Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
d068b506 | 14 | * through a control-dependency in io_get_cqe (smp_store_release to |
1e84b97b SB |
15 | * store head will do). Failure to do so could lead to reading invalid |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
ef98eb04 | 81 | #include <linux/tracehook.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
4ce8ad95 | 93 | #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8 |
65e19f54 | 94 | |
187f08c1 | 95 | /* only define max */ |
042b0d85 | 96 | #define IORING_MAX_FIXED_FILES (1U << 15) |
21b55dbc SG |
97 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
98 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 99 | |
187f08c1 | 100 | #define IO_RSRC_TAG_TABLE_SHIFT (PAGE_SHIFT - 3) |
2d091d62 PB |
101 | #define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT) |
102 | #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1) | |
103 | ||
489809e2 PB |
104 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
105 | ||
b16fed66 PB |
106 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
107 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
108 | IOSQE_BUFFER_SELECT) | |
c854357b PB |
109 | #define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \ |
110 | REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS) | |
b16fed66 | 111 | |
09899b19 PB |
112 | #define IO_TCTX_REFS_CACHE_NR (1U << 10) |
113 | ||
2b188cc1 JA |
114 | struct io_uring { |
115 | u32 head ____cacheline_aligned_in_smp; | |
116 | u32 tail ____cacheline_aligned_in_smp; | |
117 | }; | |
118 | ||
1e84b97b | 119 | /* |
75b28aff HV |
120 | * This data is shared with the application through the mmap at offsets |
121 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
122 | * |
123 | * The offsets to the member fields are published through struct | |
124 | * io_sqring_offsets when calling io_uring_setup. | |
125 | */ | |
75b28aff | 126 | struct io_rings { |
1e84b97b SB |
127 | /* |
128 | * Head and tail offsets into the ring; the offsets need to be | |
129 | * masked to get valid indices. | |
130 | * | |
75b28aff HV |
131 | * The kernel controls head of the sq ring and the tail of the cq ring, |
132 | * and the application controls tail of the sq ring and the head of the | |
133 | * cq ring. | |
1e84b97b | 134 | */ |
75b28aff | 135 | struct io_uring sq, cq; |
1e84b97b | 136 | /* |
75b28aff | 137 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
138 | * ring_entries - 1) |
139 | */ | |
75b28aff HV |
140 | u32 sq_ring_mask, cq_ring_mask; |
141 | /* Ring sizes (constant, power of 2) */ | |
142 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
143 | /* |
144 | * Number of invalid entries dropped by the kernel due to | |
145 | * invalid index stored in array | |
146 | * | |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application (i.e. get number of "new events" by comparing to | |
149 | * cached value). | |
150 | * | |
151 | * After a new SQ head value was read by the application this | |
152 | * counter includes all submissions that were dropped reaching | |
153 | * the new SQ head (and possibly more). | |
154 | */ | |
75b28aff | 155 | u32 sq_dropped; |
1e84b97b | 156 | /* |
0d9b5b3a | 157 | * Runtime SQ flags |
1e84b97b SB |
158 | * |
159 | * Written by the kernel, shouldn't be modified by the | |
160 | * application. | |
161 | * | |
162 | * The application needs a full memory barrier before checking | |
163 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
164 | */ | |
75b28aff | 165 | u32 sq_flags; |
0d9b5b3a SG |
166 | /* |
167 | * Runtime CQ flags | |
168 | * | |
169 | * Written by the application, shouldn't be modified by the | |
170 | * kernel. | |
171 | */ | |
fe7e3257 | 172 | u32 cq_flags; |
1e84b97b SB |
173 | /* |
174 | * Number of completion events lost because the queue was full; | |
175 | * this should be avoided by the application by making sure | |
0b4295b5 | 176 | * there are not more requests pending than there is space in |
1e84b97b SB |
177 | * the completion queue. |
178 | * | |
179 | * Written by the kernel, shouldn't be modified by the | |
180 | * application (i.e. get number of "new events" by comparing to | |
181 | * cached value). | |
182 | * | |
183 | * As completion events come in out of order this counter is not | |
184 | * ordered with any other data. | |
185 | */ | |
75b28aff | 186 | u32 cq_overflow; |
1e84b97b SB |
187 | /* |
188 | * Ring buffer of completion events. | |
189 | * | |
190 | * The kernel writes completion events fresh every time they are | |
191 | * produced, so the application is allowed to modify pending | |
192 | * entries. | |
193 | */ | |
75b28aff | 194 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
195 | }; |
196 | ||
45d189c6 PB |
197 | enum io_uring_cmd_flags { |
198 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 199 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
200 | }; |
201 | ||
edafccee JA |
202 | struct io_mapped_ubuf { |
203 | u64 ubuf; | |
4751f53d | 204 | u64 ubuf_end; |
edafccee | 205 | unsigned int nr_bvecs; |
de293938 | 206 | unsigned long acct_pages; |
41edf1a5 | 207 | struct bio_vec bvec[]; |
edafccee JA |
208 | }; |
209 | ||
50238531 BM |
210 | struct io_ring_ctx; |
211 | ||
6c2450ae PB |
212 | struct io_overflow_cqe { |
213 | struct io_uring_cqe cqe; | |
214 | struct list_head list; | |
215 | }; | |
216 | ||
a04b0ac0 PB |
217 | struct io_fixed_file { |
218 | /* file * with additional FFS_* flags */ | |
219 | unsigned long file_ptr; | |
220 | }; | |
221 | ||
269bbe5f BM |
222 | struct io_rsrc_put { |
223 | struct list_head list; | |
b60c8dce | 224 | u64 tag; |
50238531 BM |
225 | union { |
226 | void *rsrc; | |
227 | struct file *file; | |
bd54b6fe | 228 | struct io_mapped_ubuf *buf; |
50238531 | 229 | }; |
269bbe5f BM |
230 | }; |
231 | ||
aeca241b | 232 | struct io_file_table { |
042b0d85 | 233 | struct io_fixed_file *files; |
31b51510 JA |
234 | }; |
235 | ||
b895c9a6 | 236 | struct io_rsrc_node { |
05589553 XW |
237 | struct percpu_ref refs; |
238 | struct list_head node; | |
269bbe5f | 239 | struct list_head rsrc_list; |
b895c9a6 | 240 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 241 | struct llist_node llist; |
e297822b | 242 | bool done; |
05589553 XW |
243 | }; |
244 | ||
40ae0ff7 PB |
245 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
246 | ||
b895c9a6 | 247 | struct io_rsrc_data { |
05f3fb3c JA |
248 | struct io_ring_ctx *ctx; |
249 | ||
2d091d62 PB |
250 | u64 **tags; |
251 | unsigned int nr; | |
40ae0ff7 | 252 | rsrc_put_fn *do_put; |
3e942498 | 253 | atomic_t refs; |
05f3fb3c | 254 | struct completion done; |
8bad28d8 | 255 | bool quiesce; |
05f3fb3c JA |
256 | }; |
257 | ||
5a2e745d JA |
258 | struct io_buffer { |
259 | struct list_head list; | |
260 | __u64 addr; | |
d1f82808 | 261 | __u32 len; |
5a2e745d JA |
262 | __u16 bid; |
263 | }; | |
264 | ||
21b55dbc SG |
265 | struct io_restriction { |
266 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
267 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
268 | u8 sqe_flags_allowed; | |
269 | u8 sqe_flags_required; | |
7e84e1c7 | 270 | bool registered; |
21b55dbc SG |
271 | }; |
272 | ||
37d1e2e3 JA |
273 | enum { |
274 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
275 | IO_SQ_THREAD_SHOULD_PARK, | |
276 | }; | |
277 | ||
534ca6d6 JA |
278 | struct io_sq_data { |
279 | refcount_t refs; | |
9e138a48 | 280 | atomic_t park_pending; |
09a6f4ef | 281 | struct mutex lock; |
69fb2131 JA |
282 | |
283 | /* ctx's that are using this sqd */ | |
284 | struct list_head ctx_list; | |
69fb2131 | 285 | |
534ca6d6 JA |
286 | struct task_struct *thread; |
287 | struct wait_queue_head wait; | |
08369246 XW |
288 | |
289 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
290 | int sq_cpu; |
291 | pid_t task_pid; | |
5c2469e0 | 292 | pid_t task_tgid; |
37d1e2e3 JA |
293 | |
294 | unsigned long state; | |
37d1e2e3 | 295 | struct completion exited; |
534ca6d6 JA |
296 | }; |
297 | ||
6dd0be1e | 298 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 299 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 300 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 | 301 | |
a1ab7b35 PB |
302 | struct io_submit_link { |
303 | struct io_kiocb *head; | |
304 | struct io_kiocb *last; | |
305 | }; | |
306 | ||
258b29a9 PB |
307 | struct io_submit_state { |
308 | struct blk_plug plug; | |
a1ab7b35 | 309 | struct io_submit_link link; |
258b29a9 PB |
310 | |
311 | /* | |
312 | * io_kiocb alloc cache | |
313 | */ | |
bf019da7 | 314 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
315 | unsigned int free_reqs; |
316 | ||
317 | bool plug_started; | |
318 | ||
319 | /* | |
320 | * Batch completion logic | |
321 | */ | |
cd0ca2e0 PB |
322 | struct io_kiocb *compl_reqs[IO_COMPL_BATCH]; |
323 | unsigned int compl_nr; | |
324 | /* inline/task_work completion list, under ->uring_lock */ | |
325 | struct list_head free_list; | |
258b29a9 | 326 | |
258b29a9 PB |
327 | unsigned int ios_left; |
328 | }; | |
329 | ||
2b188cc1 | 330 | struct io_ring_ctx { |
b52ecf8c | 331 | /* const or read-mostly hot data */ |
2b188cc1 JA |
332 | struct { |
333 | struct percpu_ref refs; | |
2b188cc1 | 334 | |
b52ecf8c | 335 | struct io_rings *rings; |
2b188cc1 | 336 | unsigned int flags; |
e1d85334 | 337 | unsigned int compat: 1; |
e1d85334 RD |
338 | unsigned int drain_next: 1; |
339 | unsigned int eventfd_async: 1; | |
21b55dbc | 340 | unsigned int restricted: 1; |
f18ee4cf | 341 | unsigned int off_timeout_used: 1; |
10c66904 | 342 | unsigned int drain_active: 1; |
b52ecf8c | 343 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 344 | |
7f1129d2 | 345 | /* submission data */ |
b52ecf8c | 346 | struct { |
0499e582 PB |
347 | struct mutex uring_lock; |
348 | ||
75b28aff HV |
349 | /* |
350 | * Ring buffer of indices into array of io_uring_sqe, which is | |
351 | * mmapped by the application using the IORING_OFF_SQES offset. | |
352 | * | |
353 | * This indirection could e.g. be used to assign fixed | |
354 | * io_uring_sqe entries to operations and only submit them to | |
355 | * the queue when needed. | |
356 | * | |
357 | * The kernel modifies neither the indices array nor the entries | |
358 | * array. | |
359 | */ | |
360 | u32 *sq_array; | |
c7af47cf | 361 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
362 | unsigned cached_sq_head; |
363 | unsigned sq_entries; | |
de0617e4 | 364 | struct list_head defer_list; |
7f1129d2 PB |
365 | |
366 | /* | |
367 | * Fixed resources fast path, should be accessed only under | |
368 | * uring_lock, and updated through io_uring_register(2) | |
369 | */ | |
370 | struct io_rsrc_node *rsrc_node; | |
371 | struct io_file_table file_table; | |
372 | unsigned nr_user_files; | |
373 | unsigned nr_user_bufs; | |
374 | struct io_mapped_ubuf **user_bufs; | |
375 | ||
376 | struct io_submit_state submit_state; | |
5262f567 | 377 | struct list_head timeout_list; |
ef9dd637 | 378 | struct list_head ltimeout_list; |
1d7bb1d5 | 379 | struct list_head cq_overflow_list; |
7f1129d2 PB |
380 | struct xarray io_buffers; |
381 | struct xarray personalities; | |
382 | u32 pers_next; | |
383 | unsigned sq_thread_idle; | |
2b188cc1 JA |
384 | } ____cacheline_aligned_in_smp; |
385 | ||
d0acdee2 PB |
386 | /* IRQ completion list, under ->completion_lock */ |
387 | struct list_head locked_free_list; | |
388 | unsigned int locked_free_nr; | |
3c1a2ead | 389 | |
7c30f36a | 390 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
391 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
392 | ||
90554200 | 393 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 394 | struct list_head sqd_list; |
75b28aff | 395 | |
5ed7a37d PB |
396 | unsigned long check_cq_overflow; |
397 | ||
206aefde JA |
398 | struct { |
399 | unsigned cached_cq_tail; | |
400 | unsigned cq_entries; | |
0499e582 | 401 | struct eventfd_ctx *cq_ev_fd; |
311997b3 | 402 | struct wait_queue_head poll_wait; |
0499e582 PB |
403 | struct wait_queue_head cq_wait; |
404 | unsigned cq_extra; | |
405 | atomic_t cq_timeouts; | |
206aefde | 406 | struct fasync_struct *cq_fasync; |
0499e582 | 407 | unsigned cq_last_tm_flush; |
206aefde | 408 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 409 | |
2b188cc1 JA |
410 | struct { |
411 | spinlock_t completion_lock; | |
e94f141b | 412 | |
89850fce JA |
413 | spinlock_t timeout_lock; |
414 | ||
def596e9 | 415 | /* |
540e32a0 | 416 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
417 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
418 | * For SQPOLL, only the single threaded io_sq_thread() will | |
419 | * manipulate the list, hence no extra locking is needed there. | |
420 | */ | |
540e32a0 | 421 | struct list_head iopoll_list; |
78076bb6 JA |
422 | struct hlist_head *cancel_hash; |
423 | unsigned cancel_hash_bits; | |
915b3dde | 424 | bool poll_multi_queue; |
2b188cc1 | 425 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 426 | |
21b55dbc | 427 | struct io_restriction restrictions; |
3c1a2ead | 428 | |
b13a8918 PB |
429 | /* slow path rsrc auxilary data, used by update/register */ |
430 | struct { | |
431 | struct io_rsrc_node *rsrc_backup_node; | |
432 | struct io_mapped_ubuf *dummy_ubuf; | |
433 | struct io_rsrc_data *file_data; | |
434 | struct io_rsrc_data *buf_data; | |
435 | ||
436 | struct delayed_work rsrc_put_work; | |
437 | struct llist_head rsrc_put_llist; | |
438 | struct list_head rsrc_ref_list; | |
439 | spinlock_t rsrc_ref_lock; | |
440 | }; | |
441 | ||
3c1a2ead | 442 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
443 | struct { |
444 | #if defined(CONFIG_UNIX) | |
445 | struct socket *ring_sock; | |
446 | #endif | |
447 | /* hashed buffered write serialization */ | |
448 | struct io_wq_hash *hash_map; | |
449 | ||
450 | /* Only used for accounting purposes */ | |
451 | struct user_struct *user; | |
452 | struct mm_struct *mm_account; | |
453 | ||
454 | /* ctx exit and cancelation */ | |
9011bf9a PB |
455 | struct llist_head fallback_llist; |
456 | struct delayed_work fallback_work; | |
b986af7e PB |
457 | struct work_struct exit_work; |
458 | struct list_head tctx_list; | |
459 | struct completion ref_comp; | |
460 | }; | |
2b188cc1 JA |
461 | }; |
462 | ||
53e043b2 SM |
463 | struct io_uring_task { |
464 | /* submission side */ | |
09899b19 | 465 | int cached_refs; |
53e043b2 SM |
466 | struct xarray xa; |
467 | struct wait_queue_head wait; | |
ee53fb2b SM |
468 | const struct io_ring_ctx *last; |
469 | struct io_wq *io_wq; | |
53e043b2 | 470 | struct percpu_counter inflight; |
b303fe2e | 471 | atomic_t inflight_tracked; |
53e043b2 | 472 | atomic_t in_idle; |
53e043b2 SM |
473 | |
474 | spinlock_t task_lock; | |
475 | struct io_wq_work_list task_list; | |
53e043b2 | 476 | struct callback_head task_work; |
6294f368 | 477 | bool task_running; |
53e043b2 SM |
478 | }; |
479 | ||
09bb8394 JA |
480 | /* |
481 | * First field must be the file pointer in all the | |
482 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
483 | */ | |
221c5eb2 JA |
484 | struct io_poll_iocb { |
485 | struct file *file; | |
018043be | 486 | struct wait_queue_head *head; |
221c5eb2 | 487 | __poll_t events; |
8c838788 | 488 | bool done; |
221c5eb2 | 489 | bool canceled; |
392edb45 | 490 | struct wait_queue_entry wait; |
221c5eb2 JA |
491 | }; |
492 | ||
9d805892 | 493 | struct io_poll_update { |
018043be | 494 | struct file *file; |
9d805892 PB |
495 | u64 old_user_data; |
496 | u64 new_user_data; | |
497 | __poll_t events; | |
b69de288 JA |
498 | bool update_events; |
499 | bool update_user_data; | |
018043be PB |
500 | }; |
501 | ||
b5dba59e JA |
502 | struct io_close { |
503 | struct file *file; | |
b5dba59e JA |
504 | int fd; |
505 | }; | |
506 | ||
ad8a48ac JA |
507 | struct io_timeout_data { |
508 | struct io_kiocb *req; | |
509 | struct hrtimer timer; | |
510 | struct timespec64 ts; | |
511 | enum hrtimer_mode mode; | |
50c1df2b | 512 | u32 flags; |
ad8a48ac JA |
513 | }; |
514 | ||
8ed8d3c3 JA |
515 | struct io_accept { |
516 | struct file *file; | |
517 | struct sockaddr __user *addr; | |
518 | int __user *addr_len; | |
519 | int flags; | |
aaa4db12 | 520 | u32 file_slot; |
09952e3e | 521 | unsigned long nofile; |
8ed8d3c3 JA |
522 | }; |
523 | ||
524 | struct io_sync { | |
525 | struct file *file; | |
526 | loff_t len; | |
527 | loff_t off; | |
528 | int flags; | |
d63d1b5e | 529 | int mode; |
8ed8d3c3 JA |
530 | }; |
531 | ||
fbf23849 JA |
532 | struct io_cancel { |
533 | struct file *file; | |
534 | u64 addr; | |
535 | }; | |
536 | ||
b29472ee JA |
537 | struct io_timeout { |
538 | struct file *file; | |
bfe68a22 PB |
539 | u32 off; |
540 | u32 target_seq; | |
135fcde8 | 541 | struct list_head list; |
90cd7e42 PB |
542 | /* head of the link, used by linked timeouts only */ |
543 | struct io_kiocb *head; | |
89b263f6 JA |
544 | /* for linked completions */ |
545 | struct io_kiocb *prev; | |
b29472ee JA |
546 | }; |
547 | ||
0bdf7a2d PB |
548 | struct io_timeout_rem { |
549 | struct file *file; | |
550 | u64 addr; | |
9c8e11b3 PB |
551 | |
552 | /* timeout update */ | |
553 | struct timespec64 ts; | |
554 | u32 flags; | |
f1042b6c | 555 | bool ltimeout; |
0bdf7a2d PB |
556 | }; |
557 | ||
9adbd45d JA |
558 | struct io_rw { |
559 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
560 | struct kiocb kiocb; | |
561 | u64 addr; | |
562 | u64 len; | |
563 | }; | |
564 | ||
3fbb51c1 JA |
565 | struct io_connect { |
566 | struct file *file; | |
567 | struct sockaddr __user *addr; | |
568 | int addr_len; | |
569 | }; | |
570 | ||
e47293fd JA |
571 | struct io_sr_msg { |
572 | struct file *file; | |
fddaface | 573 | union { |
4af3417a PB |
574 | struct compat_msghdr __user *umsg_compat; |
575 | struct user_msghdr __user *umsg; | |
576 | void __user *buf; | |
fddaface | 577 | }; |
e47293fd | 578 | int msg_flags; |
bcda7baa | 579 | int bgid; |
fddaface | 580 | size_t len; |
bcda7baa | 581 | struct io_buffer *kbuf; |
e47293fd JA |
582 | }; |
583 | ||
15b71abe JA |
584 | struct io_open { |
585 | struct file *file; | |
586 | int dfd; | |
b9445598 | 587 | u32 file_slot; |
15b71abe | 588 | struct filename *filename; |
c12cedf2 | 589 | struct open_how how; |
4022e7af | 590 | unsigned long nofile; |
15b71abe JA |
591 | }; |
592 | ||
269bbe5f | 593 | struct io_rsrc_update { |
05f3fb3c JA |
594 | struct file *file; |
595 | u64 arg; | |
596 | u32 nr_args; | |
597 | u32 offset; | |
598 | }; | |
599 | ||
4840e418 JA |
600 | struct io_fadvise { |
601 | struct file *file; | |
602 | u64 offset; | |
603 | u32 len; | |
604 | u32 advice; | |
605 | }; | |
606 | ||
c1ca757b JA |
607 | struct io_madvise { |
608 | struct file *file; | |
609 | u64 addr; | |
610 | u32 len; | |
611 | u32 advice; | |
612 | }; | |
613 | ||
3e4827b0 JA |
614 | struct io_epoll { |
615 | struct file *file; | |
616 | int epfd; | |
617 | int op; | |
618 | int fd; | |
619 | struct epoll_event event; | |
e47293fd JA |
620 | }; |
621 | ||
7d67af2c PB |
622 | struct io_splice { |
623 | struct file *file_out; | |
624 | struct file *file_in; | |
625 | loff_t off_out; | |
626 | loff_t off_in; | |
627 | u64 len; | |
628 | unsigned int flags; | |
629 | }; | |
630 | ||
ddf0322d JA |
631 | struct io_provide_buf { |
632 | struct file *file; | |
633 | __u64 addr; | |
38134ada | 634 | __u32 len; |
ddf0322d JA |
635 | __u32 bgid; |
636 | __u16 nbufs; | |
637 | __u16 bid; | |
638 | }; | |
639 | ||
1d9e1288 BM |
640 | struct io_statx { |
641 | struct file *file; | |
642 | int dfd; | |
643 | unsigned int mask; | |
644 | unsigned int flags; | |
e62753e4 | 645 | const char __user *filename; |
1d9e1288 BM |
646 | struct statx __user *buffer; |
647 | }; | |
648 | ||
36f4fa68 JA |
649 | struct io_shutdown { |
650 | struct file *file; | |
651 | int how; | |
652 | }; | |
653 | ||
80a261fd JA |
654 | struct io_rename { |
655 | struct file *file; | |
656 | int old_dfd; | |
657 | int new_dfd; | |
658 | struct filename *oldpath; | |
659 | struct filename *newpath; | |
660 | int flags; | |
661 | }; | |
662 | ||
14a1143b JA |
663 | struct io_unlink { |
664 | struct file *file; | |
665 | int dfd; | |
666 | int flags; | |
667 | struct filename *filename; | |
668 | }; | |
669 | ||
e34a02dc DK |
670 | struct io_mkdir { |
671 | struct file *file; | |
672 | int dfd; | |
673 | umode_t mode; | |
674 | struct filename *filename; | |
675 | }; | |
676 | ||
7a8721f8 DK |
677 | struct io_symlink { |
678 | struct file *file; | |
679 | int new_dfd; | |
680 | struct filename *oldpath; | |
681 | struct filename *newpath; | |
682 | }; | |
683 | ||
cf30da90 DK |
684 | struct io_hardlink { |
685 | struct file *file; | |
686 | int old_dfd; | |
687 | int new_dfd; | |
688 | struct filename *oldpath; | |
689 | struct filename *newpath; | |
690 | int flags; | |
691 | }; | |
692 | ||
3ca405eb PB |
693 | struct io_completion { |
694 | struct file *file; | |
8c3f9cd1 | 695 | u32 cflags; |
3ca405eb PB |
696 | }; |
697 | ||
f499a021 JA |
698 | struct io_async_connect { |
699 | struct sockaddr_storage address; | |
700 | }; | |
701 | ||
03b1230c JA |
702 | struct io_async_msghdr { |
703 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
704 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
705 | struct iovec *free_iov; | |
03b1230c JA |
706 | struct sockaddr __user *uaddr; |
707 | struct msghdr msg; | |
b537916c | 708 | struct sockaddr_storage addr; |
03b1230c JA |
709 | }; |
710 | ||
f67676d1 JA |
711 | struct io_async_rw { |
712 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
713 | const struct iovec *free_iovec; |
714 | struct iov_iter iter; | |
cd658695 | 715 | struct iov_iter_state iter_state; |
227c0c96 | 716 | size_t bytes_done; |
bcf5a063 | 717 | struct wait_page_queue wpq; |
f67676d1 JA |
718 | }; |
719 | ||
6b47ee6e PB |
720 | enum { |
721 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
722 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
723 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
724 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
725 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 726 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 727 | |
dddca226 | 728 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 729 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
730 | REQ_F_INFLIGHT_BIT, |
731 | REQ_F_CUR_POS_BIT, | |
732 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 733 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 734 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 735 | REQ_F_POLLED_BIT, |
bcda7baa | 736 | REQ_F_BUFFER_SELECTED_BIT, |
e342c807 | 737 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 738 | REQ_F_REISSUE_BIT, |
b8e64b53 | 739 | REQ_F_CREDS_BIT, |
20e60a38 | 740 | REQ_F_REFCOUNT_BIT, |
4d13d1a4 | 741 | REQ_F_ARM_LTIMEOUT_BIT, |
7b29f92d | 742 | /* keep async read/write and isreg together and in order */ |
b191e2df PB |
743 | REQ_F_NOWAIT_READ_BIT, |
744 | REQ_F_NOWAIT_WRITE_BIT, | |
7b29f92d | 745 | REQ_F_ISREG_BIT, |
84557871 JA |
746 | |
747 | /* not a real bit, just to check we're not overflowing the space */ | |
748 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
749 | }; |
750 | ||
751 | enum { | |
752 | /* ctx owns file */ | |
753 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
754 | /* drain existing IO first */ | |
755 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
756 | /* linked sqes */ | |
757 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
758 | /* doesn't sever on completion < 0 */ | |
759 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
760 | /* IOSQE_ASYNC */ | |
761 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
762 | /* IOSQE_BUFFER_SELECT */ |
763 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 764 | |
6b47ee6e | 765 | /* fail rest of links */ |
93d2bcd2 | 766 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 767 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
768 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
769 | /* read/write uses file position */ | |
770 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
771 | /* must not punt to workers */ | |
772 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 773 | /* has or had linked timeout */ |
6b47ee6e | 774 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
775 | /* needs cleanup */ |
776 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
777 | /* already went through poll handler */ |
778 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
779 | /* buffer already selected */ |
780 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
e342c807 PB |
781 | /* completion is deferred through io_comp_state */ |
782 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
783 | /* caller should reissue async */ |
784 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
7b29f92d | 785 | /* supports async reads */ |
b191e2df | 786 | REQ_F_NOWAIT_READ = BIT(REQ_F_NOWAIT_READ_BIT), |
7b29f92d | 787 | /* supports async writes */ |
b191e2df | 788 | REQ_F_NOWAIT_WRITE = BIT(REQ_F_NOWAIT_WRITE_BIT), |
7b29f92d JA |
789 | /* regular file */ |
790 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
791 | /* has creds assigned */ |
792 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
20e60a38 PB |
793 | /* skip refcounting if not set */ |
794 | REQ_F_REFCOUNT = BIT(REQ_F_REFCOUNT_BIT), | |
4d13d1a4 PB |
795 | /* there is a linked timeout that has to be armed */ |
796 | REQ_F_ARM_LTIMEOUT = BIT(REQ_F_ARM_LTIMEOUT_BIT), | |
d7718a9d JA |
797 | }; |
798 | ||
799 | struct async_poll { | |
800 | struct io_poll_iocb poll; | |
807abcb0 | 801 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
802 | }; |
803 | ||
f237c30a | 804 | typedef void (*io_req_tw_func_t)(struct io_kiocb *req, bool *locked); |
5b0a6acc | 805 | |
7cbf1722 | 806 | struct io_task_work { |
5b0a6acc PB |
807 | union { |
808 | struct io_wq_work_node node; | |
809 | struct llist_node fallback_node; | |
810 | }; | |
811 | io_req_tw_func_t func; | |
7cbf1722 JA |
812 | }; |
813 | ||
992da01a PB |
814 | enum { |
815 | IORING_RSRC_FILE = 0, | |
816 | IORING_RSRC_BUFFER = 1, | |
817 | }; | |
818 | ||
09bb8394 JA |
819 | /* |
820 | * NOTE! Each of the iocb union members has the file pointer | |
821 | * as the first entry in their struct definition. So you can | |
822 | * access the file pointer through any of the sub-structs, | |
823 | * or directly as just 'ki_filp' in this struct. | |
824 | */ | |
2b188cc1 | 825 | struct io_kiocb { |
221c5eb2 | 826 | union { |
09bb8394 | 827 | struct file *file; |
9adbd45d | 828 | struct io_rw rw; |
221c5eb2 | 829 | struct io_poll_iocb poll; |
9d805892 | 830 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
831 | struct io_accept accept; |
832 | struct io_sync sync; | |
fbf23849 | 833 | struct io_cancel cancel; |
b29472ee | 834 | struct io_timeout timeout; |
0bdf7a2d | 835 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 836 | struct io_connect connect; |
e47293fd | 837 | struct io_sr_msg sr_msg; |
15b71abe | 838 | struct io_open open; |
b5dba59e | 839 | struct io_close close; |
269bbe5f | 840 | struct io_rsrc_update rsrc_update; |
4840e418 | 841 | struct io_fadvise fadvise; |
c1ca757b | 842 | struct io_madvise madvise; |
3e4827b0 | 843 | struct io_epoll epoll; |
7d67af2c | 844 | struct io_splice splice; |
ddf0322d | 845 | struct io_provide_buf pbuf; |
1d9e1288 | 846 | struct io_statx statx; |
36f4fa68 | 847 | struct io_shutdown shutdown; |
80a261fd | 848 | struct io_rename rename; |
14a1143b | 849 | struct io_unlink unlink; |
e34a02dc | 850 | struct io_mkdir mkdir; |
7a8721f8 | 851 | struct io_symlink symlink; |
cf30da90 | 852 | struct io_hardlink hardlink; |
3ca405eb PB |
853 | /* use only after cleaning per-op data, see io_clean_op() */ |
854 | struct io_completion compl; | |
221c5eb2 | 855 | }; |
2b188cc1 | 856 | |
e8c2bc1f JA |
857 | /* opcode allocated if it needs to store data for async defer */ |
858 | void *async_data; | |
d625c6ee | 859 | u8 opcode; |
65a6543d XW |
860 | /* polled IO has completed */ |
861 | u8 iopoll_completed; | |
2b188cc1 | 862 | |
4f4eeba8 | 863 | u16 buf_index; |
9cf7c104 | 864 | u32 result; |
4f4eeba8 | 865 | |
010e8e6b PB |
866 | struct io_ring_ctx *ctx; |
867 | unsigned int flags; | |
abc54d63 | 868 | atomic_t refs; |
010e8e6b PB |
869 | struct task_struct *task; |
870 | u64 user_data; | |
d7718a9d | 871 | |
f2f87370 | 872 | struct io_kiocb *link; |
269bbe5f | 873 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 874 | |
b303fe2e | 875 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 876 | struct list_head inflight_entry; |
5b0a6acc | 877 | struct io_task_work io_task_work; |
010e8e6b PB |
878 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
879 | struct hlist_node hash_node; | |
880 | struct async_poll *apoll; | |
881 | struct io_wq_work work; | |
fe7e3257 | 882 | const struct cred *creds; |
c10d1f98 | 883 | |
eae071c9 PB |
884 | /* store used ubuf, so we can prevent reloading */ |
885 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 886 | }; |
05589553 | 887 | |
13bf43f5 PB |
888 | struct io_tctx_node { |
889 | struct list_head ctx_node; | |
890 | struct task_struct *task; | |
13bf43f5 PB |
891 | struct io_ring_ctx *ctx; |
892 | }; | |
893 | ||
27dc8338 PB |
894 | struct io_defer_entry { |
895 | struct list_head list; | |
896 | struct io_kiocb *req; | |
9cf7c104 | 897 | u32 seq; |
2b188cc1 JA |
898 | }; |
899 | ||
d3656344 | 900 | struct io_op_def { |
d3656344 JA |
901 | /* needs req->file assigned */ |
902 | unsigned needs_file : 1; | |
d3656344 JA |
903 | /* hash wq insertion if file is a regular file */ |
904 | unsigned hash_reg_file : 1; | |
905 | /* unbound wq insertion if file is a non-regular file */ | |
906 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
907 | /* opcode is not supported by this kernel */ |
908 | unsigned not_supported : 1; | |
8a72758c JA |
909 | /* set if opcode supports polled "wait" */ |
910 | unsigned pollin : 1; | |
911 | unsigned pollout : 1; | |
bcda7baa JA |
912 | /* op supports buffer selection */ |
913 | unsigned buffer_select : 1; | |
26f0505a PB |
914 | /* do prep async if is going to be punted */ |
915 | unsigned needs_async_setup : 1; | |
27926b68 JA |
916 | /* should block plug */ |
917 | unsigned plug : 1; | |
e8c2bc1f JA |
918 | /* size of async data needed, if any */ |
919 | unsigned short async_size; | |
d3656344 JA |
920 | }; |
921 | ||
0918682b | 922 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
923 | [IORING_OP_NOP] = {}, |
924 | [IORING_OP_READV] = { | |
d3656344 JA |
925 | .needs_file = 1, |
926 | .unbound_nonreg_file = 1, | |
8a72758c | 927 | .pollin = 1, |
4d954c25 | 928 | .buffer_select = 1, |
26f0505a | 929 | .needs_async_setup = 1, |
27926b68 | 930 | .plug = 1, |
e8c2bc1f | 931 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 932 | }, |
0463b6c5 | 933 | [IORING_OP_WRITEV] = { |
d3656344 JA |
934 | .needs_file = 1, |
935 | .hash_reg_file = 1, | |
936 | .unbound_nonreg_file = 1, | |
8a72758c | 937 | .pollout = 1, |
26f0505a | 938 | .needs_async_setup = 1, |
27926b68 | 939 | .plug = 1, |
e8c2bc1f | 940 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 941 | }, |
0463b6c5 | 942 | [IORING_OP_FSYNC] = { |
d3656344 JA |
943 | .needs_file = 1, |
944 | }, | |
0463b6c5 | 945 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
946 | .needs_file = 1, |
947 | .unbound_nonreg_file = 1, | |
8a72758c | 948 | .pollin = 1, |
27926b68 | 949 | .plug = 1, |
e8c2bc1f | 950 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 951 | }, |
0463b6c5 | 952 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
953 | .needs_file = 1, |
954 | .hash_reg_file = 1, | |
955 | .unbound_nonreg_file = 1, | |
8a72758c | 956 | .pollout = 1, |
27926b68 | 957 | .plug = 1, |
e8c2bc1f | 958 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 959 | }, |
0463b6c5 | 960 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
961 | .needs_file = 1, |
962 | .unbound_nonreg_file = 1, | |
963 | }, | |
0463b6c5 PB |
964 | [IORING_OP_POLL_REMOVE] = {}, |
965 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
966 | .needs_file = 1, |
