7887 | } |
c3a31e60 | 7888 | } |
05f3fb3c JA |
7889 | } |
7890 | ||
a7f0ed5a PB |
7891 | if (needs_switch) |
7892 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7893 | return done ? done : err; |
7894 | } | |
05589553 | 7895 | |
5280f7e5 | 7896 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7897 | { |
7898 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7899 | ||
5280f7e5 PB |
7900 | req = io_put_req_find_next(req); |
7901 | return req ? &req->work : NULL; | |
7d723065 JA |
7902 | } |
7903 | ||
685fe7fe JA |
7904 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7905 | struct task_struct *task) | |
24369c2e | 7906 | { |
e941894e | 7907 | struct io_wq_hash *hash; |
24369c2e | 7908 | struct io_wq_data data; |
24369c2e | 7909 | unsigned int concurrency; |
24369c2e | 7910 | |
e941894e JA |
7911 | hash = ctx->hash_map; |
7912 | if (!hash) { | |
7913 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7914 | if (!hash) | |
7915 | return ERR_PTR(-ENOMEM); | |
7916 | refcount_set(&hash->refs, 1); | |
7917 | init_waitqueue_head(&hash->wait); | |
7918 | ctx->hash_map = hash; | |
24369c2e PB |
7919 | } |
7920 | ||
e941894e | 7921 | data.hash = hash; |
685fe7fe | 7922 | data.task = task; |
e9fd9396 | 7923 | data.free_work = io_free_work; |
f5fa38c5 | 7924 | data.do_work = io_wq_submit_work; |
24369c2e | 7925 | |
d25e3a3d JA |
7926 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7927 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7928 | |
5aa75ed5 | 7929 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7930 | } |
7931 | ||
5aa75ed5 JA |
7932 | static int io_uring_alloc_task_context(struct task_struct *task, |
7933 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7934 | { |
7935 | struct io_uring_task *tctx; | |
d8a6df10 | 7936 | int ret; |
0f212204 | 7937 | |
09899b19 | 7938 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
7939 | if (unlikely(!tctx)) |
7940 | return -ENOMEM; | |
7941 | ||
d8a6df10 JA |
7942 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7943 | if (unlikely(ret)) { | |
7944 | kfree(tctx); | |
7945 | return ret; | |
7946 | } | |
7947 | ||
685fe7fe | 7948 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7949 | if (IS_ERR(tctx->io_wq)) { |
7950 | ret = PTR_ERR(tctx->io_wq); | |
7951 | percpu_counter_destroy(&tctx->inflight); | |
7952 | kfree(tctx); | |
7953 | return ret; | |
7954 | } | |
7955 | ||
0f212204 JA |
7956 | xa_init(&tctx->xa); |
7957 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 7958 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7959 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7960 | task->io_uring = tctx; |
7cbf1722 JA |
7961 | spin_lock_init(&tctx->task_lock); |
7962 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 7963 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
7964 | return 0; |
7965 | } | |
7966 | ||
7967 | void __io_uring_free(struct task_struct *tsk) | |
7968 | { | |
7969 | struct io_uring_task *tctx = tsk->io_uring; | |
7970 | ||
7971 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 7972 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 7973 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 7974 | |
d8a6df10 | 7975 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7976 | kfree(tctx); |
7977 | tsk->io_uring = NULL; | |
7978 | } | |
7979 | ||
7e84e1c7 SG |
7980 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7981 | struct io_uring_params *p) | |
2b188cc1 JA |
7982 | { |
7983 | int ret; | |
7984 | ||
d25e3a3d JA |
7985 | /* Retain compatibility with failing for an invalid attach attempt */ |
7986 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7987 | IORING_SETUP_ATTACH_WQ) { | |
7988 | struct fd f; | |
7989 | ||
7990 | f = fdget(p->wq_fd); | |
7991 | if (!f.file) | |
7992 | return -ENXIO; | |
d25e3a3d | 7993 | fdput(f); |
f2a48dd0 PB |
7994 | if (f.file->f_op != &io_uring_fops) |
7995 | return -EINVAL; | |
d25e3a3d | 7996 | } |
6c271ce2 | 7997 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7998 | struct task_struct *tsk; |
534ca6d6 | 7999 | struct io_sq_data *sqd; |
26984fbf | 8000 | bool attached; |
534ca6d6 | 8001 | |
26984fbf | 8002 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8003 | if (IS_ERR(sqd)) { |
8004 | ret = PTR_ERR(sqd); | |
8005 | goto err; | |
8006 | } | |
69fb2131 | 8007 | |
7c30f36a | 8008 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8009 | ctx->sq_data = sqd; |
917257da JA |
8010 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8011 | if (!ctx->sq_thread_idle) | |
8012 | ctx->sq_thread_idle = HZ; | |
8013 | ||
78d7f6ba | 8014 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8015 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8016 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8017 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8018 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8019 | io_sq_thread_unpark(sqd); |
8020 | ||
de75a3d3 PB |
8021 | if (ret < 0) |
8022 | goto err; | |
8023 | if (attached) | |
5aa75ed5 | 8024 | return 0; |
aa06165d | 8025 | |
6c271ce2 | 8026 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8027 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8028 | |
917257da | 8029 | ret = -EINVAL; |
f2a48dd0 | 8030 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8031 | goto err_sqpoll; |
37d1e2e3 | 8032 | sqd->sq_cpu = cpu; |
6c271ce2 | 8033 | } else { |
37d1e2e3 | 8034 | sqd->sq_cpu = -1; |
6c271ce2 | 8035 | } |
37d1e2e3 JA |
8036 | |
8037 | sqd->task_pid = current->pid; | |
5c2469e0 | 8038 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8039 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8040 | if (IS_ERR(tsk)) { | |
8041 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8042 | goto err_sqpoll; |
6c271ce2 | 8043 | } |
97a73a0f | 8044 | |
46fe18b1 | 8045 | sqd->thread = tsk; |
97a73a0f | 8046 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8047 | wake_up_new_task(tsk); |
0f212204 JA |
8048 | if (ret) |
8049 | goto err; | |
6c271ce2 JA |
8050 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8051 | /* Can't have SQ_AFF without SQPOLL */ | |
8052 | ret = -EINVAL; | |
8053 | goto err; | |
8054 | } | |
8055 | ||
2b188cc1 | 8056 | return 0; |
f2a48dd0 PB |
8057 | err_sqpoll: |
8058 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8059 | err: |
37d1e2e3 | 8060 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8061 | return ret; |
8062 | } | |
8063 | ||
a087e2b5 BM |
8064 | static inline void __io_unaccount_mem(struct user_struct *user, |
8065 | unsigned long nr_pages) | |
2b188cc1 JA |
8066 | { |
8067 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8068 | } | |
8069 | ||
a087e2b5 BM |
8070 | static inline int __io_account_mem(struct user_struct *user, |
8071 | unsigned long nr_pages) | |
2b188cc1 JA |
8072 | { |
8073 | unsigned long page_limit, cur_pages, new_pages; | |
8074 | ||
8075 | /* Don't allow more pages than we can safely lock */ | |
8076 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8077 | ||
8078 | do { | |
8079 | cur_pages = atomic_long_read(&user->locked_vm); | |
8080 | new_pages = cur_pages + nr_pages; | |
8081 | if (new_pages > page_limit) | |
8082 | return -ENOMEM; | |
8083 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8084 | new_pages) != cur_pages); | |
8085 | ||
8086 | return 0; | |
8087 | } | |
8088 | ||
26bfa89e | 8089 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8090 | { |
62e398be | 8091 | if (ctx->user) |
a087e2b5 | 8092 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8093 | |
26bfa89e JA |
8094 | if (ctx->mm_account) |
8095 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8096 | } |
8097 | ||
26bfa89e | 8098 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8099 | { |
30975825 BM |
8100 | int ret; |
8101 | ||
62e398be | 8102 | if (ctx->user) { |
30975825 BM |
8103 | ret = __io_account_mem(ctx->user, nr_pages); |
8104 | if (ret) | |
8105 | return ret; | |
8106 | } | |
8107 | ||
26bfa89e JA |
8108 | if (ctx->mm_account) |
8109 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8110 | |
8111 | return 0; | |
8112 | } | |
8113 | ||
2b188cc1 JA |
8114 | static void io_mem_free(void *ptr) |
8115 | { | |
52e04ef4 MR |
8116 | struct page *page; |
8117 | ||
8118 | if (!ptr) | |
8119 | return; | |
2b188cc1 | 8120 | |
52e04ef4 | 8121 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8122 | if (put_page_testzero(page)) |
8123 | free_compound_page(page); | |
8124 | } | |
8125 | ||
8126 | static void *io_mem_alloc(size_t size) | |
8127 | { | |
8128 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8129 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8130 | |
8131 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8132 | } | |
8133 | ||
75b28aff HV |
8134 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8135 | size_t *sq_offset) | |
8136 | { | |
8137 | struct io_rings *rings; | |
8138 | size_t off, sq_array_size; | |
8139 | ||
8140 | off = struct_size(rings, cqes, cq_entries); | |
8141 | if (off == SIZE_MAX) | |
8142 | return SIZE_MAX; | |
8143 | ||
8144 | #ifdef CONFIG_SMP | |
8145 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8146 | if (off == 0) | |
8147 | return SIZE_MAX; | |
8148 | #endif | |
8149 | ||
b36200f5 DV |
8150 | if (sq_offset) |
8151 | *sq_offset = off; | |
8152 | ||
75b28aff HV |
8153 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8154 | if (sq_array_size == SIZE_MAX) | |
8155 | return SIZE_MAX; | |
8156 | ||
8157 | if (check_add_overflow(off, sq_array_size, &off)) | |
8158 | return SIZE_MAX; | |
8159 | ||
75b28aff HV |
8160 | return off; |
8161 | } | |
8162 | ||
41edf1a5 | 8163 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8164 | { |
41edf1a5 | 8165 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8166 | unsigned int i; |
8167 | ||
6224843d PB |
8168 | if (imu != ctx->dummy_ubuf) { |
8169 | for (i = 0; i < imu->nr_bvecs; i++) | |
8170 | unpin_user_page(imu->bvec[i].bv_page); | |
8171 | if (imu->acct_pages) | |
8172 | io_unaccount_mem(ctx, imu->acct_pages); | |
8173 | kvfree(imu); | |
8174 | } | |
41edf1a5 | 8175 | *slot = NULL; |
7f61a1e9 PB |
8176 | } |
8177 | ||
bd54b6fe | 8178 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8179 | { |
634d00df PB |
8180 | io_buffer_unmap(ctx, &prsrc->buf); |
8181 | prsrc->buf = NULL; | |
bd54b6fe | 8182 | } |
edafccee | 8183 | |
bd54b6fe BM |
8184 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8185 | { | |
8186 | unsigned int i; | |
edafccee | 8187 | |
7f61a1e9 PB |
8188 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8189 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8190 | kfree(ctx->user_bufs); |
bb6659cc | 8191 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8192 | ctx->user_bufs = NULL; |
bd54b6fe | 8193 | ctx->buf_data = NULL; |
edafccee | 8194 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8195 | } |
8196 | ||
0a96bbe4 | 8197 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8198 | { |
bd54b6fe | 8199 | int ret; |
edafccee | 8200 | |
bd54b6fe | 8201 | if (!ctx->buf_data) |
edafccee JA |
8202 | return -ENXIO; |
8203 | ||
bd54b6fe BM |
8204 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8205 | if (!ret) | |
8206 | __io_sqe_buffers_unregister(ctx); | |
8207 | return ret; | |
edafccee JA |
8208 | } |
8209 | ||
8210 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8211 | void __user *arg, unsigned index) | |
8212 | { | |
8213 | struct iovec __user *src; | |
8214 | ||
8215 | #ifdef CONFIG_COMPAT | |
8216 | if (ctx->compat) { | |
8217 | struct compat_iovec __user *ciovs; | |
8218 | struct compat_iovec ciov; | |
8219 | ||
8220 | ciovs = (struct compat_iovec __user *) arg; | |
8221 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8222 | return -EFAULT; | |
8223 | ||
d55e5f5b | 8224 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8225 | dst->iov_len = ciov.iov_len; |
8226 | return 0; | |
8227 | } | |
8228 | #endif | |
8229 | src = (struct iovec __user *) arg; | |
8230 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8231 | return -EFAULT; | |
8232 | return 0; | |
8233 | } | |
8234 | ||
de293938 JA |
8235 | /* |
8236 | * Not super efficient, but this is just a registration time. And we do cache | |
8237 | * the last compound head, so generally we'll only do a full search if we don't | |
8238 | * match that one. | |
8239 | * | |
8240 | * We check if the given compound head page has already been accounted, to | |
8241 | * avoid double accounting it. This allows us to account the full size of the | |
8242 | * page, not just the constituent pages of a huge page. | |
8243 | */ | |
8244 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8245 | int nr_pages, struct page *hpage) | |
8246 | { | |
8247 | int i, j; | |
8248 | ||
8249 | /* check current page array */ | |
8250 | for (i = 0; i < nr_pages; i++) { | |
8251 | if (!PageCompound(pages[i])) | |
8252 | continue; | |
8253 | if (compound_head(pages[i]) == hpage) | |
8254 | return true; | |
8255 | } | |
8256 | ||
8257 | /* check previously registered pages */ | |
8258 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8259 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8260 | |
8261 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8262 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8263 | continue; | |
8264 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8265 | return true; | |
8266 | } | |
8267 | } | |
8268 | ||
8269 | return false; | |
8270 | } | |
8271 | ||
8272 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8273 | int nr_pages, struct io_mapped_ubuf *imu, | |
8274 | struct page **last_hpage) | |
8275 | { | |
8276 | int i, ret; | |
8277 | ||
216e5835 | 8278 | imu->acct_pages = 0; |
de293938 JA |
8279 | for (i = 0; i < nr_pages; i++) { |
8280 | if (!PageCompound(pages[i])) { | |
8281 | imu->acct_pages++; | |
8282 | } else { | |
8283 | struct page *hpage; | |
8284 | ||
8285 | hpage = compound_head(pages[i]); | |
8286 | if (hpage == *last_hpage) | |
8287 | continue; | |
8288 | *last_hpage = hpage; | |
8289 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8290 | continue; | |
8291 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8292 | } | |
8293 | } | |
8294 | ||
8295 | if (!imu->acct_pages) | |
8296 | return 0; | |
8297 | ||
26bfa89e | 8298 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8299 | if (ret) |
8300 | imu->acct_pages = 0; | |
8301 | return ret; | |
8302 | } | |
8303 | ||
0a96bbe4 | 8304 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8305 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8306 | struct page **last_hpage) |
edafccee | 8307 | { |
41edf1a5 | 8308 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8309 | struct vm_area_struct **vmas = NULL; |
8310 | struct page **pages = NULL; | |
0a96bbe4 BM |
8311 | unsigned long off, start, end, ubuf; |
8312 | size_t size; | |
8313 | int ret, pret, nr_pages, i; | |
8314 | ||
6224843d PB |
8315 | if (!iov->iov_base) { |
8316 | *pimu = ctx->dummy_ubuf; | |
8317 | return 0; | |
8318 | } | |
8319 | ||
0a96bbe4 BM |
8320 | ubuf = (unsigned long) iov->iov_base; |
8321 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8322 | start = ubuf >> PAGE_SHIFT; | |
8323 | nr_pages = end - start; | |
8324 | ||
41edf1a5 | 8325 | *pimu = NULL; |
0a96bbe4 BM |
8326 | ret = -ENOMEM; |
8327 | ||
8328 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8329 | if (!pages) | |
8330 | goto done; | |
8331 | ||
8332 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8333 | GFP_KERNEL); | |
8334 | if (!vmas) | |
8335 | goto done; | |
edafccee | 8336 | |
41edf1a5 | 8337 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8338 | if (!imu) |
0a96bbe4 BM |
8339 | goto done; |
8340 | ||
8341 | ret = 0; | |
8342 | mmap_read_lock(current->mm); | |
8343 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8344 | pages, vmas); | |
8345 | if (pret == nr_pages) { | |
8346 | /* don't support file backed memory */ | |
8347 | for (i = 0; i < nr_pages; i++) { | |
8348 | struct vm_area_struct *vma = vmas[i]; | |
8349 | ||
40dad765 PB |
8350 | if (vma_is_shmem(vma)) |
8351 | continue; | |
0a96bbe4 BM |
8352 | if (vma->vm_file && |
8353 | !is_file_hugepages(vma->vm_file)) { | |
8354 | ret = -EOPNOTSUPP; | |
8355 | break; | |
8356 | } | |
8357 | } | |
8358 | } else { | |
8359 | ret = pret < 0 ? pret : -EFAULT; | |
8360 | } | |
8361 | mmap_read_unlock(current->mm); | |
8362 | if (ret) { | |
8363 | /* | |
8364 | * if we did partial map, or found file backed vmas, | |
8365 | * release any pages we did get | |
8366 | */ | |
8367 | if (pret > 0) | |
8368 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8369 | goto done; |
8370 | } | |
8371 | ||
8372 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8373 | if (ret) { | |
8374 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8375 | goto done; |
8376 | } | |
8377 | ||
8378 | off = ubuf & ~PAGE_MASK; | |
8379 | size = iov->iov_len; | |
8380 | for (i = 0; i < nr_pages; i++) { | |
8381 | size_t vec_len; | |
8382 | ||
8383 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8384 | imu->bvec[i].bv_page = pages[i]; | |
8385 | imu->bvec[i].bv_len = vec_len; | |
8386 | imu->bvec[i].bv_offset = off; | |
8387 | off = 0; | |
8388 | size -= vec_len; | |
8389 | } | |
8390 | /* store original address for later verification */ | |
8391 | imu->ubuf = ubuf; | |
4751f53d | 8392 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8393 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8394 | *pimu = imu; |
0a96bbe4 BM |
8395 | ret = 0; |
8396 | done: | |
41edf1a5 PB |
8397 | if (ret) |
8398 | kvfree(imu); | |
0a96bbe4 BM |
8399 | kvfree(pages); |
8400 | kvfree(vmas); | |
8401 | return ret; | |
8402 | } | |
8403 | ||
2b358604 | 8404 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8405 | { |
87094465 PB |
8406 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8407 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8408 | } |
edafccee | 8409 | |
2b358604 BM |
8410 | static int io_buffer_validate(struct iovec *iov) |
8411 | { | |
50e96989 PB |
8412 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8413 | ||
2b358604 BM |
8414 | /* |
8415 | * Don't impose further limits on the size and buffer | |
8416 | * constraints here, we'll -EINVAL later when IO is | |
8417 | * submitted if they are wrong. | |
8418 | */ | |
6224843d PB |
8419 | if (!iov->iov_base) |
8420 | return iov->iov_len ? -EFAULT : 0; | |
8421 | if (!iov->iov_len) | |
2b358604 | 8422 | return -EFAULT; |
edafccee | 8423 | |
2b358604 BM |
8424 | /* arbitrary limit, but we need something */ |
8425 | if (iov->iov_len > SZ_1G) | |
8426 | return -EFAULT; | |
edafccee | 8427 | |
50e96989 PB |
8428 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8429 | return -EOVERFLOW; | |
8430 | ||
2b358604 BM |
8431 | return 0; |
8432 | } | |
edafccee | 8433 | |
2b358604 | 8434 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 8435 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 8436 | { |
bd54b6fe BM |
8437 | struct page *last_hpage = NULL; |
8438 | struct io_rsrc_data *data; | |
2b358604 BM |
8439 | int i, ret; |
8440 | struct iovec iov; | |
edafccee | 8441 | |
87094465 PB |
8442 | if (ctx->user_bufs) |
8443 | return -EBUSY; | |
489809e2 | 8444 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 8445 | return -EINVAL; |
bd54b6fe | 8446 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8447 | if (ret) |
8448 | return ret; | |
d878c816 PB |
8449 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
8450 | if (ret) | |
8451 | return ret; | |
bd54b6fe BM |
8452 | ret = io_buffers_map_alloc(ctx, nr_args); |
8453 | if (ret) { | |
bb6659cc | 8454 | io_rsrc_data_free(data); |
bd54b6fe BM |
8455 | return ret; |
8456 | } | |
edafccee | 8457 | |
87094465 | 8458 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
8459 | ret = io_copy_iov(ctx, &iov, arg, i); |
8460 | if (ret) | |
0a96bbe4 | 8461 | break; |
2b358604 BM |
8462 | ret = io_buffer_validate(&iov); |
8463 | if (ret) | |
0a96bbe4 | 8464 | break; |
2d091d62 | 8465 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
8466 | ret = -EINVAL; |
8467 | break; | |
8468 | } | |
edafccee | 8469 | |
41edf1a5 PB |
8470 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
8471 | &last_hpage); | |
0a96bbe4 BM |
8472 | if (ret) |
8473 | break; | |
edafccee | 8474 | } |
0a96bbe4 | 8475 | |
bd54b6fe | 8476 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8477 | |
bd54b6fe BM |
8478 | ctx->buf_data = data; |
8479 | if (ret) | |
8480 | __io_sqe_buffers_unregister(ctx); | |
8481 | else | |
8482 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
8483 | return ret; |
8484 | } | |
8485 | ||
634d00df PB |
8486 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
8487 | struct io_uring_rsrc_update2 *up, | |
8488 | unsigned int nr_args) | |
8489 | { | |
8490 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
8491 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
8492 | struct page *last_hpage = NULL; |
8493 | bool needs_switch = false; | |
8494 | __u32 done; | |
8495 | int i, err; | |
8496 | ||
8497 | if (!ctx->buf_data) | |
8498 | return -ENXIO; | |
8499 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
8500 | return -EINVAL; | |
8501 | ||
8502 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