967 | }, | |
0463b6c5 | 968 | [IORING_OP_SENDMSG] = { |
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_msghdr), |
d3656344 | 974 | }, |
0463b6c5 | 975 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
976 | .needs_file = 1, |
977 | .unbound_nonreg_file = 1, | |
8a72758c | 978 | .pollin = 1, |
52de1fe1 | 979 | .buffer_select = 1, |
26f0505a | 980 | .needs_async_setup = 1, |
e8c2bc1f | 981 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 982 | }, |
0463b6c5 | 983 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 984 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 985 | }, |
9c8e11b3 PB |
986 | [IORING_OP_TIMEOUT_REMOVE] = { |
987 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 988 | }, |
0463b6c5 | 989 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
990 | .needs_file = 1, |
991 | .unbound_nonreg_file = 1, | |
8a72758c | 992 | .pollin = 1, |
d3656344 | 993 | }, |
0463b6c5 PB |
994 | [IORING_OP_ASYNC_CANCEL] = {}, |
995 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 996 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 997 | }, |
0463b6c5 | 998 | [IORING_OP_CONNECT] = { |
d3656344 JA |
999 | .needs_file = 1, |
1000 | .unbound_nonreg_file = 1, | |
8a72758c | 1001 | .pollout = 1, |
26f0505a | 1002 | .needs_async_setup = 1, |
e8c2bc1f | 1003 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 1004 | }, |
0463b6c5 | 1005 | [IORING_OP_FALLOCATE] = { |
d3656344 | 1006 | .needs_file = 1, |
d3656344 | 1007 | }, |
44526bed JA |
1008 | [IORING_OP_OPENAT] = {}, |
1009 | [IORING_OP_CLOSE] = {}, | |
1010 | [IORING_OP_FILES_UPDATE] = {}, | |
1011 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 1012 | [IORING_OP_READ] = { |
3a6820f2 JA |
1013 | .needs_file = 1, |
1014 | .unbound_nonreg_file = 1, | |
8a72758c | 1015 | .pollin = 1, |
bcda7baa | 1016 | .buffer_select = 1, |
27926b68 | 1017 | .plug = 1, |
e8c2bc1f | 1018 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1019 | }, |
0463b6c5 | 1020 | [IORING_OP_WRITE] = { |
3a6820f2 | 1021 | .needs_file = 1, |
7b3188e7 | 1022 | .hash_reg_file = 1, |
3a6820f2 | 1023 | .unbound_nonreg_file = 1, |
8a72758c | 1024 | .pollout = 1, |
27926b68 | 1025 | .plug = 1, |
e8c2bc1f | 1026 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1027 | }, |
0463b6c5 | 1028 | [IORING_OP_FADVISE] = { |
4840e418 | 1029 | .needs_file = 1, |
c1ca757b | 1030 | }, |
44526bed | 1031 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1032 | [IORING_OP_SEND] = { |
fddaface JA |
1033 | .needs_file = 1, |
1034 | .unbound_nonreg_file = 1, | |
8a72758c | 1035 | .pollout = 1, |
fddaface | 1036 | }, |
0463b6c5 | 1037 | [IORING_OP_RECV] = { |
fddaface JA |
1038 | .needs_file = 1, |
1039 | .unbound_nonreg_file = 1, | |
8a72758c | 1040 | .pollin = 1, |
bcda7baa | 1041 | .buffer_select = 1, |
fddaface | 1042 | }, |
0463b6c5 | 1043 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1044 | }, |
3e4827b0 JA |
1045 | [IORING_OP_EPOLL_CTL] = { |
1046 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1047 | }, |
7d67af2c PB |
1048 | [IORING_OP_SPLICE] = { |
1049 | .needs_file = 1, | |
1050 | .hash_reg_file = 1, | |
1051 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1052 | }, |
1053 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1054 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1055 | [IORING_OP_TEE] = { |
1056 | .needs_file = 1, | |
1057 | .hash_reg_file = 1, | |
1058 | .unbound_nonreg_file = 1, | |
1059 | }, | |
36f4fa68 JA |
1060 | [IORING_OP_SHUTDOWN] = { |
1061 | .needs_file = 1, | |
1062 | }, | |
44526bed JA |
1063 | [IORING_OP_RENAMEAT] = {}, |
1064 | [IORING_OP_UNLINKAT] = {}, | |
e34a02dc | 1065 | [IORING_OP_MKDIRAT] = {}, |
7a8721f8 | 1066 | [IORING_OP_SYMLINKAT] = {}, |
cf30da90 | 1067 | [IORING_OP_LINKAT] = {}, |
d3656344 JA |
1068 | }; |
1069 | ||
0756a869 PB |
1070 | /* requests with any of those set should undergo io_disarm_next() */ |
1071 | #define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL) | |
1072 | ||
7a612350 | 1073 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1074 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1075 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1076 | struct task_struct *task, | |
3dd0c97a | 1077 | bool cancel_all); |
78cc687b | 1078 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
1ffc5422 | 1079 | |
d4d19c19 PB |
1080 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1081 | long res, unsigned int cflags); | |
ec9c02ad | 1082 | static void io_put_req(struct io_kiocb *req); |
91c2f697 | 1083 | static void io_put_req_deferred(struct io_kiocb *req); |
c7dae4ba | 1084 | static void io_dismantle_req(struct io_kiocb *req); |
94ae5e77 | 1085 | static void io_queue_linked_timeout(struct io_kiocb *req); |
fdecb662 | 1086 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1087 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1088 | unsigned nr_args); |
68fb8979 | 1089 | static void io_clean_op(struct io_kiocb *req); |
ac177053 | 1090 | static struct file *io_file_get(struct io_ring_ctx *ctx, |
8371adf5 | 1091 | struct io_kiocb *req, int fd, bool fixed); |
c5eef2b9 | 1092 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1093 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1094 | |
907d1df3 | 1095 | static void io_req_task_queue(struct io_kiocb *req); |
2a2758f2 | 1096 | static void io_submit_flush_completions(struct io_ring_ctx *ctx); |
179ae0d1 | 1097 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1098 | |
b9445598 PB |
1099 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
1100 | unsigned int issue_flags, u32 slot_index); | |
f1042b6c | 1101 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); |
b9445598 | 1102 | |
2b188cc1 JA |
1103 | static struct kmem_cache *req_cachep; |
1104 | ||
0918682b | 1105 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1106 | |
1107 | struct sock *io_uring_get_socket(struct file *file) | |
1108 | { | |
1109 | #if defined(CONFIG_UNIX) | |
1110 | if (file->f_op == &io_uring_fops) { | |
1111 | struct io_ring_ctx *ctx = file->private_data; | |
1112 | ||
1113 | return ctx->ring_sock->sk; | |
1114 | } | |
1115 | #endif | |
1116 | return NULL; | |
1117 | } | |
1118 | EXPORT_SYMBOL(io_uring_get_socket); | |
1119 | ||
f237c30a PB |
1120 | static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) |
1121 | { | |
1122 | if (!*locked) { | |
1123 | mutex_lock(&ctx->uring_lock); | |
1124 | *locked = true; | |
1125 | } | |
1126 | } | |
1127 | ||
f2f87370 PB |
1128 | #define io_for_each_link(pos, head) \ |
1129 | for (pos = (head); pos; pos = pos->link) | |
1130 | ||
21c843d5 PB |
1131 | /* |
1132 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1133 | * see commit f958d7b528b1 for details. | |
1134 | */ | |
1135 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1136 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1137 | ||
1138 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) | |
1139 | { | |
20e60a38 | 1140 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1141 | return atomic_inc_not_zero(&req->refs); |
1142 | } | |
1143 | ||
21c843d5 PB |
1144 | static inline bool req_ref_put_and_test(struct io_kiocb *req) |
1145 | { | |
20e60a38 PB |
1146 | if (likely(!(req->flags & REQ_F_REFCOUNT))) |
1147 | return true; | |
1148 | ||
21c843d5 PB |
1149 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1150 | return atomic_dec_and_test(&req->refs); | |
1151 | } | |
1152 | ||
1153 | static inline void req_ref_put(struct io_kiocb *req) | |
1154 | { | |
20e60a38 | 1155 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1156 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
1157 | } | |
1158 | ||
1159 | static inline void req_ref_get(struct io_kiocb *req) | |
1160 | { | |
20e60a38 | 1161 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1162 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1163 | atomic_inc(&req->refs); | |
1164 | } | |
1165 | ||
48dcd38d | 1166 | static inline void __io_req_set_refcount(struct io_kiocb *req, int nr) |
20e60a38 PB |
1167 | { |
1168 | if (!(req->flags & REQ_F_REFCOUNT)) { | |
1169 | req->flags |= REQ_F_REFCOUNT; | |
48dcd38d | 1170 | atomic_set(&req->refs, nr); |
20e60a38 PB |
1171 | } |
1172 | } | |
1173 | ||
48dcd38d PB |
1174 | static inline void io_req_set_refcount(struct io_kiocb *req) |
1175 | { | |
1176 | __io_req_set_refcount(req, 1); | |
1177 | } | |
1178 | ||
b895c9a6 | 1179 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1180 | { |
1181 | struct io_ring_ctx *ctx = req->ctx; | |
1182 | ||
269bbe5f | 1183 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1184 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1185 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1186 | } |
1187 | } | |
1188 | ||
f70865db PB |
1189 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1190 | { | |
1191 | bool got = percpu_ref_tryget(ref); | |
1192 | ||
1193 | /* already at zero, wait for ->release() */ | |
1194 | if (!got) | |
1195 | wait_for_completion(compl); | |
1196 | percpu_ref_resurrect(ref); | |
1197 | if (got) | |
1198 | percpu_ref_put(ref); | |
1199 | } | |
1200 | ||
3dd0c97a PB |
1201 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1202 | bool cancel_all) | |
08d23634 PB |
1203 | { |
1204 | struct io_kiocb *req; | |
1205 | ||
68207680 | 1206 | if (task && head->task != task) |
08d23634 | 1207 | return false; |
3dd0c97a | 1208 | if (cancel_all) |
08d23634 PB |
1209 | return true; |
1210 | ||
1211 | io_for_each_link(req, head) { | |
b05a1bcd | 1212 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1213 | return true; |
08d23634 PB |
1214 | } |
1215 | return false; | |
1216 | } | |
1217 | ||
93d2bcd2 | 1218 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1219 | { |
93d2bcd2 | 1220 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1221 | } |
4a38aed2 | 1222 | |
a8295b98 HX |
1223 | static inline void req_fail_link_node(struct io_kiocb *req, int res) |
1224 | { | |
1225 | req_set_fail(req); | |
1226 | req->result = res; | |
1227 | } | |
1228 | ||
2b188cc1 JA |
1229 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1230 | { | |
1231 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1232 | ||
0f158b4c | 1233 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1234 | } |
1235 | ||
8eb7e2d0 PB |
1236 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1237 | { | |
1238 | return !req->timeout.off; | |
1239 | } | |
1240 | ||
f56165e6 PB |
1241 | static void io_fallback_req_func(struct work_struct *work) |
1242 | { | |
1243 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
1244 | fallback_work.work); | |
1245 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
1246 | struct io_kiocb *req, *tmp; | |
f237c30a | 1247 | bool locked = false; |
f56165e6 PB |
1248 | |
1249 | percpu_ref_get(&ctx->refs); | |
1250 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) | |
f237c30a | 1251 | req->io_task_work.func(req, &locked); |
5636c00d | 1252 | |
f237c30a PB |
1253 | if (locked) { |
1254 | if (ctx->submit_state.compl_nr) | |
1255 | io_submit_flush_completions(ctx); | |
1256 | mutex_unlock(&ctx->uring_lock); | |
1257 | } | |
f56165e6 | 1258 | percpu_ref_put(&ctx->refs); |
f237c30a | 1259 | |
f56165e6 PB |
1260 | } |
1261 | ||
2b188cc1 JA |
1262 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1263 | { | |
1264 | struct io_ring_ctx *ctx; | |
78076bb6 | 1265 | int hash_bits; |
2b188cc1 JA |
1266 | |
1267 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1268 | if (!ctx) | |
1269 | return NULL; | |
1270 | ||
78076bb6 JA |
1271 | /* |
1272 | * Use 5 bits less than the max cq entries, that should give us around | |
1273 | * 32 entries per hash list if totally full and uniformly spread. | |
1274 | */ | |
1275 | hash_bits = ilog2(p->cq_entries); | |
1276 | hash_bits -= 5; | |
1277 | if (hash_bits <= 0) | |
1278 | hash_bits = 1; | |
1279 | ctx->cancel_hash_bits = hash_bits; | |
1280 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1281 | GFP_KERNEL); | |
1282 | if (!ctx->cancel_hash) | |
1283 | goto err; | |
1284 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1285 | ||
6224843d PB |
1286 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1287 | if (!ctx->dummy_ubuf) | |
1288 | goto err; | |
1289 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1290 | ctx->dummy_ubuf->ubuf = -1UL; | |
1291 | ||
21482896 | 1292 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1293 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1294 | goto err; | |
2b188cc1 JA |
1295 | |
1296 | ctx->flags = p->flags; | |
90554200 | 1297 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1298 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1299 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1300 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1301 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1302 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1303 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1304 | mutex_init(&ctx->uring_lock); |
311997b3 | 1305 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1306 | spin_lock_init(&ctx->completion_lock); |
89850fce | 1307 | spin_lock_init(&ctx->timeout_lock); |
540e32a0 | 1308 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1309 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1310 | INIT_LIST_HEAD(&ctx->timeout_list); |
ef9dd637 | 1311 | INIT_LIST_HEAD(&ctx->ltimeout_list); |
d67d2263 BM |
1312 | spin_lock_init(&ctx->rsrc_ref_lock); |
1313 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1314 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1315 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1316 | INIT_LIST_HEAD(&ctx->tctx_list); |
cd0ca2e0 | 1317 | INIT_LIST_HEAD(&ctx->submit_state.free_list); |
d0acdee2 | 1318 | INIT_LIST_HEAD(&ctx->locked_free_list); |
9011bf9a | 1319 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
2b188cc1 | 1320 | return ctx; |
206aefde | 1321 | err: |
6224843d | 1322 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1323 | kfree(ctx->cancel_hash); |
206aefde JA |
1324 | kfree(ctx); |
1325 | return NULL; | |
2b188cc1 JA |
1326 | } |
1327 | ||
8f6ed49a PB |
1328 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1329 | { | |
1330 | struct io_rings *r = ctx->rings; | |
1331 | ||
1332 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1333 | ctx->cq_extra--; | |
1334 | } | |
1335 | ||
9cf7c104 | 1336 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1337 | { |
2bc9930e JA |
1338 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1339 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1340 | |
8f6ed49a | 1341 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1342 | } |
de0617e4 | 1343 | |
9d858b21 | 1344 | return false; |
de0617e4 JA |
1345 | } |
1346 | ||
c97d8a0f PB |
1347 | #define FFS_ASYNC_READ 0x1UL |
1348 | #define FFS_ASYNC_WRITE 0x2UL | |
1349 | #ifdef CONFIG_64BIT | |
1350 | #define FFS_ISREG 0x4UL | |
1351 | #else | |
1352 | #define FFS_ISREG 0x0UL | |
1353 | #endif | |
1354 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
1355 | ||
1356 | static inline bool io_req_ffs_set(struct io_kiocb *req) | |
1357 | { | |
1358 | return IS_ENABLED(CONFIG_64BIT) && (req->flags & REQ_F_FIXED_FILE); | |
1359 | } | |
1360 | ||
ce3d5aae PB |
1361 | static void io_req_track_inflight(struct io_kiocb *req) |
1362 | { | |
ce3d5aae | 1363 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1364 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1365 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1366 | } |
1367 | } | |
1368 | ||
906c6caa PB |
1369 | static inline void io_unprep_linked_timeout(struct io_kiocb *req) |
1370 | { | |
1371 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
1372 | } | |
1373 | ||
fd08e530 PB |
1374 | static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req) |
1375 | { | |
906c6caa PB |
1376 | if (WARN_ON_ONCE(!req->link)) |
1377 | return NULL; | |
1378 | ||
4d13d1a4 PB |
1379 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
1380 | req->flags |= REQ_F_LINK_TIMEOUT; | |
fd08e530 PB |
1381 | |
1382 | /* linked timeouts should have two refs once prep'ed */ | |
48dcd38d | 1383 | io_req_set_refcount(req); |
4d13d1a4 PB |
1384 | __io_req_set_refcount(req->link, 2); |
1385 | return req->link; | |
fd08e530 PB |
1386 | } |
1387 | ||
1388 | static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) | |
1389 | { | |
4d13d1a4 | 1390 | if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) |
fd08e530 PB |
1391 | return NULL; |
1392 | return __io_prep_linked_timeout(req); | |
1393 | } | |
1394 | ||
1e6fa521 JA |
1395 | static void io_prep_async_work(struct io_kiocb *req) |
1396 | { | |
1397 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1398 | struct io_ring_ctx *ctx = req->ctx; |
1399 | ||
b8e64b53 PB |
1400 | if (!(req->flags & REQ_F_CREDS)) { |
1401 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1402 | req->creds = get_current_cred(); |
b8e64b53 | 1403 | } |
003e8dcc | 1404 | |
e1d675df PB |
1405 | req->work.list.next = NULL; |
1406 | req->work.flags = 0; | |
feaadc4f PB |
1407 | if (req->flags & REQ_F_FORCE_ASYNC) |
1408 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1409 | ||
1e6fa521 JA |
1410 | if (req->flags & REQ_F_ISREG) { |
1411 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1412 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1413 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1414 | if (def->unbound_nonreg_file) |
1415 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1416 | } | |
e1d675df PB |
1417 | |
1418 | switch (req->opcode) { | |
1419 | case IORING_OP_SPLICE: | |
1420 | case IORING_OP_TEE: | |
e1d675df PB |
1421 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1422 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1423 | break; | |
1424 | } | |
561fb04a | 1425 | } |
cccf0ee8 | 1426 | |
cbdcb435 | 1427 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1428 | { |
cbdcb435 | 1429 | struct io_kiocb *cur; |
54a91f3b | 1430 | |
44eff40a PB |
1431 | if (req->flags & REQ_F_LINK_TIMEOUT) { |
1432 | struct io_ring_ctx *ctx = req->ctx; | |
1433 | ||
79ebeaee | 1434 | spin_lock(&ctx->completion_lock); |
44eff40a PB |
1435 | io_for_each_link(cur, req) |
1436 | io_prep_async_work(cur); | |
79ebeaee | 1437 | spin_unlock(&ctx->completion_lock); |
44eff40a PB |
1438 | } else { |
1439 | io_for_each_link(cur, req) | |
1440 | io_prep_async_work(cur); | |
1441 | } | |
561fb04a JA |
1442 | } |
1443 | ||
f237c30a | 1444 | static void io_queue_async_work(struct io_kiocb *req, bool *locked) |
561fb04a | 1445 | { |
a197f664 | 1446 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1447 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1448 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1449 | |
f237c30a PB |
1450 | /* must not take the lock, NULL it as a precaution */ |
1451 | locked = NULL; | |
1452 | ||
3bfe6106 JA |
1453 | BUG_ON(!tctx); |
1454 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1455 | |
cbdcb435 PB |
1456 | /* init ->work of the whole link before punting */ |
1457 | io_prep_async_link(req); | |
991468dc JA |
1458 | |
1459 | /* | |
1460 | * Not expected to happen, but if we do have a bug where this _can_ | |
1461 | * happen, catch it here and ensure the request is marked as | |
1462 | * canceled. That will make io-wq go through the usual work cancel | |
1463 | * procedure rather than attempt to run this request (or create a new | |
1464 | * worker for it). | |
1465 | */ | |
1466 | if (WARN_ON_ONCE(!same_thread_group(req->task, current))) | |
1467 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1468 | ||
d07f1e8a PB |
1469 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1470 | &req->work, req->flags); | |
ebf93667 | 1471 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1472 | if (link) |
1473 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1474 | } |
1475 | ||
1ee4160c | 1476 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1477 | __must_hold(&req->ctx->completion_lock) |
89850fce | 1478 | __must_hold(&req->ctx->timeout_lock) |
5262f567 | 1479 | { |
e8c2bc1f | 1480 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1481 | |
fd9c7bc5 | 1482 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
2ae2eb9d PB |
1483 | if (status) |
1484 | req_set_fail(req); | |
01cec8c1 PB |
1485 | atomic_set(&req->ctx->cq_timeouts, |
1486 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1487 | list_del_init(&req->timeout.list); |
d4d19c19 | 1488 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
91c2f697 | 1489 | io_put_req_deferred(req); |
5262f567 JA |
1490 | } |
1491 | } | |
1492 | ||
441b8a78 | 1493 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1494 | { |
441b8a78 | 1495 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1496 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1497 | struct io_defer_entry, list); | |
de0617e4 | 1498 | |
9cf7c104 | 1499 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1500 | break; |
27dc8338 | 1501 | list_del_init(&de->list); |
907d1df3 | 1502 | io_req_task_queue(de->req); |
27dc8338 | 1503 | kfree(de); |
441b8a78 | 1504 | } |
04518945 PB |
1505 | } |
1506 | ||
360428f8 | 1507 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
89850fce | 1508 | __must_hold(&ctx->completion_lock) |
de0617e4 | 1509 | { |
441b8a78 | 1510 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1511 | |
79ebeaee | 1512 | spin_lock_irq(&ctx->timeout_lock); |
f18ee4cf | 1513 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1514 | u32 events_needed, events_got; |
360428f8 | 1515 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1516 | struct io_kiocb, timeout.list); |
de0617e4 | 1517 | |
8eb7e2d0 | 1518 | if (io_is_timeout_noseq(req)) |
360428f8 | 1519 | break; |
f010505b MDG |
1520 | |
1521 | /* | |
1522 | * Since seq can easily wrap around over time, subtract | |
1523 | * the last seq at which timeouts were flushed before comparing. | |
1524 | * Assuming not more than 2^31-1 events have happened since, | |
1525 | * these subtractions won't have wrapped, so we can check if | |
1526 | * target is in [last_seq, current_seq] by comparing the two. | |
1527 | */ | |
1528 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1529 | events_got = seq - ctx->cq_last_tm_flush; | |
1530 | if (events_got < events_needed) | |
360428f8 | 1531 | break; |
bfe68a22 | 1532 | |
135fcde8 | 1533 | list_del_init(&req->timeout.list); |
1ee4160c | 1534 | io_kill_timeout(req, 0); |
f18ee4cf | 1535 | } |
f010505b | 1536 | ctx->cq_last_tm_flush = seq; |
79ebeaee | 1537 | spin_unlock_irq(&ctx->timeout_lock); |
360428f8 | 1538 | } |
5262f567 | 1539 | |
2335f6f5 | 1540 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1541 | { |
2335f6f5 PB |
1542 | if (ctx->off_timeout_used) |
1543 | io_flush_timeouts(ctx); | |
1544 | if (ctx->drain_active) | |
1545 | io_queue_deferred(ctx); | |
1546 | } | |
1547 | ||
1548 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) | |
1549 | { | |
1550 | if (unlikely(ctx->off_timeout_used || ctx->drain_active)) | |
1551 | __io_commit_cqring_flush(ctx); | |
ec30e04b PB |
1552 | /* order cqe stores with ring update */ |
1553 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1554 | } |
1555 | ||
90554200 JA |
1556 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1557 | { | |
1558 | struct io_rings *r = ctx->rings; | |
1559 | ||
a566c556 | 1560 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1561 | } |
1562 | ||
888aae2e PB |
1563 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1564 | { | |
1565 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1566 | } | |
1567 | ||
d068b506 | 1568 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1569 | { |
75b28aff | 1570 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1571 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1572 | |
115e12e5 SB |
1573 | /* |
1574 | * writes to the cq entry need to come after reading head; the | |
1575 | * control dependency is enough as we're using WRITE_ONCE to | |
1576 | * fill the cq entry | |
1577 | */ | |
a566c556 | 1578 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1579 | return NULL; |
1580 | ||
888aae2e | 1581 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1582 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1583 | } |
1584 | ||
f2842ab5 JA |
1585 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1586 | { | |
44c769de | 1587 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1588 | return false; |
7e55a19c SG |
1589 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1590 | return false; | |
44c769de | 1591 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1592 | } |
1593 | ||
2c5d763c JA |
1594 | /* |
1595 | * This should only get called when at least one event has been posted. | |
1596 | * Some applications rely on the eventfd notification count only changing | |
1597 | * IFF a new CQE has been added to the CQ ring. There's no depedency on | |
1598 | * 1:1 relationship between how many times this function is called (and | |
1599 | * hence the eventfd count) and number of CQEs posted to the CQ ring. | |
1600 | */ | |
b41e9852 | 1601 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1602 | { |
5fd46178 JA |
1603 | /* |
1604 | * wake_up_all() may seem excessive, but io_wake_function() and | |
1605 | * io_should_wake() handle the termination of the loop and only | |
1606 | * wake as many waiters as we need to. | |
1607 | */ | |
1608 | if (wq_has_sleeper(&ctx->cq_wait)) | |
1609 | wake_up_all(&ctx->cq_wait); | |
534ca6d6 JA |
1610 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1611 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1612 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1613 | eventfd_signal(ctx->cq_ev_fd, 1); |
311997b3 PB |
1614 | if (waitqueue_active(&ctx->poll_wait)) { |
1615 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1616 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1617 | } | |
1d7bb1d5 JA |
1618 | } |
1619 | ||
80c18e4a PB |
1620 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1621 | { | |
c57a91fb PB |
1622 | /* see waitqueue_active() comment */ |
1623 | smp_mb(); | |
1624 | ||
80c18e4a | 1625 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c57a91fb | 1626 | if (waitqueue_active(&ctx->cq_wait)) |
5fd46178 | 1627 | wake_up_all(&ctx->cq_wait); |
80c18e4a PB |
1628 | } |
1629 | if (io_should_trigger_evfd(ctx)) | |
1630 | eventfd_signal(ctx->cq_ev_fd, 1); | |
311997b3 PB |
1631 | if (waitqueue_active(&ctx->poll_wait)) { |
1632 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1633 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1634 | } | |
80c18e4a PB |
1635 | } |
1636 | ||
c4a2ed72 | 1637 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1638 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1639 | { |
b18032bb | 1640 | bool all_flushed, posted; |
1d7bb1d5 | 1641 | |
a566c556 | 1642 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1643 | return false; |
1d7bb1d5 | 1644 | |
b18032bb | 1645 | posted = false; |
79ebeaee | 1646 | spin_lock(&ctx->completion_lock); |
6c2450ae | 1647 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1648 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1649 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1650 | |
1d7bb1d5 JA |
1651 | if (!cqe && !force) |
1652 | break; | |
6c2450ae PB |
1653 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1654 | struct io_overflow_cqe, list); | |
1655 | if (cqe) | |
1656 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1657 | else | |
8f6ed49a PB |
1658 | io_account_cq_overflow(ctx); |
1659 | ||
b18032bb | 1660 | posted = true; |
6c2450ae PB |
1661 | list_del(&ocqe->list); |
1662 | kfree(ocqe); | |
1d7bb1d5 JA |
1663 | } |
1664 | ||
09e88404 PB |
1665 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1666 | if (all_flushed) { | |
5ed7a37d | 1667 | clear_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1668 | WRITE_ONCE(ctx->rings->sq_flags, |
1669 | ctx->rings->sq_flags & ~IORING_SQ_CQ_OVERFLOW); | |
09e88404 | 1670 | } |
46930143 | 1671 | |
b18032bb JA |
1672 | if (posted) |
1673 | io_commit_cqring(ctx); | |
79ebeaee | 1674 | spin_unlock(&ctx->completion_lock); |
b18032bb JA |
1675 | if (posted) |
1676 | io_cqring_ev_posted(ctx); | |
09e88404 | 1677 | return all_flushed; |
1d7bb1d5 JA |
1678 | } |
1679 | ||
90f67366 | 1680 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx) |
6c503150 | 1681 | { |
ca0a2651 JA |
1682 | bool ret = true; |
1683 | ||
5ed7a37d | 1684 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1685 | /* iopoll syncs against uring_lock, not completion_lock */ |
1686 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1687 | mutex_lock(&ctx->uring_lock); | |
90f67366 | 1688 | ret = __io_cqring_overflow_flush(ctx, false); |
6c503150 PB |
1689 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1690 | mutex_unlock(&ctx->uring_lock); | |
1691 | } | |
ca0a2651 JA |
1692 | |
1693 | return ret; | |
6c503150 PB |
1694 | } |
1695 | ||
6a290a14 PB |
1696 | /* must to be called somewhat shortly after putting a request */ |
1697 | static inline void io_put_task(struct task_struct *task, int nr) | |
1698 | { | |
1699 | struct io_uring_task *tctx = task->io_uring; | |
1700 | ||
e98e49b2 PB |
1701 | if (likely(task == current)) { |
1702 | tctx->cached_refs += nr; | |
1703 | } else { | |
1704 | percpu_counter_sub(&tctx->inflight, nr); | |
1705 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1706 | wake_up(&tctx->wait); | |
1707 | put_task_struct_many(task, nr); | |
1708 | } | |
6a290a14 PB |
1709 | } |
1710 | ||
9a10867a PB |
1711 | static void io_task_refs_refill(struct io_uring_task *tctx) |
1712 | { | |
1713 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
1714 | ||
1715 | percpu_counter_add(&tctx->inflight, refill); | |
1716 | refcount_add(refill, ¤t->usage); | |
1717 | tctx->cached_refs += refill; | |
1718 | } | |
1719 | ||
1720 | static inline void io_get_task_refs(int nr) | |
1721 | { | |
1722 | struct io_uring_task *tctx = current->io_uring; | |
1723 | ||
1724 | tctx->cached_refs -= nr; | |
1725 | if (unlikely(tctx->cached_refs < 0)) | |
1726 | io_task_refs_refill(tctx); | |
1727 | } | |
1728 | ||
d4d19c19 PB |
1729 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1730 | long res, unsigned int cflags) | |
2b188cc1 | 1731 | { |
cce4b8b0 | 1732 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1733 | |
cce4b8b0 PB |
1734 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1735 | if (!ocqe) { | |
1736 | /* | |
1737 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1738 | * or cannot allocate an overflow entry, then we need to drop it | |
1739 | * on the floor. | |
1740 | */ | |
8f6ed49a | 1741 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1742 | return false; |
2b188cc1 | 1743 | } |
cce4b8b0 | 1744 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1745 | set_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1746 | WRITE_ONCE(ctx->rings->sq_flags, |
1747 | ctx->rings->sq_flags | IORING_SQ_CQ_OVERFLOW); | |
1748 | ||
cce4b8b0 | 1749 | } |
d4d19c19 | 1750 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1751 | ocqe->cqe.res = res; |
1752 | ocqe->cqe.flags = cflags; | |
1753 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1754 | return true; | |
2b188cc1 JA |
1755 | } |
1756 | ||
d4d19c19 PB |
1757 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1758 | long res, unsigned int cflags) | |
2b188cc1 JA |
1759 | { |
1760 | struct io_uring_cqe *cqe; | |
1761 | ||
d4d19c19 | 1762 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1763 | |
2b188cc1 JA |
1764 | /* |
1765 | * If we can't get a cq entry, userspace overflowed the | |
1766 | * submission (by quite a lot). Increment the overflow count in | |
1767 | * the ring. | |
1768 | */ | |
d068b506 | 1769 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1770 | if (likely(cqe)) { |
d4d19c19 | 1771 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1772 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1773 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1774 | return true; |