8503 | struct io_mapped_ubuf *imu; |
8504 | int offset = up->offset + done; | |
634d00df PB |
8505 | u64 tag = 0; |
8506 | ||
8507 | err = io_copy_iov(ctx, &iov, iovs, done); | |
8508 | if (err) | |
8509 | break; | |
8510 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
8511 | err = -EFAULT; | |
8512 | break; | |
8513 | } | |
0b8c0e7c PB |
8514 | err = io_buffer_validate(&iov); |
8515 | if (err) | |
8516 | break; | |
cf3770e7 CIK |
8517 | if (!iov.iov_base && tag) { |
8518 | err = -EINVAL; | |
8519 | break; | |
8520 | } | |
0b8c0e7c PB |
8521 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
8522 | if (err) | |
8523 | break; | |
634d00df | 8524 | |
0b8c0e7c | 8525 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 8526 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
8527 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
8528 | ctx->rsrc_node, ctx->user_bufs[i]); | |
8529 | if (unlikely(err)) { | |
8530 | io_buffer_unmap(ctx, &imu); | |
634d00df | 8531 | break; |
0b8c0e7c | 8532 | } |
634d00df PB |
8533 | ctx->user_bufs[i] = NULL; |
8534 | needs_switch = true; | |
8535 | } | |
8536 | ||
0b8c0e7c | 8537 | ctx->user_bufs[i] = imu; |
2d091d62 | 8538 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
8539 | } |
8540 | ||
8541 | if (needs_switch) | |
8542 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
8543 | return done ? done : err; | |
8544 | } | |
8545 | ||
9b402849 JA |
8546 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8547 | { | |
8548 | __s32 __user *fds = arg; | |
8549 | int fd; | |
8550 | ||
8551 | if (ctx->cq_ev_fd) | |
8552 | return -EBUSY; | |
8553 | ||
8554 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8555 | return -EFAULT; | |
8556 | ||
8557 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8558 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8559 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8560 | ctx->cq_ev_fd = NULL; | |
8561 | return ret; | |
8562 | } | |
8563 | ||
8564 | return 0; | |
8565 | } | |
8566 | ||
8567 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8568 | { | |
8569 | if (ctx->cq_ev_fd) { | |
8570 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8571 | ctx->cq_ev_fd = NULL; | |
8572 | return 0; | |
8573 | } | |
8574 | ||
8575 | return -ENXIO; | |
8576 | } | |
8577 | ||
5a2e745d JA |
8578 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8579 | { | |
9e15c3a0 JA |
8580 | struct io_buffer *buf; |
8581 | unsigned long index; | |
8582 | ||
8583 | xa_for_each(&ctx->io_buffers, index, buf) | |
8584 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8585 | } |
8586 | ||
68e68ee6 | 8587 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8588 | { |
68e68ee6 | 8589 | struct io_kiocb *req, *nxt; |
1b4c351f | 8590 | |
68e68ee6 JA |
8591 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8592 | if (tsk && req->task != tsk) | |
8593 | continue; | |
1b4c351f JA |
8594 | list_del(&req->compl.list); |
8595 | kmem_cache_free(req_cachep, req); | |
8596 | } | |
8597 | } | |
8598 | ||
4010fec4 | 8599 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8600 | { |
bf019da7 | 8601 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8602 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8603 | |
9a4fdbd8 JA |
8604 | mutex_lock(&ctx->uring_lock); |
8605 | ||
8e5c66c4 | 8606 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8607 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8608 | submit_state->reqs); | |
8e5c66c4 PB |
8609 | submit_state->free_reqs = 0; |
8610 | } | |
9a4fdbd8 | 8611 | |
dac7a098 | 8612 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8613 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8614 | mutex_unlock(&ctx->uring_lock); |
8615 | } | |
8616 | ||
bd54b6fe | 8617 | static bool io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 8618 | { |
bd54b6fe BM |
8619 | if (!data) |
8620 | return false; | |
8621 | if (!atomic_dec_and_test(&data->refs)) | |
8622 | wait_for_completion(&data->done); | |
8623 | return true; | |
8624 | } | |
04fc6c80 | 8625 | |
2b188cc1 JA |
8626 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8627 | { | |
37d1e2e3 | 8628 | io_sq_thread_finish(ctx); |
2aede0e4 | 8629 | |
37d1e2e3 | 8630 | if (ctx->mm_account) { |
2aede0e4 JA |
8631 | mmdrop(ctx->mm_account); |
8632 | ctx->mm_account = NULL; | |
30975825 | 8633 | } |
def596e9 | 8634 | |
8bad28d8 | 8635 | mutex_lock(&ctx->uring_lock); |
bd54b6fe BM |
8636 | if (io_wait_rsrc_data(ctx->buf_data)) |
8637 | __io_sqe_buffers_unregister(ctx); | |
8638 | if (io_wait_rsrc_data(ctx->file_data)) | |
08480400 | 8639 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
8640 | if (ctx->rings) |
8641 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8642 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8643 | io_eventfd_unregister(ctx); |
5a2e745d | 8644 | io_destroy_buffers(ctx); |
07db298a PB |
8645 | if (ctx->sq_creds) |
8646 | put_cred(ctx->sq_creds); | |
def596e9 | 8647 | |
a7f0ed5a PB |
8648 | /* there are no registered resources left, nobody uses it */ |
8649 | if (ctx->rsrc_node) | |
8650 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8651 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8652 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8653 | flush_delayed_work(&ctx->rsrc_put_work); |
8654 | ||
8655 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8656 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 8657 | |
2b188cc1 | 8658 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8659 | if (ctx->ring_sock) { |
8660 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8661 | sock_release(ctx->ring_sock); |
355e8d26 | 8662 | } |
2b188cc1 JA |
8663 | #endif |
8664 | ||
75b28aff | 8665 | io_mem_free(ctx->rings); |
2b188cc1 | 8666 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8667 | |
8668 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8669 | free_uid(ctx->user); |
4010fec4 | 8670 | io_req_caches_free(ctx); |
e941894e JA |
8671 | if (ctx->hash_map) |
8672 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8673 | kfree(ctx->cancel_hash); |
6224843d | 8674 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
8675 | kfree(ctx); |
8676 | } | |
8677 | ||
8678 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8679 | { | |
8680 | struct io_ring_ctx *ctx = file->private_data; | |
8681 | __poll_t mask = 0; | |
8682 | ||
311997b3 | 8683 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
8684 | /* |
8685 | * synchronizes with barrier from wq_has_sleeper call in | |
8686 | * io_commit_cqring | |
8687 | */ | |
2b188cc1 | 8688 | smp_rmb(); |
90554200 | 8689 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8690 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8691 | |
8692 | /* | |
8693 | * Don't flush cqring overflow list here, just do a simple check. | |
8694 | * Otherwise there could possible be ABBA deadlock: | |
8695 | * CPU0 CPU1 | |
8696 | * ---- ---- | |
8697 | * lock(&ctx->uring_lock); | |
8698 | * lock(&ep->mtx); | |
8699 | * lock(&ctx->uring_lock); | |
8700 | * lock(&ep->mtx); | |
8701 | * | |
8702 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8703 | * pushs them to do the flush. | |
8704 | */ | |
5ed7a37d | 8705 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
8706 | mask |= EPOLLIN | EPOLLRDNORM; |
8707 | ||
8708 | return mask; | |
8709 | } | |
8710 | ||
8711 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8712 | { | |
8713 | struct io_ring_ctx *ctx = file->private_data; | |
8714 | ||
8715 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8716 | } | |
8717 | ||
0bead8cd | 8718 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8719 | { |
4379bf8b | 8720 | const struct cred *creds; |
071698e1 | 8721 | |
61cf9370 | 8722 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8723 | if (creds) { |
8724 | put_cred(creds); | |
0bead8cd | 8725 | return 0; |
1e6fa521 | 8726 | } |
0bead8cd YD |
8727 | |
8728 | return -EINVAL; | |
8729 | } | |
8730 | ||
9b465711 | 8731 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8732 | { |
9b465711 | 8733 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8734 | } |
8735 | ||
d56d938b PB |
8736 | struct io_tctx_exit { |
8737 | struct callback_head task_work; | |
8738 | struct completion completion; | |
baf186c4 | 8739 | struct io_ring_ctx *ctx; |
d56d938b PB |
8740 | }; |
8741 | ||
8742 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8743 | { | |
8744 | struct io_uring_task *tctx = current->io_uring; | |
8745 | struct io_tctx_exit *work; | |
8746 | ||
8747 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8748 | /* | |
8749 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8750 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8751 | */ | |
8752 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 8753 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
8754 | complete(&work->completion); |
8755 | } | |
8756 | ||
28090c13 PB |
8757 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8758 | { | |
8759 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8760 | ||
8761 | return req->ctx == data; | |
8762 | } | |
8763 | ||
85faa7b8 JA |
8764 | static void io_ring_exit_work(struct work_struct *work) |
8765 | { | |
d56d938b | 8766 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8767 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8768 | struct io_tctx_exit exit; |
8769 | struct io_tctx_node *node; | |
8770 | int ret; | |
85faa7b8 | 8771 | |
56952e91 JA |
8772 | /* |
8773 | * If we're doing polled IO and end up having requests being | |
8774 | * submitted async (out-of-line), then completions can come in while | |
8775 | * we're waiting for refs to drop. We need to reap these manually, | |
8776 | * as nobody else will be looking for them. | |
8777 | */ | |
b2edc0a7 | 8778 | do { |
3dd0c97a | 8779 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
8780 | if (ctx->sq_data) { |
8781 | struct io_sq_data *sqd = ctx->sq_data; | |
8782 | struct task_struct *tsk; | |
8783 | ||
8784 | io_sq_thread_park(sqd); | |
8785 | tsk = sqd->thread; | |
8786 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
8787 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
8788 | io_cancel_ctx_cb, ctx, true); | |
8789 | io_sq_thread_unpark(sqd); | |
8790 | } | |
b5bb3a24 PB |
8791 | |
8792 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8793 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8794 | |
7f00651a PB |
8795 | init_completion(&exit.completion); |
8796 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8797 | exit.ctx = ctx; | |
89b5066e PB |
8798 | /* |