2b188cc1 | 1775 | } |
d4d19c19 | 1776 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1777 | } |
1778 | ||
8d13326e | 1779 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1780 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1781 | long res, unsigned int cflags) | |
bcda7baa | 1782 | { |
d4d19c19 | 1783 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1784 | } |
1785 | ||
7a612350 PB |
1786 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1787 | unsigned int cflags) | |
2b188cc1 | 1788 | { |
78e19bbe | 1789 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 | 1790 | |
79ebeaee | 1791 | spin_lock(&ctx->completion_lock); |
d4d19c19 | 1792 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1793 | /* |
1794 | * If we're the last reference to this request, add to our locked | |
1795 | * free_list cache. | |
1796 | */ | |
de9b4cca | 1797 | if (req_ref_put_and_test(req)) { |
7a612350 | 1798 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
0756a869 | 1799 | if (req->flags & IO_DISARM_MASK) |
7a612350 PB |
1800 | io_disarm_next(req); |
1801 | if (req->link) { | |
1802 | io_req_task_queue(req->link); | |
1803 | req->link = NULL; | |
1804 | } | |
1805 | } | |
c7dae4ba JA |
1806 | io_dismantle_req(req); |
1807 | io_put_task(req->task, 1); | |
bb943b82 | 1808 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
d0acdee2 | 1809 | ctx->locked_free_nr++; |
180f829f PB |
1810 | } else { |
1811 | if (!percpu_ref_tryget(&ctx->refs)) | |
1812 | req = NULL; | |
1813 | } | |
7a612350 | 1814 | io_commit_cqring(ctx); |
79ebeaee | 1815 | spin_unlock(&ctx->completion_lock); |
7a612350 | 1816 | |
180f829f PB |
1817 | if (req) { |
1818 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1819 | percpu_ref_put(&ctx->refs); |
180f829f | 1820 | } |
229a7b63 JA |
1821 | } |
1822 | ||
4e3d9ff9 JA |
1823 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1824 | { | |
c854357b | 1825 | return req->flags & IO_REQ_CLEAN_FLAGS; |
4e3d9ff9 JA |
1826 | } |
1827 | ||
a38d68db | 1828 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1829 | unsigned int cflags) |
229a7b63 | 1830 | { |
4e3d9ff9 | 1831 | if (io_req_needs_clean(req)) |
68fb8979 | 1832 | io_clean_op(req); |
a38d68db PB |
1833 | req->result = res; |
1834 | req->compl.cflags = cflags; | |
e342c807 | 1835 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1836 | } |
1837 | ||
889fca73 PB |
1838 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1839 | long res, unsigned cflags) | |
bcda7baa | 1840 | { |
889fca73 PB |
1841 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1842 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1843 | else |
c7dae4ba | 1844 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1845 | } |
1846 | ||
a38d68db | 1847 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1848 | { |
889fca73 | 1849 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1850 | } |
1851 | ||
f41db273 PB |
1852 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1853 | { | |
93d2bcd2 | 1854 | req_set_fail(req); |
f41db273 PB |
1855 | io_req_complete_post(req, res, 0); |
1856 | } | |
1857 | ||
c6d3d9cb PB |
1858 | static void io_req_complete_fail_submit(struct io_kiocb *req) |
1859 | { | |
1860 | /* | |
1861 | * We don't submit, fail them all, for that replace hardlinks with | |
1862 | * normal links. Extra REQ_F_LINK is tolerated. | |
1863 | */ | |
1864 | req->flags &= ~REQ_F_HARDLINK; | |
1865 | req->flags |= REQ_F_LINK; | |
1866 | io_req_complete_failed(req, req->result); | |
1867 | } | |
1868 | ||
864ea921 PB |
1869 | /* |
1870 | * Don't initialise the fields below on every allocation, but do that in | |
1871 | * advance and keep them valid across allocations. | |
1872 | */ | |
1873 | static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
1874 | { | |
1875 | req->ctx = ctx; | |
1876 | req->link = NULL; | |
1877 | req->async_data = NULL; | |
1878 | /* not necessary, but safer to zero */ | |
1879 | req->result = 0; | |
1880 | } | |
1881 | ||
dac7a098 | 1882 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
cd0ca2e0 | 1883 | struct io_submit_state *state) |
dac7a098 | 1884 | { |
79ebeaee | 1885 | spin_lock(&ctx->completion_lock); |
cd0ca2e0 | 1886 | list_splice_init(&ctx->locked_free_list, &state->free_list); |
d0acdee2 | 1887 | ctx->locked_free_nr = 0; |
79ebeaee | 1888 | spin_unlock(&ctx->completion_lock); |
dac7a098 PB |
1889 | } |
1890 | ||
dd78f492 | 1891 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1892 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1893 | { |
c7dae4ba | 1894 | struct io_submit_state *state = &ctx->submit_state; |
dd78f492 | 1895 | int nr; |
0ddf92e8 | 1896 | |
c7dae4ba JA |
1897 | /* |
1898 | * If we have more than a batch's worth of requests in our IRQ side | |
1899 | * locked cache, grab the lock and move them over to our submission | |
1900 | * side cache. | |
1901 | */ | |
d0acdee2 | 1902 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
cd0ca2e0 | 1903 | io_flush_cached_locked_reqs(ctx, state); |
0ddf92e8 | 1904 | |
dd78f492 | 1905 | nr = state->free_reqs; |
cd0ca2e0 PB |
1906 | while (!list_empty(&state->free_list)) { |
1907 | struct io_kiocb *req = list_first_entry(&state->free_list, | |
bb943b82 | 1908 | struct io_kiocb, inflight_entry); |
dd78f492 | 1909 | |
bb943b82 | 1910 | list_del(&req->inflight_entry); |
dd78f492 PB |
1911 | state->reqs[nr++] = req; |
1912 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1913 | break; |
1b4c351f JA |
1914 | } |
1915 | ||
dd78f492 PB |
1916 | state->free_reqs = nr; |
1917 | return nr != 0; | |
0ddf92e8 JA |
1918 | } |
1919 | ||
5d5901a3 PB |
1920 | /* |
1921 | * A request might get retired back into the request caches even before opcode | |
1922 | * handlers and io_issue_sqe() are done with it, e.g. inline completion path. | |
1923 | * Because of that, io_alloc_req() should be called only under ->uring_lock | |
1924 | * and with extra caution to not get a request that is still worked on. | |
1925 | */ | |
e5d1bc0a | 1926 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
5d5901a3 | 1927 | __must_hold(&ctx->uring_lock) |
2b188cc1 | 1928 | { |
e5d1bc0a | 1929 | struct io_submit_state *state = &ctx->submit_state; |
864ea921 PB |
1930 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
1931 | int ret, i; | |
e5d1bc0a | 1932 | |
fe7e3257 | 1933 | BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); |
e5d1bc0a | 1934 | |
864ea921 PB |
1935 | if (likely(state->free_reqs || io_flush_cached_reqs(ctx))) |
1936 | goto got_req; | |
e5d1bc0a | 1937 | |
864ea921 PB |
1938 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1939 | state->reqs); | |
fd6fab2c | 1940 | |
864ea921 PB |
1941 | /* |
1942 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1943 | * retry single alloc to be on the safe side. | |
1944 | */ | |
1945 | if (unlikely(ret <= 0)) { | |
1946 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1947 | if (!state->reqs[0]) | |
1948 | return NULL; | |
1949 | ret = 1; | |
2b188cc1 | 1950 | } |
864ea921 PB |
1951 | |
1952 | for (i = 0; i < ret; i++) | |
1953 | io_preinit_req(state->reqs[i], ctx); | |
1954 | state->free_reqs = ret; | |
e5d1bc0a | 1955 | got_req: |
291b2821 PB |
1956 | state->free_reqs--; |
1957 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1958 | } |
1959 | ||
e1d767f0 | 1960 | static inline void io_put_file(struct file *file) |
8da11c19 | 1961 | { |
e1d767f0 | 1962 | if (file) |
8da11c19 PB |
1963 | fput(file); |
1964 | } | |
1965 | ||
4edf20f9 | 1966 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1967 | { |
094bae49 | 1968 | unsigned int flags = req->flags; |
929a3af9 | 1969 | |
3a0a6902 PB |
1970 | if (io_req_needs_clean(req)) |
1971 | io_clean_op(req); | |
e1d767f0 PB |
1972 | if (!(flags & REQ_F_FIXED_FILE)) |
1973 | io_put_file(req->file); | |
269bbe5f BM |
1974 | if (req->fixed_rsrc_refs) |
1975 | percpu_ref_put(req->fixed_rsrc_refs); | |
99ebe4ef | 1976 | if (req->async_data) { |
094bae49 | 1977 | kfree(req->async_data); |
99ebe4ef PB |
1978 | req->async_data = NULL; |
1979 | } | |
e65ef56d JA |
1980 | } |
1981 | ||
216578e5 | 1982 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1983 | { |
51a4cc11 | 1984 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1985 | |
216578e5 | 1986 | io_dismantle_req(req); |
7c660731 | 1987 | io_put_task(req->task, 1); |
c6ca97b3 | 1988 | |
79ebeaee | 1989 | spin_lock(&ctx->completion_lock); |
bb943b82 | 1990 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
c34b025f | 1991 | ctx->locked_free_nr++; |
79ebeaee | 1992 | spin_unlock(&ctx->completion_lock); |
c34b025f | 1993 | |
ecfc5177 | 1994 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1995 | } |
1996 | ||
f2f87370 PB |
1997 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1998 | { | |
1999 | struct io_kiocb *nxt = req->link; | |
2000 | ||
2001 | req->link = nxt->link; | |
2002 | nxt->link = NULL; | |
2003 | } | |
2004 | ||
33cc89a9 PB |
2005 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
2006 | __must_hold(&req->ctx->completion_lock) | |
89b263f6 | 2007 | __must_hold(&req->ctx->timeout_lock) |
2665abfd | 2008 | { |
33cc89a9 | 2009 | struct io_kiocb *link = req->link; |
f2f87370 | 2010 | |
b97e736a | 2011 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
c9abd7ad | 2012 | struct io_timeout_data *io = link->async_data; |
7c86ffee | 2013 | |
f2f87370 | 2014 | io_remove_next_linked(req); |
90cd7e42 | 2015 | link->timeout.head = NULL; |
fd9c7bc5 | 2016 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ef9dd637 | 2017 | list_del(&link->timeout.list); |
d4d19c19 PB |
2018 | io_cqring_fill_event(link->ctx, link->user_data, |
2019 | -ECANCELED, 0); | |
91c2f697 | 2020 | io_put_req_deferred(link); |
d4729fbd | 2021 | return true; |
c9abd7ad PB |
2022 | } |
2023 | } | |
d4729fbd | 2024 | return false; |
7c86ffee PB |
2025 | } |
2026 | ||
d148ca4b | 2027 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 2028 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 2029 | { |
33cc89a9 | 2030 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 2031 | |
f2f87370 | 2032 | req->link = NULL; |
f2f87370 | 2033 | while (link) { |
a8295b98 HX |
2034 | long res = -ECANCELED; |
2035 | ||
2036 | if (link->flags & REQ_F_FAIL) | |
2037 | res = link->result; | |
2038 | ||
f2f87370 PB |
2039 | nxt = link->link; |
2040 | link->link = NULL; | |
2665abfd | 2041 | |
f2f87370 | 2042 | trace_io_uring_fail_link(req, link); |
a8295b98 | 2043 | io_cqring_fill_event(link->ctx, link->user_data, res, 0); |
91c2f697 | 2044 | io_put_req_deferred(link); |
f2f87370 | 2045 | link = nxt; |
9e645e11 | 2046 | } |
33cc89a9 | 2047 | } |
9e645e11 | 2048 | |
33cc89a9 PB |
2049 | static bool io_disarm_next(struct io_kiocb *req) |
2050 | __must_hold(&req->ctx->completion_lock) | |
2051 | { | |
2052 | bool posted = false; | |
2053 | ||
0756a869 PB |
2054 | if (req->flags & REQ_F_ARM_LTIMEOUT) { |
2055 | struct io_kiocb *link = req->link; | |
2056 | ||
906c6caa | 2057 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
0756a869 PB |
2058 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
2059 | io_remove_next_linked(req); | |
2060 | io_cqring_fill_event(link->ctx, link->user_data, | |
2061 | -ECANCELED, 0); | |
2062 | io_put_req_deferred(link); | |
2063 | posted = true; | |
2064 | } | |
2065 | } else if (req->flags & REQ_F_LINK_TIMEOUT) { | |
89b263f6 JA |
2066 | struct io_ring_ctx *ctx = req->ctx; |
2067 | ||
2068 | spin_lock_irq(&ctx->timeout_lock); | |
33cc89a9 | 2069 | posted = io_kill_linked_timeout(req); |
89b263f6 JA |
2070 | spin_unlock_irq(&ctx->timeout_lock); |
2071 | } | |
93d2bcd2 | 2072 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 2073 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
2074 | posted |= (req->link != NULL); |
2075 | io_fail_links(req); | |
2076 | } | |
2077 | return posted; | |
9e645e11 JA |
2078 | } |
2079 | ||
3fa5e0f3 | 2080 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2081 | { |
33cc89a9 | 2082 | struct io_kiocb *nxt; |
944e58bf | 2083 | |
9e645e11 JA |
2084 | /* |
2085 | * If LINK is set, we have dependent requests in this chain. If we | |
2086 | * didn't fail this request, queue the first one up, moving any other | |
2087 | * dependencies to the next request. In case of failure, fail the rest | |
2088 | * of the chain. | |
2089 | */ | |
0756a869 | 2090 | if (req->flags & IO_DISARM_MASK) { |
33cc89a9 | 2091 | struct io_ring_ctx *ctx = req->ctx; |
33cc89a9 PB |
2092 | bool posted; |
2093 | ||
79ebeaee | 2094 | spin_lock(&ctx->completion_lock); |
33cc89a9 PB |
2095 | posted = io_disarm_next(req); |
2096 | if (posted) | |
2097 | io_commit_cqring(req->ctx); | |
79ebeaee | 2098 | spin_unlock(&ctx->completion_lock); |
33cc89a9 PB |
2099 | if (posted) |
2100 | io_cqring_ev_posted(ctx); | |
f2f87370 | 2101 | } |
33cc89a9 PB |
2102 | nxt = req->link; |
2103 | req->link = NULL; | |
2104 | return nxt; | |
4d7dd462 | 2105 | } |
9e645e11 | 2106 | |
f2f87370 | 2107 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2108 | { |
cdbff982 | 2109 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2110 | return NULL; |
2111 | return __io_req_find_next(req); | |
2112 | } | |
2113 | ||
f237c30a | 2114 | static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked) |
2c32395d PB |
2115 | { |
2116 | if (!ctx) | |
2117 | return; | |
f237c30a | 2118 | if (*locked) { |
99c8bc52 HX |
2119 | if (ctx->submit_state.compl_nr) |
2120 | io_submit_flush_completions(ctx); | |
2c32395d | 2121 | mutex_unlock(&ctx->uring_lock); |
f237c30a | 2122 | *locked = false; |
2c32395d PB |
2123 | } |
2124 | percpu_ref_put(&ctx->refs); | |
2125 | } | |
2126 | ||
7cbf1722 | 2127 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2128 | { |
f237c30a | 2129 | bool locked = false; |
ebd0df2e | 2130 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
2131 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
2132 | task_work); | |
c40f6379 | 2133 | |
16f72070 | 2134 | while (1) { |
3f18407d PB |
2135 | struct io_wq_work_node *node; |
2136 | ||
8d4ad41e PB |
2137 | if (!tctx->task_list.first && locked && ctx->submit_state.compl_nr) |
2138 | io_submit_flush_completions(ctx); | |
2139 | ||
3f18407d | 2140 | spin_lock_irq(&tctx->task_lock); |
c6538be9 | 2141 | node = tctx->task_list.first; |
3f18407d | 2142 | INIT_WQ_LIST(&tctx->task_list); |
6294f368 PB |
2143 | if (!node) |
2144 | tctx->task_running = false; | |
3f18407d | 2145 | spin_unlock_irq(&tctx->task_lock); |
6294f368 PB |
2146 | if (!node) |
2147 | break; | |
3f18407d | 2148 | |
6294f368 | 2149 | do { |
3f18407d PB |
2150 | struct io_wq_work_node *next = node->next; |
2151 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2152 | io_task_work.node); | |
2153 | ||
2154 | if (req->ctx != ctx) { | |
f237c30a | 2155 | ctx_flush_and_put(ctx, &locked); |
3f18407d | 2156 | ctx = req->ctx; |
126180b9 PB |
2157 | /* if not contended, grab and improve batching */ |
2158 | locked = mutex_trylock(&ctx->uring_lock); | |
3f18407d PB |
2159 | percpu_ref_get(&ctx->refs); |
2160 | } | |
f237c30a | 2161 | req->io_task_work.func(req, &locked); |
3f18407d | 2162 | node = next; |
6294f368 PB |
2163 | } while (node); |
2164 | ||
7cbf1722 | 2165 | cond_resched(); |
3f18407d | 2166 | } |
ebd0df2e | 2167 | |
f237c30a | 2168 | ctx_flush_and_put(ctx, &locked); |
7cbf1722 JA |
2169 | } |
2170 | ||
e09ee510 | 2171 | static void io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 2172 | { |
c15b79de | 2173 | struct task_struct *tsk = req->task; |
7cbf1722 | 2174 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 2175 | enum task_work_notify_mode notify; |
e09ee510 | 2176 | struct io_wq_work_node *node; |
0b81e80c | 2177 | unsigned long flags; |
6294f368 | 2178 | bool running; |
7cbf1722 JA |
2179 | |
2180 | WARN_ON_ONCE(!tctx); | |
2181 | ||
0b81e80c | 2182 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2183 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
6294f368 PB |
2184 | running = tctx->task_running; |
2185 | if (!running) | |
2186 | tctx->task_running = true; | |
0b81e80c | 2187 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2188 | |
2189 | /* task_work already pending, we're done */ | |
6294f368 | 2190 | if (running) |
e09ee510 | 2191 | return; |
7cbf1722 | 2192 | |
c15b79de PB |
2193 | /* |
2194 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
2195 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2196 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2197 | * will do the job. | |
2198 | */ | |
2199 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
c15b79de PB |
2200 | if (!task_work_add(tsk, &tctx->task_work, notify)) { |
2201 | wake_up_process(tsk); | |
e09ee510 | 2202 | return; |
c15b79de | 2203 | } |
2215bed9 | 2204 | |
0b81e80c | 2205 | spin_lock_irqsave(&tctx->task_lock, flags); |
6294f368 | 2206 | tctx->task_running = false; |
e09ee510 PB |
2207 | node = tctx->task_list.first; |
2208 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2209 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 2210 | |
e09ee510 PB |
2211 | while (node) { |
2212 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
2213 | node = node->next; | |
2214 | if (llist_add(&req->io_task_work.fallback_node, | |
2215 | &req->ctx->fallback_llist)) | |
2216 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
2217 | } | |
eab30c4d PB |
2218 | } |
2219 | ||
f237c30a | 2220 | static void io_req_task_cancel(struct io_kiocb *req, bool *locked) |
c40f6379 | 2221 | { |
87ceb6a6 | 2222 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2223 | |
b18a1a45 | 2224 | /* not needed for normal modes, but SQPOLL depends on it */ |
f237c30a | 2225 | io_tw_lock(ctx, locked); |
2593553a | 2226 | io_req_complete_failed(req, req->result); |
c40f6379 JA |
2227 | } |
2228 | ||
f237c30a | 2229 | static void io_req_task_submit(struct io_kiocb *req, bool *locked) |
c40f6379 JA |
2230 | { |
2231 | struct io_ring_ctx *ctx = req->ctx; | |
2232 | ||
f237c30a | 2233 | io_tw_lock(ctx, locked); |
316319e8 | 2234 | /* req->task == current here, checking PF_EXITING is safe */ |
af066f31 | 2235 | if (likely(!(req->task->flags & PF_EXITING))) |
c5eef2b9 | 2236 | __io_queue_sqe(req); |
81b6d05c | 2237 | else |
2593553a | 2238 | io_req_complete_failed(req, -EFAULT); |
c40f6379 JA |
2239 | } |
2240 | ||
2c4b8eb6 | 2241 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2242 | { |
2c4b8eb6 | 2243 | req->result = ret; |
5b0a6acc | 2244 | req->io_task_work.func = io_req_task_cancel; |
e09ee510 | 2245 | io_req_task_work_add(req); |
c40f6379 JA |
2246 | } |
2247 | ||
2c4b8eb6 | 2248 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2249 | { |
5b0a6acc | 2250 | req->io_task_work.func = io_req_task_submit; |
e09ee510 | 2251 | io_req_task_work_add(req); |
a3df7698 PB |
2252 | } |
2253 | ||
773af691 JA |
2254 | static void io_req_task_queue_reissue(struct io_kiocb *req) |
2255 | { | |
2256 | req->io_task_work.func = io_queue_async_work; | |
2257 | io_req_task_work_add(req); | |
2258 | } | |
2259 | ||
f2f87370 | 2260 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2261 | { |
9b5f7bd9 | 2262 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2263 | |
2264 | if (nxt) | |
906a8c3f | 2265 | io_req_task_queue(nxt); |
c69f8dbe JL |
2266 | } |
2267 | ||
c3524383 | 2268 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2269 | { |
c3524383 PB |
2270 | io_queue_next(req); |
2271 | __io_free_req(req); | |
2272 | } | |
8766dd51 | 2273 | |
f237c30a PB |
2274 | static void io_free_req_work(struct io_kiocb *req, bool *locked) |
2275 | { | |
2276 | io_free_req(req); | |
2277 | } | |
2278 | ||
2d6500d4 | 2279 | struct req_batch { |
5af1d13e PB |
2280 | struct task_struct *task; |
2281 | int task_refs; | |
1b4c351f | 2282 | int ctx_refs; |
2d6500d4 PB |
2283 | }; |
2284 | ||
5af1d13e PB |
2285 | static inline void io_init_req_batch(struct req_batch *rb) |
2286 | { | |
5af1d13e | 2287 | rb->task_refs = 0; |
9ae72463 | 2288 | rb->ctx_refs = 0; |
5af1d13e PB |
2289 | rb->task = NULL; |
2290 | } | |
2291 | ||
2d6500d4 PB |
2292 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2293 | struct req_batch *rb) | |
2294 | { | |
9ae72463 PB |
2295 | if (rb->ctx_refs) |
2296 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
e98e49b2 | 2297 | if (rb->task) |
e9dbe221 | 2298 | io_put_task(rb->task, rb->task_refs); |
2d6500d4 PB |
2299 | } |
2300 | ||
6ff119a6 PB |
2301 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2302 | struct io_submit_state *state) | |
2d6500d4 | 2303 | { |
f2f87370 | 2304 | io_queue_next(req); |
96670657 | 2305 | io_dismantle_req(req); |
2d6500d4 | 2306 | |
e3bc8e9d | 2307 | if (req->task != rb->task) { |
7c660731 PB |
2308 | if (rb->task) |
2309 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2310 | rb->task = req->task; |
2311 | rb->task_refs = 0; | |
5af1d13e | 2312 | } |
e3bc8e9d | 2313 | rb->task_refs++; |
9ae72463 | 2314 | rb->ctx_refs++; |
5af1d13e | 2315 | |
bd759045 | 2316 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2317 | state->reqs[state->free_reqs++] = req; |
bd759045 | 2318 | else |
cd0ca2e0 | 2319 | list_add(&req->inflight_entry, &state->free_list); |
7a743e22 PB |
2320 | } |
2321 | ||
2a2758f2 | 2322 | static void io_submit_flush_completions(struct io_ring_ctx *ctx) |
a141dd89 | 2323 | __must_hold(&ctx->uring_lock) |
905c172f | 2324 | { |
cd0ca2e0 PB |
2325 | struct io_submit_state *state = &ctx->submit_state; |
2326 | int i, nr = state->compl_nr; | |
905c172f PB |
2327 | struct req_batch rb; |
2328 | ||
79ebeaee | 2329 | spin_lock(&ctx->completion_lock); |
905c172f | 2330 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2331 | struct io_kiocb *req = state->compl_reqs[i]; |
5182ed2e | 2332 | |
d4d19c19 PB |
2333 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2334 | req->compl.cflags); | |
905c172f PB |
2335 | } |
2336 | io_commit_cqring(ctx); | |
79ebeaee | 2337 | spin_unlock(&ctx->completion_lock); |
905c172f | 2338 | io_cqring_ev_posted(ctx); |
5182ed2e PB |
2339 | |
2340 | io_init_req_batch(&rb); | |
905c172f | 2341 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2342 | struct io_kiocb *req = state->compl_reqs[i]; |
905c172f | 2343 | |
91c2f697 | 2344 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2345 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2346 | } |
2347 | ||
2348 | io_req_free_batch_finish(ctx, &rb); | |
cd0ca2e0 | 2349 | state->compl_nr = 0; |
7a743e22 PB |
2350 | } |
2351 | ||
ba816ad6 JA |
2352 | /* |
2353 | * Drop reference to request, return next in chain (if there is one) if this | |
2354 | * was the last reference to this request. | |
2355 | */ | |
0d85035a | 2356 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2357 | { |
9b5f7bd9 PB |
2358 | struct io_kiocb *nxt = NULL; |
2359 | ||
de9b4cca | 2360 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2361 | nxt = io_req_find_next(req); |
4d7dd462 | 2362 | __io_free_req(req); |
2a44f467 | 2363 | } |
9b5f7bd9 | 2364 | return nxt; |
2b188cc1 JA |
2365 | } |
2366 | ||
0d85035a | 2367 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2368 | { |
de9b4cca | 2369 | if (req_ref_put_and_test(req)) |
e65ef56d | 2370 | io_free_req(req); |
2b188cc1 JA |
2371 | } |
2372 | ||
91c2f697 | 2373 | static inline void io_put_req_deferred(struct io_kiocb *req) |
216578e5 | 2374 | { |
91c2f697 | 2375 | if (req_ref_put_and_test(req)) { |
f237c30a | 2376 | req->io_task_work.func = io_free_req_work; |
543af3a1 PB |
2377 | io_req_task_work_add(req); |
2378 | } | |
216578e5 PB |
2379 | } |
2380 | ||
6c503150 | 2381 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2382 | { |
2383 | /* See comment at the top of this file */ | |
2384 | smp_rmb(); | |
e23de15f | 2385 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2386 | } |
2387 | ||
fb5ccc98 PB |
2388 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2389 | { | |
2390 | struct io_rings *rings = ctx->rings; | |
2391 | ||
2392 | /* make sure SQ entry isn't read before tail */ | |
2393 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2394 | } | |
2395 | ||
8ff069bf | 2396 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2397 | { |
8ff069bf | 2398 | unsigned int cflags; |
e94f141b | 2399 | |
bcda7baa JA |
2400 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2401 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2402 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2403 | kfree(kbuf); |
2404 | return cflags; | |
e94f141b JA |
2405 | } |
2406 | ||
8ff069bf | 2407 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2408 | { |
4d954c25 | 2409 | struct io_buffer *kbuf; |
bcda7baa | 2410 | |
ae421d93 PB |
2411 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) |
2412 | return 0; | |
4d954c25 | 2413 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2414 | return io_put_kbuf(req, kbuf); |
2415 | } | |
2416 | ||
4c6e277c JA |
2417 | static inline bool io_run_task_work(void) |
2418 | { | |
ef98eb04 | 2419 | if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) { |
4c6e277c | 2420 | __set_current_state(TASK_RUNNING); |
ef98eb04 | 2421 | tracehook_notify_signal(); |
4c6e277c JA |
2422 | return true; |
2423 | } | |
2424 | ||
2425 | return false; | |
bcda7baa JA |
2426 | } |
2427 | ||
def596e9 JA |
2428 | /* |
2429 | * Find and free completed poll iocbs | |
2430 | */ | |
2431 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
a8576af9 | 2432 | struct list_head *done) |
def596e9 | 2433 | { |
8237e045 | 2434 | struct req_batch rb; |
def596e9 | 2435 | struct io_kiocb *req; |
bbde017a XW |
2436 | |
2437 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2438 | smp_rmb(); | |
def596e9 | 2439 | |
5af1d13e | 2440 | io_init_req_batch(&rb); |
def596e9 | 2441 | while (!list_empty(done)) { |
d21ffe7e | 2442 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2443 | list_del(&req->inflight_entry); |
2444 | ||
ae421d93 PB |
2445 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2446 | io_put_rw_kbuf(req)); | |
def596e9 JA |
2447 | (*nr_events)++; |
2448 | ||
de9b4cca | 2449 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2450 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2451 | } |
def596e9 | 2452 | |
09bb8394 | 2453 | io_commit_cqring(ctx); |
80c18e4a | 2454 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2455 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2456 | } |
2457 | ||
def596e9 | 2458 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
a8576af9 | 2459 | long min) |
def596e9 JA |
2460 | { |
2461 | struct io_kiocb *req, *tmp; | |
2462 | LIST_HEAD(done); | |
2463 | bool spin; | |
def596e9 JA |
2464 | |
2465 | /* | |
2466 | * Only spin for completions if we don't have multiple devices hanging | |
2467 | * off our complete list, and we're under the requested amount. | |
2468 | */ | |
915b3dde | 2469 | spin = !ctx->poll_multi_queue && *nr_events < min; |
def596e9 | 2470 | |
d21ffe7e | 2471 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2472 | struct kiocb *kiocb = &req->rw.kiocb; |
a2416e1e | 2473 | int ret; |
def596e9 JA |
2474 | |
2475 | /* | |
581f9810 BM |
2476 | * Move completed and retryable entries to our local lists. |
2477 | * If we find a request that requires polling, break out | |
2478 | * and complete those lists first, if we have entries there. | |
def596e9 | 2479 | */ |
65a6543d | 2480 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2481 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2482 | continue; |
2483 | } | |
2484 | if (!list_empty(&done)) | |
2485 | break; | |
2486 | ||
2487 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
a2416e1e PB |
2488 | if (unlikely(ret < 0)) |
2489 | return ret; | |
2490 | else if (ret) | |
2491 | spin = false; | |
def596e9 | 2492 | |
3aadc23e PB |
2493 | /* iopoll may have completed current req */ |
2494 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2495 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2496 | } |
2497 | ||
2498 | if (!list_empty(&done)) | |
a8576af9 | 2499 | io_iopoll_complete(ctx, nr_events, &done); |
def596e9 | 2500 | |
a2416e1e | 2501 | return 0; |
def596e9 JA |
2502 | } |
2503 | ||
def596e9 JA |
2504 | /* |
2505 | * We can't just wait for polled events to come to us, we have to actively | |
2506 | * find and complete them. | |
2507 | */ | |
b2edc0a7 | 2508 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2509 | { |
2510 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2511 | return; | |
2512 | ||
2513 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2514 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2515 | unsigned int nr_events = 0; |
2516 | ||
a8576af9 | 2517 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2518 | |
b2edc0a7 PB |
2519 | /* let it sleep and repeat later if can't complete a request */ |
2520 | if (nr_events == 0) | |
2521 | break; | |
08f5439f JA |
2522 | /* |
2523 | * Ensure we allow local-to-the-cpu processing to take place, | |
2524 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2525 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2526 | */ |
3fcee5a6 PB |
2527 | if (need_resched()) { |
2528 | mutex_unlock(&ctx->uring_lock); | |
2529 | cond_resched(); | |
2530 | mutex_lock(&ctx->uring_lock); | |
2531 | } | |
def596e9 JA |
2532 | } |
2533 | mutex_unlock(&ctx->uring_lock); | |
2534 | } | |
2535 | ||
7668b92a | 2536 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2537 | { |
7668b92a | 2538 | unsigned int nr_events = 0; |
e9979b36 | 2539 | int ret = 0; |
500f9fba | 2540 | |
c7849be9 XW |
2541 | /* |
2542 | * We disallow the app entering submit/complete with polling, but we | |
2543 | * still need to lock the ring to prevent racing with polled issue | |
2544 | * that got punted to a workqueue. | |
2545 | */ | |
2546 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2547 | /* |
2548 | * Don't enter poll loop if we already have events pending. | |
2549 | * If we do, we can potentially be spinning for commands that | |
2550 | * already triggered a CQE (eg in error). | |
2551 | */ | |
5ed7a37d | 2552 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2553 | __io_cqring_overflow_flush(ctx, false); |
2554 | if (io_cqring_events(ctx)) | |
2555 | goto out; | |
def596e9 | 2556 | do { |
500f9fba JA |
2557 | /* |
2558 | * If a submit got punted to a workqueue, we can have the | |
2559 | * application entering polling for a command before it gets | |
2560 | * issued. That app will hold the uring_lock for the duration | |
2561 | * of the poll right here, so we need to take a breather every | |
2562 | * now and then to ensure that the issue has a chance to add | |
2563 | * the poll to the issued list. Otherwise we can spin here | |
2564 | * forever, while the workqueue is stuck trying to acquire the | |
2565 | * very same mutex. | |
2566 | */ | |
e9979b36 | 2567 | if (list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
2568 | u32 tail = ctx->cached_cq_tail; |
2569 | ||
500f9fba | 2570 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2571 | io_run_task_work(); |
500f9fba | 2572 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2573 | |
8f487ef2 PB |
2574 | /* some requests don't go through iopoll_list */ |
2575 | if (tail != ctx->cached_cq_tail || | |
2576 | list_empty(&ctx->iopoll_list)) | |
e9979b36 | 2577 | break; |
500f9fba | 2578 | } |
a8576af9 | 2579 | ret = io_do_iopoll(ctx, &nr_events, min); |
f39c8a5b PB |
2580 | } while (!ret && nr_events < min && !need_resched()); |
2581 | out: | |
500f9fba | 2582 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2583 | return ret; |
2584 | } | |
2585 | ||
491381ce | 2586 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2587 | { |
491381ce JA |
2588 | /* |
2589 | * Tell lockdep we inherited freeze protection from submission | |
2590 | * thread. | |
2591 | */ | |
2592 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2593 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2594 | |
1c98679d PB |
2595 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2596 | sb_end_write(sb); | |
2b188cc1 JA |
2597 | } |
2598 | } | |
2599 | ||
b63534c4 | 2600 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2601 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2602 | { |
ab454438 | 2603 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2604 | |
ab454438 PB |
2605 | if (!rw) |
2606 | return !io_req_prep_async(req); | |
cd658695 | 2607 | iov_iter_restore(&rw->iter, &rw->iter_state); |
ab454438 | 2608 | return true; |
b63534c4 | 2609 | } |
b63534c4 | 2610 | |
3e6a0d3c | 2611 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2612 | { |
355afaeb | 2613 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2614 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2615 | |
355afaeb JA |
2616 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2617 | return false; | |
3e6a0d3c JA |
2618 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2619 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2620 | return false; |
7c977a58 JA |
2621 | /* |
2622 | * If ref is dying, we might be running poll reap from the exit work. | |
2623 | * Don't attempt to reissue from that path, just let it fail with | |
2624 | * -EAGAIN. | |
2625 | */ | |
3e6a0d3c JA |
2626 | if (percpu_ref_is_dying(&ctx->refs)) |
2627 | return false; | |
ef046888 JA |
2628 | /* |
2629 | * Play it safe and assume not safe to re-import and reissue if we're | |
2630 | * not in the original thread group (or in task context). | |
2631 | */ | |
2632 | if (!same_thread_group(req->task, current) || !in_task()) | |
2633 | return false; | |
3e6a0d3c JA |
2634 | return true; |
2635 | } | |
e82ad485 | 2636 | #else |
a1ff1e3f | 2637 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2638 | { |
2639 | return false; | |
2640 | } | |
e82ad485 | 2641 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2642 | { |
b63534c4 JA |
2643 | return false; |
2644 | } | |
3e6a0d3c | 2645 | #endif |
b63534c4 | 2646 | |
8ef12efe | 2647 | static bool __io_complete_rw_common(struct io_kiocb *req, long res) |
a1d7c393 | 2648 | { |
b65c128f PB |
2649 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2650 | kiocb_end_write(req); | |
9532b99b PB |
2651 | if (res != req->result) { |
2652 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2653 | io_rw_should_reissue(req)) { | |
2654 | req->flags |= REQ_F_REISSUE; | |
8ef12efe | 2655 | return true; |
9532b99b | 2656 | } |
93d2bcd2 | 2657 | req_set_fail(req); |
8ef12efe | 2658 | req->result = res; |
9532b99b | 2659 | } |
8ef12efe JA |
2660 | return false; |
2661 | } | |
2662 | ||
f237c30a | 2663 | static void io_req_task_complete(struct io_kiocb *req, bool *locked) |
8ef12efe | 2664 | { |
126180b9 PB |
2665 | unsigned int cflags = io_put_rw_kbuf(req); |
2666 | long res = req->result; | |
2667 | ||
2668 | if (*locked) { | |
2669 | struct io_ring_ctx *ctx = req->ctx; | |
2670 | struct io_submit_state *state = &ctx->submit_state; | |
2671 | ||
2672 | io_req_complete_state(req, res, cflags); | |
2673 | state->compl_reqs[state->compl_nr++] = req; | |
2674 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2675 | io_submit_flush_completions(ctx); | |
2676 | } else { | |
2677 | io_req_complete_post(req, res, cflags); | |
2678 | } | |
8ef12efe JA |
2679 | } |
2680 | ||
2681 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, | |
2682 | unsigned int issue_flags) | |
2683 | { | |
2684 | if (__io_complete_rw_common(req, res)) | |
2685 | return; | |
63637853 | 2686 | __io_req_complete(req, issue_flags, req->result, io_put_rw_kbuf(req)); |
ba816ad6 JA |
2687 | } |
2688 | ||
2689 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2690 | { | |
9adbd45d | 2691 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2692 | |
8ef12efe JA |
2693 | if (__io_complete_rw_common(req, res)) |
2694 | return; | |
2695 | req->result = res; | |
2696 | req->io_task_work.func = io_req_task_complete; | |
2697 | io_req_task_work_add(req); | |
2b188cc1 JA |
2698 | } |
2699 | ||
def596e9 JA |
2700 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2701 | { | |
9adbd45d | 2702 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2703 | |
491381ce JA |
2704 | if (kiocb->ki_flags & IOCB_WRITE) |
2705 | kiocb_end_write(req); | |
9532b99b | 2706 | if (unlikely(res != req->result)) { |
b66ceaf3 PB |
2707 | if (res == -EAGAIN && io_rw_should_reissue(req)) { |
2708 | req->flags |= REQ_F_REISSUE; | |
2709 | return; | |
9532b99b | 2710 | } |
8c130827 | 2711 | } |
bbde017a XW |
2712 | |
2713 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2714 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2715 | smp_wmb(); |
2716 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2717 | } |
2718 | ||
2719 | /* | |
2720 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2721 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2722 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2723 | * accessing the kiocb cookie. |
2724 | */ | |
cb3d8972 | 2725 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2726 | { |
2727 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2728 | const bool in_async = io_wq_current_is_worker(); |
2729 | ||
2730 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2731 | if (unlikely(in_async)) | |
2732 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2733 | |
2734 | /* | |
2735 | * Track whether we have multiple files in our lists. This will impact | |
2736 | * how we do polling eventually, not spinning if we're on potentially | |
2737 | * different devices. | |
2738 | */ | |
540e32a0 | 2739 | if (list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
2740 | ctx->poll_multi_queue = false; |
2741 | } else if (!ctx->poll_multi_queue) { | |
def596e9 | 2742 | struct io_kiocb *list_req; |
915b3dde | 2743 | unsigned int queue_num0, queue_num1; |
def596e9 | 2744 | |
540e32a0 | 2745 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2746 | inflight_entry); |
915b3dde HX |
2747 | |
2748 | if (list_req->file != req->file) { | |
2749 | ctx->poll_multi_queue = true; | |
2750 | } else { | |
2751 | queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie); | |
2752 | queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie); | |
2753 | if (queue_num0 != queue_num1) | |
2754 | ctx->poll_multi_queue = true; | |
2755 | } | |
def596e9 JA |
2756 | } |
2757 | ||
2758 | /* | |
2759 | * For fast devices, IO may have already completed. If it has, add | |
2760 | * it to the front so we find it first. | |
2761 | */ | |
65a6543d | 2762 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2763 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2764 | else |
d21ffe7e | 2765 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2766 | |
cb3d8972 PB |
2767 | if (unlikely(in_async)) { |
2768 | /* | |
2769 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2770 | * in sq thread task context or in io worker task context. If | |
2771 | * current task context is sq thread, we don't need to check | |
2772 | * whether should wake up sq thread. | |
2773 | */ | |
2774 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2775 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2776 | wake_up(&ctx->sq_data->wait); | |
2777 | ||
2778 | mutex_unlock(&ctx->uring_lock); | |
2779 | } | |
def596e9 JA |
2780 | } |
2781 | ||
4503b767 JA |
2782 | static bool io_bdev_nowait(struct block_device *bdev) |
2783 | { | |
9ba0d0c8 | 2784 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2785 | } |
2786 | ||
2b188cc1 JA |
2787 | /* |
2788 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2789 | * any file. For now, just ensure that anything potentially problematic is done | |
2790 | * inline. | |
2791 | */ | |
b191e2df | 2792 | static bool __io_file_supports_nowait(struct file *file, int rw) |
2b188cc1 JA |
2793 | { |
2794 | umode_t mode = file_inode(file)->i_mode; | |
2795 | ||
4503b767 | 2796 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2797 | if (IS_ENABLED(CONFIG_BLOCK) && |
2798 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2799 | return true; |
2800 | return false; | |
2801 | } | |
976517f1 | 2802 | if (S_ISSOCK(mode)) |
2b188cc1 | 2803 | return true; |
4503b767 | 2804 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2805 | if (IS_ENABLED(CONFIG_BLOCK) && |
2806 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2807 | file->f_op != &io_uring_fops) |
2808 | return true; | |
2809 | return false; | |
2810 | } | |
2b188cc1 | 2811 | |
c5b85625 JA |
2812 | /* any ->read/write should understand O_NONBLOCK */ |
2813 | if (file->f_flags & O_NONBLOCK) | |
2814 | return true; | |
2815 | ||
af197f50 JA |
2816 | if (!(file->f_mode & FMODE_NOWAIT)) |
2817 | return false; | |
2818 | ||
2819 | if (rw == READ) | |
2820 | return file->f_op->read_iter != NULL; | |
2821 | ||
2822 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2823 | } |
2824 | ||
b191e2df | 2825 | static bool io_file_supports_nowait(struct io_kiocb *req, int rw) |
7b29f92d | 2826 | { |
b191e2df | 2827 | if (rw == READ && (req->flags & REQ_F_NOWAIT_READ)) |
7b29f92d | 2828 | return true; |
b191e2df | 2829 | else if (rw == WRITE && (req->flags & REQ_F_NOWAIT_WRITE)) |
7b29f92d JA |
2830 | return true; |
2831 | ||
b191e2df | 2832 | return __io_file_supports_nowait(req->file, rw); |
7b29f92d JA |
2833 | } |
2834 | ||
5d329e12 JA |
2835 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2836 | int rw) | |
2b188cc1 | 2837 | { |
def596e9 | 2838 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2839 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2840 | struct file *file = req->file; |
09bb8394 JA |
2841 | unsigned ioprio; |
2842 | int ret; | |
2b188cc1 | 2843 | |
c97d8a0f | 2844 | if (!io_req_ffs_set(req) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2845 | req->flags |= REQ_F_ISREG; |
2846 | ||
2b188cc1 | 2847 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2848 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2849 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2850 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2851 | } |
2b188cc1 | 2852 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2853 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2854 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2855 | if (unlikely(ret)) | |
2856 | return ret; | |
2b188cc1 | 2857 | |
5d329e12 JA |
2858 | /* |
2859 | * If the file is marked O_NONBLOCK, still allow retry for it if it | |
2860 | * supports async. Otherwise it's impossible to use O_NONBLOCK files | |
2861 | * reliably. If not, or it IOCB_NOWAIT is set, don't retry. | |
2862 | */ | |
2863 | if ((kiocb->ki_flags & IOCB_NOWAIT) || | |
2864 | ((file->f_flags & O_NONBLOCK) && !io_file_supports_nowait(req, rw))) | |
75c668cd PB |
2865 | req->flags |= REQ_F_NOWAIT; |
2866 | ||
2b188cc1 JA |
2867 | ioprio = READ_ONCE(sqe->ioprio); |
2868 | if (ioprio) { | |
2869 | ret = ioprio_check_cap(ioprio); | |
2870 | if (ret) | |
09bb8394 | 2871 | return ret; |
2b188cc1 JA |
2872 | |
2873 | kiocb->ki_ioprio = ioprio; | |
2874 | } else | |
2875 | kiocb->ki_ioprio = get_current_ioprio(); | |
2876 | ||
def596e9 | 2877 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2878 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2879 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2880 | return -EOPNOTSUPP; |
2b188cc1 | 2881 | |
394918eb | 2882 | kiocb->ki_flags |= IOCB_HIPRI | IOCB_ALLOC_CACHE; |
def596e9 | 2883 | kiocb->ki_complete = io_complete_rw_iopoll; |
65a6543d | 2884 | req->iopoll_completed = 0; |
def596e9 | 2885 | } else { |
09bb8394 JA |
2886 | if (kiocb->ki_flags & IOCB_HIPRI) |
2887 | return -EINVAL; | |
def596e9 JA |
2888 | kiocb->ki_complete = io_complete_rw; |
2889 | } | |
9adbd45d | 2890 | |
eae071c9 PB |
2891 | if (req->opcode == IORING_OP_READ_FIXED || |
2892 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2893 | req->imu = NULL; | |
2894 | io_req_set_rsrc_node(req); | |
2895 | } | |
2896 | ||
3529d8c2 JA |
2897 | req->rw.addr = READ_ONCE(sqe->addr); |
2898 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2899 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2900 | return 0; |
2b188cc1 JA |
2901 | } |
2902 | ||
2903 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2904 | { | |
2905 | switch (ret) { | |
2906 | case -EIOCBQUEUED: | |
2907 | break; | |
2908 | case -ERESTARTSYS: | |
2909 | case -ERESTARTNOINTR: | |
2910 | case -ERESTARTNOHAND: | |
2911 | case -ERESTART_RESTARTBLOCK: | |
2912 | /* | |
2913 | * We can't just restart the syscall, since previously | |
2914 | * submitted sqes may already be in progress. Just fail this | |
2915 | * IO with EINTR. | |
2916 | */ | |
2917 | ret = -EINTR; | |
df561f66 | 2918 | fallthrough; |
2b188cc1 JA |
2919 | default: |
2920 | kiocb->ki_complete(kiocb, ret, 0); | |
2921 | } | |
2922 | } | |
2923 | ||
a1d7c393 | 2924 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2925 | unsigned int issue_flags) |
ba816ad6 | 2926 | { |
ba04291e | 2927 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2928 | struct io_async_rw *io = req->async_data; |
ba04291e | 2929 | |
227c0c96 | 2930 | /* add previously done IO, if any */ |
e8c2bc1f | 2931 | if (io && io->bytes_done > 0) { |
227c0c96 | 2932 | if (ret < 0) |
e8c2bc1f | 2933 | ret = io->bytes_done; |
227c0c96 | 2934 | else |
e8c2bc1f | 2935 | ret += io->bytes_done; |
227c0c96 JA |
2936 | } |
2937 | ||
ba04291e JA |
2938 | if (req->flags & REQ_F_CUR_POS) |
2939 | req->file->f_pos = kiocb->ki_pos; | |
b66ceaf3 | 2940 | if (ret >= 0 && (kiocb->ki_complete == io_complete_rw)) |
889fca73 | 2941 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2942 | else |
2943 | io_rw_done(kiocb, ret); | |
97284637 | 2944 | |
b66ceaf3 | 2945 | if (req->flags & REQ_F_REISSUE) { |
97284637 | 2946 | req->flags &= ~REQ_F_REISSUE; |
a7be7c23 | 2947 | if (io_resubmit_prep(req)) { |
773af691 | 2948 | io_req_task_queue_reissue(req); |
8c130827 | 2949 | } else { |
b66ceaf3 PB |
2950 | unsigned int cflags = io_put_rw_kbuf(req); |
2951 | struct io_ring_ctx *ctx = req->ctx; | |
2952 | ||
93d2bcd2 | 2953 | req_set_fail(req); |
b66ceaf3 PB |
2954 | if (issue_flags & IO_URING_F_NONBLOCK) { |
2955 | mutex_lock(&ctx->uring_lock); | |
2956 | __io_req_complete(req, issue_flags, ret, cflags); | |
2957 | mutex_unlock(&ctx->uring_lock); | |
2958 | } else { | |
2959 | __io_req_complete(req, issue_flags, ret, cflags); | |
2960 | } | |
97284637 PB |
2961 | } |
2962 | } | |
ba816ad6 JA |
2963 | } |
2964 | ||
eae071c9 PB |
2965 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2966 | struct io_mapped_ubuf *imu) | |
edafccee | 2967 | { |
9adbd45d | 2968 | size_t len = req->rw.len; |
75769e3f | 2969 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2970 | size_t offset; |
edafccee | 2971 | |
75769e3f | 2972 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2973 | return -EFAULT; |
2974 | /* not inside the mapped region */ | |
4751f53d | 2975 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2976 | return -EFAULT; |
2977 | ||
2978 | /* | |
2979 | * May not be a start of buffer, set size appropriately | |
2980 | * and advance us to the beginning. | |
2981 | */ | |
2982 | offset = buf_addr - imu->ubuf; | |
2983 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2984 | |
2985 | if (offset) { | |
2986 | /* | |
2987 | * Don't use iov_iter_advance() here, as it's really slow for | |
2988 | * using the latter parts of a big fixed buffer - it iterates | |
2989 | * over each segment manually. We can cheat a bit here, because | |
2990 | * we know that: | |
2991 | * | |
2992 | * 1) it's a BVEC iter, we set it up | |
2993 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2994 | * first and last bvec | |
2995 | * | |
2996 | * So just find our index, and adjust the iterator afterwards. | |
2997 | * If the offset is within the first bvec (or the whole first | |
2998 | * bvec, just use iov_iter_advance(). This makes it easier | |
2999 | * since we can just skip the first segment, which may not | |
3000 | * be PAGE_SIZE aligned. | |
3001 | */ | |
3002 | const struct bio_vec *bvec = imu->bvec; | |
3003 | ||
3004 | if (offset <= bvec->bv_len) { | |
3005 | iov_iter_advance(iter, offset); | |
3006 | } else { | |
3007 | unsigned long seg_skip; | |
3008 | ||
3009 | /* skip first vec */ | |
3010 | offset -= bvec->bv_len; | |
3011 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3012 | ||
3013 | iter->bvec = bvec + seg_skip; | |
3014 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3015 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3016 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3017 | } |
3018 | } | |
3019 | ||
847595de | 3020 | return 0; |
edafccee JA |
3021 | } |
3022 | ||
eae071c9 PB |
3023 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
3024 | { | |
3025 | struct io_ring_ctx *ctx = req->ctx; | |
3026 | struct io_mapped_ubuf *imu = req->imu; | |
3027 | u16 index, buf_index = req->buf_index; | |
3028 | ||
3029 | if (likely(!imu)) { | |
3030 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
3031 | return -EFAULT; | |
3032 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
3033 | imu = READ_ONCE(ctx->user_bufs[index]); | |
3034 | req->imu = imu; | |
3035 | } | |
3036 | return __io_import_fixed(req, rw, iter, imu); | |
3037 | } | |
3038 | ||
bcda7baa JA |
3039 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3040 | { | |
3041 | if (needs_lock) | |
3042 | mutex_unlock(&ctx->uring_lock); | |
3043 | } | |
3044 | ||
3045 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3046 | { | |
3047 | /* | |
3048 | * "Normal" inline submissions always hold the uring_lock, since we | |
3049 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3050 | * The only exception is when we've detached the request and issue it | |
3051 | * from an async worker thread, grab the lock for that case. | |
3052 | */ | |
3053 | if (needs_lock) | |
3054 | mutex_lock(&ctx->uring_lock); | |
3055 | } | |
3056 | ||
3057 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3058 | int bgid, struct io_buffer *kbuf, | |
3059 | bool needs_lock) | |
3060 | { | |
3061 | struct io_buffer *head; | |
3062 | ||
3063 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3064 | return kbuf; | |
3065 | ||
3066 | io_ring_submit_lock(req->ctx, needs_lock); | |
3067 | ||
3068 | lockdep_assert_held(&req->ctx->uring_lock); | |
3069 | ||
9e15c3a0 | 3070 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
3071 | if (head) { |
3072 | if (!list_empty(&head->list)) { | |
3073 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3074 | list); | |
3075 | list_del(&kbuf->list); | |
3076 | } else { | |
3077 | kbuf = head; | |
9e15c3a0 | 3078 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
3079 | } |
3080 | if (*len > kbuf->len) | |
3081 | *len = kbuf->len; | |
3082 | } else { | |
3083 | kbuf = ERR_PTR(-ENOBUFS); | |
3084 | } | |
3085 | ||
3086 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3087 | ||
3088 | return kbuf; | |
3089 | } | |
3090 | ||
4d954c25 JA |
3091 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3092 | bool needs_lock) | |
3093 | { | |
3094 | struct io_buffer *kbuf; | |
4f4eeba8 | 3095 | u16 bgid; |
4d954c25 JA |
3096 | |
3097 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3098 | bgid = req->buf_index; |
4d954c25 JA |
3099 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3100 | if (IS_ERR(kbuf)) | |
3101 | return kbuf; | |
3102 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3103 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3104 | return u64_to_user_ptr(kbuf->addr); | |
3105 | } | |
3106 | ||
3107 | #ifdef CONFIG_COMPAT | |
3108 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3109 | bool needs_lock) | |
3110 | { | |
3111 | struct compat_iovec __user *uiov; | |
3112 | compat_ssize_t clen; | |
3113 | void __user *buf; | |
3114 | ssize_t len; | |
3115 | ||
3116 | uiov = u64_to_user_ptr(req->rw.addr); | |
3117 | if (!access_ok(uiov, sizeof(*uiov))) | |
3118 | return -EFAULT; | |
3119 | if (__get_user(clen, &uiov->iov_len)) | |
3120 | return -EFAULT; | |
3121 | if (clen < 0) | |
3122 | return -EINVAL; | |
3123 | ||
3124 | len = clen; | |
3125 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3126 | if (IS_ERR(buf)) | |
3127 | return PTR_ERR(buf); | |
3128 | iov[0].iov_base = buf; | |
3129 | iov[0].iov_len = (compat_size_t) len; | |
3130 | return 0; | |
3131 | } | |
3132 | #endif | |
3133 | ||
3134 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3135 | bool needs_lock) | |
3136 | { | |
3137 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3138 | void __user *buf; | |
3139 | ssize_t len; | |
3140 | ||
3141 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3142 | return -EFAULT; | |
3143 | ||
3144 | len = iov[0].iov_len; | |
3145 | if (len < 0) | |
3146 | return -EINVAL; | |
3147 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3148 | if (IS_ERR(buf)) | |
3149 | return PTR_ERR(buf); | |
3150 | iov[0].iov_base = buf; | |
3151 | iov[0].iov_len = len; | |
3152 | return 0; | |
3153 | } | |
3154 | ||
3155 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3156 | bool needs_lock) | |
3157 | { | |
dddb3e26 JA |
3158 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3159 | struct io_buffer *kbuf; | |
3160 | ||
3161 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3162 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3163 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3164 | return 0; |
dddb3e26 | 3165 | } |
dd201662 | 3166 | if (req->rw.len != 1) |
4d954c25 JA |
3167 | return -EINVAL; |
3168 | ||
3169 | #ifdef CONFIG_COMPAT | |
3170 | if (req->ctx->compat) | |
3171 | return io_compat_import(req, iov, needs_lock); | |
3172 | #endif | |
3173 | ||
3174 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3175 | } | |
3176 | ||
847595de PB |
3177 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3178 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3179 | { |
9adbd45d JA |
3180 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3181 | size_t sqe_len = req->rw.len; | |
847595de | 3182 | u8 opcode = req->opcode; |
4d954c25 | 3183 | ssize_t ret; |
edafccee | 3184 | |
7d009165 | 3185 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3186 | *iovec = NULL; |
9adbd45d | 3187 | return io_import_fixed(req, rw, iter); |
edafccee | 3188 | } |
2b188cc1 | 3189 | |
bcda7baa | 3190 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3191 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3192 | return -EINVAL; |
3193 | ||
3a6820f2 | 3194 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3195 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3196 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3197 | if (IS_ERR(buf)) |
4d954c25 | 3198 | return PTR_ERR(buf); |
3f9d6441 | 3199 | req->rw.len = sqe_len; |
bcda7baa JA |
3200 | } |
3201 | ||
3a6820f2 JA |
3202 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3203 | *iovec = NULL; | |
10fc72e4 | 3204 | return ret; |
3a6820f2 JA |
3205 | } |
3206 | ||
4d954c25 JA |
3207 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3208 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3209 | if (!ret) |
3210 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3211 | *iovec = NULL; |
3212 | return ret; | |
3213 | } | |
3214 | ||
89cd35c5 CH |
3215 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3216 | req->ctx->compat); | |
2b188cc1 JA |
3217 | } |
3218 | ||
0fef9483 JA |
3219 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3220 | { | |
5b09e37e | 3221 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3222 | } |
3223 | ||
31b51510 | 3224 | /* |
32960613 JA |
3225 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3226 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3227 | */ |
4017eb91 | 3228 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3229 | { |
4017eb91 JA |
3230 | struct kiocb *kiocb = &req->rw.kiocb; |
3231 | struct file *file = req->file; | |
32960613 JA |
3232 | ssize_t ret = 0; |
3233 | ||
3234 | /* | |
3235 | * Don't support polled IO through this interface, and we can't | |
3236 | * support non-blocking either. For the latter, this just causes | |
3237 | * the kiocb to be handled from an async context. | |
3238 | */ | |
3239 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3240 | return -EOPNOTSUPP; | |
3241 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3242 | return -EAGAIN; | |
3243 | ||
3244 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3245 | struct iovec iovec; |
32960613 JA |
3246 | ssize_t nr; |
3247 | ||
311ae9e1 PB |
3248 | if (!iov_iter_is_bvec(iter)) { |
3249 | iovec = iov_iter_iovec(iter); | |
3250 | } else { | |
4017eb91 JA |
3251 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3252 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3253 | } |
3254 | ||
32960613 JA |
3255 | if (rw == READ) { |
3256 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3257 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3258 | } else { |
3259 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3260 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3261 | } |
3262 | ||
3263 | if (nr < 0) { | |
3264 | if (!ret) | |
3265 | ret = nr; | |
3266 | break; | |
3267 | } | |
16c8d2df JA |
3268 | if (!iov_iter_is_bvec(iter)) { |
3269 | iov_iter_advance(iter, nr); | |
3270 | } else { | |
3271 | req->rw.len -= nr; | |
3272 | req->rw.addr += nr; | |
3273 | } | |
32960613 JA |
3274 | ret += nr; |
3275 | if (nr != iovec.iov_len) | |
3276 | break; | |
32960613 JA |
3277 | } |
3278 | ||
3279 | return ret; | |
3280 | } | |
3281 | ||
ff6165b2 JA |
3282 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3283 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3284 | { |
e8c2bc1f | 3285 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3286 | |
ff6165b2 | 3287 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3288 | rw->free_iovec = iovec; |
227c0c96 | 3289 | rw->bytes_done = 0; |
ff6165b2 | 3290 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3291 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3292 | return; |
b64e3444 | 3293 | if (!iovec) { |
ff6165b2 JA |
3294 | unsigned iov_off = 0; |
3295 | ||
3296 | rw->iter.iov = rw->fast_iov; | |
3297 | if (iter->iov != fast_iov) { | |
3298 | iov_off = iter->iov - fast_iov; | |
3299 | rw->iter.iov += iov_off; | |
3300 | } | |
3301 | if (rw->fast_iov != fast_iov) | |
3302 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3303 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3304 | } else { |
3305 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3306 | } |
3307 | } | |
3308 | ||
6cb78689 | 3309 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3310 | { |
e8c2bc1f JA |
3311 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3312 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3313 | return req->async_data == NULL; | |
3d9932a8 XW |
3314 | } |
3315 | ||
ff6165b2 JA |
3316 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3317 | const struct iovec *fast_iov, | |
227c0c96 | 3318 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3319 | { |
26f0505a | 3320 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3321 | return 0; |
e8c2bc1f | 3322 | if (!req->async_data) { |
cd658695 JA |
3323 | struct io_async_rw *iorw; |
3324 | ||
6cb78689 | 3325 | if (io_alloc_async_data(req)) { |
6bf985dc | 3326 | kfree(iovec); |
5d204bcf | 3327 | return -ENOMEM; |
6bf985dc | 3328 | } |
b7bb4f7d | 3329 | |
ff6165b2 | 3330 | io_req_map_rw(req, iovec, fast_iov, iter); |
cd658695 JA |
3331 | iorw = req->async_data; |
3332 | /* we've copied and mapped the iter, ensure state is saved */ | |
3333 | iov_iter_save_state(&iorw->iter, &iorw->iter_state); | |
5d204bcf | 3334 | } |
b7bb4f7d | 3335 | return 0; |
f67676d1 JA |
3336 | } |
3337 | ||
73debe68 | 3338 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3339 | { |
e8c2bc1f | 3340 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3341 | struct iovec *iov = iorw->fast_iov; |
847595de | 3342 | int ret; |
c3e330a4 | 3343 | |
2846c481 | 3344 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3345 | if (unlikely(ret < 0)) |
3346 | return ret; | |
3347 | ||
ab0b196c PB |
3348 | iorw->bytes_done = 0; |
3349 | iorw->free_iovec = iov; | |
3350 | if (iov) | |
3351 | req->flags |= REQ_F_NEED_CLEANUP; | |
cd658695 | 3352 | iov_iter_save_state(&iorw->iter, &iorw->iter_state); |
c3e330a4 PB |
3353 | return 0; |
3354 | } | |
3355 | ||
73debe68 | 3356 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3357 | { |
3529d8c2 JA |
3358 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3359 | return -EBADF; | |
5d329e12 | 3360 | return io_prep_rw(req, sqe, READ); |
f67676d1 JA |
3361 | } |
3362 | ||
c1dd91d1 JA |
3363 | /* |
3364 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3365 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3366 | * This gets called when the page is unlocked, and we generally expect that to | |
3367 | * happen when the page IO is completed and the page is now uptodate. This will | |
3368 | * queue a task_work based retry of the operation, attempting to copy the data | |
3369 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3370 | * do a thread based blocking retry of the operation. That's the unexpected | |
3371 | * slow path. | |
3372 | */ | |
bcf5a063 JA |
3373 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3374 | int sync, void *arg) | |
3375 | { | |
3376 | struct wait_page_queue *wpq; | |
3377 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3378 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3379 | |
3380 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3381 | ||
cdc8fcb4 LT |
3382 | if (!wake_page_match(wpq, key)) |
3383 | return 0; | |
3384 | ||
c8d317aa | 3385 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 | 3386 | list_del_init(&wait->entry); |
921b9054 | 3387 | io_req_task_queue(req); |
bcf5a063 JA |
3388 | return 1; |
3389 | } | |
3390 | ||
c1dd91d1 JA |
3391 | /* |
3392 | * This controls whether a given IO request should be armed for async page | |
3393 | * based retry. If we return false here, the request is handed to the async | |
3394 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3395 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3396 | * will either succeed because the page is now uptodate and unlocked, or it | |
3397 | * will register a callback when the page is unlocked at IO completion. Through | |
3398 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3399 | * That retry will attempt the buffered read again. The retry will generally | |
3400 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3401 | * async worker threads for a blocking retry. | |
3402 | */ | |
227c0c96 | 3403 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3404 | { |
e8c2bc1f JA |
3405 | struct io_async_rw *rw = req->async_data; |
3406 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3407 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3408 | |
bcf5a063 JA |
3409 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3410 | if (req->flags & REQ_F_NOWAIT) | |
3411 | return false; | |
f67676d1 | 3412 | |
227c0c96 | 3413 | /* Only for buffered IO */ |
3b2a4439 | 3414 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3415 | return false; |
3b2a4439 | 3416 | |
bcf5a063 JA |
3417 | /* |
3418 | * just use poll if we can, and don't attempt if the fs doesn't | |
3419 | * support callback based unlocks | |
3420 | */ | |
3421 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3422 | return false; | |
f67676d1 | 3423 | |
3b2a4439 JA |
3424 | wait->wait.func = io_async_buf_func; |
3425 | wait->wait.private = req; | |
3426 | wait->wait.flags = 0; | |
3427 | INIT_LIST_HEAD(&wait->wait.entry); | |
3428 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3429 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3430 | kiocb->ki_waitq = wait; |
3b2a4439 | 3431 | return true; |
bcf5a063 JA |
3432 | } |
3433 | ||
aeab9506 | 3434 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3435 | { |
3436 | if (req->file->f_op->read_iter) | |
3437 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3438 | else if (req->file->f_op->read) |