8799 | * Some may use context even when all refs and requests have been put, | |
8800 | * and they are free to do so while still holding uring_lock or | |
8801 | * completion_lock, see __io_req_task_submit(). Apart from other work, | |
8802 | * this lock/unlock section also waits them to finish. | |
8803 | */ | |
d56d938b PB |
8804 | mutex_lock(&ctx->uring_lock); |
8805 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8806 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8807 | ||
d56d938b PB |
8808 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8809 | ctx_node); | |
7f00651a PB |
8810 | /* don't spin on a single task if cancellation failed */ |
8811 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8812 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8813 | if (WARN_ON_ONCE(ret)) | |
8814 | continue; | |
8815 | wake_up_process(node->task); | |
8816 | ||
8817 | mutex_unlock(&ctx->uring_lock); | |
8818 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8819 | mutex_lock(&ctx->uring_lock); |
8820 | } | |
8821 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8822 | spin_lock_irq(&ctx->completion_lock); |
8823 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8824 | |
85faa7b8 JA |
8825 | io_ring_ctx_free(ctx); |
8826 | } | |
8827 | ||
80c4cbdb PB |
8828 | /* Returns true if we found and killed one or more timeouts */ |
8829 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 8830 | bool cancel_all) |
80c4cbdb PB |
8831 | { |
8832 | struct io_kiocb *req, *tmp; | |
8833 | int canceled = 0; | |
8834 | ||
8835 | spin_lock_irq(&ctx->completion_lock); | |
8836 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
3dd0c97a | 8837 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
8838 | io_kill_timeout(req, -ECANCELED); |
8839 | canceled++; | |
8840 | } | |
8841 | } | |
51520426 PB |
8842 | if (canceled != 0) |
8843 | io_commit_cqring(ctx); | |
80c4cbdb | 8844 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8845 | if (canceled != 0) |
8846 | io_cqring_ev_posted(ctx); | |
8847 | return canceled != 0; | |
8848 | } | |
8849 | ||
2b188cc1 JA |
8850 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8851 | { | |
61cf9370 MWO |
8852 | unsigned long index; |
8853 | struct creds *creds; | |
8854 | ||
2b188cc1 JA |
8855 | mutex_lock(&ctx->uring_lock); |
8856 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8857 | if (ctx->rings) |
6c2450ae | 8858 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8859 | xa_for_each(&ctx->personalities, index, creds) |
8860 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8861 | mutex_unlock(&ctx->uring_lock); |
8862 | ||
3dd0c97a PB |
8863 | io_kill_timeouts(ctx, NULL, true); |
8864 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 8865 | |
15dff286 | 8866 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8867 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8868 | |
85faa7b8 | 8869 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8870 | /* |
8871 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8872 | * if we're exiting a ton of rings at the same time. It just adds | |
8873 | * noise and overhead, there's no discernable change in runtime | |
8874 | * over using system_wq. | |
8875 | */ | |
8876 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8877 | } |
8878 | ||
8879 | static int io_uring_release(struct inode *inode, struct file *file) | |
8880 | { | |
8881 | struct io_ring_ctx *ctx = file->private_data; | |
8882 | ||
8883 | file->private_data = NULL; | |
8884 | io_ring_ctx_wait_and_kill(ctx); | |
8885 | return 0; | |
8886 | } | |
8887 | ||
f6edbabb PB |
8888 | struct io_task_cancel { |
8889 | struct task_struct *task; | |
3dd0c97a | 8890 | bool all; |
f6edbabb | 8891 | }; |
f254ac04 | 8892 | |
f6edbabb | 8893 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8894 | { |
9a472ef7 | 8895 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8896 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8897 | bool ret; |
8898 | ||
3dd0c97a | 8899 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8900 | unsigned long flags; |
8901 | struct io_ring_ctx *ctx = req->ctx; | |
8902 | ||
8903 | /* protect against races with linked timeouts */ | |
8904 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
3dd0c97a | 8905 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8906 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8907 | } else { | |
3dd0c97a | 8908 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8909 | } |
8910 | return ret; | |
b711d4ea JA |
8911 | } |
8912 | ||
e1915f76 | 8913 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 8914 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 8915 | { |
e1915f76 | 8916 | struct io_defer_entry *de; |
b7ddce3c PB |
8917 | LIST_HEAD(list); |
8918 | ||
8919 | spin_lock_irq(&ctx->completion_lock); | |
8920 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
3dd0c97a | 8921 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
8922 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8923 | break; | |
8924 | } | |
8925 | } | |
8926 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8927 | if (list_empty(&list)) |
8928 | return false; | |
b7ddce3c PB |
8929 | |
8930 | while (!list_empty(&list)) { | |
8931 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8932 | list_del_init(&de->list); | |
f41db273 | 8933 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8934 | kfree(de); |
8935 | } | |
e1915f76 | 8936 | return true; |
b7ddce3c PB |
8937 | } |
8938 | ||
1b00764f PB |
8939 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
8940 | { | |
8941 | struct io_tctx_node *node; | |
8942 | enum io_wq_cancel cret; | |
8943 | bool ret = false; | |
8944 | ||
8945 | mutex_lock(&ctx->uring_lock); | |
8946 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8947 | struct io_uring_task *tctx = node->task->io_uring; | |
8948 | ||
8949 | /* | |
8950 | * io_wq will stay alive while we hold uring_lock, because it's | |
8951 | * killed after ctx nodes, which requires to take the lock. | |
8952 | */ | |
8953 | if (!tctx || !tctx->io_wq) | |
8954 | continue; | |
8955 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8956 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8957 | } | |
8958 | mutex_unlock(&ctx->uring_lock); | |
8959 | ||
8960 | return ret; | |
8961 | } | |
8962 | ||
9936c7c2 PB |
8963 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8964 | struct task_struct *task, | |
3dd0c97a | 8965 | bool cancel_all) |
9936c7c2 | 8966 | { |
3dd0c97a | 8967 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 8968 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8969 | |
8970 | while (1) { | |
8971 | enum io_wq_cancel cret; | |
8972 | bool ret = false; | |
8973 | ||
1b00764f PB |
8974 | if (!task) { |
8975 | ret |= io_uring_try_cancel_iowq(ctx); | |
8976 | } else if (tctx && tctx->io_wq) { | |
8977 | /* | |
8978 | * Cancels requests of all rings, not only @ctx, but | |
8979 | * it's fine as the task is in exit/exec. | |
8980 | */ | |
5aa75ed5 | 8981 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8982 | &cancel, true); |
8983 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8984 | } | |
8985 | ||
8986 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 8987 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 8988 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
8989 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8990 | io_iopoll_try_reap_events(ctx); | |
8991 | ret = true; | |
8992 | } | |
8993 | } | |
8994 | ||
3dd0c97a PB |
8995 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
8996 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
8997 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
9936c7c2 | 8998 | ret |= io_run_task_work(); |
ba50a036 | 8999 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
9000 | if (!ret) |
9001 | break; | |
9002 | cond_resched(); | |
9003 | } | |
9004 | } | |
9005 | ||
eef51daa | 9006 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 9007 | { |
236434c3 | 9008 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 9009 | struct io_tctx_node *node; |
a528b04e | 9010 | int ret; |
236434c3 MWO |
9011 | |
9012 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9013 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9014 | if (unlikely(ret)) |
9015 | return ret; | |
236434c3 | 9016 | tctx = current->io_uring; |
0f212204 | 9017 | } |
cf27f3b1 PB |
9018 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9019 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9020 | if (!node) | |
9021 | return -ENOMEM; | |
9022 | node->ctx = ctx; | |
9023 | node->task = current; | |
13bf43f5 | 9024 | |
cf27f3b1 PB |
9025 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9026 | node, GFP_KERNEL)); | |
9027 | if (ret) { | |
9028 | kfree(node); | |
9029 | return ret; | |
0f212204 | 9030 | } |
cf27f3b1 PB |
9031 | |
9032 | mutex_lock(&ctx->uring_lock); | |
9033 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9034 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9035 | } |
cf27f3b1 | 9036 | tctx->last = ctx; |
0f212204 JA |
9037 | return 0; |
9038 | } | |
9039 | ||
cf27f3b1 PB |
9040 | /* |
9041 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9042 | */ | |
eef51daa | 9043 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9044 | { |
9045 | struct io_uring_task *tctx = current->io_uring; | |
9046 | ||
9047 | if (likely(tctx && tctx->last == ctx)) | |
9048 | return 0; | |
eef51daa | 9049 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9050 | } |
9051 | ||
0f212204 JA |
9052 | /* |
9053 | * Remove this io_uring_file -> task mapping. | |
9054 | */ | |
eef51daa | 9055 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9056 | { |
9057 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9058 | struct io_tctx_node *node; |
2941267b | 9059 | |
eebd2e37 PB |
9060 | if (!tctx) |
9061 | return; | |
13bf43f5 PB |
9062 | node = xa_erase(&tctx->xa, index); |
9063 | if (!node) | |
2941267b | 9064 | return; |
0f212204 | 9065 | |
13bf43f5 PB |
9066 | WARN_ON_ONCE(current != node->task); |
9067 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9068 | ||
9069 | mutex_lock(&node->ctx->uring_lock); | |
9070 | list_del(&node->ctx_node); | |
9071 | mutex_unlock(&node->ctx->uring_lock); | |
9072 | ||
baf186c4 | 9073 | if (tctx->last == node->ctx) |
0f212204 | 9074 | tctx->last = NULL; |
13bf43f5 | 9075 | kfree(node); |
0f212204 JA |
9076 | } |
9077 | ||
8452d4a6 | 9078 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9079 | { |
ba5ef6dc | 9080 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9081 | struct io_tctx_node *node; |
de7f1d9e PB |
9082 | unsigned long index; |
9083 | ||
13bf43f5 | 9084 | xa_for_each(&tctx->xa, index, node) |
eef51daa | 9085 | io_uring_del_tctx_node(index); |
b16ef427 ME |
9086 | if (wq) { |
9087 | /* | |
9088 | * Must be after io_uring_del_task_file() (removes nodes under | |
9089 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9090 | */ | |
9091 | tctx->io_wq = NULL; | |
ba5ef6dc | 9092 | io_wq_put_and_exit(wq); |
b16ef427 | 9093 | } |
de7f1d9e PB |
9094 | } |
9095 | ||
3f48cf18 | 9096 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9097 | { |
3f48cf18 PB |
9098 | if (tracked) |
9099 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9100 | return percpu_counter_sum(&tctx->inflight); |
9101 | } | |
9102 | ||
09899b19 PB |
9103 | static void io_uring_drop_tctx_refs(struct task_struct *task) |
9104 | { | |
9105 | struct io_uring_task *tctx = task->io_uring; | |
9106 | unsigned int refs = tctx->cached_refs; | |
9107 | ||
9108 | tctx->cached_refs = 0; | |
9109 | percpu_counter_sub(&tctx->inflight, refs); | |
9110 | put_task_struct_many(task, refs); | |
9111 | } | |
9112 | ||
78cc687b PB |
9113 | /* |
9114 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
9115 | * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation. | |
9116 | */ | |
9117 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9118 | { |
521d6a73 | 9119 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9120 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9121 | s64 inflight; |
9122 | DEFINE_WAIT(wait); | |
fdaf083c | 9123 | |
78cc687b PB |
9124 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9125 | ||
6d042ffb PO |
9126 | if (!current->io_uring) |
9127 | return; | |
17a91051 PB |
9128 | if (tctx->io_wq) |
9129 | io_wq_exit_start(tctx->io_wq); | |
9130 | ||
09899b19 | 9131 | io_uring_drop_tctx_refs(current); |
0e9ddb39 PB |
9132 | atomic_inc(&tctx->in_idle); |
9133 | do { | |
9134 | /* read completions before cancelations */ | |
78cc687b | 9135 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9136 | if (!inflight) |
9137 | break; | |
fdaf083c | 9138 | |
78cc687b PB |
9139 | if (!sqd) { |
9140 | struct io_tctx_node *node; | |
9141 | unsigned long index; | |
0f212204 | 9142 | |
78cc687b PB |
9143 | xa_for_each(&tctx->xa, index, node) { |
9144 | /* sqpoll task will cancel all its requests */ | |
9145 | if (node->ctx->sq_data) | |
9146 | continue; | |
9147 | io_uring_try_cancel_requests(node->ctx, current, | |
9148 | cancel_all); | |
9149 | } | |
9150 | } else { | |
9151 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9152 | io_uring_try_cancel_requests(ctx, current, | |
9153 | cancel_all); | |
9154 | } | |
17a91051 | 9155 | |
0f212204 | 9156 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
0f212204 | 9157 | /* |
a1bb3cd5 PB |
9158 | * If we've seen completions, retry without waiting. This |
9159 | * avoids a race where a completion comes in before we did | |
9160 | * prepare_to_wait(). | |
0f212204 | 9161 | */ |
3dd0c97a | 9162 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9163 | schedule(); |
f57555ed | 9164 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9165 | } while (1); |
fdaf083c | 9166 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9167 | |
8452d4a6 | 9168 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9169 | if (cancel_all) { |
3f48cf18 PB |
9170 | /* for exec all current's requests should be gone, kill tctx */ |
9171 | __io_uring_free(current); | |
9172 | } | |
44e728b8 PB |
9173 | } |
9174 | ||
78cc687b PB |
9175 | void __io_uring_cancel(struct files_struct *files) |
9176 | { | |
9177 | io_uring_cancel_generic(!files, NULL); | |
9178 | } | |
9179 | ||
6c5c240e RP |
9180 | static void *io_uring_validate_mmap_request(struct file *file, |
9181 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9182 | { |
2b188cc1 | 9183 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9184 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9185 | struct page *page; |
9186 | void *ptr; | |
9187 | ||
9188 | switch (offset) { | |
9189 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9190 | case IORING_OFF_CQ_RING: |
9191 | ptr = ctx->rings; | |
2b188cc1 JA |
9192 | break; |
9193 | case IORING_OFF_SQES: | |
9194 | ptr = ctx->sq_sqes; | |
9195 | break; | |
2b188cc1 | 9196 | default: |
6c5c240e | 9197 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9198 | } |
9199 | ||
9200 | page = virt_to_head_page(ptr); | |
a50b854e | 9201 | if (sz > page_size(page)) |
6c5c240e RP |
9202 | return ERR_PTR(-EINVAL); |
9203 | ||
9204 | return ptr; | |
9205 | } | |
9206 | ||
9207 | #ifdef CONFIG_MMU | |
9208 | ||
9209 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9210 | { | |
9211 | size_t sz = vma->vm_end - vma->vm_start; | |
9212 | unsigned long pfn; | |
9213 | void *ptr; | |
9214 | ||
9215 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9216 | if (IS_ERR(ptr)) | |
9217 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9218 | |
9219 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9220 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9221 | } | |
9222 | ||
6c5c240e RP |
9223 | #else /* !CONFIG_MMU */ |
9224 | ||
9225 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9226 | { | |
9227 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9228 | } | |
9229 | ||
9230 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9231 | { | |
9232 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9233 | } | |
9234 | ||
9235 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9236 | unsigned long addr, unsigned long len, | |
9237 | unsigned long pgoff, unsigned long flags) | |
9238 | { | |
9239 | void *ptr; | |
9240 | ||
9241 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9242 | if (IS_ERR(ptr)) | |
9243 | return PTR_ERR(ptr); | |
9244 | ||
9245 | return (unsigned long) ptr; | |
9246 | } | |
9247 | ||
9248 | #endif /* !CONFIG_MMU */ | |
9249 | ||
d9d05217 | 9250 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9251 | { |
9252 | DEFINE_WAIT(wait); | |
9253 | ||
9254 | do { | |
9255 | if (!io_sqring_full(ctx)) | |
9256 | break; | |
90554200 JA |
9257 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9258 | ||
9259 | if (!io_sqring_full(ctx)) | |
9260 | break; | |
90554200 JA |
9261 | schedule(); |
9262 | } while (!signal_pending(current)); | |
9263 | ||
9264 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9265 | return 0; |
90554200 JA |
9266 | } |
9267 | ||
c73ebb68 HX |
9268 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9269 | struct __kernel_timespec __user **ts, | |
9270 | const sigset_t __user **sig) | |
9271 | { | |
9272 | struct io_uring_getevents_arg arg; | |
9273 | ||
9274 | /* | |
9275 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9276 | * is just a pointer to the sigset_t. | |
9277 | */ | |
9278 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9279 | *sig = (const sigset_t __user *) argp; | |
9280 | *ts = NULL; | |
9281 | return 0; | |
9282 | } | |
9283 | ||
9284 | /* | |
9285 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9286 | * timespec and sigset_t pointers if good. | |
9287 | */ | |
9288 | if (*argsz != sizeof(arg)) | |
9289 | return -EINVAL; | |
9290 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9291 | return -EFAULT; | |
9292 | *sig = u64_to_user_ptr(arg.sigmask); | |
9293 | *argsz = arg.sigmask_sz; | |
9294 | *ts = u64_to_user_ptr(arg.ts); | |
9295 | return 0; | |
9296 | } | |
9297 | ||
2b188cc1 | 9298 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9299 | u32, min_complete, u32, flags, const void __user *, argp, |
9300 | size_t, argsz) | |
2b188cc1 JA |
9301 | { |
9302 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9303 | int submitted = 0; |
9304 | struct fd f; | |
33f993da | 9305 | long ret; |
2b188cc1 | 9306 | |
4c6e277c | 9307 | io_run_task_work(); |
b41e9852 | 9308 | |
33f993da PB |
9309 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9310 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9311 | return -EINVAL; |
9312 | ||
9313 | f = fdget(fd); | |
33f993da | 9314 | if (unlikely(!f.file)) |
2b188cc1 JA |
9315 | return -EBADF; |
9316 | ||
9317 | ret = -EOPNOTSUPP; | |
33f993da | 9318 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9319 | goto out_fput; |
9320 | ||
9321 | ret = -ENXIO; | |
9322 | ctx = f.file->private_data; | |
33f993da | 9323 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9324 | goto out_fput; |
9325 | ||
7e84e1c7 | 9326 | ret = -EBADFD; |
33f993da | 9327 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9328 | goto out; |
9329 | ||
6c271ce2 JA |
9330 | /* |
9331 | * For SQ polling, the thread will do all submissions and completions. | |
9332 | * Just return the requested submit count, and wake the thread if | |
9333 | * we were asked to. | |
9334 | */ | |
b2a9eada | 9335 | ret = 0; |
6c271ce2 | 9336 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9337 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9338 | |
d9d05217 | 9339 | ret = -EOWNERDEAD; |
04147488 SM |
9340 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9341 | goto out; | |
9342 | } | |
6c271ce2 | 9343 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9344 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9345 | if (flags & IORING_ENTER_SQ_WAIT) { |
9346 | ret = io_sqpoll_wait_sq(ctx); | |
9347 | if (ret) | |
9348 | goto out; | |
9349 | } | |
6c271ce2 | 9350 | submitted = to_submit; |
b2a9eada | 9351 | } else if (to_submit) { |
eef51daa | 9352 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
9353 | if (unlikely(ret)) |
9354 | goto out; | |
2b188cc1 | 9355 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9356 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9357 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9358 | |
9359 | if (submitted != to_submit) | |
9360 | goto out; | |
2b188cc1 JA |
9361 | } |
9362 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9363 | const sigset_t __user *sig; |
9364 | struct __kernel_timespec __user *ts; | |
9365 | ||
9366 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9367 | if (unlikely(ret)) | |
9368 | goto out; | |
9369 | ||
2b188cc1 JA |
9370 | min_complete = min(min_complete, ctx->cq_entries); |
9371 | ||
32b2244a XW |
9372 | /* |
9373 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9374 | * space applications don't need to do io completion events | |
9375 | * polling again, they can rely on io_sq_thread to do polling | |
9376 | * work, which can reduce cpu usage and uring_lock contention. | |
9377 | */ | |
9378 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9379 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9380 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9381 | } else { |