4017eb91 | 3439 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3440 | else |
3441 | return -EINVAL; | |
f67676d1 JA |
3442 | } |
3443 | ||
7db30437 ML |
3444 | static bool need_read_all(struct io_kiocb *req) |
3445 | { | |
3446 | return req->flags & REQ_F_ISREG || | |
3447 | S_ISBLK(file_inode(req->file)->i_mode); | |
3448 | } | |
3449 | ||
889fca73 | 3450 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3451 | { |
3452 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3453 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3454 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3455 | struct io_async_rw *rw = req->async_data; |
45d189c6 | 3456 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
cd658695 JA |
3457 | struct iov_iter_state __state, *state; |
3458 | ssize_t ret, ret2; | |
ff6165b2 | 3459 | |
2846c481 | 3460 | if (rw) { |
e8c2bc1f | 3461 | iter = &rw->iter; |
cd658695 JA |
3462 | state = &rw->iter_state; |
3463 | /* | |
3464 | * We come here from an earlier attempt, restore our state to | |
3465 | * match in case it doesn't. It's cheap enough that we don't | |
3466 | * need to make this conditional. | |
3467 | */ | |
3468 | iov_iter_restore(iter, state); | |
2846c481 PB |
3469 | iovec = NULL; |
3470 | } else { | |
3471 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3472 | if (ret < 0) | |
3473 | return ret; | |
cd658695 JA |
3474 | state = &__state; |
3475 | iov_iter_save_state(iter, state); | |
2846c481 | 3476 | } |
cd658695 | 3477 | req->result = iov_iter_count(iter); |
2b188cc1 | 3478 | |
fd6c2e4c JA |
3479 | /* Ensure we clear previously set non-block flag */ |
3480 | if (!force_nonblock) | |
29de5f6a | 3481 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3482 | else |
3483 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3484 | ||
24c74678 | 3485 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3486 | if (force_nonblock && !io_file_supports_nowait(req, READ)) { |
6713e7a6 | 3487 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3488 | return ret ?: -EAGAIN; |
6713e7a6 | 3489 | } |
9e645e11 | 3490 | |
cd658695 | 3491 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), req->result); |
5ea5dd45 PB |
3492 | if (unlikely(ret)) { |
3493 | kfree(iovec); | |
3494 | return ret; | |
3495 | } | |
2b188cc1 | 3496 | |
227c0c96 | 3497 | ret = io_iter_do_read(req, iter); |
32960613 | 3498 | |
230d50d4 | 3499 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3500 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3501 | /* IOPOLL retry should happen for io-wq threads */ |
3502 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3503 | goto done; |
75c668cd PB |
3504 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3505 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3506 | goto done; |
f38c7e3a | 3507 | ret = 0; |
230d50d4 JA |
3508 | } else if (ret == -EIOCBQUEUED) { |
3509 | goto out_free; | |
cd658695 | 3510 | } else if (ret <= 0 || ret == req->result || !force_nonblock || |
7db30437 | 3511 | (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { |
7335e3bf | 3512 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3513 | goto done; |
227c0c96 JA |
3514 | } |
3515 | ||
cd658695 JA |
3516 | /* |
3517 | * Don't depend on the iter state matching what was consumed, or being | |
3518 | * untouched in case of error. Restore it and we'll advance it | |
3519 | * manually if we need to. | |
3520 | */ | |
3521 | iov_iter_restore(iter, state); | |
3522 | ||
227c0c96 | 3523 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3524 | if (ret2) |
3525 | return ret2; | |
3526 | ||
fe1cdd55 | 3527 | iovec = NULL; |
e8c2bc1f | 3528 | rw = req->async_data; |
cd658695 JA |
3529 | /* |
3530 | * Now use our persistent iterator and state, if we aren't already. | |
3531 | * We've restored and mapped the iter to match. | |
3532 | */ | |
3533 | if (iter != &rw->iter) { | |
3534 | iter = &rw->iter; | |
3535 | state = &rw->iter_state; | |
3536 | } | |
227c0c96 | 3537 | |
b23df91b | 3538 | do { |
cd658695 JA |
3539 | /* |
3540 | * We end up here because of a partial read, either from | |
3541 | * above or inside this loop. Advance the iter by the bytes | |
3542 | * that were consumed. | |
3543 | */ | |
3544 | iov_iter_advance(iter, ret); | |
3545 | if (!iov_iter_count(iter)) | |
3546 | break; | |
b23df91b | 3547 | rw->bytes_done += ret; |
cd658695 JA |
3548 | iov_iter_save_state(iter, state); |
3549 | ||
b23df91b PB |
3550 | /* if we can retry, do so with the callbacks armed */ |
3551 | if (!io_rw_should_retry(req)) { | |
3552 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3553 | return -EAGAIN; | |
3554 | } | |
3555 | ||
3556 | /* | |
3557 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3558 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3559 | * desired page gets unlocked. We can also get a partial read | |
3560 | * here, and if we do, then just retry at the new offset. | |
3561 | */ | |
3562 | ret = io_iter_do_read(req, iter); | |
3563 | if (ret == -EIOCBQUEUED) | |
3564 | return 0; | |
227c0c96 | 3565 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3566 | kiocb->ki_flags &= ~IOCB_WAITQ; |
cd658695 JA |
3567 | iov_iter_restore(iter, state); |
3568 | } while (ret > 0); | |
227c0c96 | 3569 | done: |
889fca73 | 3570 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3571 | out_free: |
3572 | /* it's faster to check here then delegate to kfree */ | |
3573 | if (iovec) | |
3574 | kfree(iovec); | |
5ea5dd45 | 3575 | return 0; |
2b188cc1 JA |
3576 | } |
3577 | ||
73debe68 | 3578 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3579 | { |
3529d8c2 JA |
3580 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3581 | return -EBADF; | |
5d329e12 | 3582 | return io_prep_rw(req, sqe, WRITE); |
f67676d1 JA |
3583 | } |
3584 | ||
889fca73 | 3585 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3586 | { |
3587 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3588 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3589 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3590 | struct io_async_rw *rw = req->async_data; |
45d189c6 | 3591 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
cd658695 JA |
3592 | struct iov_iter_state __state, *state; |
3593 | ssize_t ret, ret2; | |
2b188cc1 | 3594 | |
2846c481 | 3595 | if (rw) { |
e8c2bc1f | 3596 | iter = &rw->iter; |
cd658695 JA |
3597 | state = &rw->iter_state; |
3598 | iov_iter_restore(iter, state); | |
2846c481 PB |
3599 | iovec = NULL; |
3600 | } else { | |
3601 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3602 | if (ret < 0) | |
3603 | return ret; | |
cd658695 JA |
3604 | state = &__state; |
3605 | iov_iter_save_state(iter, state); | |
2846c481 | 3606 | } |
cd658695 JA |
3607 | req->result = iov_iter_count(iter); |
3608 | ret2 = 0; | |
2b188cc1 | 3609 | |
fd6c2e4c JA |
3610 | /* Ensure we clear previously set non-block flag */ |
3611 | if (!force_nonblock) | |
a88fc400 PB |
3612 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3613 | else | |
3614 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3615 | |
24c74678 | 3616 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3617 | if (force_nonblock && !io_file_supports_nowait(req, WRITE)) |
f67676d1 | 3618 | goto copy_iov; |
31b51510 | 3619 | |
10d59345 JA |
3620 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3621 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3622 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3623 | goto copy_iov; |
31b51510 | 3624 | |
cd658695 | 3625 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), req->result); |
fa15bafb PB |
3626 | if (unlikely(ret)) |
3627 | goto out_free; | |
4ed734b0 | 3628 | |
fa15bafb PB |
3629 | /* |
3630 | * Open-code file_start_write here to grab freeze protection, | |
3631 | * which will be released by another thread in | |
3632 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3633 | * released so that it doesn't complain about the held lock when | |
3634 | * we return to userspace. | |
3635 | */ | |
3636 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3637 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3638 | __sb_writers_release(file_inode(req->file)->i_sb, |
3639 | SB_FREEZE_WRITE); | |
3640 | } | |
3641 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3642 | |
fa15bafb | 3643 | if (req->file->f_op->write_iter) |
ff6165b2 | 3644 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3645 | else if (req->file->f_op->write) |
4017eb91 | 3646 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3647 | else |
3648 | ret2 = -EINVAL; | |
4ed734b0 | 3649 | |
6ad7f233 PB |
3650 | if (req->flags & REQ_F_REISSUE) { |
3651 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3652 | ret2 = -EAGAIN; |
6ad7f233 | 3653 | } |
230d50d4 | 3654 | |
fa15bafb PB |
3655 | /* |
3656 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3657 | * retry them without IOCB_NOWAIT. | |
3658 | */ | |
3659 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3660 | ret2 = -EAGAIN; | |
75c668cd PB |
3661 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3662 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3663 | goto done; |
fa15bafb | 3664 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3665 | /* IOPOLL retry should happen for io-wq threads */ |
3666 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3667 | goto copy_iov; | |
355afaeb | 3668 | done: |
889fca73 | 3669 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3670 | } else { |
f67676d1 | 3671 | copy_iov: |
cd658695 JA |
3672 | iov_iter_restore(iter, state); |
3673 | if (ret2 > 0) | |
3674 | iov_iter_advance(iter, ret2); | |
227c0c96 | 3675 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3676 | return ret ?: -EAGAIN; |
2b188cc1 | 3677 | } |
31b51510 | 3678 | out_free: |
f261c168 | 3679 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3680 | if (iovec) |
6f2cc166 | 3681 | kfree(iovec); |
2b188cc1 JA |
3682 | return ret; |
3683 | } | |
3684 | ||
80a261fd JA |
3685 | static int io_renameat_prep(struct io_kiocb *req, |
3686 | const struct io_uring_sqe *sqe) | |
3687 | { | |
3688 | struct io_rename *ren = &req->rename; | |
3689 | const char __user *oldf, *newf; | |
3690 | ||
ed7eb259 JA |
3691 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3692 | return -EINVAL; | |
26578cda | 3693 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
ed7eb259 | 3694 | return -EINVAL; |
80a261fd JA |
3695 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3696 | return -EBADF; | |
3697 | ||
3698 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3699 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3700 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3701 | ren->new_dfd = READ_ONCE(sqe->len); | |
3702 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3703 | ||
3704 | ren->oldpath = getname(oldf); | |
3705 | if (IS_ERR(ren->oldpath)) | |
3706 | return PTR_ERR(ren->oldpath); | |
3707 | ||
3708 | ren->newpath = getname(newf); | |
3709 | if (IS_ERR(ren->newpath)) { | |
3710 | putname(ren->oldpath); | |
3711 | return PTR_ERR(ren->newpath); | |
3712 | } | |
3713 | ||
3714 | req->flags |= REQ_F_NEED_CLEANUP; | |
3715 | return 0; | |
3716 | } | |
3717 | ||
45d189c6 | 3718 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3719 | { |
3720 | struct io_rename *ren = &req->rename; | |
3721 | int ret; | |
3722 | ||
45d189c6 | 3723 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3724 | return -EAGAIN; |
3725 | ||
3726 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3727 | ren->newpath, ren->flags); | |
3728 | ||
3729 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3730 | if (ret < 0) | |
93d2bcd2 | 3731 | req_set_fail(req); |
80a261fd JA |
3732 | io_req_complete(req, ret); |
3733 | return 0; | |
3734 | } | |
3735 | ||
14a1143b JA |
3736 | static int io_unlinkat_prep(struct io_kiocb *req, |
3737 | const struct io_uring_sqe *sqe) | |
3738 | { | |
3739 | struct io_unlink *un = &req->unlink; | |
3740 | const char __user *fname; | |
3741 | ||
22634bc5 JA |
3742 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3743 | return -EINVAL; | |
26578cda PB |
3744 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || |
3745 | sqe->splice_fd_in) | |
22634bc5 | 3746 | return -EINVAL; |
14a1143b JA |
3747 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3748 | return -EBADF; | |
3749 | ||
3750 | un->dfd = READ_ONCE(sqe->fd); | |
3751 | ||
3752 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3753 | if (un->flags & ~AT_REMOVEDIR) | |
3754 | return -EINVAL; | |
3755 | ||
3756 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3757 | un->filename = getname(fname); | |
3758 | if (IS_ERR(un->filename)) | |
3759 | return PTR_ERR(un->filename); | |
3760 | ||
3761 | req->flags |= REQ_F_NEED_CLEANUP; | |
3762 | return 0; | |
3763 | } | |
3764 | ||
45d189c6 | 3765 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3766 | { |
3767 | struct io_unlink *un = &req->unlink; | |
3768 | int ret; | |
3769 | ||
45d189c6 | 3770 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3771 | return -EAGAIN; |
3772 | ||
3773 | if (un->flags & AT_REMOVEDIR) | |
3774 | ret = do_rmdir(un->dfd, un->filename); | |
3775 | else | |
3776 | ret = do_unlinkat(un->dfd, un->filename); | |
3777 | ||
3778 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3779 | if (ret < 0) | |
93d2bcd2 | 3780 | req_set_fail(req); |
14a1143b JA |
3781 | io_req_complete(req, ret); |
3782 | return 0; | |
3783 | } | |
3784 | ||
e34a02dc DK |
3785 | static int io_mkdirat_prep(struct io_kiocb *req, |
3786 | const struct io_uring_sqe *sqe) | |
3787 | { | |
3788 | struct io_mkdir *mkd = &req->mkdir; | |
3789 | const char __user *fname; | |
3790 | ||
3791 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3792 | return -EINVAL; | |
3793 | if (sqe->ioprio || sqe->off || sqe->rw_flags || sqe->buf_index || | |
3794 | sqe->splice_fd_in) | |
3795 | return -EINVAL; | |
3796 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3797 | return -EBADF; | |
3798 | ||
3799 | mkd->dfd = READ_ONCE(sqe->fd); | |
3800 | mkd->mode = READ_ONCE(sqe->len); | |
3801 | ||
3802 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3803 | mkd->filename = getname(fname); | |
3804 | if (IS_ERR(mkd->filename)) | |
3805 | return PTR_ERR(mkd->filename); | |
3806 | ||
3807 | req->flags |= REQ_F_NEED_CLEANUP; | |
3808 | return 0; | |
3809 | } | |
3810 | ||
3811 | static int io_mkdirat(struct io_kiocb *req, int issue_flags) | |
3812 | { | |
3813 | struct io_mkdir *mkd = &req->mkdir; | |
3814 | int ret; | |
3815 | ||
3816 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3817 | return -EAGAIN; | |
3818 | ||
3819 | ret = do_mkdirat(mkd->dfd, mkd->filename, mkd->mode); | |
3820 | ||
3821 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3822 | if (ret < 0) | |
3823 | req_set_fail(req); | |
3824 | io_req_complete(req, ret); | |
3825 | return 0; | |
3826 | } | |
3827 | ||
7a8721f8 DK |
3828 | static int io_symlinkat_prep(struct io_kiocb *req, |
3829 | const struct io_uring_sqe *sqe) | |
3830 | { | |
3831 | struct io_symlink *sl = &req->symlink; | |
3832 | const char __user *oldpath, *newpath; | |
3833 | ||
3834 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3835 | return -EINVAL; | |
3836 | if (sqe->ioprio || sqe->len || sqe->rw_flags || sqe->buf_index || | |
3837 | sqe->splice_fd_in) | |
3838 | return -EINVAL; | |
3839 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3840 | return -EBADF; | |
3841 | ||
3842 | sl->new_dfd = READ_ONCE(sqe->fd); | |
3843 | oldpath = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3844 | newpath = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3845 | ||
3846 | sl->oldpath = getname(oldpath); | |
3847 | if (IS_ERR(sl->oldpath)) | |
3848 | return PTR_ERR(sl->oldpath); | |
3849 | ||
3850 | sl->newpath = getname(newpath); | |
3851 | if (IS_ERR(sl->newpath)) { | |
3852 | putname(sl->oldpath); | |
3853 | return PTR_ERR(sl->newpath); | |
3854 | } | |
3855 | ||
3856 | req->flags |= REQ_F_NEED_CLEANUP; | |
3857 | return 0; | |
3858 | } | |
3859 | ||
3860 | static int io_symlinkat(struct io_kiocb *req, int issue_flags) | |
3861 | { | |
3862 | struct io_symlink *sl = &req->symlink; | |
3863 | int ret; | |
3864 | ||
3865 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3866 | return -EAGAIN; | |
3867 | ||
3868 | ret = do_symlinkat(sl->oldpath, sl->new_dfd, sl->newpath); | |
3869 | ||
3870 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3871 | if (ret < 0) | |
3872 | req_set_fail(req); | |
3873 | io_req_complete(req, ret); | |
3874 | return 0; | |
3875 | } | |
3876 | ||
cf30da90 DK |
3877 | static int io_linkat_prep(struct io_kiocb *req, |
3878 | const struct io_uring_sqe *sqe) | |
3879 | { | |
3880 | struct io_hardlink *lnk = &req->hardlink; | |
3881 | const char __user *oldf, *newf; | |
3882 | ||
3883 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3884 | return -EINVAL; | |
3885 | if (sqe->ioprio || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) | |
3886 | return -EINVAL; | |
3887 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3888 | return -EBADF; | |
3889 | ||
3890 | lnk->old_dfd = READ_ONCE(sqe->fd); | |
3891 | lnk->new_dfd = READ_ONCE(sqe->len); | |
3892 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3893 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3894 | lnk->flags = READ_ONCE(sqe->hardlink_flags); | |
3895 | ||
3896 | lnk->oldpath = getname(oldf); | |
3897 | if (IS_ERR(lnk->oldpath)) | |
3898 | return PTR_ERR(lnk->oldpath); | |
3899 | ||
3900 | lnk->newpath = getname(newf); | |
3901 | if (IS_ERR(lnk->newpath)) { | |
3902 | putname(lnk->oldpath); | |
3903 | return PTR_ERR(lnk->newpath); | |
3904 | } | |
3905 | ||
3906 | req->flags |= REQ_F_NEED_CLEANUP; | |
3907 | return 0; | |
3908 | } | |
3909 | ||
3910 | static int io_linkat(struct io_kiocb *req, int issue_flags) | |
3911 | { | |
3912 | struct io_hardlink *lnk = &req->hardlink; | |
3913 | int ret; | |
3914 | ||
3915 | if (issue_flags & IO_URING_F_NONBLOCK) | |
3916 | return -EAGAIN; | |
3917 | ||
3918 | ret = do_linkat(lnk->old_dfd, lnk->oldpath, lnk->new_dfd, | |
3919 | lnk->newpath, lnk->flags); | |
3920 | ||
3921 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3922 | if (ret < 0) | |
3923 | req_set_fail(req); | |
3924 | io_req_complete(req, ret); | |
3925 | return 0; | |
3926 | } | |
3927 | ||
36f4fa68 JA |
3928 | static int io_shutdown_prep(struct io_kiocb *req, |
3929 | const struct io_uring_sqe *sqe) | |
3930 | { | |
3931 | #if defined(CONFIG_NET) | |
3932 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3933 | return -EINVAL; | |
26578cda PB |
3934 | if (unlikely(sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || |
3935 | sqe->buf_index || sqe->splice_fd_in)) | |
36f4fa68 JA |
3936 | return -EINVAL; |
3937 | ||
3938 | req->shutdown.how = READ_ONCE(sqe->len); | |
3939 | return 0; | |
3940 | #else | |
3941 | return -EOPNOTSUPP; | |
3942 | #endif | |
3943 | } | |
3944 | ||
45d189c6 | 3945 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3946 | { |
3947 | #if defined(CONFIG_NET) | |
3948 | struct socket *sock; | |
3949 | int ret; | |
3950 | ||
45d189c6 | 3951 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3952 | return -EAGAIN; |
3953 | ||
48aba79b | 3954 | sock = sock_from_file(req->file); |
36f4fa68 | 3955 | if (unlikely(!sock)) |
48aba79b | 3956 | return -ENOTSOCK; |
36f4fa68 JA |
3957 | |
3958 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3959 | if (ret < 0) |
93d2bcd2 | 3960 | req_set_fail(req); |
36f4fa68 JA |
3961 | io_req_complete(req, ret); |
3962 | return 0; | |
3963 | #else | |
3964 | return -EOPNOTSUPP; | |
3965 | #endif | |
3966 | } | |
3967 | ||
f2a8d5c7 PB |
3968 | static int __io_splice_prep(struct io_kiocb *req, |
3969 | const struct io_uring_sqe *sqe) | |
7d67af2c | 3970 | { |
fe7e3257 | 3971 | struct io_splice *sp = &req->splice; |
7d67af2c | 3972 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 3973 | |
3232dd02 PB |
3974 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3975 | return -EINVAL; | |
7d67af2c PB |
3976 | |
3977 | sp->file_in = NULL; | |
7d67af2c PB |
3978 | sp->len = READ_ONCE(sqe->len); |
3979 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3980 | ||
3981 | if (unlikely(sp->flags & ~valid_flags)) | |
3982 | return -EINVAL; | |
3983 | ||
62906e89 | 3984 | sp->file_in = io_file_get(req->ctx, req, READ_ONCE(sqe->splice_fd_in), |
8371adf5 PB |
3985 | (sp->flags & SPLICE_F_FD_IN_FIXED)); |
3986 | if (!sp->file_in) | |
3987 | return -EBADF; | |
7d67af2c | 3988 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3989 | return 0; |
3990 | } | |
3991 | ||
f2a8d5c7 PB |
3992 | static int io_tee_prep(struct io_kiocb *req, |
3993 | const struct io_uring_sqe *sqe) | |
3994 | { | |
3995 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3996 | return -EINVAL; | |
3997 | return __io_splice_prep(req, sqe); | |
3998 | } | |
3999 | ||
45d189c6 | 4000 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
4001 | { |
4002 | struct io_splice *sp = &req->splice; | |
4003 | struct file *in = sp->file_in; | |
4004 | struct file *out = sp->file_out; | |
4005 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
4006 | long ret = 0; | |
4007 | ||
45d189c6 | 4008 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
4009 | return -EAGAIN; |
4010 | if (sp->len) | |
4011 | ret = do_tee(in, out, sp->len, flags); | |
4012 | ||
e1d767f0 PB |
4013 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4014 | io_put_file(in); | |
f2a8d5c7 PB |
4015 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4016 | ||
f2a8d5c7 | 4017 | if (ret != sp->len) |
93d2bcd2 | 4018 | req_set_fail(req); |
e1e16097 | 4019 | io_req_complete(req, ret); |
f2a8d5c7 PB |
4020 | return 0; |
4021 | } | |
4022 | ||
4023 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4024 | { | |
fe7e3257 | 4025 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
4026 | |
4027 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
4028 | sp->off_out = READ_ONCE(sqe->off); | |
4029 | return __io_splice_prep(req, sqe); | |
4030 | } | |
4031 | ||
45d189c6 | 4032 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
4033 | { |
4034 | struct io_splice *sp = &req->splice; | |
4035 | struct file *in = sp->file_in; | |
4036 | struct file *out = sp->file_out; | |
4037 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
4038 | loff_t *poff_in, *poff_out; | |
c9687426 | 4039 | long ret = 0; |
7d67af2c | 4040 | |
45d189c6 | 4041 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 4042 | return -EAGAIN; |
7d67af2c PB |
4043 | |
4044 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
4045 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 4046 | |
948a7749 | 4047 | if (sp->len) |
c9687426 | 4048 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 4049 | |
e1d767f0 PB |
4050 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4051 | io_put_file(in); | |
7d67af2c PB |
4052 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4053 | ||
7d67af2c | 4054 | if (ret != sp->len) |
93d2bcd2 | 4055 | req_set_fail(req); |
e1e16097 | 4056 | io_req_complete(req, ret); |
7d67af2c PB |
4057 | return 0; |
4058 | } | |
4059 | ||
2b188cc1 JA |
4060 | /* |
4061 | * IORING_OP_NOP just posts a completion event, nothing else. | |
4062 | */ | |
889fca73 | 4063 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
4064 | { |
4065 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4066 | |
def596e9 JA |
4067 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4068 | return -EINVAL; | |
4069 | ||
889fca73 | 4070 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4071 | return 0; |
4072 | } | |
4073 | ||
1155c76a | 4074 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4075 | { |
6b06314c | 4076 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4077 | |
09bb8394 JA |
4078 | if (!req->file) |
4079 | return -EBADF; | |
c992fe29 | 4080 | |
6b06314c | 4081 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4082 | return -EINVAL; |
26578cda PB |
4083 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4084 | sqe->splice_fd_in)) | |
c992fe29 CH |
4085 | return -EINVAL; |
4086 | ||
8ed8d3c3 JA |
4087 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4088 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4089 | return -EINVAL; | |
4090 | ||
4091 | req->sync.off = READ_ONCE(sqe->off); | |
4092 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4093 | return 0; |
4094 | } | |
4095 | ||
45d189c6 | 4096 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4097 | { |
8ed8d3c3 | 4098 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4099 | int ret; |
4100 | ||
ac45abc0 | 4101 | /* fsync always requires a blocking context */ |
45d189c6 | 4102 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4103 | return -EAGAIN; |
4104 | ||
9adbd45d | 4105 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4106 | end > 0 ? end : LLONG_MAX, |
4107 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4108 | if (ret < 0) | |
93d2bcd2 | 4109 | req_set_fail(req); |
e1e16097 | 4110 | io_req_complete(req, ret); |
c992fe29 CH |
4111 | return 0; |
4112 | } | |
4113 | ||
d63d1b5e JA |
4114 | static int io_fallocate_prep(struct io_kiocb *req, |
4115 | const struct io_uring_sqe *sqe) | |
4116 | { | |
26578cda PB |
4117 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags || |
4118 | sqe->splice_fd_in) | |
d63d1b5e | 4119 | return -EINVAL; |
3232dd02 PB |
4120 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4121 | return -EINVAL; | |
d63d1b5e JA |
4122 | |
4123 | req->sync.off = READ_ONCE(sqe->off); | |
4124 | req->sync.len = READ_ONCE(sqe->addr); | |
4125 | req->sync.mode = READ_ONCE(sqe->len); | |
4126 | return 0; | |
4127 | } | |
4128 | ||
45d189c6 | 4129 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4130 | { |
ac45abc0 PB |
4131 | int ret; |
4132 | ||
d63d1b5e | 4133 | /* fallocate always requiring blocking context */ |
45d189c6 | 4134 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4135 | return -EAGAIN; |
ac45abc0 PB |
4136 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4137 | req->sync.len); | |
ac45abc0 | 4138 | if (ret < 0) |
93d2bcd2 | 4139 | req_set_fail(req); |
e1e16097 | 4140 | io_req_complete(req, ret); |
5d17b4a4 JA |
4141 | return 0; |
4142 | } | |
4143 | ||
ec65fea5 | 4144 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4145 | { |
f8748881 | 4146 | const char __user *fname; |
15b71abe | 4147 | int ret; |
b7bb4f7d | 4148 | |
d3fddf6d PB |
4149 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4150 | return -EINVAL; | |
b9445598 | 4151 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4152 | return -EINVAL; |
ec65fea5 | 4153 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4154 | return -EBADF; |
03b1230c | 4155 | |
ec65fea5 PB |
4156 | /* open.how should be already initialised */ |
4157 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4158 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4159 | |
25e72d10 PB |
4160 | req->open.dfd = READ_ONCE(sqe->fd); |
4161 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4162 | req->open.filename = getname(fname); |
15b71abe JA |
4163 | if (IS_ERR(req->open.filename)) { |
4164 | ret = PTR_ERR(req->open.filename); | |
4165 | req->open.filename = NULL; | |
4166 | return ret; | |
4167 | } | |
b9445598 PB |
4168 | |
4169 | req->open.file_slot = READ_ONCE(sqe->file_index); | |
4170 | if (req->open.file_slot && (req->open.how.flags & O_CLOEXEC)) | |
4171 | return -EINVAL; | |
4172 | ||
4022e7af | 4173 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4174 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4175 | return 0; |
03b1230c JA |
4176 | } |
4177 | ||
ec65fea5 PB |
4178 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4179 | { | |
d3fddf6d PB |
4180 | u64 mode = READ_ONCE(sqe->len); |
4181 | u64 flags = READ_ONCE(sqe->open_flags); | |
ec65fea5 | 4182 | |
ec65fea5 PB |
4183 | req->open.how = build_open_how(flags, mode); |
4184 | return __io_openat_prep(req, sqe); | |
4185 | } | |
4186 | ||
cebdb986 | 4187 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4188 | { |
cebdb986 | 4189 | struct open_how __user *how; |
cebdb986 | 4190 | size_t len; |
0fa03c62 JA |
4191 | int ret; |
4192 | ||
cebdb986 JA |
4193 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4194 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4195 | if (len < OPEN_HOW_SIZE_VER0) |
4196 | return -EINVAL; | |
3529d8c2 | 4197 | |
cebdb986 JA |
4198 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4199 | len); | |
4200 | if (ret) | |
4201 | return ret; | |
3529d8c2 | 4202 | |
ec65fea5 | 4203 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4204 | } |
4205 | ||
45d189c6 | 4206 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4207 | { |
4208 | struct open_flags op; | |
15b71abe | 4209 | struct file *file; |
b9445598 PB |
4210 | bool resolve_nonblock, nonblock_set; |
4211 | bool fixed = !!req->open.file_slot; | |
15b71abe JA |
4212 | int ret; |
4213 | ||
cebdb986 | 4214 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4215 | if (ret) |
4216 | goto err; | |
3a81fd02 JA |
4217 | nonblock_set = op.open_flag & O_NONBLOCK; |
4218 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4219 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4220 | /* |
4221 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4222 | * it'll always -EAGAIN | |
4223 | */ | |
4224 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4225 | return -EAGAIN; | |
4226 | op.lookup_flags |= LOOKUP_CACHED; | |
4227 | op.open_flag |= O_NONBLOCK; | |
4228 | } | |
15b71abe | 4229 | |
b9445598 PB |
4230 | if (!fixed) { |
4231 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); | |
4232 | if (ret < 0) | |
4233 | goto err; | |
4234 | } | |
15b71abe JA |
4235 | |
4236 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 4237 | if (IS_ERR(file)) { |
944d1444 | 4238 | /* |
12dcb58a PB |
4239 | * We could hang on to this 'fd' on retrying, but seems like |
4240 | * marginal gain for something that is now known to be a slower | |
4241 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 4242 | */ |
b9445598 PB |
4243 | if (!fixed) |
4244 | put_unused_fd(ret); | |
3a81fd02 | 4245 | |
15b71abe | 4246 | ret = PTR_ERR(file); |
12dcb58a PB |
4247 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
4248 | if (ret == -EAGAIN && | |
4249 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
4250 | return -EAGAIN; | |
4251 | goto err; | |
15b71abe | 4252 | } |
12dcb58a PB |
4253 | |
4254 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
4255 | file->f_flags &= ~O_NONBLOCK; | |
4256 | fsnotify_open(file); | |
b9445598 PB |
4257 | |
4258 | if (!fixed) | |
4259 | fd_install(ret, file); | |
4260 | else | |
4261 | ret = io_install_fixed_file(req, file, issue_flags, | |
4262 | req->open.file_slot - 1); | |
15b71abe JA |
4263 | err: |
4264 | putname(req->open.filename); | |
8fef80bf | 4265 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 4266 | if (ret < 0) |
93d2bcd2 | 4267 | req_set_fail(req); |
0bdf3398 | 4268 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
4269 | return 0; |
4270 | } | |
4271 | ||
45d189c6 | 4272 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4273 | { |
e45cff58 | 4274 | return io_openat2(req, issue_flags); |
cebdb986 JA |
4275 | } |
4276 | ||
067524e9 JA |
4277 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4278 | const struct io_uring_sqe *sqe) | |
4279 | { | |
4280 | struct io_provide_buf *p = &req->pbuf; | |
4281 | u64 tmp; | |
4282 | ||
26578cda PB |
4283 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off || |
4284 | sqe->splice_fd_in) | |
067524e9 JA |
4285 | return -EINVAL; |
4286 | ||
4287 | tmp = READ_ONCE(sqe->fd); | |
4288 | if (!tmp || tmp > USHRT_MAX) | |
4289 | return -EINVAL; | |
4290 | ||
4291 | memset(p, 0, sizeof(*p)); | |
4292 | p->nbufs = tmp; | |
4293 | p->bgid = READ_ONCE(sqe->buf_group); | |
4294 | return 0; | |
4295 | } | |
4296 | ||
4297 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4298 | int bgid, unsigned nbufs) | |
4299 | { | |
4300 | unsigned i = 0; | |