c73ebb68 | 9382 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9383 | } |
2b188cc1 JA |
9384 | } |
9385 | ||
7c504e65 | 9386 | out: |
6805b32e | 9387 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9388 | out_fput: |
9389 | fdput(f); | |
9390 | return submitted ? submitted : ret; | |
9391 | } | |
9392 | ||
bebdb65e | 9393 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9394 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9395 | const struct cred *cred) | |
87ce955b | 9396 | { |
87ce955b JA |
9397 | struct user_namespace *uns = seq_user_ns(m); |
9398 | struct group_info *gi; | |
9399 | kernel_cap_t cap; | |
9400 | unsigned __capi; | |
9401 | int g; | |
9402 | ||
9403 | seq_printf(m, "%5d\n", id); | |
9404 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9405 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9406 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9407 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9408 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9409 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9410 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9411 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9412 | seq_puts(m, "\n\tGroups:\t"); | |
9413 | gi = cred->group_info; | |
9414 | for (g = 0; g < gi->ngroups; g++) { | |
9415 | seq_put_decimal_ull(m, g ? " " : "", | |
9416 | from_kgid_munged(uns, gi->gid[g])); | |
9417 | } | |
9418 | seq_puts(m, "\n\tCapEff:\t"); | |
9419 | cap = cred->cap_effective; | |
9420 | CAP_FOR_EACH_U32(__capi) | |
9421 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9422 | seq_putc(m, '\n'); | |
9423 | return 0; | |
9424 | } | |
9425 | ||
9426 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9427 | { | |
dbbe9c64 | 9428 | struct io_sq_data *sq = NULL; |
fad8e0de | 9429 | bool has_lock; |
87ce955b JA |
9430 | int i; |
9431 | ||
fad8e0de JA |
9432 | /* |
9433 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9434 | * since fdinfo case grabs it in the opposite direction of normal use | |
9435 | * cases. If we fail to get the lock, we just don't iterate any | |
9436 | * structures that could be going away outside the io_uring mutex. | |
9437 | */ | |
9438 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9439 | ||
5f3f26f9 | 9440 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9441 | sq = ctx->sq_data; |
5f3f26f9 JA |
9442 | if (!sq->thread) |
9443 | sq = NULL; | |
9444 | } | |
dbbe9c64 JQ |
9445 | |
9446 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9447 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9448 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9449 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9450 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9451 | |
87ce955b JA |
9452 | if (f) |
9453 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9454 | else | |
9455 | seq_printf(m, "%5u: <none>\n", i); | |
9456 | } | |
9457 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9458 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9459 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9460 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9461 | |
4751f53d | 9462 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9463 | } |
61cf9370 MWO |
9464 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9465 | unsigned long index; | |
9466 | const struct cred *cred; | |
9467 | ||
87ce955b | 9468 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9469 | xa_for_each(&ctx->personalities, index, cred) |
9470 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9471 | } |
d7718a9d JA |
9472 | seq_printf(m, "PollList:\n"); |
9473 | spin_lock_irq(&ctx->completion_lock); | |
9474 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9475 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9476 | struct io_kiocb *req; | |
9477 | ||
9478 | hlist_for_each_entry(req, list, hash_node) | |
9479 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9480 | req->task->task_works != NULL); | |
9481 | } | |
9482 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9483 | if (has_lock) |
9484 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9485 | } |
9486 | ||
9487 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9488 | { | |
9489 | struct io_ring_ctx *ctx = f->private_data; | |
9490 | ||
9491 | if (percpu_ref_tryget(&ctx->refs)) { | |
9492 | __io_uring_show_fdinfo(ctx, m); | |
9493 | percpu_ref_put(&ctx->refs); | |
9494 | } | |
9495 | } | |
bebdb65e | 9496 | #endif |
87ce955b | 9497 | |
2b188cc1 JA |
9498 | static const struct file_operations io_uring_fops = { |
9499 | .release = io_uring_release, | |
9500 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9501 | #ifndef CONFIG_MMU |
9502 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9503 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9504 | #endif | |
2b188cc1 JA |
9505 | .poll = io_uring_poll, |
9506 | .fasync = io_uring_fasync, | |
bebdb65e | 9507 | #ifdef CONFIG_PROC_FS |
87ce955b | 9508 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9509 | #endif |
2b188cc1 JA |
9510 | }; |
9511 | ||
9512 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9513 | struct io_uring_params *p) | |
9514 | { | |
75b28aff HV |
9515 | struct io_rings *rings; |
9516 | size_t size, sq_array_offset; | |
2b188cc1 | 9517 | |
bd740481 JA |
9518 | /* make sure these are sane, as we already accounted them */ |
9519 | ctx->sq_entries = p->sq_entries; | |
9520 | ctx->cq_entries = p->cq_entries; | |
9521 | ||
75b28aff HV |
9522 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9523 | if (size == SIZE_MAX) | |
9524 | return -EOVERFLOW; | |
9525 | ||
9526 | rings = io_mem_alloc(size); | |
9527 | if (!rings) | |
2b188cc1 JA |
9528 | return -ENOMEM; |
9529 | ||
75b28aff HV |
9530 | ctx->rings = rings; |
9531 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9532 | rings->sq_ring_mask = p->sq_entries - 1; | |
9533 | rings->cq_ring_mask = p->cq_entries - 1; | |
9534 | rings->sq_ring_entries = p->sq_entries; | |
9535 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
9536 | |
9537 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9538 | if (size == SIZE_MAX) { |
9539 | io_mem_free(ctx->rings); | |
9540 | ctx->rings = NULL; | |
2b188cc1 | 9541 | return -EOVERFLOW; |
eb065d30 | 9542 | } |
2b188cc1 JA |
9543 | |
9544 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9545 | if (!ctx->sq_sqes) { |
9546 | io_mem_free(ctx->rings); | |
9547 | ctx->rings = NULL; | |
2b188cc1 | 9548 | return -ENOMEM; |
eb065d30 | 9549 | } |
2b188cc1 | 9550 | |
2b188cc1 JA |
9551 | return 0; |
9552 | } | |
9553 | ||
9faadcc8 PB |
9554 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9555 | { | |
9556 | int ret, fd; | |
9557 | ||
9558 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9559 | if (fd < 0) | |
9560 | return fd; | |
9561 | ||
eef51daa | 9562 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
9563 | if (ret) { |
9564 | put_unused_fd(fd); | |
9565 | return ret; | |
9566 | } | |
9567 | fd_install(fd, file); | |
9568 | return fd; | |
9569 | } | |
9570 | ||
2b188cc1 JA |
9571 | /* |
9572 | * Allocate an anonymous fd, this is what constitutes the application | |
9573 | * visible backing of an io_uring instance. The application mmaps this | |
9574 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9575 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9576 | */ | |
9faadcc8 | 9577 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9578 | { |
9579 | struct file *file; | |
9faadcc8 | 9580 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9581 | int ret; |
9582 | ||
2b188cc1 JA |
9583 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9584 | &ctx->ring_sock); | |
9585 | if (ret) | |
9faadcc8 | 9586 | return ERR_PTR(ret); |
2b188cc1 JA |
9587 | #endif |
9588 | ||
2b188cc1 JA |
9589 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9590 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9591 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9592 | if (IS_ERR(file)) { |
9593 | sock_release(ctx->ring_sock); | |
9594 | ctx->ring_sock = NULL; | |
9595 | } else { | |
9596 | ctx->ring_sock->file = file; | |
0f212204 | 9597 | } |
2b188cc1 | 9598 | #endif |
9faadcc8 | 9599 | return file; |
2b188cc1 JA |
9600 | } |
9601 | ||
7f13657d XW |
9602 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9603 | struct io_uring_params __user *params) | |
2b188cc1 | 9604 | { |
2b188cc1 | 9605 | struct io_ring_ctx *ctx; |
9faadcc8 | 9606 | struct file *file; |
2b188cc1 JA |
9607 | int ret; |
9608 | ||
8110c1a6 | 9609 | if (!entries) |
2b188cc1 | 9610 | return -EINVAL; |
8110c1a6 JA |
9611 | if (entries > IORING_MAX_ENTRIES) { |
9612 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9613 | return -EINVAL; | |
9614 | entries = IORING_MAX_ENTRIES; | |
9615 | } | |
2b188cc1 JA |
9616 | |
9617 | /* | |
9618 | * Use twice as many entries for the CQ ring. It's possible for the | |
9619 | * application to drive a higher depth than the size of the SQ ring, | |
9620 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9621 | * some flexibility in overcommitting a bit. If the application has |
9622 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9623 | * of CQ ring entries manually. | |
2b188cc1 JA |
9624 | */ |
9625 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9626 | if (p->flags & IORING_SETUP_CQSIZE) { |
9627 | /* | |
9628 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9629 | * to a power-of-two, if it isn't already. We do NOT impose | |
9630 | * any cq vs sq ring sizing. | |
9631 | */ | |
eb2667b3 | 9632 | if (!p->cq_entries) |
33a107f0 | 9633 | return -EINVAL; |
8110c1a6 JA |
9634 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9635 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9636 | return -EINVAL; | |
9637 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9638 | } | |
eb2667b3 JQ |
9639 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9640 | if (p->cq_entries < p->sq_entries) | |
9641 | return -EINVAL; | |
33a107f0 JA |
9642 | } else { |
9643 | p->cq_entries = 2 * p->sq_entries; | |
9644 | } | |
2b188cc1 | 9645 | |
2b188cc1 | 9646 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9647 | if (!ctx) |
2b188cc1 | 9648 | return -ENOMEM; |
2b188cc1 | 9649 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9650 | if (!capable(CAP_IPC_LOCK)) |
9651 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9652 | |
9653 | /* | |
9654 | * This is just grabbed for accounting purposes. When a process exits, | |
9655 | * the mm is exited and dropped before the files, hence we need to hang | |
9656 | * on to this mm purely for the purposes of being able to unaccount | |
9657 | * memory (locked/pinned vm). It's not used for anything else. | |
9658 | */ | |
6b7898eb | 9659 | mmgrab(current->mm); |
2aede0e4 | 9660 | ctx->mm_account = current->mm; |
6b7898eb | 9661 | |
2b188cc1 JA |
9662 | ret = io_allocate_scq_urings(ctx, p); |
9663 | if (ret) | |
9664 | goto err; | |
9665 | ||
7e84e1c7 | 9666 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9667 | if (ret) |
9668 | goto err; | |
eae071c9 | 9669 | /* always set a rsrc node */ |
47b228ce PB |
9670 | ret = io_rsrc_node_switch_start(ctx); |
9671 | if (ret) | |
9672 | goto err; | |
eae071c9 | 9673 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 9674 | |
2b188cc1 | 9675 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9676 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9677 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9678 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9679 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9680 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9681 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9682 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9683 | |
9684 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9685 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9686 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9687 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9688 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9689 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9690 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9691 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9692 | |
7f13657d XW |
9693 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9694 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9695 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9696 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
9697 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
9698 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
9699 | |
9700 | if (copy_to_user(params, p, sizeof(*p))) { | |
9701 | ret = -EFAULT; | |
9702 | goto err; | |
9703 | } | |
d1719f70 | 9704 | |
9faadcc8 PB |
9705 | file = io_uring_get_file(ctx); |
9706 | if (IS_ERR(file)) { | |
9707 | ret = PTR_ERR(file); | |
9708 | goto err; | |
9709 | } | |
9710 | ||
044c1ab3 JA |
9711 | /* |
9712 | * Install ring fd as the very last thing, so we don't risk someone | |
9713 | * having closed it before we finish setup | |
9714 | */ | |
9faadcc8 PB |
9715 | ret = io_uring_install_fd(ctx, file); |
9716 | if (ret < 0) { | |
9717 | /* fput will clean it up */ | |
9718 | fput(file); | |
9719 | return ret; | |
9720 | } | |
044c1ab3 | 9721 | |
c826bd7a | 9722 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9723 | return ret; |
9724 | err: | |
9725 | io_ring_ctx_wait_and_kill(ctx); | |
9726 | return ret; | |
9727 | } | |
9728 | ||
9729 | /* | |
9730 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9731 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9732 | * params structure passed in. | |
9733 | */ | |
9734 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9735 | { | |
9736 | struct io_uring_params p; | |
2b188cc1 JA |
9737 | int i; |
9738 | ||
9739 | if (copy_from_user(&p, params, sizeof(p))) | |
9740 | return -EFAULT; | |
9741 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9742 | if (p.resv[i]) | |
9743 | return -EINVAL; | |
9744 | } | |
9745 | ||
6c271ce2 | 9746 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9747 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9748 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9749 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9750 | return -EINVAL; |
9751 | ||
7f13657d | 9752 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9753 | } |
9754 | ||
9755 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9756 | struct io_uring_params __user *, params) | |
9757 | { | |
9758 | return io_uring_setup(entries, params); | |
9759 | } | |
9760 | ||
66f4af93 JA |
9761 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9762 | { | |
9763 | struct io_uring_probe *p; | |
9764 | size_t size; | |
9765 | int i, ret; | |
9766 | ||
9767 | size = struct_size(p, ops, nr_args); | |
9768 | if (size == SIZE_MAX) | |
9769 | return -EOVERFLOW; | |
9770 | p = kzalloc(size, GFP_KERNEL); | |
9771 | if (!p) | |
9772 | return -ENOMEM; | |
9773 | ||
9774 | ret = -EFAULT; | |
9775 | if (copy_from_user(p, arg, size)) | |
9776 | goto out; | |
9777 | ret = -EINVAL; | |
9778 | if (memchr_inv(p, 0, size)) | |
9779 | goto out; | |
9780 | ||
9781 | p->last_op = IORING_OP_LAST - 1; | |
9782 | if (nr_args > IORING_OP_LAST) | |
9783 | nr_args = IORING_OP_LAST; | |
9784 | ||
9785 | for (i = 0; i < nr_args; i++) { | |
9786 | p->ops[i].op = i; | |
9787 | if (!io_op_defs[i].not_supported) | |
9788 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9789 | } | |
9790 | p->ops_len = i; | |
9791 | ||
9792 | ret = 0; | |
9793 | if (copy_to_user(arg, p, size)) | |
9794 | ret = -EFAULT; | |
9795 | out: | |
9796 | kfree(p); | |
9797 | return ret; | |
9798 | } | |
9799 | ||
071698e1 JA |
9800 | static int io_register_personality(struct io_ring_ctx *ctx) |
9801 | { | |
4379bf8b | 9802 | const struct cred *creds; |
61cf9370 | 9803 | u32 id; |
1e6fa521 | 9804 | int ret; |
071698e1 | 9805 | |
4379bf8b | 9806 | creds = get_current_cred(); |
1e6fa521 | 9807 | |
61cf9370 MWO |
9808 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9809 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9810 | if (!ret) | |
9811 | return id; | |
9812 | put_cred(creds); | |
1e6fa521 | 9813 | return ret; |
071698e1 JA |
9814 | } |
9815 | ||
21b55dbc SG |
9816 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9817 | unsigned int nr_args) | |
9818 | { | |
9819 | struct io_uring_restriction *res; | |
9820 | size_t size; | |
9821 | int i, ret; | |
9822 | ||
7e84e1c7 SG |
9823 | /* Restrictions allowed only if rings started disabled */ |
9824 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9825 | return -EBADFD; | |
9826 | ||
21b55dbc | 9827 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9828 | if (ctx->restrictions.registered) |
21b55dbc SG |
9829 | return -EBUSY; |
9830 | ||
9831 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9832 | return -EINVAL; | |
9833 | ||
9834 | size = array_size(nr_args, sizeof(*res)); | |
9835 | if (size == SIZE_MAX) | |
9836 | return -EOVERFLOW; | |
9837 | ||
9838 | res = memdup_user(arg, size); | |
9839 | if (IS_ERR(res)) | |
9840 | return PTR_ERR(res); | |
9841 | ||
9842 | ret = 0; | |
9843 | ||
9844 | for (i = 0; i < nr_args; i++) { | |
9845 | switch (res[i].opcode) { | |
9846 | case IORING_RESTRICTION_REGISTER_OP: | |
9847 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9848 | ret = -EINVAL; | |
9849 | goto out; | |
9850 | } | |
9851 | ||
9852 | __set_bit(res[i].register_op, | |
9853 | ctx->restrictions.register_op); | |
9854 | break; | |
9855 | case IORING_RESTRICTION_SQE_OP: | |
9856 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9857 | ret = -EINVAL; | |
9858 | goto out; | |
9859 | } | |
9860 | ||
9861 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9862 | break; | |
9863 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9864 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9865 | break; | |
9866 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9867 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9868 | break; | |
9869 | default: | |
9870 | ret = -EINVAL; | |
9871 | goto out; | |
9872 | } | |
9873 | } | |
9874 | ||
9875 | out: | |
9876 | /* Reset all restrictions if an error happened */ | |
9877 | if (ret != 0) | |
9878 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9879 | else | |
7e84e1c7 | 9880 | ctx->restrictions.registered = true; |
21b55dbc SG |
9881 | |
9882 | kfree(res); | |
9883 | return ret; | |
9884 | } | |
9885 | ||
7e84e1c7 SG |
9886 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9887 | { | |
9888 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9889 | return -EBADFD; | |
9890 | ||
9891 | if (ctx->restrictions.registered) | |
9892 | ctx->restricted = 1; | |
9893 | ||
0298ef96 PB |
9894 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9895 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9896 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9897 | return 0; |
9898 | } | |
9899 | ||
fdecb662 | 9900 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 9901 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
9902 | unsigned nr_args) |
9903 | { | |
9904 | __u32 tmp; | |
9905 | int err; | |
9906 | ||
c3bdad02 PB |
9907 | if (up->resv) |
9908 | return -EINVAL; | |
98f0b3b4 PB |
9909 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
9910 | return -EOVERFLOW; | |
9911 | err = io_rsrc_node_switch_start(ctx); | |
9912 | if (err) | |
9913 | return err; | |
9914 | ||
fdecb662 PB |
9915 | switch (type) { |
9916 | case IORING_RSRC_FILE: | |
98f0b3b4 | 9917 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
9918 | case IORING_RSRC_BUFFER: |
9919 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
9920 | } |
9921 | return -EINVAL; | |
9922 | } | |
9923 | ||
c3bdad02 PB |
9924 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
9925 | unsigned nr_args) | |
98f0b3b4 | 9926 | { |
c3bdad02 | 9927 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
9928 | |
9929 | if (!nr_args) | |
9930 | return -EINVAL; | |
c3bdad02 PB |
9931 | memset(&up, 0, sizeof(up)); |
9932 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
9933 | return -EFAULT; | |
9934 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
9935 | } | |
9936 | ||
9937 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 9938 | unsigned size, unsigned type) |
c3bdad02 PB |
9939 | { |
9940 | struct io_uring_rsrc_update2 up; | |
9941 | ||
9942 | if (size != sizeof(up)) | |
9943 | return -EINVAL; | |
98f0b3b4 PB |
9944 | if (copy_from_user(&up, arg, sizeof(up))) |
9945 | return -EFAULT; | |
992da01a | 9946 | if (!up.nr || up.resv) |
98f0b3b4 | 9947 | return -EINVAL; |
992da01a | 9948 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
9949 | } |
9950 | ||
792e3582 | 9951 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 9952 | unsigned int size, unsigned int type) |
792e3582 PB |
9953 | { |
9954 | struct io_uring_rsrc_register rr; | |
9955 | ||
9956 | /* keep it extendible */ | |
9957 | if (size != sizeof(rr)) | |
9958 | return -EINVAL; | |
9959 | ||
9960 | memset(&rr, 0, sizeof(rr)); | |
9961 | if (copy_from_user(&rr, arg, size)) | |