4301 | ||
4302 | /* shouldn't happen */ | |
4303 | if (!nbufs) | |
4304 | return 0; | |
4305 | ||
4306 | /* the head kbuf is the list itself */ | |
4307 | while (!list_empty(&buf->list)) { | |
4308 | struct io_buffer *nxt; | |
4309 | ||
4310 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4311 | list_del(&nxt->list); | |
4312 | kfree(nxt); | |
4313 | if (++i == nbufs) | |
4314 | return i; | |
4315 | } | |
4316 | i++; | |
4317 | kfree(buf); | |
9e15c3a0 | 4318 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
4319 | |
4320 | return i; | |
4321 | } | |
4322 | ||
889fca73 | 4323 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4324 | { |
4325 | struct io_provide_buf *p = &req->pbuf; | |
4326 | struct io_ring_ctx *ctx = req->ctx; | |
4327 | struct io_buffer *head; | |
4328 | int ret = 0; | |
45d189c6 | 4329 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4330 | |
4331 | io_ring_submit_lock(ctx, !force_nonblock); | |
4332 | ||
4333 | lockdep_assert_held(&ctx->uring_lock); | |
4334 | ||
4335 | ret = -ENOENT; | |
9e15c3a0 | 4336 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
4337 | if (head) |
4338 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 4339 | if (ret < 0) |
93d2bcd2 | 4340 | req_set_fail(req); |
067524e9 | 4341 | |
9fb8cb49 PB |
4342 | /* complete before unlock, IOPOLL may need the lock */ |
4343 | __io_req_complete(req, issue_flags, ret, 0); | |
4344 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
4345 | return 0; |
4346 | } | |
4347 | ||
ddf0322d JA |
4348 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4349 | const struct io_uring_sqe *sqe) | |
4350 | { | |
38134ada | 4351 | unsigned long size, tmp_check; |
ddf0322d JA |
4352 | struct io_provide_buf *p = &req->pbuf; |
4353 | u64 tmp; | |
4354 | ||
26578cda | 4355 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
ddf0322d JA |
4356 | return -EINVAL; |
4357 | ||
4358 | tmp = READ_ONCE(sqe->fd); | |
4359 | if (!tmp || tmp > USHRT_MAX) | |
4360 | return -E2BIG; | |
4361 | p->nbufs = tmp; | |
4362 | p->addr = READ_ONCE(sqe->addr); | |
4363 | p->len = READ_ONCE(sqe->len); | |
4364 | ||
38134ada PB |
4365 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
4366 | &size)) | |
4367 | return -EOVERFLOW; | |
4368 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
4369 | return -EOVERFLOW; | |
4370 | ||
d81269fe PB |
4371 | size = (unsigned long)p->len * p->nbufs; |
4372 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
4373 | return -EFAULT; |
4374 | ||
4375 | p->bgid = READ_ONCE(sqe->buf_group); | |
4376 | tmp = READ_ONCE(sqe->off); | |
4377 | if (tmp > USHRT_MAX) | |
4378 | return -E2BIG; | |
4379 | p->bid = tmp; | |
4380 | return 0; | |
4381 | } | |
4382 | ||
4383 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4384 | { | |
4385 | struct io_buffer *buf; | |
4386 | u64 addr = pbuf->addr; | |
4387 | int i, bid = pbuf->bid; | |
4388 | ||
4389 | for (i = 0; i < pbuf->nbufs; i++) { | |
9990da93 | 4390 | buf = kmalloc(sizeof(*buf), GFP_KERNEL_ACCOUNT); |
ddf0322d JA |
4391 | if (!buf) |
4392 | break; | |
4393 | ||
4394 | buf->addr = addr; | |
d1f82808 | 4395 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
4396 | buf->bid = bid; |
4397 | addr += pbuf->len; | |
4398 | bid++; | |
4399 | if (!*head) { | |
4400 | INIT_LIST_HEAD(&buf->list); | |
4401 | *head = buf; | |
4402 | } else { | |
4403 | list_add_tail(&buf->list, &(*head)->list); | |
4404 | } | |
4405 | } | |
4406 | ||
4407 | return i ? i : -ENOMEM; | |
4408 | } | |
4409 | ||
889fca73 | 4410 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4411 | { |
4412 | struct io_provide_buf *p = &req->pbuf; | |
4413 | struct io_ring_ctx *ctx = req->ctx; | |
4414 | struct io_buffer *head, *list; | |
4415 | int ret = 0; | |
45d189c6 | 4416 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4417 | |
4418 | io_ring_submit_lock(ctx, !force_nonblock); | |
4419 | ||
4420 | lockdep_assert_held(&ctx->uring_lock); | |
4421 | ||
9e15c3a0 | 4422 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4423 | |
4424 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4425 | if (ret >= 0 && !list) { |
4426 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4427 | if (ret < 0) | |
067524e9 | 4428 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4429 | } |
ddf0322d | 4430 | if (ret < 0) |
93d2bcd2 | 4431 | req_set_fail(req); |
9fb8cb49 PB |
4432 | /* complete before unlock, IOPOLL may need the lock */ |
4433 | __io_req_complete(req, issue_flags, ret, 0); | |
4434 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4435 | return 0; |
cebdb986 JA |
4436 | } |
4437 | ||
3e4827b0 JA |
4438 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4439 | const struct io_uring_sqe *sqe) | |
4440 | { | |
4441 | #if defined(CONFIG_EPOLL) | |
26578cda | 4442 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
3e4827b0 | 4443 | return -EINVAL; |
2d74d042 | 4444 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4445 | return -EINVAL; |
3e4827b0 JA |
4446 | |
4447 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4448 | req->epoll.op = READ_ONCE(sqe->len); | |
4449 | req->epoll.fd = READ_ONCE(sqe->off); | |
4450 | ||
4451 | if (ep_op_has_event(req->epoll.op)) { | |
4452 | struct epoll_event __user *ev; | |
4453 | ||
4454 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4455 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4456 | return -EFAULT; | |
4457 | } | |
4458 | ||
4459 | return 0; | |
4460 | #else | |
4461 | return -EOPNOTSUPP; | |
4462 | #endif | |
4463 | } | |
4464 | ||
889fca73 | 4465 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4466 | { |
4467 | #if defined(CONFIG_EPOLL) | |
4468 | struct io_epoll *ie = &req->epoll; | |
4469 | int ret; | |
45d189c6 | 4470 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4471 | |
4472 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4473 | if (force_nonblock && ret == -EAGAIN) | |
4474 | return -EAGAIN; | |
4475 | ||
4476 | if (ret < 0) | |
93d2bcd2 | 4477 | req_set_fail(req); |
889fca73 | 4478 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4479 | return 0; |
4480 | #else | |
4481 | return -EOPNOTSUPP; | |
4482 | #endif | |
4483 | } | |
4484 | ||
c1ca757b JA |
4485 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4486 | { | |
4487 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
26578cda | 4488 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->splice_fd_in) |
c1ca757b | 4489 | return -EINVAL; |
3232dd02 PB |
4490 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4491 | return -EINVAL; | |
c1ca757b JA |
4492 | |
4493 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4494 | req->madvise.len = READ_ONCE(sqe->len); | |
4495 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4496 | return 0; | |
4497 | #else | |
4498 | return -EOPNOTSUPP; | |
4499 | #endif | |
4500 | } | |
4501 | ||
45d189c6 | 4502 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4503 | { |
4504 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4505 | struct io_madvise *ma = &req->madvise; | |
4506 | int ret; | |
4507 | ||
45d189c6 | 4508 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4509 | return -EAGAIN; |
4510 | ||
0726b01e | 4511 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4512 | if (ret < 0) |
93d2bcd2 | 4513 | req_set_fail(req); |
e1e16097 | 4514 | io_req_complete(req, ret); |
c1ca757b JA |
4515 | return 0; |
4516 | #else | |
4517 | return -EOPNOTSUPP; | |
4518 | #endif | |
4519 | } | |
4520 | ||
4840e418 JA |
4521 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4522 | { | |
26578cda | 4523 | if (sqe->ioprio || sqe->buf_index || sqe->addr || sqe->splice_fd_in) |
4840e418 | 4524 | return -EINVAL; |
3232dd02 PB |
4525 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4526 | return -EINVAL; | |
4840e418 JA |
4527 | |
4528 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4529 | req->fadvise.len = READ_ONCE(sqe->len); | |
4530 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4531 | return 0; | |
4532 | } | |
4533 | ||
45d189c6 | 4534 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4535 | { |
4536 | struct io_fadvise *fa = &req->fadvise; | |
4537 | int ret; | |
4538 | ||
45d189c6 | 4539 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4540 | switch (fa->advice) { |
4541 | case POSIX_FADV_NORMAL: | |
4542 | case POSIX_FADV_RANDOM: | |
4543 | case POSIX_FADV_SEQUENTIAL: | |
4544 | break; | |
4545 | default: | |
4546 | return -EAGAIN; | |
4547 | } | |
4548 | } | |
4840e418 JA |
4549 | |
4550 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4551 | if (ret < 0) | |
93d2bcd2 | 4552 | req_set_fail(req); |
0bdf3398 | 4553 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4554 | return 0; |
4555 | } | |
4556 | ||
eddc7ef5 JA |
4557 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4558 | { | |
2d74d042 | 4559 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4560 | return -EINVAL; |
26578cda | 4561 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
eddc7ef5 | 4562 | return -EINVAL; |
9c280f90 | 4563 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4564 | return -EBADF; |
eddc7ef5 | 4565 | |
1d9e1288 BM |
4566 | req->statx.dfd = READ_ONCE(sqe->fd); |
4567 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4568 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4569 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4570 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4571 | |
4572 | return 0; | |
4573 | } | |
4574 | ||
45d189c6 | 4575 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4576 | { |
1d9e1288 | 4577 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4578 | int ret; |
4579 | ||
59d70013 | 4580 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4581 | return -EAGAIN; |
4582 | ||
e62753e4 BM |
4583 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4584 | ctx->buffer); | |
eddc7ef5 | 4585 | |
eddc7ef5 | 4586 | if (ret < 0) |
93d2bcd2 | 4587 | req_set_fail(req); |
e1e16097 | 4588 | io_req_complete(req, ret); |
eddc7ef5 JA |
4589 | return 0; |
4590 | } | |
4591 | ||
b5dba59e JA |
4592 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4593 | { | |
14587a46 | 4594 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4595 | return -EINVAL; |
b5dba59e | 4596 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
26578cda | 4597 | sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) |
b5dba59e | 4598 | return -EINVAL; |
9c280f90 | 4599 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4600 | return -EBADF; |
b5dba59e JA |
4601 | |
4602 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4603 | return 0; |
b5dba59e JA |
4604 | } |
4605 | ||
889fca73 | 4606 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4607 | { |
9eac1904 | 4608 | struct files_struct *files = current->files; |
3af73b28 | 4609 | struct io_close *close = &req->close; |
9eac1904 | 4610 | struct fdtable *fdt; |
a1fde923 PB |
4611 | struct file *file = NULL; |
4612 | int ret = -EBADF; | |
b5dba59e | 4613 | |
9eac1904 JA |
4614 | spin_lock(&files->file_lock); |
4615 | fdt = files_fdtable(files); | |
4616 | if (close->fd >= fdt->max_fds) { | |
4617 | spin_unlock(&files->file_lock); | |
4618 | goto err; | |
4619 | } | |
4620 | file = fdt->fd[close->fd]; | |
a1fde923 | 4621 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4622 | spin_unlock(&files->file_lock); |
4623 | file = NULL; | |
4624 | goto err; | |
3af73b28 | 4625 | } |
b5dba59e JA |
4626 | |
4627 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4628 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4629 | spin_unlock(&files->file_lock); |
0bf0eefd | 4630 | return -EAGAIN; |
a2100672 | 4631 | } |
b5dba59e | 4632 | |
9eac1904 JA |
4633 | ret = __close_fd_get_file(close->fd, &file); |
4634 | spin_unlock(&files->file_lock); | |
4635 | if (ret < 0) { | |
4636 | if (ret == -ENOENT) | |
4637 | ret = -EBADF; | |
4638 | goto err; | |
4639 | } | |
4640 | ||
3af73b28 | 4641 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4642 | ret = filp_close(file, current->files); |
4643 | err: | |
3af73b28 | 4644 | if (ret < 0) |
93d2bcd2 | 4645 | req_set_fail(req); |
9eac1904 JA |
4646 | if (file) |
4647 | fput(file); | |
889fca73 | 4648 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4649 | return 0; |
b5dba59e JA |
4650 | } |
4651 | ||
1155c76a | 4652 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4653 | { |
4654 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4655 | |
5d17b4a4 JA |
4656 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4657 | return -EINVAL; | |
26578cda PB |
4658 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4659 | sqe->splice_fd_in)) | |
5d17b4a4 JA |
4660 | return -EINVAL; |
4661 | ||
8ed8d3c3 JA |
4662 | req->sync.off = READ_ONCE(sqe->off); |
4663 | req->sync.len = READ_ONCE(sqe->len); | |
4664 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4665 | return 0; |
4666 | } | |
4667 | ||
45d189c6 | 4668 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4669 | { |
8ed8d3c3 JA |
4670 | int ret; |
4671 | ||
ac45abc0 | 4672 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4673 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4674 | return -EAGAIN; |
4675 | ||
9adbd45d | 4676 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4677 | req->sync.flags); |
4678 | if (ret < 0) | |
93d2bcd2 | 4679 | req_set_fail(req); |
e1e16097 | 4680 | io_req_complete(req, ret); |
5d17b4a4 JA |
4681 | return 0; |
4682 | } | |
4683 | ||
469956e8 | 4684 | #if defined(CONFIG_NET) |
02d27d89 PB |
4685 | static int io_setup_async_msg(struct io_kiocb *req, |
4686 | struct io_async_msghdr *kmsg) | |
4687 | { | |
e8c2bc1f JA |
4688 | struct io_async_msghdr *async_msg = req->async_data; |
4689 | ||
4690 | if (async_msg) | |
02d27d89 | 4691 | return -EAGAIN; |
e8c2bc1f | 4692 | if (io_alloc_async_data(req)) { |
257e84a5 | 4693 | kfree(kmsg->free_iov); |
02d27d89 PB |
4694 | return -ENOMEM; |
4695 | } | |
e8c2bc1f | 4696 | async_msg = req->async_data; |
02d27d89 | 4697 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4698 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4699 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4700 | /* if were using fast_iov, set it to the new one */ |
4701 | if (!async_msg->free_iov) | |
4702 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4703 | ||
02d27d89 PB |
4704 | return -EAGAIN; |
4705 | } | |
4706 | ||
2ae523ed PB |
4707 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4708 | struct io_async_msghdr *iomsg) | |
4709 | { | |
2ae523ed | 4710 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4711 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4712 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4713 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4714 | } |
4715 | ||
93642ef8 PB |
4716 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4717 | { | |
4718 | int ret; | |
4719 | ||
93642ef8 PB |
4720 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4721 | if (!ret) | |
4722 | req->flags |= REQ_F_NEED_CLEANUP; | |
4723 | return ret; | |
4724 | } | |
4725 | ||
3529d8c2 | 4726 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4727 | { |
e47293fd | 4728 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4729 | |
d2b6f48b PB |
4730 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4731 | return -EINVAL; | |
4732 | ||
270a5940 | 4733 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4734 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4735 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4736 | if (sr->msg_flags & MSG_DONTWAIT) | |
4737 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4738 | |
d8768362 JA |
4739 | #ifdef CONFIG_COMPAT |
4740 | if (req->ctx->compat) | |
4741 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4742 | #endif | |
93642ef8 | 4743 | return 0; |
03b1230c JA |
4744 | } |
4745 | ||
889fca73 | 4746 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4747 | { |
6b754c8b | 4748 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4749 | struct socket *sock; |
7a7cacba | 4750 | unsigned flags; |
0031275d | 4751 | int min_ret = 0; |
0fa03c62 JA |
4752 | int ret; |
4753 | ||
dba4a925 | 4754 | sock = sock_from_file(req->file); |
7a7cacba | 4755 | if (unlikely(!sock)) |
dba4a925 | 4756 | return -ENOTSOCK; |
3529d8c2 | 4757 | |
257e84a5 PB |
4758 | kmsg = req->async_data; |
4759 | if (!kmsg) { | |
7a7cacba PB |
4760 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4761 | if (ret) | |
4762 | return ret; | |
4763 | kmsg = &iomsg; | |
0fa03c62 | 4764 | } |
0fa03c62 | 4765 | |
04411806 PB |
4766 | flags = req->sr_msg.msg_flags; |
4767 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4768 | flags |= MSG_DONTWAIT; |
0031275d SM |
4769 | if (flags & MSG_WAITALL) |
4770 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4771 | ||
7a7cacba | 4772 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4773 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4774 | return io_setup_async_msg(req, kmsg); |
4775 | if (ret == -ERESTARTSYS) | |
4776 | ret = -EINTR; | |
0fa03c62 | 4777 | |
257e84a5 PB |
4778 | /* fast path, check for non-NULL to avoid function call */ |
4779 | if (kmsg->free_iov) | |
4780 | kfree(kmsg->free_iov); | |
99bc4c38 | 4781 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4782 | if (ret < min_ret) |
93d2bcd2 | 4783 | req_set_fail(req); |
889fca73 | 4784 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4785 | return 0; |
03b1230c | 4786 | } |
aa1fa28f | 4787 | |
889fca73 | 4788 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4789 | { |
7a7cacba PB |
4790 | struct io_sr_msg *sr = &req->sr_msg; |
4791 | struct msghdr msg; | |
4792 | struct iovec iov; | |
fddaface | 4793 | struct socket *sock; |
7a7cacba | 4794 | unsigned flags; |
0031275d | 4795 | int min_ret = 0; |
fddaface JA |
4796 | int ret; |
4797 | ||
dba4a925 | 4798 | sock = sock_from_file(req->file); |
7a7cacba | 4799 | if (unlikely(!sock)) |
dba4a925 | 4800 | return -ENOTSOCK; |
fddaface | 4801 | |
7a7cacba PB |
4802 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4803 | if (unlikely(ret)) | |
14db8411 | 4804 | return ret; |
fddaface | 4805 | |
7a7cacba PB |
4806 | msg.msg_name = NULL; |
4807 | msg.msg_control = NULL; | |
4808 | msg.msg_controllen = 0; | |
4809 | msg.msg_namelen = 0; | |
fddaface | 4810 | |
04411806 PB |
4811 | flags = req->sr_msg.msg_flags; |
4812 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4813 | flags |= MSG_DONTWAIT; |
0031275d SM |
4814 | if (flags & MSG_WAITALL) |
4815 | min_ret = iov_iter_count(&msg.msg_iter); | |
4816 | ||
7a7cacba PB |
4817 | msg.msg_flags = flags; |
4818 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4819 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4820 | return -EAGAIN; |
4821 | if (ret == -ERESTARTSYS) | |
4822 | ret = -EINTR; | |
fddaface | 4823 | |
0031275d | 4824 | if (ret < min_ret) |
93d2bcd2 | 4825 | req_set_fail(req); |
889fca73 | 4826 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4827 | return 0; |
fddaface JA |
4828 | } |
4829 | ||
1400e697 PB |
4830 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4831 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4832 | { |
4833 | struct io_sr_msg *sr = &req->sr_msg; | |
4834 | struct iovec __user *uiov; | |
4835 | size_t iov_len; | |
4836 | int ret; | |
4837 | ||
1400e697 PB |
4838 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4839 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4840 | if (ret) |
4841 | return ret; | |
4842 | ||
4843 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4844 | if (iov_len > 1) | |
4845 | return -EINVAL; | |
5476dfed | 4846 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4847 | return -EFAULT; |
5476dfed | 4848 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4849 | iomsg->free_iov = NULL; |
52de1fe1 | 4850 | } else { |
257e84a5 | 4851 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4852 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4853 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4854 | false); |
52de1fe1 JA |
4855 | if (ret > 0) |
4856 | ret = 0; | |
4857 | } | |
4858 | ||
4859 | return ret; | |
4860 | } | |
4861 | ||
4862 | #ifdef CONFIG_COMPAT | |
4863 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4864 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4865 | { |
52de1fe1 JA |
4866 | struct io_sr_msg *sr = &req->sr_msg; |
4867 | struct compat_iovec __user *uiov; | |
4868 | compat_uptr_t ptr; | |
4869 | compat_size_t len; | |
4870 | int ret; | |
4871 | ||
4af3417a PB |
4872 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4873 | &ptr, &len); | |
52de1fe1 JA |
4874 | if (ret) |
4875 | return ret; | |
4876 | ||
4877 | uiov = compat_ptr(ptr); | |
4878 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4879 | compat_ssize_t clen; | |
4880 | ||
4881 | if (len > 1) | |
4882 | return -EINVAL; | |
4883 | if (!access_ok(uiov, sizeof(*uiov))) | |
4884 | return -EFAULT; | |
4885 | if (__get_user(clen, &uiov->iov_len)) | |
4886 | return -EFAULT; | |
4887 | if (clen < 0) | |
4888 | return -EINVAL; | |
2d280bc8 | 4889 | sr->len = clen; |
257e84a5 | 4890 | iomsg->free_iov = NULL; |
52de1fe1 | 4891 | } else { |
257e84a5 | 4892 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4893 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4894 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4895 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4896 | if (ret < 0) |
4897 | return ret; | |
4898 | } | |
4899 | ||
4900 | return 0; | |
4901 | } | |
4902 | #endif | |
4903 | ||
1400e697 PB |
4904 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4905 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4906 | { |
1400e697 | 4907 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4908 | |
4909 | #ifdef CONFIG_COMPAT | |
4910 | if (req->ctx->compat) | |
1400e697 | 4911 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4912 | #endif |
52de1fe1 | 4913 | |
1400e697 | 4914 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4915 | } |
4916 | ||
bcda7baa | 4917 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4918 | bool needs_lock) |
bcda7baa JA |
4919 | { |
4920 | struct io_sr_msg *sr = &req->sr_msg; | |
4921 | struct io_buffer *kbuf; | |
4922 | ||
bcda7baa JA |
4923 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4924 | if (IS_ERR(kbuf)) | |
4925 | return kbuf; | |
4926 | ||
4927 | sr->kbuf = kbuf; | |
4928 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4929 | return kbuf; |
fddaface JA |
4930 | } |
4931 | ||
7fbb1b54 PB |
4932 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4933 | { | |
4934 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4935 | } | |
4936 | ||
93642ef8 | 4937 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4938 | { |
99bc4c38 | 4939 | int ret; |
3529d8c2 | 4940 | |
93642ef8 PB |
4941 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4942 | if (!ret) | |
4943 | req->flags |= REQ_F_NEED_CLEANUP; | |
4944 | return ret; | |
4945 | } | |
4946 | ||
4947 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4948 | { | |
4949 | struct io_sr_msg *sr = &req->sr_msg; | |
4950 | ||
d2b6f48b PB |
4951 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4952 | return -EINVAL; | |
4953 | ||
270a5940 | 4954 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4955 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4956 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4957 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4958 | if (sr->msg_flags & MSG_DONTWAIT) | |
4959 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4960 | |
d8768362 JA |
4961 | #ifdef CONFIG_COMPAT |
4962 | if (req->ctx->compat) | |
4963 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4964 | #endif | |
93642ef8 | 4965 | return 0; |
aa1fa28f JA |
4966 | } |
4967 | ||
889fca73 | 4968 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4969 | { |
6b754c8b | 4970 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4971 | struct socket *sock; |
7fbb1b54 | 4972 | struct io_buffer *kbuf; |
7a7cacba | 4973 | unsigned flags; |
0031275d | 4974 | int min_ret = 0; |
52de1fe1 | 4975 | int ret, cflags = 0; |
45d189c6 | 4976 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4977 | |
dba4a925 | 4978 | sock = sock_from_file(req->file); |
7a7cacba | 4979 | if (unlikely(!sock)) |
dba4a925 | 4980 | return -ENOTSOCK; |
3529d8c2 | 4981 | |
257e84a5 PB |
4982 | kmsg = req->async_data; |
4983 | if (!kmsg) { | |
7a7cacba PB |
4984 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4985 | if (ret) | |
681fda8d | 4986 | return ret; |
7a7cacba PB |
4987 | kmsg = &iomsg; |
4988 | } | |
03b1230c | 4989 | |
bc02ef33 | 4990 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4991 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4992 | if (IS_ERR(kbuf)) |
52de1fe1 | 4993 | return PTR_ERR(kbuf); |
7a7cacba | 4994 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4995 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4996 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4997 | 1, req->sr_msg.len); |
4998 | } | |
52de1fe1 | 4999 | |
04411806 PB |
5000 | flags = req->sr_msg.msg_flags; |
5001 | if (force_nonblock) | |
7a7cacba | 5002 | flags |= MSG_DONTWAIT; |
0031275d SM |
5003 | if (flags & MSG_WAITALL) |
5004 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
5005 | ||
7a7cacba PB |
5006 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
5007 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
5008 | if (force_nonblock && ret == -EAGAIN) |
5009 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
5010 | if (ret == -ERESTARTSYS) |
5011 | ret = -EINTR; | |
03b1230c | 5012 | |
7fbb1b54 PB |
5013 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5014 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
5015 | /* fast path, check for non-NULL to avoid function call */ |
5016 | if (kmsg->free_iov) | |
5017 | kfree(kmsg->free_iov); | |
99bc4c38 | 5018 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 5019 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 5020 | req_set_fail(req); |
889fca73 | 5021 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 5022 | return 0; |
0fa03c62 | 5023 | } |
5d17b4a4 | 5024 | |
889fca73 | 5025 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 5026 | { |
6b754c8b | 5027 | struct io_buffer *kbuf; |
7a7cacba PB |
5028 | struct io_sr_msg *sr = &req->sr_msg; |
5029 | struct msghdr msg; | |
5030 | void __user *buf = sr->buf; | |
fddaface | 5031 | struct socket *sock; |
7a7cacba PB |
5032 | struct iovec iov; |
5033 | unsigned flags; | |
0031275d | 5034 | int min_ret = 0; |
bcda7baa | 5035 | int ret, cflags = 0; |
45d189c6 | 5036 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 5037 | |
dba4a925 | 5038 | sock = sock_from_file(req->file); |
7a7cacba | 5039 | if (unlikely(!sock)) |
dba4a925 | 5040 | return -ENOTSOCK; |
fddaface | 5041 | |
bc02ef33 | 5042 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 5043 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
5044 | if (IS_ERR(kbuf)) |
5045 | return PTR_ERR(kbuf); | |
7a7cacba | 5046 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 5047 | } |
bcda7baa | 5048 | |
7a7cacba | 5049 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
5050 | if (unlikely(ret)) |
5051 | goto out_free; | |
fddaface | 5052 | |
7a7cacba PB |
5053 | msg.msg_name = NULL; |
5054 | msg.msg_control = NULL; | |
5055 | msg.msg_controllen = 0; | |
5056 | msg.msg_namelen = 0; | |
5057 | msg.msg_iocb = NULL; | |
5058 | msg.msg_flags = 0; | |
fddaface | 5059 | |
04411806 PB |
5060 | flags = req->sr_msg.msg_flags; |
5061 | if (force_nonblock) | |
7a7cacba | 5062 | flags |= MSG_DONTWAIT; |
0031275d SM |
5063 | if (flags & MSG_WAITALL) |
5064 | min_ret = iov_iter_count(&msg.msg_iter); | |
5065 | ||
7a7cacba PB |
5066 | ret = sock_recvmsg(sock, &msg, flags); |
5067 | if (force_nonblock && ret == -EAGAIN) | |
5068 | return -EAGAIN; | |
5069 | if (ret == -ERESTARTSYS) | |
5070 | ret = -EINTR; | |
14c32eee | 5071 | out_free: |
7fbb1b54 PB |
5072 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5073 | cflags = io_put_recv_kbuf(req); | |
0031275d | 5074 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 5075 | req_set_fail(req); |
889fca73 | 5076 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5077 | return 0; |
fddaface JA |
5078 | } |
5079 | ||
3529d8c2 | 5080 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5081 | { |
8ed8d3c3 JA |
5082 | struct io_accept *accept = &req->accept; |
5083 | ||
14587a46 | 5084 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5085 | return -EINVAL; |
aaa4db12 | 5086 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5087 | return -EINVAL; |
5088 | ||
d55e5f5b JA |
5089 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5090 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5091 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5092 | accept->nofile = rlimit(RLIMIT_NOFILE); |
a7083ad5 | 5093 | |
aaa4db12 PB |
5094 | accept->file_slot = READ_ONCE(sqe->file_index); |
5095 | if (accept->file_slot && ((req->open.how.flags & O_CLOEXEC) || | |
5096 | (accept->flags & SOCK_CLOEXEC))) | |
5097 | return -EINVAL; | |
a7083ad5 PB |
5098 | if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) |
5099 | return -EINVAL; | |
5100 | if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) | |
5101 | accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; | |
8ed8d3c3 | 5102 | return 0; |
8ed8d3c3 | 5103 | } |
17f2fe35 | 5104 | |
889fca73 | 5105 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5106 | { |
5107 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5108 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5109 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
aaa4db12 | 5110 | bool fixed = !!accept->file_slot; |
a7083ad5 PB |
5111 | struct file *file; |
5112 | int ret, fd; | |
8ed8d3c3 | 5113 | |
e697deed JX |
5114 | if (req->file->f_flags & O_NONBLOCK) |
5115 | req->flags |= REQ_F_NOWAIT; | |
5116 | ||
aaa4db12 PB |
5117 | if (!fixed) { |
5118 | fd = __get_unused_fd_flags(accept->flags, accept->nofile); | |
5119 | if (unlikely(fd < 0)) | |
5120 | return fd; | |
5121 | } | |
a7083ad5 PB |
5122 | file = do_accept(req->file, file_flags, accept->addr, accept->addr_len, |
5123 | accept->flags); | |
5124 | if (IS_ERR(file)) { | |
aaa4db12 PB |
5125 | if (!fixed) |
5126 | put_unused_fd(fd); | |
a7083ad5 PB |
5127 | ret = PTR_ERR(file); |
5128 | if (ret == -EAGAIN && force_nonblock) | |
5129 | return -EAGAIN; | |
ac45abc0 PB |
5130 | if (ret == -ERESTARTSYS) |
5131 | ret = -EINTR; | |
93d2bcd2 | 5132 | req_set_fail(req); |
aaa4db12 | 5133 | } else if (!fixed) { |
a7083ad5 PB |
5134 | fd_install(fd, file); |
5135 | ret = fd; | |
aaa4db12 PB |
5136 | } else { |
5137 | ret = io_install_fixed_file(req, file, issue_flags, | |
5138 | accept->file_slot - 1); | |
ac45abc0 | 5139 | } |
889fca73 | 5140 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5141 | return 0; |
8ed8d3c3 JA |
5142 | } |
5143 | ||
93642ef8 PB |
5144 | static int io_connect_prep_async(struct io_kiocb *req) |
5145 | { | |
5146 | struct io_async_connect *io = req->async_data; | |
5147 | struct io_connect *conn = &req->connect; | |
5148 | ||
5149 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
5150 | } | |
5151 | ||
3529d8c2 | 5152 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5153 | { |
3529d8c2 | 5154 | struct io_connect *conn = &req->connect; |
f499a021 | 5155 | |
14587a46 | 5156 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 | 5157 | return -EINVAL; |
26578cda PB |
5158 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags || |
5159 | sqe->splice_fd_in) | |
3fbb51c1 JA |
5160 | return -EINVAL; |
5161 | ||
3529d8c2 JA |
5162 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5163 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 5164 | return 0; |
f499a021 JA |
5165 | } |
5166 | ||
889fca73 | 5167 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5168 | { |
e8c2bc1f | 5169 | struct io_async_connect __io, *io; |