9962 | return -EFAULT; | |
992da01a | 9963 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
9964 | return -EINVAL; |
9965 | ||
992da01a | 9966 | switch (type) { |
792e3582 PB |
9967 | case IORING_RSRC_FILE: |
9968 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
9969 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
9970 | case IORING_RSRC_BUFFER: |
9971 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
9972 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
9973 | } |
9974 | return -EINVAL; | |
9975 | } | |
9976 | ||
071698e1 JA |
9977 | static bool io_register_op_must_quiesce(int op) |
9978 | { | |
9979 | switch (op) { | |
bd54b6fe BM |
9980 | case IORING_REGISTER_BUFFERS: |
9981 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 9982 | case IORING_REGISTER_FILES: |
071698e1 JA |
9983 | case IORING_UNREGISTER_FILES: |
9984 | case IORING_REGISTER_FILES_UPDATE: | |
9985 | case IORING_REGISTER_PROBE: | |
9986 | case IORING_REGISTER_PERSONALITY: | |
9987 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
9988 | case IORING_REGISTER_FILES2: |
9989 | case IORING_REGISTER_FILES_UPDATE2: | |
9990 | case IORING_REGISTER_BUFFERS2: | |
9991 | case IORING_REGISTER_BUFFERS_UPDATE: | |
071698e1 JA |
9992 | return false; |
9993 | default: | |
9994 | return true; | |
9995 | } | |
9996 | } | |
9997 | ||
edafccee JA |
9998 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9999 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10000 | __releases(ctx->uring_lock) |
10001 | __acquires(ctx->uring_lock) | |
edafccee JA |
10002 | { |
10003 | int ret; | |
10004 | ||
35fa71a0 JA |
10005 | /* |
10006 | * We're inside the ring mutex, if the ref is already dying, then | |
10007 | * someone else killed the ctx or is already going through | |
10008 | * io_uring_register(). | |
10009 | */ | |
10010 | if (percpu_ref_is_dying(&ctx->refs)) | |
10011 | return -ENXIO; | |
10012 | ||
75c4021a PB |
10013 | if (ctx->restricted) { |
10014 | if (opcode >= IORING_REGISTER_LAST) | |
10015 | return -EINVAL; | |
10016 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10017 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10018 | return -EACCES; | |
10019 | } | |
10020 | ||
071698e1 | 10021 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10022 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10023 | |
05f3fb3c JA |
10024 | /* |
10025 | * Drop uring mutex before waiting for references to exit. If | |
10026 | * another thread is currently inside io_uring_enter() it might | |
10027 | * need to grab the uring_lock to make progress. If we hold it | |
10028 | * here across the drain wait, then we can deadlock. It's safe | |
10029 | * to drop the mutex here, since no new references will come in | |
10030 | * after we've killed the percpu ref. | |
10031 | */ | |
10032 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10033 | do { |
10034 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10035 | if (!ret) | |
10036 | break; | |
ed6930c9 JA |
10037 | ret = io_run_task_work_sig(); |
10038 | if (ret < 0) | |
10039 | break; | |
af9c1a44 | 10040 | } while (1); |
05f3fb3c | 10041 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10042 | |
c150368b | 10043 | if (ret) { |
f70865db PB |
10044 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
10045 | return ret; | |
c150368b | 10046 | } |
05f3fb3c | 10047 | } |
edafccee JA |
10048 | |
10049 | switch (opcode) { | |
10050 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10051 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10052 | break; |
10053 | case IORING_UNREGISTER_BUFFERS: | |
10054 | ret = -EINVAL; | |
10055 | if (arg || nr_args) | |
10056 | break; | |
0a96bbe4 | 10057 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10058 | break; |
6b06314c | 10059 | case IORING_REGISTER_FILES: |
792e3582 | 10060 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10061 | break; |
10062 | case IORING_UNREGISTER_FILES: | |
10063 | ret = -EINVAL; | |
10064 | if (arg || nr_args) | |
10065 | break; | |
10066 | ret = io_sqe_files_unregister(ctx); | |
10067 | break; | |
c3a31e60 | 10068 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10069 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10070 | break; |
9b402849 | 10071 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10072 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10073 | ret = -EINVAL; |
10074 | if (nr_args != 1) | |
10075 | break; | |
10076 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10077 | if (ret) |
10078 | break; | |
10079 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10080 | ctx->eventfd_async = 1; | |
10081 | else | |
10082 | ctx->eventfd_async = 0; | |
9b402849 JA |
10083 | break; |
10084 | case IORING_UNREGISTER_EVENTFD: | |
10085 | ret = -EINVAL; | |
10086 | if (arg || nr_args) | |
10087 | break; | |
10088 | ret = io_eventfd_unregister(ctx); | |
10089 | break; | |
66f4af93 JA |
10090 | case IORING_REGISTER_PROBE: |
10091 | ret = -EINVAL; | |
10092 | if (!arg || nr_args > 256) | |
10093 | break; | |
10094 | ret = io_probe(ctx, arg, nr_args); | |
10095 | break; | |
071698e1 JA |
10096 | case IORING_REGISTER_PERSONALITY: |
10097 | ret = -EINVAL; | |
10098 | if (arg || nr_args) | |
10099 | break; | |
10100 | ret = io_register_personality(ctx); | |
10101 | break; | |
10102 | case IORING_UNREGISTER_PERSONALITY: | |
10103 | ret = -EINVAL; | |
10104 | if (arg) | |
10105 | break; | |
10106 | ret = io_unregister_personality(ctx, nr_args); | |
10107 | break; | |
7e84e1c7 SG |
10108 | case IORING_REGISTER_ENABLE_RINGS: |
10109 | ret = -EINVAL; | |
10110 | if (arg || nr_args) | |
10111 | break; | |
10112 | ret = io_register_enable_rings(ctx); | |
10113 | break; | |
21b55dbc SG |
10114 | case IORING_REGISTER_RESTRICTIONS: |
10115 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10116 | break; | |
992da01a PB |
10117 | case IORING_REGISTER_FILES2: |
10118 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10119 | break; | |
10120 | case IORING_REGISTER_FILES_UPDATE2: | |
10121 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10122 | IORING_RSRC_FILE); | |
10123 | break; | |
10124 | case IORING_REGISTER_BUFFERS2: | |
10125 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10126 | break; |
992da01a PB |
10127 | case IORING_REGISTER_BUFFERS_UPDATE: |
10128 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10129 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10130 | break; |
edafccee JA |
10131 | default: |
10132 | ret = -EINVAL; | |
10133 | break; | |
10134 | } | |
10135 | ||
071698e1 | 10136 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10137 | /* bring the ctx back to life */ |
05f3fb3c | 10138 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10139 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10140 | } |
edafccee JA |
10141 | return ret; |
10142 | } | |
10143 | ||
10144 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10145 | void __user *, arg, unsigned int, nr_args) | |
10146 | { | |
10147 | struct io_ring_ctx *ctx; | |
10148 | long ret = -EBADF; | |
10149 | struct fd f; | |
10150 | ||
10151 | f = fdget(fd); | |
10152 | if (!f.file) | |
10153 | return -EBADF; | |
10154 | ||
10155 | ret = -EOPNOTSUPP; | |
10156 | if (f.file->f_op != &io_uring_fops) | |
10157 | goto out_fput; | |
10158 | ||
10159 | ctx = f.file->private_data; | |
10160 | ||
b6c23dd5 PB |
10161 | io_run_task_work(); |
10162 | ||
edafccee JA |
10163 | mutex_lock(&ctx->uring_lock); |
10164 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10165 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10166 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10167 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10168 | out_fput: |
10169 | fdput(f); | |
10170 | return ret; | |
10171 | } | |
10172 | ||
2b188cc1 JA |
10173 | static int __init io_uring_init(void) |
10174 | { | |
d7f62e82 SM |
10175 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10176 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10177 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10178 | } while (0) | |
10179 | ||
10180 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10181 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10182 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10183 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10184 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10185 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10186 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10187 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10188 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10189 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10190 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10191 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10192 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10193 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10194 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10195 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10196 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10197 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10198 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10199 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10200 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10201 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10202 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10203 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10204 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10205 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10206 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10207 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10208 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10209 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10210 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10211 | |
b0d658ec PB |
10212 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10213 | sizeof(struct io_uring_rsrc_update)); | |
10214 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10215 | sizeof(struct io_uring_rsrc_update2)); | |
10216 | /* should fit into one byte */ | |
10217 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10218 | ||
d3656344 | 10219 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10220 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10221 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10222 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10223 | return 0; |
10224 | }; | |
10225 | __initcall(io_uring_init); |