f8e85cf2 | 5170 | unsigned file_flags; |
3fbb51c1 | 5171 | int ret; |
45d189c6 | 5172 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5173 | |
e8c2bc1f JA |
5174 | if (req->async_data) { |
5175 | io = req->async_data; | |
f499a021 | 5176 | } else { |
3529d8c2 JA |
5177 | ret = move_addr_to_kernel(req->connect.addr, |
5178 | req->connect.addr_len, | |
e8c2bc1f | 5179 | &__io.address); |
f499a021 JA |
5180 | if (ret) |
5181 | goto out; | |
5182 | io = &__io; | |
5183 | } | |
5184 | ||
3fbb51c1 JA |
5185 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5186 | ||
e8c2bc1f | 5187 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5188 | req->connect.addr_len, file_flags); |
87f80d62 | 5189 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5190 | if (req->async_data) |
b7bb4f7d | 5191 | return -EAGAIN; |
e8c2bc1f | 5192 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5193 | ret = -ENOMEM; |
5194 | goto out; | |
5195 | } | |
e8c2bc1f | 5196 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 5197 | return -EAGAIN; |
f499a021 | 5198 | } |
f8e85cf2 JA |
5199 | if (ret == -ERESTARTSYS) |
5200 | ret = -EINTR; | |
f499a021 | 5201 | out: |
4e88d6e7 | 5202 | if (ret < 0) |
93d2bcd2 | 5203 | req_set_fail(req); |
889fca73 | 5204 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5205 | return 0; |
469956e8 Y |
5206 | } |
5207 | #else /* !CONFIG_NET */ | |
99a10081 JA |
5208 | #define IO_NETOP_FN(op) \ |
5209 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
5210 | { \ | |
5211 | return -EOPNOTSUPP; \ | |
5212 | } | |
5213 | ||
5214 | #define IO_NETOP_PREP(op) \ | |
5215 | IO_NETOP_FN(op) \ | |
5216 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
5217 | { \ | |
5218 | return -EOPNOTSUPP; \ | |
5219 | } \ | |
5220 | ||
5221 | #define IO_NETOP_PREP_ASYNC(op) \ | |
5222 | IO_NETOP_PREP(op) \ | |
5223 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
5224 | { \ | |
5225 | return -EOPNOTSUPP; \ | |
5226 | } | |
5227 | ||
5228 | IO_NETOP_PREP_ASYNC(sendmsg); | |
5229 | IO_NETOP_PREP_ASYNC(recvmsg); | |
5230 | IO_NETOP_PREP_ASYNC(connect); | |
5231 | IO_NETOP_PREP(accept); | |
5232 | IO_NETOP_FN(send); | |
5233 | IO_NETOP_FN(recv); | |
469956e8 | 5234 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5235 | |
d7718a9d JA |
5236 | struct io_poll_table { |
5237 | struct poll_table_struct pt; | |
5238 | struct io_kiocb *req; | |
68b11e8b | 5239 | int nr_entries; |
d7718a9d JA |
5240 | int error; |
5241 | }; | |
ce593a6c | 5242 | |
d7718a9d | 5243 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5b0a6acc | 5244 | __poll_t mask, io_req_tw_func_t func) |
d7718a9d | 5245 | { |
d7718a9d JA |
5246 | /* for instances that support it check for an event match first: */ |
5247 | if (mask && !(mask & poll->events)) | |
5248 | return 0; | |
5249 | ||
5250 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5251 | ||
5252 | list_del_init(&poll->wait.entry); | |
5253 | ||
d7718a9d | 5254 | req->result = mask; |
5b0a6acc | 5255 | req->io_task_work.func = func; |
6d816e08 | 5256 | |
d7718a9d | 5257 | /* |
e3aabf95 JA |
5258 | * If this fails, then the task is exiting. When a task exits, the |
5259 | * work gets canceled, so just cancel this request as well instead | |
5260 | * of executing it. We can't safely execute it anyway, as we may not | |
5261 | * have the needed state needed for it anyway. | |
d7718a9d | 5262 | */ |
e09ee510 | 5263 | io_req_task_work_add(req); |
d7718a9d JA |
5264 | return 1; |
5265 | } | |
5266 | ||
74ce6ce4 JA |
5267 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5268 | __acquires(&req->ctx->completion_lock) | |
5269 | { | |
5270 | struct io_ring_ctx *ctx = req->ctx; | |
5271 | ||
316319e8 | 5272 | /* req->task == current here, checking PF_EXITING is safe */ |
e09ee510 PB |
5273 | if (unlikely(req->task->flags & PF_EXITING)) |
5274 | WRITE_ONCE(poll->canceled, true); | |
5275 | ||
74ce6ce4 JA |
5276 | if (!req->result && !READ_ONCE(poll->canceled)) { |
5277 | struct poll_table_struct pt = { ._key = poll->events }; | |
5278 | ||
5279 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5280 | } | |
5281 | ||
79ebeaee | 5282 | spin_lock(&ctx->completion_lock); |
74ce6ce4 JA |
5283 | if (!req->result && !READ_ONCE(poll->canceled)) { |
5284 | add_wait_queue(poll->head, &poll->wait); | |
5285 | return true; | |
5286 | } | |
5287 | ||
5288 | return false; | |
5289 | } | |
5290 | ||
d4e7cd36 | 5291 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5292 | { |
e8c2bc1f | 5293 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5294 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5295 | return req->async_data; |
d4e7cd36 JA |
5296 | return req->apoll->double_poll; |
5297 | } | |
5298 | ||
5299 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5300 | { | |
5301 | if (req->opcode == IORING_OP_POLL_ADD) | |
5302 | return &req->poll; | |
5303 | return &req->apoll->poll; | |
5304 | } | |
5305 | ||
5306 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 5307 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
5308 | { |
5309 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5310 | |
5311 | lockdep_assert_held(&req->ctx->completion_lock); | |
5312 | ||
5313 | if (poll && poll->head) { | |
5314 | struct wait_queue_head *head = poll->head; | |
5315 | ||
79ebeaee | 5316 | spin_lock_irq(&head->lock); |
18bceab1 JA |
5317 | list_del_init(&poll->wait.entry); |
5318 | if (poll->wait.private) | |
de9b4cca | 5319 | req_ref_put(req); |
18bceab1 | 5320 | poll->head = NULL; |
79ebeaee | 5321 | spin_unlock_irq(&head->lock); |
18bceab1 JA |
5322 | } |
5323 | } | |
5324 | ||
31efe48e | 5325 | static bool __io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 5326 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
5327 | { |
5328 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 5329 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 5330 | int error; |
18bceab1 | 5331 | |
e27414be | 5332 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 5333 | error = -ECANCELED; |
88e41cf9 | 5334 | req->poll.events |= EPOLLONESHOT; |
e27414be | 5335 | } else { |
5082620f | 5336 | error = mangle_poll(mask); |
e27414be | 5337 | } |
b69de288 JA |
5338 | if (req->poll.events & EPOLLONESHOT) |
5339 | flags = 0; | |
a62682f9 HX |
5340 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5341 | req->poll.events |= EPOLLONESHOT; | |
88e41cf9 | 5342 | flags = 0; |
a62682f9 | 5343 | } |
7b289c38 HX |
5344 | if (flags & IORING_CQE_F_MORE) |
5345 | ctx->cq_extra++; | |
18bceab1 | 5346 | |
88e41cf9 | 5347 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
5348 | } |
5349 | ||
31efe48e XW |
5350 | static inline bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
5351 | __must_hold(&req->ctx->completion_lock) | |
5352 | { | |
5353 | bool done; | |
5354 | ||
5355 | done = __io_poll_complete(req, mask); | |
5356 | io_commit_cqring(req->ctx); | |
5357 | return done; | |
5358 | } | |
5359 | ||
f237c30a | 5360 | static void io_poll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 JA |
5361 | { |
5362 | struct io_ring_ctx *ctx = req->ctx; | |
dd221f46 | 5363 | struct io_kiocb *nxt; |
18bceab1 JA |
5364 | |
5365 | if (io_poll_rewait(req, &req->poll)) { | |
79ebeaee | 5366 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5367 | } else { |
f40b964a | 5368 | bool done; |
18bceab1 | 5369 | |
5b7aa38d HX |
5370 | if (req->poll.done) { |
5371 | spin_unlock(&ctx->completion_lock); | |
5372 | return; | |
5373 | } | |
31efe48e | 5374 | done = __io_poll_complete(req, req->result); |
88e41cf9 | 5375 | if (done) { |
a890d01e | 5376 | io_poll_remove_double(req); |
88e41cf9 | 5377 | hash_del(&req->hash_node); |
bd99c71b | 5378 | req->poll.done = true; |
f40b964a | 5379 | } else { |
88e41cf9 JA |
5380 | req->result = 0; |
5381 | add_wait_queue(req->poll.head, &req->poll.wait); | |
5382 | } | |
31efe48e | 5383 | io_commit_cqring(ctx); |
79ebeaee | 5384 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5385 | io_cqring_ev_posted(ctx); |
18bceab1 | 5386 | |
88e41cf9 JA |
5387 | if (done) { |
5388 | nxt = io_put_req_find_next(req); | |
5389 | if (nxt) | |
f237c30a | 5390 | io_req_task_submit(nxt, locked); |
88e41cf9 | 5391 | } |
dd221f46 | 5392 | } |
18bceab1 JA |
5393 | } |
5394 | ||
5395 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5396 | int sync, void *key) | |
5397 | { | |
5398 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5399 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 | 5400 | __poll_t mask = key_to_poll(key); |
79ebeaee | 5401 | unsigned long flags; |
18bceab1 JA |
5402 | |
5403 | /* for instances that support it check for an event match first: */ | |
5404 | if (mask && !(mask & poll->events)) | |
5405 | return 0; | |
88e41cf9 JA |
5406 | if (!(poll->events & EPOLLONESHOT)) |
5407 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 5408 | |
8706e04e JA |
5409 | list_del_init(&wait->entry); |
5410 | ||
9ce85ef2 | 5411 | if (poll->head) { |
18bceab1 JA |
5412 | bool done; |
5413 | ||
79ebeaee | 5414 | spin_lock_irqsave(&poll->head->lock, flags); |
807abcb0 | 5415 | done = list_empty(&poll->wait.entry); |
18bceab1 | 5416 | if (!done) |
807abcb0 | 5417 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5418 | /* make sure double remove sees this as being gone */ |
5419 | wait->private = NULL; | |
79ebeaee | 5420 | spin_unlock_irqrestore(&poll->head->lock, flags); |
c8b5e260 JA |
5421 | if (!done) { |
5422 | /* use wait func handler, so it matches the rq type */ | |
5423 | poll->wait.func(&poll->wait, mode, sync, key); | |
5424 | } | |
18bceab1 | 5425 | } |
de9b4cca | 5426 | req_ref_put(req); |
18bceab1 JA |
5427 | return 1; |
5428 | } | |
5429 | ||
5430 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5431 | wait_queue_func_t wake_func) | |
5432 | { | |
5433 | poll->head = NULL; | |
5434 | poll->done = false; | |
5435 | poll->canceled = false; | |
464dca61 JA |
5436 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5437 | /* mask in events that we always want/need */ | |
5438 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5439 | INIT_LIST_HEAD(&poll->wait.entry); |
5440 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5441 | } | |
5442 | ||
5443 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5444 | struct wait_queue_head *head, |
5445 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5446 | { |
5447 | struct io_kiocb *req = pt->req; | |
5448 | ||
5449 | /* | |
68b11e8b PB |
5450 | * The file being polled uses multiple waitqueues for poll handling |
5451 | * (e.g. one for read, one for write). Setup a separate io_poll_iocb | |
5452 | * if this happens. | |
18bceab1 | 5453 | */ |
68b11e8b | 5454 | if (unlikely(pt->nr_entries)) { |
58852d4d PB |
5455 | struct io_poll_iocb *poll_one = poll; |
5456 | ||
23a65db8 PB |
5457 | /* double add on the same waitqueue head, ignore */ |
5458 | if (poll_one->head == head) | |
5459 | return; | |
18bceab1 | 5460 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5461 | if (*poll_ptr) { |
23a65db8 PB |
5462 | if ((*poll_ptr)->head == head) |
5463 | return; | |
18bceab1 JA |
5464 | pt->error = -EINVAL; |
5465 | return; | |
5466 | } | |
ea6a693d JA |
5467 | /* |
5468 | * Can't handle multishot for double wait for now, turn it | |
5469 | * into one-shot mode. | |
5470 | */ | |
7a274727 PB |
5471 | if (!(poll_one->events & EPOLLONESHOT)) |
5472 | poll_one->events |= EPOLLONESHOT; | |
18bceab1 JA |
5473 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5474 | if (!poll) { | |
5475 | pt->error = -ENOMEM; | |
5476 | return; | |
5477 | } | |
58852d4d | 5478 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5479 | req_ref_get(req); |
18bceab1 | 5480 | poll->wait.private = req; |
807abcb0 | 5481 | *poll_ptr = poll; |
18bceab1 JA |
5482 | } |
5483 | ||
68b11e8b | 5484 | pt->nr_entries++; |
18bceab1 | 5485 | poll->head = head; |
a31eb4a2 JX |
5486 | |
5487 | if (poll->events & EPOLLEXCLUSIVE) | |
5488 | add_wait_queue_exclusive(head, &poll->wait); | |
5489 | else | |
5490 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5491 | } |
5492 | ||
5493 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5494 | struct poll_table_struct *p) | |
5495 | { | |
5496 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5497 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5498 | |
807abcb0 | 5499 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5500 | } |
5501 | ||
f237c30a | 5502 | static void io_async_task_func(struct io_kiocb *req, bool *locked) |
d7718a9d | 5503 | { |
d7718a9d JA |
5504 | struct async_poll *apoll = req->apoll; |
5505 | struct io_ring_ctx *ctx = req->ctx; | |
5506 | ||
236daeae | 5507 | trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data); |
d7718a9d | 5508 | |
74ce6ce4 | 5509 | if (io_poll_rewait(req, &apoll->poll)) { |
79ebeaee | 5510 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5511 | return; |
d7718a9d JA |
5512 | } |
5513 | ||
0ea13b44 | 5514 | hash_del(&req->hash_node); |
d4e7cd36 | 5515 | io_poll_remove_double(req); |
bd99c71b | 5516 | apoll->poll.done = true; |
79ebeaee | 5517 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5518 | |
0be0b0e3 | 5519 | if (!READ_ONCE(apoll->poll.canceled)) |
f237c30a | 5520 | io_req_task_submit(req, locked); |
0be0b0e3 | 5521 | else |
2593553a | 5522 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5523 | } |
5524 | ||
5525 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5526 | void *key) | |
5527 | { | |
5528 | struct io_kiocb *req = wait->private; | |
5529 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5530 | ||
5531 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5532 | key_to_poll(key)); | |
5533 | ||
5534 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5535 | } | |
5536 | ||
5537 | static void io_poll_req_insert(struct io_kiocb *req) | |
5538 | { | |
5539 | struct io_ring_ctx *ctx = req->ctx; | |
5540 | struct hlist_head *list; | |
5541 | ||
5542 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5543 | hlist_add_head(&req->hash_node, list); | |
5544 | } | |
5545 | ||
5546 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5547 | struct io_poll_iocb *poll, | |
5548 | struct io_poll_table *ipt, __poll_t mask, | |
5549 | wait_queue_func_t wake_func) | |
5550 | __acquires(&ctx->completion_lock) | |
5551 | { | |
5552 | struct io_ring_ctx *ctx = req->ctx; | |
5553 | bool cancel = false; | |
5554 | ||
4d52f338 | 5555 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5556 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5557 | poll->file = req->file; |
18bceab1 | 5558 | poll->wait.private = req; |
d7718a9d JA |
5559 | |
5560 | ipt->pt._key = mask; | |
5561 | ipt->req = req; | |
68b11e8b PB |
5562 | ipt->error = 0; |
5563 | ipt->nr_entries = 0; | |
d7718a9d | 5564 | |
d7718a9d | 5565 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
68b11e8b PB |
5566 | if (unlikely(!ipt->nr_entries) && !ipt->error) |
5567 | ipt->error = -EINVAL; | |
d7718a9d | 5568 | |
79ebeaee | 5569 | spin_lock(&ctx->completion_lock); |
a890d01e | 5570 | if (ipt->error || (mask && (poll->events & EPOLLONESHOT))) |
46fee9ab | 5571 | io_poll_remove_double(req); |
d7718a9d | 5572 | if (likely(poll->head)) { |
79ebeaee | 5573 | spin_lock_irq(&poll->head->lock); |
d7718a9d JA |
5574 | if (unlikely(list_empty(&poll->wait.entry))) { |
5575 | if (ipt->error) | |
5576 | cancel = true; | |
5577 | ipt->error = 0; | |
5578 | mask = 0; | |
5579 | } | |
88e41cf9 | 5580 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5581 | list_del_init(&poll->wait.entry); |
5582 | else if (cancel) | |
5583 | WRITE_ONCE(poll->canceled, true); | |
5584 | else if (!poll->done) /* actually waiting for an event */ | |
5585 | io_poll_req_insert(req); | |
79ebeaee | 5586 | spin_unlock_irq(&poll->head->lock); |
d7718a9d JA |
5587 | } |
5588 | ||
5589 | return mask; | |
5590 | } | |
5591 | ||
59b735ae OL |
5592 | enum { |
5593 | IO_APOLL_OK, | |
5594 | IO_APOLL_ABORTED, | |
5595 | IO_APOLL_READY | |
5596 | }; | |
5597 | ||
5598 | static int io_arm_poll_handler(struct io_kiocb *req) | |
d7718a9d JA |
5599 | { |
5600 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5601 | struct io_ring_ctx *ctx = req->ctx; | |
5602 | struct async_poll *apoll; | |
5603 | struct io_poll_table ipt; | |
b2d9c3da | 5604 | __poll_t ret, mask = EPOLLONESHOT | POLLERR | POLLPRI; |
9dab14b8 | 5605 | int rw; |
d7718a9d JA |
5606 | |
5607 | if (!req->file || !file_can_poll(req->file)) | |
59b735ae | 5608 | return IO_APOLL_ABORTED; |
24c74678 | 5609 | if (req->flags & REQ_F_POLLED) |
59b735ae | 5610 | return IO_APOLL_ABORTED; |
b2d9c3da PB |
5611 | if (!def->pollin && !def->pollout) |
5612 | return IO_APOLL_ABORTED; | |
5613 | ||
5614 | if (def->pollin) { | |
9dab14b8 | 5615 | rw = READ; |
b2d9c3da PB |
5616 | mask |= POLLIN | POLLRDNORM; |
5617 | ||
5618 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5619 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5620 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5621 | mask &= ~POLLIN; | |
5622 | } else { | |
9dab14b8 | 5623 | rw = WRITE; |
b2d9c3da PB |
5624 | mask |= POLLOUT | POLLWRNORM; |
5625 | } | |
5626 | ||
9dab14b8 | 5627 | /* if we can't nonblock try, then no point in arming a poll handler */ |
b191e2df | 5628 | if (!io_file_supports_nowait(req, rw)) |
59b735ae | 5629 | return IO_APOLL_ABORTED; |
d7718a9d JA |
5630 | |
5631 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5632 | if (unlikely(!apoll)) | |
59b735ae | 5633 | return IO_APOLL_ABORTED; |
807abcb0 | 5634 | apoll->double_poll = NULL; |
d7718a9d | 5635 | req->apoll = apoll; |
b2d9c3da | 5636 | req->flags |= REQ_F_POLLED; |
d7718a9d | 5637 | ipt.pt._qproc = io_async_queue_proc; |
48dcd38d | 5638 | io_req_set_refcount(req); |
d7718a9d JA |
5639 | |
5640 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5641 | io_async_wake); | |
79ebeaee | 5642 | spin_unlock(&ctx->completion_lock); |
41a5169c HX |
5643 | if (ret || ipt.error) |
5644 | return ret ? IO_APOLL_READY : IO_APOLL_ABORTED; | |
5645 | ||
236daeae OL |
5646 | trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, |
5647 | mask, apoll->poll.events); | |
59b735ae | 5648 | return IO_APOLL_OK; |
d7718a9d JA |
5649 | } |
5650 | ||
5651 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5652 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5653 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5654 | { |
b41e9852 | 5655 | bool do_complete = false; |
221c5eb2 | 5656 | |
5082620f JA |
5657 | if (!poll->head) |
5658 | return false; | |
79ebeaee | 5659 | spin_lock_irq(&poll->head->lock); |
b2e720ac JA |
5660 | if (do_cancel) |
5661 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5662 | if (!list_empty(&poll->wait.entry)) { |
5663 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5664 | do_complete = true; |
221c5eb2 | 5665 | } |
79ebeaee | 5666 | spin_unlock_irq(&poll->head->lock); |
3bfa5bcb | 5667 | hash_del(&req->hash_node); |
d7718a9d JA |
5668 | return do_complete; |
5669 | } | |
5670 | ||
5d709043 | 5671 | static bool io_poll_remove_one(struct io_kiocb *req) |
e07785b0 | 5672 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5673 | { |
5674 | bool do_complete; | |
5675 | ||
d4e7cd36 | 5676 | io_poll_remove_double(req); |
e31001a3 | 5677 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5678 | |
b41e9852 | 5679 | if (do_complete) { |
d4d19c19 | 5680 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5681 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5682 | req_set_fail(req); |
91c2f697 | 5683 | io_put_req_deferred(req); |
5d709043 | 5684 | } |
b41e9852 | 5685 | return do_complete; |
221c5eb2 JA |
5686 | } |
5687 | ||
76e1b642 JA |
5688 | /* |
5689 | * Returns true if we found and killed one or more poll requests | |
5690 | */ | |
6b81928d | 5691 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5692 | bool cancel_all) |
221c5eb2 | 5693 | { |
78076bb6 | 5694 | struct hlist_node *tmp; |
221c5eb2 | 5695 | struct io_kiocb *req; |
8e2e1faf | 5696 | int posted = 0, i; |
221c5eb2 | 5697 | |
79ebeaee | 5698 | spin_lock(&ctx->completion_lock); |
78076bb6 JA |
5699 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5700 | struct hlist_head *list; | |
5701 | ||
5702 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5703 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5704 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5705 | posted += io_poll_remove_one(req); |
5706 | } | |
221c5eb2 | 5707 | } |
79ebeaee | 5708 | spin_unlock(&ctx->completion_lock); |
b41e9852 | 5709 | |
8e2e1faf JA |
5710 | if (posted) |
5711 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5712 | |
5713 | return posted != 0; | |
221c5eb2 JA |
5714 | } |
5715 | ||
9ba5fac8 PB |
5716 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5717 | bool poll_only) | |
e07785b0 | 5718 | __must_hold(&ctx->completion_lock) |
47f46768 | 5719 | { |
78076bb6 | 5720 | struct hlist_head *list; |
47f46768 JA |
5721 | struct io_kiocb *req; |
5722 | ||
78076bb6 JA |
5723 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5724 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5725 | if (sqe_addr != req->user_data) |
5726 | continue; | |
9ba5fac8 PB |
5727 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5728 | continue; | |
b2cb805f | 5729 | return req; |
47f46768 | 5730 | } |
b2cb805f JA |
5731 | return NULL; |
5732 | } | |
5733 | ||
9ba5fac8 PB |
5734 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5735 | bool poll_only) | |
e07785b0 | 5736 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5737 | { |
5738 | struct io_kiocb *req; | |
5739 | ||
9ba5fac8 | 5740 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5741 | if (!req) |
5742 | return -ENOENT; | |
5743 | if (io_poll_remove_one(req)) | |
5744 | return 0; | |
5745 | ||
5746 | return -EALREADY; | |
47f46768 JA |
5747 | } |
5748 | ||
9096af3e PB |
5749 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5750 | unsigned int flags) | |
5751 | { | |
5752 | u32 events; | |
47f46768 | 5753 | |
9096af3e PB |
5754 | events = READ_ONCE(sqe->poll32_events); |
5755 | #ifdef __BIG_ENDIAN | |
5756 | events = swahw32(events); | |
5757 | #endif | |
5758 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5759 | events |= EPOLLONESHOT; | |
5760 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5761 | } |
5762 | ||
c5de0036 | 5763 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5764 | const struct io_uring_sqe *sqe) |
0969e783 | 5765 | { |
c5de0036 PB |
5766 | struct io_poll_update *upd = &req->poll_update; |
5767 | u32 flags; | |
5768 | ||
0969e783 JA |
5769 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5770 | return -EINVAL; | |
26578cda | 5771 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
c5de0036 PB |
5772 | return -EINVAL; |
5773 | flags = READ_ONCE(sqe->len); | |
5774 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5775 | IORING_POLL_ADD_MULTI)) | |
5776 | return -EINVAL; | |
5777 | /* meaningless without update */ | |
5778 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5779 | return -EINVAL; |
5780 | ||
c5de0036 PB |
5781 | upd->old_user_data = READ_ONCE(sqe->addr); |
5782 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5783 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5784 | |
c5de0036 PB |
5785 | upd->new_user_data = READ_ONCE(sqe->off); |
5786 | if (!upd->update_user_data && upd->new_user_data) | |
5787 | return -EINVAL; | |
5788 | if (upd->update_events) | |
5789 | upd->events = io_poll_parse_events(sqe, flags); | |
5790 | else if (sqe->poll32_events) | |
5791 | return -EINVAL; | |
221c5eb2 | 5792 | |
221c5eb2 JA |
5793 | return 0; |
5794 | } | |
5795 | ||
221c5eb2 JA |
5796 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5797 | void *key) | |
5798 | { | |
c2f2eb7d JA |
5799 | struct io_kiocb *req = wait->private; |
5800 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5801 | |
d7718a9d | 5802 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5803 | } |
5804 | ||
221c5eb2 JA |
5805 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5806 | struct poll_table_struct *p) | |
5807 | { | |
5808 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5809 | ||
e8c2bc1f | 5810 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5811 | } |
5812 | ||
3529d8c2 | 5813 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5814 | { |
5815 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5816 | u32 flags; |
221c5eb2 JA |
5817 | |
5818 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5819 | return -EINVAL; | |
c5de0036 | 5820 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5821 | return -EINVAL; |
5822 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5823 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5824 | return -EINVAL; |
5825 | ||
48dcd38d | 5826 | io_req_set_refcount(req); |
c5de0036 | 5827 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5828 | return 0; |
5829 | } | |
5830 | ||
61e98203 | 5831 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5832 | { |
5833 | struct io_poll_iocb *poll = &req->poll; | |
5834 | struct io_ring_ctx *ctx = req->ctx; | |
5835 | struct io_poll_table ipt; | |
0969e783 | 5836 | __poll_t mask; |
5b7aa38d | 5837 | bool done; |
0969e783 | 5838 | |
d7718a9d | 5839 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5840 | |
d7718a9d JA |
5841 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5842 | io_poll_wake); | |
221c5eb2 | 5843 | |
8c838788 | 5844 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5845 | ipt.error = 0; |
5b7aa38d | 5846 | done = io_poll_complete(req, mask); |
221c5eb2 | 5847 | } |
79ebeaee | 5848 | spin_unlock(&ctx->completion_lock); |
221c5eb2 | 5849 | |
8c838788 JA |
5850 | if (mask) { |
5851 | io_cqring_ev_posted(ctx); | |
5b7aa38d | 5852 | if (done) |
88e41cf9 | 5853 | io_put_req(req); |
221c5eb2 | 5854 | } |
8c838788 | 5855 | return ipt.error; |
221c5eb2 JA |
5856 | } |
5857 | ||
c5de0036 | 5858 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5859 | { |
5860 | struct io_ring_ctx *ctx = req->ctx; | |
5861 | struct io_kiocb *preq; | |
cb3b200e | 5862 | bool completing; |
b69de288 JA |
5863 | int ret; |
5864 | ||
79ebeaee | 5865 | spin_lock(&ctx->completion_lock); |
9ba5fac8 | 5866 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5867 | if (!preq) { |
5868 | ret = -ENOENT; | |
5869 | goto err; | |
b69de288 | 5870 | } |
cb3b200e | 5871 | |
c5de0036 PB |
5872 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5873 | completing = true; | |
5874 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5875 | goto err; | |
5876 | } | |
5877 | ||
cb3b200e JA |
5878 | /* |
5879 | * Don't allow racy completion with singleshot, as we cannot safely | |
5880 | * update those. For multishot, if we're racing with completion, just | |
5881 | * let completion re-add it. | |
5882 | */ | |
5883 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5884 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5885 | ret = -EALREADY; | |
5886 | goto err; | |
b69de288 JA |
5887 | } |
5888 | /* we now have a detached poll request. reissue. */ | |
5889 | ret = 0; | |
5890 | err: | |
b69de288 | 5891 | if (ret < 0) { |
79ebeaee | 5892 | spin_unlock(&ctx->completion_lock); |
93d2bcd2 | 5893 | req_set_fail(req); |
b69de288 JA |
5894 | io_req_complete(req, ret); |
5895 | return 0; | |
5896 | } | |
5897 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5898 | if (req->poll_update.update_events) { |
b69de288 | 5899 | preq->poll.events &= ~0xffff; |
9d805892 | 5900 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5901 | preq->poll.events |= IO_POLL_UNMASK; |
5902 | } | |
9d805892 PB |
5903 | if (req->poll_update.update_user_data) |
5904 | preq->user_data = req->poll_update.new_user_data; | |
79ebeaee | 5905 | spin_unlock(&ctx->completion_lock); |
cb3b200e | 5906 | |
b69de288 JA |
5907 | /* complete update request, we're done with it */ |
5908 | io_req_complete(req, ret); | |
5909 | ||
cb3b200e | 5910 | if (!completing) { |
c5de0036 | 5911 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5912 | if (ret < 0) { |
93d2bcd2 | 5913 | req_set_fail(preq); |
cb3b200e JA |
5914 | io_req_complete(preq, ret); |
5915 | } | |
b69de288 JA |
5916 | } |
5917 | return 0; | |
5918 | } | |
5919 | ||
f237c30a | 5920 | static void io_req_task_timeout(struct io_kiocb *req, bool *locked) |
89850fce | 5921 | { |
89850fce | 5922 | req_set_fail(req); |
505657bc | 5923 | io_req_complete_post(req, -ETIME, 0); |
89850fce JA |
5924 | } |
5925 | ||
5262f567 JA |
5926 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5927 | { | |
ad8a48ac JA |
5928 | struct io_timeout_data *data = container_of(timer, |
5929 | struct io_timeout_data, timer); | |
5930 | struct io_kiocb *req = data->req; | |
5931 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5932 | unsigned long flags; |
5933 | ||
89850fce | 5934 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
a71976f3 | 5935 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5936 | atomic_set(&req->ctx->cq_timeouts, |
5937 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
89850fce | 5938 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); |
01cec8c1 | 5939 | |
89850fce JA |
5940 | req->io_task_work.func = io_req_task_timeout; |
5941 | io_req_task_work_add(req); | |
5262f567 JA |
5942 | return HRTIMER_NORESTART; |
5943 | } | |
5944 | ||
fbd15848 PB |
5945 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5946 | __u64 user_data) | |
89850fce | 5947 | __must_hold(&ctx->timeout_lock) |
f254ac04 | 5948 | { |
fbd15848 | 5949 | struct io_timeout_data *io; |
47f46768 | 5950 | struct io_kiocb *req; |
fd9c7bc5 | 5951 | bool found = false; |
f254ac04 | 5952 | |
135fcde8 | 5953 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5954 | found = user_data == req->user_data; |
5955 | if (found) | |
47f46768 | 5956 | break; |
47f46768 | 5957 | } |
fd9c7bc5 PB |
5958 | if (!found) |
5959 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5960 | |
5961 | io = req->async_data; | |
fd9c7bc5 | 5962 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5963 | return ERR_PTR(-EALREADY); |
a71976f3 | 5964 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5965 | return req; |
5966 | } | |
47f46768 | 5967 | |
fbd15848 | 5968 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
ec3c3d0f | 5969 | __must_hold(&ctx->completion_lock) |
89850fce | 5970 | __must_hold(&ctx->timeout_lock) |
fbd15848 PB |
5971 | { |
5972 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5973 | ||
5974 | if (IS_ERR(req)) | |
5975 | return PTR_ERR(req); | |
f254ac04 | 5976 | |
93d2bcd2 | 5977 | req_set_fail(req); |
d4d19c19 | 5978 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
91c2f697 | 5979 | io_put_req_deferred(req); |
f254ac04 JA |
5980 | return 0; |
5981 | } | |
5982 | ||
50c1df2b JA |
5983 | static clockid_t io_timeout_get_clock(struct io_timeout_data *data) |
5984 | { | |
5985 | switch (data->flags & IORING_TIMEOUT_CLOCK_MASK) { | |
5986 | case IORING_TIMEOUT_BOOTTIME: | |
5987 | return CLOCK_BOOTTIME; | |
5988 | case IORING_TIMEOUT_REALTIME: | |
5989 | return CLOCK_REALTIME; | |
5990 | default: | |
5991 | /* can't happen, vetted at prep time */ | |
5992 | WARN_ON_ONCE(1); | |
5993 | fallthrough; | |
5994 | case 0: | |
5995 | return CLOCK_MONOTONIC; | |
5996 | } | |
5997 | } | |
5998 | ||
f1042b6c PB |
5999 | static int io_linked_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
6000 | struct timespec64 *ts, enum hrtimer_mode mode) | |
6001 | __must_hold(&ctx->timeout_lock) | |
6002 | { | |
6003 | struct io_timeout_data *io; | |
6004 | struct io_kiocb *req; | |
6005 | bool found = false; | |
6006 | ||
6007 | list_for_each_entry(req, &ctx->ltimeout_list, timeout.list) { | |
6008 | found = user_data == req->user_data; | |
6009 | if (found) | |
6010 | break; | |
6011 | } | |
6012 | if (!found) | |
6013 | return -ENOENT; | |
6014 | ||
6015 | io = req->async_data; | |
6016 | if (hrtimer_try_to_cancel(&io->timer) == -1) | |
6017 | return -EALREADY; | |
6018 | hrtimer_init(&io->timer, io_timeout_get_clock(io), mode); | |
6019 | io->timer.function = io_link_timeout_fn; | |
6020 | hrtimer_start(&io->timer, timespec64_to_ktime(*ts), mode); | |
6021 | return 0; | |
6022 | } | |
6023 | ||
9c8e11b3 PB |
6024 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
6025 | struct timespec64 *ts, enum hrtimer_mode mode) | |
89850fce | 6026 | __must_hold(&ctx->timeout_lock) |
47f46768 | 6027 | { |
9c8e11b3 PB |
6028 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
6029 | struct io_timeout_data *data; | |
47f46768 | 6030 | |
9c8e11b3 PB |
6031 | if (IS_ERR(req)) |
6032 | return PTR_ERR(req); | |
47f46768 | 6033 | |
9c8e11b3 PB |
6034 | req->timeout.off = 0; /* noseq */ |
6035 | data = req->async_data; | |
6036 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
50c1df2b | 6037 | hrtimer_init(&data->timer, io_timeout_get_clock(data), mode); |
9c8e11b3 PB |
6038 | data->timer.function = io_timeout_fn; |
6039 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
6040 | return 0; | |
47f46768 JA |
6041 | } |
6042 | ||
3529d8c2 JA |
6043 | static int io_timeout_remove_prep(struct io_kiocb *req, |
6044 | const struct io_uring_sqe *sqe) | |
b29472ee | 6045 | { |
9c8e11b3 PB |
6046 | struct io_timeout_rem *tr = &req->timeout_rem; |
6047 | ||
b29472ee JA |
6048 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
6049 | return -EINVAL; | |
61710e43 DA |
6050 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6051 | return -EINVAL; | |
26578cda | 6052 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->splice_fd_in) |
b29472ee JA |
6053 | return -EINVAL; |
6054 | ||
f1042b6c | 6055 | tr->ltimeout = false; |
9c8e11b3 PB |
6056 | tr->addr = READ_ONCE(sqe->addr); |
6057 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
f1042b6c PB |
6058 | if (tr->flags & IORING_TIMEOUT_UPDATE_MASK) { |
6059 | if (hweight32(tr->flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
6060 | return -EINVAL; | |
6061 | if (tr->flags & IORING_LINK_TIMEOUT_UPDATE) | |
6062 | tr->ltimeout = true; | |
6063 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE_MASK|IORING_TIMEOUT_ABS)) | |
9c8e11b3 PB |
6064 | return -EINVAL; |
6065 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
6066 | return -EFAULT; | |
6067 | } else if (tr->flags) { | |
6068 | /* timeout removal doesn't support flags */ | |
b29472ee | 6069 | return -EINVAL; |
9c8e11b3 | 6070 | } |
b29472ee | 6071 | |
b29472ee JA |
6072 | return 0; |
6073 | } | |
6074 | ||
8662daec PB |
6075 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
6076 | { | |
6077 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
6078 | : HRTIMER_MODE_REL; | |
6079 | } | |
6080 | ||
11365043 JA |
6081 | /* |
6082 | * Remove or update an existing timeout command | |
6083 | */ | |
61e98203 | 6084 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 6085 | { |
9c8e11b3 | 6086 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 6087 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 6088 | int ret; |
11365043 | 6089 | |
ec3c3d0f PB |
6090 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { |
6091 | spin_lock(&ctx->completion_lock); | |
6092 | spin_lock_irq(&ctx->timeout_lock); | |
9c8e11b3 | 6093 | ret = io_timeout_cancel(ctx, tr->addr); |
ec3c3d0f PB |
6094 | spin_unlock_irq(&ctx->timeout_lock); |
6095 | spin_unlock(&ctx->completion_lock); | |
6096 | } else { | |
f1042b6c PB |
6097 | enum hrtimer_mode mode = io_translate_timeout_mode(tr->flags); |
6098 | ||
ec3c3d0f | 6099 | spin_lock_irq(&ctx->timeout_lock); |
f1042b6c PB |
6100 | if (tr->ltimeout) |
6101 | ret = io_linked_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
6102 | else | |
6103 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
ec3c3d0f PB |
6104 | spin_unlock_irq(&ctx->timeout_lock); |
6105 | } | |
11365043 | 6106 | |
4e88d6e7 | 6107 | if (ret < 0) |
93d2bcd2 | 6108 | req_set_fail(req); |
505657bc | 6109 | io_req_complete_post(req, ret, 0); |
11365043 | 6110 | return 0; |
5262f567 JA |
6111 | } |
6112 | ||
3529d8c2 | 6113 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 6114 | bool is_timeout_link) |
5262f567 | 6115 | { |
ad8a48ac | 6116 | struct io_timeout_data *data; |
a41525ab | 6117 | unsigned flags; |
56080b02 | 6118 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 6119 | |
ad8a48ac | 6120 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 6121 | return -EINVAL; |
26578cda PB |
6122 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1 || |
6123 | sqe->splice_fd_in) | |
a41525ab | 6124 | return -EINVAL; |
56080b02 | 6125 | if (off && is_timeout_link) |
2d28390a | 6126 | return -EINVAL; |
a41525ab | 6127 | flags = READ_ONCE(sqe->timeout_flags); |
50c1df2b JA |
6128 | if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_CLOCK_MASK)) |
6129 | return -EINVAL; | |
6130 | /* more than one clock specified is invalid, obviously */ | |
6131 | if (hweight32(flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
5262f567 | 6132 | return -EINVAL; |
bdf20073 | 6133 | |
ef9dd637 | 6134 | INIT_LIST_HEAD(&req->timeout.list); |
bfe68a22 | 6135 | req->timeout.off = off; |
f18ee4cf PB |
6136 | if (unlikely(off && !req->ctx->off_timeout_used)) |
6137 | req->ctx->off_timeout_used = true; | |
26a61679 | 6138 | |
e8c2bc1f | 6139 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
6140 | return -ENOMEM; |
6141 | ||
e8c2bc1f | 6142 | data = req->async_data; |
ad8a48ac | 6143 | data->req = req; |
50c1df2b | 6144 | data->flags = flags; |
ad8a48ac JA |
6145 | |
6146 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
6147 | return -EFAULT; |
6148 | ||
8662daec | 6149 | data->mode = io_translate_timeout_mode(flags); |
50c1df2b | 6150 | hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode); |
b97e736a PB |
6151 | |
6152 | if (is_timeout_link) { | |
6153 | struct io_submit_link *link = &req->ctx->submit_state.link; | |
6154 | ||
6155 | if (!link->head) | |
6156 | return -EINVAL; | |
6157 | if (link->last->opcode == IORING_OP_LINK_TIMEOUT) | |
6158 | return -EINVAL; | |
4d13d1a4 PB |
6159 | req->timeout.head = link->last; |
6160 | link->last->flags |= REQ_F_ARM_LTIMEOUT; | |
b97e736a | 6161 | } |
ad8a48ac JA |
6162 | return 0; |
6163 | } | |
6164 | ||
61e98203 | 6165 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 6166 | { |
ad8a48ac | 6167 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 6168 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 6169 | struct list_head *entry; |
bfe68a22 | 6170 | u32 tail, off = req->timeout.off; |
ad8a48ac | 6171 | |
89850fce | 6172 | spin_lock_irq(&ctx->timeout_lock); |
93bd25bb | 6173 | |
5262f567 JA |
6174 | /* |
6175 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
6176 | * timeout event to be satisfied. If it isn't set, then this is |
6177 | * a pure timeout request, sequence isn't used. | |
5262f567 | 6178 | */ |
8eb7e2d0 | 6179 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
6180 | entry = ctx->timeout_list.prev; |
6181 | goto add; | |
6182 | } | |
5262f567 | 6183 | |
bfe68a22 PB |
6184 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
6185 | req->timeout.target_seq = tail + off; | |
5262f567 | 6186 | |
f010505b MDG |
6187 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
6188 | * This is safe because ->completion_lock is held, and submissions | |
6189 | * and completions are never mixed in the same ->completion_lock section. | |
6190 | */ | |
6191 | ctx->cq_last_tm_flush = tail; | |
6192 | ||
5262f567 JA |
6193 | /* |
6194 | * Insertion sort, ensuring the first entry in the list is always | |
6195 | * the one we need first. | |
6196 | */ | |
5262f567 | 6197 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
6198 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
6199 | timeout.list); | |
5262f567 | 6200 | |
8eb7e2d0 | 6201 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 6202 | continue; |
bfe68a22 PB |
6203 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
6204 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
6205 | break; |
6206 | } | |
93bd25bb | 6207 | add: |
135fcde8 | 6208 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
6209 | data->timer.function = io_timeout_fn; |
6210 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
89850fce | 6211 | spin_unlock_irq(&ctx->timeout_lock); |
5262f567 JA |
6212 | return 0; |
6213 | } | |
5262f567 | 6214 | |
f458dd84 PB |
6215 | struct io_cancel_data { |
6216 | struct io_ring_ctx *ctx; | |
6217 | u64 user_data; | |
6218 | }; | |
6219 | ||
62755e35 JA |
6220 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
6221 | { | |
6222 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 6223 | struct io_cancel_data *cd = data; |
62755e35 | 6224 | |
f458dd84 | 6225 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
6226 | } |
6227 | ||
f458dd84 PB |
6228 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
6229 | struct io_ring_ctx *ctx) | |
62755e35 | 6230 | { |
f458dd84 | 6231 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 6232 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
6233 | int ret = 0; |
6234 | ||
f458dd84 | 6235 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
6236 | return -ENOENT; |
6237 | ||
f458dd84 | 6238 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
6239 | switch (cancel_ret) { |
6240 | case IO_WQ_CANCEL_OK: | |
6241 | ret = 0; | |
6242 | break; | |
6243 | case IO_WQ_CANCEL_RUNNING: | |
6244 | ret = -EALREADY; | |
6245 | break; | |
6246 | case IO_WQ_CANCEL_NOTFOUND: | |
6247 | ret = -ENOENT; | |
6248 | break; | |
6249 | } | |
6250 | ||
e977d6d3 JA |
6251 | return ret; |
6252 | } | |
6253 | ||
8cb01fac | 6254 | static int io_try_cancel_userdata(struct io_kiocb *req, u64 sqe_addr) |
47f46768 | 6255 | { |
8cb01fac | 6256 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 JA |
6257 | int ret; |
6258 | ||
dadebc35 | 6259 | WARN_ON_ONCE(!io_wq_current_is_worker() && req->task != current); |
8cb01fac | 6260 | |
f458dd84 | 6261 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
df9727af | 6262 | if (ret != -ENOENT) |
8cb01fac | 6263 | return ret; |
505657bc PB |
6264 | |
6265 | spin_lock(&ctx->completion_lock); | |
79ebeaee | 6266 | spin_lock_irq(&ctx->timeout_lock); |
47f46768 | 6267 | ret = io_timeout_cancel(ctx, sqe_addr); |
79ebeaee | 6268 | spin_unlock_irq(&ctx->timeout_lock); |
47f46768 | 6269 | if (ret != -ENOENT) |
505657bc PB |
6270 | goto out; |
6271 | ret = io_poll_cancel(ctx, sqe_addr, false); | |
6272 | out: | |
6273 | spin_unlock(&ctx->completion_lock); | |
6274 | return ret; | |
47f46768 JA |
6275 | } |
6276 | ||
3529d8c2 JA |
6277 | static int io_async_cancel_prep(struct io_kiocb *req, |
6278 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6279 | { |
fbf23849 | 6280 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6281 | return -EINVAL; |
61710e43 DA |
6282 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6283 | return -EINVAL; | |
26578cda PB |
6284 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags || |
6285 | sqe->splice_fd_in) | |
e977d6d3 JA |
6286 | return -EINVAL; |
6287 | ||
fbf23849 JA |
6288 | req->cancel.addr = READ_ONCE(sqe->addr); |
6289 | return 0; | |
6290 | } | |
6291 | ||
61e98203 | 6292 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6293 | { |
6294 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
6295 | u64 sqe_addr = req->cancel.addr; |
6296 | struct io_tctx_node *node; | |
6297 | int ret; | |
6298 | ||
8cb01fac | 6299 | ret = io_try_cancel_userdata(req, sqe_addr); |
58f99373 PB |
6300 | if (ret != -ENOENT) |
6301 | goto done; | |
58f99373 PB |
6302 | |
6303 | /* slow path, try all io-wq's */ | |
6304 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
6305 | ret = -ENOENT; | |
6306 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
6307 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 6308 | |
58f99373 PB |
6309 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
6310 | if (ret != -ENOENT) | |
6311 | break; | |
6312 | } | |
6313 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
58f99373 | 6314 | done: |
58f99373 | 6315 | if (ret < 0) |
93d2bcd2 | 6316 | req_set_fail(req); |
505657bc | 6317 | io_req_complete_post(req, ret, 0); |
5262f567 JA |
6318 | return 0; |
6319 | } | |
6320 | ||
269bbe5f | 6321 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6322 | const struct io_uring_sqe *sqe) |
6323 | { | |
61710e43 DA |
6324 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6325 | return -EINVAL; | |
26578cda | 6326 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
05f3fb3c JA |
6327 | return -EINVAL; |
6328 | ||
269bbe5f BM |
6329 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6330 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6331 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6332 | return -EINVAL; |
269bbe5f | 6333 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6334 | return 0; |
6335 | } | |
6336 | ||
889fca73 | 6337 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6338 | { |
6339 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 6340 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 6341 | int ret; |
fbf23849 | 6342 | |
45d189c6 | 6343 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 6344 | return -EAGAIN; |
05f3fb3c | 6345 | |
269bbe5f BM |
6346 | up.offset = req->rsrc_update.offset; |
6347 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
6348 | up.nr = 0; |
6349 | up.tags = 0; | |
615cee49 | 6350 | up.resv = 0; |
05f3fb3c JA |
6351 | |
6352 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 6353 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 6354 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
6355 | mutex_unlock(&ctx->uring_lock); |
6356 | ||
6357 | if (ret < 0) | |
93d2bcd2 | 6358 | req_set_fail(req); |
889fca73 | 6359 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6360 | return 0; |
6361 | } | |
6362 | ||
bfe76559 | 6363 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6364 | { |
d625c6ee | 6365 | switch (req->opcode) { |
e781573e | 6366 | case IORING_OP_NOP: |
bfe76559 | 6367 | return 0; |
f67676d1 JA |
6368 | case IORING_OP_READV: |
6369 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6370 | case IORING_OP_READ: |
bfe76559 | 6371 | return io_read_prep(req, sqe); |
f67676d1 JA |
6372 | case IORING_OP_WRITEV: |
6373 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6374 | case IORING_OP_WRITE: |
bfe76559 | 6375 | return io_write_prep(req, sqe); |
0969e783 | 6376 | case IORING_OP_POLL_ADD: |
bfe76559 | 6377 | return io_poll_add_prep(req, sqe); |
0969e783 | 6378 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 6379 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 6380 | case IORING_OP_FSYNC: |
1155c76a | 6381 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6382 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6383 | return io_sfr_prep(req, sqe); |
03b1230c | 6384 | case IORING_OP_SENDMSG: |
fddaface | 6385 | case IORING_OP_SEND: |
bfe76559 | 6386 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6387 | case IORING_OP_RECVMSG: |
fddaface | 6388 | case IORING_OP_RECV: |
bfe76559 | 6389 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6390 | case IORING_OP_CONNECT: |
bfe76559 | 6391 | return io_connect_prep(req, sqe); |
2d28390a | 6392 | case IORING_OP_TIMEOUT: |
bfe76559 | 6393 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6394 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6395 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6396 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6397 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6398 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6399 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6400 | case IORING_OP_ACCEPT: |
bfe76559 | 6401 | return io_accept_prep(req, sqe); |
d63d1b5e | 6402 | case IORING_OP_FALLOCATE: |
bfe76559 | 6403 | return io_fallocate_prep(req, sqe); |
15b71abe | 6404 | case IORING_OP_OPENAT: |
bfe76559 | 6405 | return io_openat_prep(req, sqe); |
b5dba59e | 6406 | case IORING_OP_CLOSE: |
bfe76559 | 6407 | return io_close_prep(req, sqe); |
05f3fb3c | 6408 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6409 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6410 | case IORING_OP_STATX: |
bfe76559 | 6411 | return io_statx_prep(req, sqe); |
4840e418 | 6412 | case IORING_OP_FADVISE: |
bfe76559 | 6413 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6414 | case IORING_OP_MADVISE: |
bfe76559 | 6415 | return io_madvise_prep(req, sqe); |
cebdb986 | 6416 | case IORING_OP_OPENAT2: |
bfe76559 | 6417 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6418 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6419 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6420 | case IORING_OP_SPLICE: |
bfe76559 | 6421 | return io_splice_prep(req, sqe); |
ddf0322d | 6422 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6423 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6424 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6425 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6426 | case IORING_OP_TEE: |
bfe76559 | 6427 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6428 | case IORING_OP_SHUTDOWN: |
6429 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6430 | case IORING_OP_RENAMEAT: |
6431 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6432 | case IORING_OP_UNLINKAT: |
6433 | return io_unlinkat_prep(req, sqe); | |
e34a02dc DK |
6434 | case IORING_OP_MKDIRAT: |
6435 | return io_mkdirat_prep(req, sqe); | |
7a8721f8 DK |
6436 | case IORING_OP_SYMLINKAT: |
6437 | return io_symlinkat_prep(req, sqe); | |
cf30da90 DK |
6438 | case IORING_OP_LINKAT: |
6439 | return io_linkat_prep(req, sqe); | |
f67676d1 JA |
6440 | } |
6441 | ||
bfe76559 PB |
6442 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6443 | req->opcode); | |
bd54b6fe | 6444 | return -EINVAL; |
bfe76559 PB |
6445 | } |
6446 | ||
93642ef8 | 6447 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 6448 | { |
b7e298d2 PB |
6449 | if (!io_op_defs[req->opcode].needs_async_setup) |
6450 | return 0; | |
6451 | if (WARN_ON_ONCE(req->async_data)) | |
6452 | return -EFAULT; | |
6453 | if (io_alloc_async_data(req)) | |
6454 | return -EAGAIN; | |
6455 | ||
93642ef8 PB |
6456 | switch (req->opcode) { |
6457 | case IORING_OP_READV: | |
93642ef8 PB |
6458 | return io_rw_prep_async(req, READ); |
6459 | case IORING_OP_WRITEV: | |
93642ef8 PB |
6460 | return io_rw_prep_async(req, WRITE); |
6461 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
6462 | return io_sendmsg_prep_async(req); |
6463 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
6464 | return io_recvmsg_prep_async(req); |
6465 | case IORING_OP_CONNECT: | |
6466 | return io_connect_prep_async(req); | |
6467 | } | |
b7e298d2 PB |
6468 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
6469 | req->opcode); | |
6470 | return -EFAULT; | |
f67676d1 JA |
6471 | } |
6472 | ||
9cf7c104 PB |
6473 | static u32 io_get_sequence(struct io_kiocb *req) |
6474 | { | |
a3dbdf54 | 6475 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 6476 | |
a3dbdf54 PB |
6477 | /* need original cached_sq_head, but it was increased for each req */ |
6478 | io_for_each_link(req, req) | |
6479 | seq--; | |
6480 | return seq; | |
9cf7c104 PB |
6481 | } |
6482 | ||
76cc33d7 | 6483 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 6484 | { |
3c19966d | 6485 | struct io_kiocb *pos; |
a197f664 | 6486 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6487 | struct io_defer_entry *de; |
f67676d1 | 6488 | int ret; |
9cf7c104 | 6489 | u32 seq; |
de0617e4 | 6490 | |
b8ce1b9d PB |
6491 | if (req->flags & REQ_F_FAIL) { |
6492 | io_req_complete_fail_submit(req); | |
6493 | return true; | |
6494 | } | |
6495 | ||
3c19966d PB |
6496 | /* |
6497 | * If we need to drain a request in the middle of a link, drain the | |
6498 | * head request and the next request/link after the current link. | |
6499 | * Considering sequential execution of links, IOSQE_IO_DRAIN will be | |
6500 | * maintained for every request of our link. | |
6501 | */ | |
6502 | if (ctx->drain_next) { | |
6503 | req->flags |= REQ_F_IO_DRAIN; | |
6504 | ctx->drain_next = false; | |
6505 | } | |
6506 | /* not interested in head, start from the first linked */ | |
6507 | io_for_each_link(pos, req->link) { | |
6508 | if (pos->flags & REQ_F_IO_DRAIN) { | |
6509 | ctx->drain_next = true; | |
6510 | req->flags |= REQ_F_IO_DRAIN; | |
6511 | break; | |
6512 | } | |
6513 | } | |
6514 | ||
9d858b21 | 6515 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 | 6516 | if (likely(list_empty_careful(&ctx->defer_list) && |
10c66904 PB |
6517 | !(req->flags & REQ_F_IO_DRAIN))) { |
6518 | ctx->drain_active = false; | |
76cc33d7 | 6519 | return false; |
10c66904 | 6520 | } |
9cf7c104 PB |
6521 | |
6522 | seq = io_get_sequence(req); | |
6523 | /* Still a chance to pass the sequence check */ | |
6524 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6525 | return false; |
de0617e4 | 6526 | |
b7e298d2 | 6527 | ret = io_req_prep_async(req); |
be7053b7 | 6528 | if (ret) |
1b48773f | 6529 | goto fail; |
cbdcb435 | 6530 | io_prep_async_link(req); |
27dc8338 | 6531 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 6532 | if (!de) { |
1b48773f PB |
6533 | ret = -ENOMEM; |
6534 | fail: | |
6535 | io_req_complete_failed(req, ret); | |
76cc33d7 PB |
6536 | return true; |
6537 | } | |
2d28390a | 6538 | |
79ebeaee | 6539 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 6540 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
79ebeaee | 6541 | spin_unlock(&ctx->completion_lock); |
27dc8338 | 6542 | kfree(de); |
f237c30a | 6543 | io_queue_async_work(req, NULL); |
76cc33d7 | 6544 | return true; |
de0617e4 JA |
6545 | } |
6546 | ||
915967f6 | 6547 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6548 | de->req = req; |
9cf7c104 | 6549 | de->seq = seq; |
27dc8338 | 6550 | list_add_tail(&de->list, &ctx->defer_list); |
79ebeaee | 6551 | spin_unlock(&ctx->completion_lock); |
76cc33d7 | 6552 | return true; |
de0617e4 JA |
6553 | } |
6554 | ||
68fb8979 | 6555 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6556 | { |
0e1b6fe3 PB |
6557 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6558 | switch (req->opcode) { | |
6559 | case IORING_OP_READV: | |
6560 | case IORING_OP_READ_FIXED: | |
6561 | case IORING_OP_READ: | |
bcda7baa | 6562 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6563 | break; |
6564 | case IORING_OP_RECVMSG: | |
6565 | case IORING_OP_RECV: | |
bcda7baa | 6566 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6567 | break; |
6568 | } | |
99bc4c38 PB |
6569 | } |
6570 | ||
0e1b6fe3 PB |
6571 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6572 | switch (req->opcode) { | |
6573 | case IORING_OP_READV: | |
6574 | case IORING_OP_READ_FIXED: | |
6575 | case IORING_OP_READ: | |
6576 | case IORING_OP_WRITEV: | |
6577 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6578 | case IORING_OP_WRITE: { |
6579 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
6580 | |
6581 | kfree(io->free_iovec); | |
0e1b6fe3 | 6582 | break; |
e8c2bc1f | 6583 | } |
0e1b6fe3 | 6584 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6585 | case IORING_OP_SENDMSG: { |
6586 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6587 | |
6588 | kfree(io->free_iov); | |
0e1b6fe3 | 6589 | break; |
e8c2bc1f | 6590 | } |
0e1b6fe3 PB |
6591 | case IORING_OP_SPLICE: |
6592 | case IORING_OP_TEE: | |
e1d767f0 PB |
6593 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6594 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6595 | break; |
f3cd4850 JA |
6596 | case IORING_OP_OPENAT: |
6597 | case IORING_OP_OPENAT2: | |
6598 | if (req->open.filename) | |
6599 | putname(req->open.filename); | |
6600 | break; | |
80a261fd JA |
6601 | case IORING_OP_RENAMEAT: |
6602 | putname(req->rename.oldpath); | |
6603 | putname(req->rename.newpath); | |
6604 | break; | |
14a1143b JA |
6605 | case IORING_OP_UNLINKAT: |
6606 | putname(req->unlink.filename); | |
6607 | break; | |
e34a02dc DK |
6608 | case IORING_OP_MKDIRAT: |
6609 | putname(req->mkdir.filename); | |
6610 | break; | |
7a8721f8 DK |
6611 | case IORING_OP_SYMLINKAT: |
6612 | putname(req->symlink.oldpath); | |
6613 | putname(req->symlink.newpath); | |
6614 | break; | |
cf30da90 DK |
6615 | case IORING_OP_LINKAT: |
6616 | putname(req->hardlink.oldpath); | |
6617 | putname(req->hardlink.newpath); | |
6618 | break; | |
0e1b6fe3 | 6619 | } |
99bc4c38 | 6620 | } |
75652a30 JA |
6621 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6622 | kfree(req->apoll->double_poll); | |
6623 | kfree(req->apoll); | |
6624 | req->apoll = NULL; | |
6625 | } | |
3a0a6902 PB |
6626 | if (req->flags & REQ_F_INFLIGHT) { |
6627 | struct io_uring_task *tctx = req->task->io_uring; | |
6628 | ||
6629 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 6630 | } |
c854357b | 6631 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 6632 | put_cred(req->creds); |
c854357b PB |
6633 | |
6634 | req->flags &= ~IO_REQ_CLEAN_FLAGS; | |
99bc4c38 PB |
6635 | } |
6636 | ||
889fca73 | 6637 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6638 | { |
a197f664 | 6639 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6640 | const struct cred *creds = NULL; |
d625c6ee | 6641 | int ret; |
2b188cc1 | 6642 | |
b8e64b53 | 6643 | if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) |
c10d1f98 | 6644 | creds = override_creds(req->creds); |
5730b27e | 6645 | |
d625c6ee | 6646 | switch (req->opcode) { |
2b188cc1 | 6647 | case IORING_OP_NOP: |
889fca73 | 6648 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6649 | break; |
6650 | case IORING_OP_READV: | |
edafccee | 6651 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6652 | case IORING_OP_READ: |
889fca73 | 6653 | ret = io_read(req, issue_flags); |
edafccee | 6654 | break; |
3529d8c2 | 6655 | case IORING_OP_WRITEV: |
edafccee | 6656 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6657 | case IORING_OP_WRITE: |
889fca73 | 6658 | ret = io_write(req, issue_flags); |
2b188cc1 | 6659 | break; |
c992fe29 | 6660 | case IORING_OP_FSYNC: |
45d189c6 | 6661 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6662 | break; |
221c5eb2 | 6663 | case IORING_OP_POLL_ADD: |
61e98203 | 6664 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6665 | break; |
6666 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6667 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6668 | break; |
5d17b4a4 | 6669 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6670 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6671 | break; |
0fa03c62 | 6672 | case IORING_OP_SENDMSG: |
889fca73 | 6673 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6674 | break; |
fddaface | 6675 | case IORING_OP_SEND: |
889fca73 | 6676 | ret = io_send(req, issue_flags); |
0fa03c62 | 6677 | break; |
aa1fa28f | 6678 | case IORING_OP_RECVMSG: |
889fca73 | 6679 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6680 | break; |
fddaface | 6681 | case IORING_OP_RECV: |
889fca73 | 6682 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6683 | break; |
5262f567 | 6684 | case IORING_OP_TIMEOUT: |
61e98203 | 6685 | ret = io_timeout(req, issue_flags); |
5262f567 | 6686 | break; |
11365043 | 6687 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6688 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6689 | break; |
17f2fe35 | 6690 | case IORING_OP_ACCEPT: |
889fca73 | 6691 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6692 | break; |
f8e85cf2 | 6693 | case IORING_OP_CONNECT: |
889fca73 | 6694 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6695 | break; |
62755e35 | 6696 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6697 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6698 | break; |
d63d1b5e | 6699 | case IORING_OP_FALLOCATE: |
45d189c6 | 6700 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6701 | break; |
15b71abe | 6702 | case IORING_OP_OPENAT: |
45d189c6 | 6703 | ret = io_openat(req, issue_flags); |
15b71abe | 6704 | break; |
b5dba59e | 6705 | case IORING_OP_CLOSE: |
889fca73 | 6706 | ret = io_close(req, issue_flags); |
b5dba59e | 6707 | break; |
05f3fb3c | 6708 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6709 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6710 | break; |
eddc7ef5 | 6711 | case IORING_OP_STATX: |
45d189c6 | 6712 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6713 | break; |
4840e418 | 6714 | case IORING_OP_FADVISE: |
45d189c6 | 6715 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6716 | break; |
c1ca757b | 6717 | case IORING_OP_MADVISE: |
45d189c6 | 6718 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6719 | break; |
cebdb986 | 6720 | case IORING_OP_OPENAT2: |
45d189c6 | 6721 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6722 | break; |
3e4827b0 | 6723 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6724 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6725 | break; |
7d67af2c | 6726 | case IORING_OP_SPLICE: |
45d189c6 | 6727 | ret = io_splice(req, issue_flags); |
7d67af2c | 6728 | break; |
ddf0322d | 6729 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6730 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6731 | break; |
067524e9 | 6732 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6733 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6734 | break; |
f2a8d5c7 | 6735 | case IORING_OP_TEE: |
45d189c6 | 6736 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6737 | break; |
36f4fa68 | 6738 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6739 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6740 | break; |
80a261fd | 6741 | case IORING_OP_RENAMEAT: |
45d189c6 | 6742 | ret = io_renameat(req, issue_flags); |
80a261fd | 6743 | break; |
14a1143b | 6744 | case IORING_OP_UNLINKAT: |
45d189c6 | 6745 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6746 | break; |
e34a02dc DK |
6747 | case IORING_OP_MKDIRAT: |
6748 | ret = io_mkdirat(req, issue_flags); | |
6749 | break; | |
7a8721f8 DK |
6750 | case IORING_OP_SYMLINKAT: |
6751 | ret = io_symlinkat(req, issue_flags); | |
6752 | break; | |
cf30da90 DK |
6753 | case IORING_OP_LINKAT: |
6754 | ret = io_linkat(req, issue_flags); | |
6755 | break; | |
2b188cc1 JA |
6756 | default: |
6757 | ret = -EINVAL; | |
6758 | break; | |
6759 | } | |
6760 | ||
5730b27e JA |
6761 | if (creds) |
6762 | revert_creds(creds); | |
def596e9 JA |
6763 | if (ret) |
6764 | return ret; | |
b532576e | 6765 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6766 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6767 | io_iopoll_req_issued(req); | |
def596e9 JA |
6768 | |
6769 | return 0; | |
2b188cc1 JA |
6770 | } |
6771 | ||
ebc11b6c PB |
6772 | static struct io_wq_work *io_wq_free_work(struct io_wq_work *work) |
6773 | { | |
6774 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6775 | ||
6776 | req = io_put_req_find_next(req); | |
6777 | return req ? &req->work : NULL; | |
6778 | } | |
6779 | ||
5280f7e5 | 6780 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6781 | { |
6782 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6783 | struct io_kiocb *timeout; |
561fb04a | 6784 | int ret = 0; |
2b188cc1 | 6785 | |
48dcd38d PB |
6786 | /* one will be dropped by ->io_free_work() after returning to io-wq */ |
6787 | if (!(req->flags & REQ_F_REFCOUNT)) | |
6788 | __io_req_set_refcount(req, 2); | |
6789 | else | |
6790 | req_ref_get(req); | |
5d5901a3 | 6791 | |
6df1db6b PB |
6792 | timeout = io_prep_linked_timeout(req); |
6793 | if (timeout) | |
6794 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6795 | |
dadebc35 | 6796 | /* either cancelled or io-wq is dying, so don't touch tctx->iowq */ |
4014d943 | 6797 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6798 | ret = -ECANCELED; |
31b51510 | 6799 | |
561fb04a | 6800 | if (!ret) { |
561fb04a | 6801 | do { |
889fca73 | 6802 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6803 | /* |
6804 | * We can get EAGAIN for polled IO even though we're | |
6805 | * forcing a sync submission from here, since we can't | |
6806 | * wait for request slots on the block side. | |
6807 | */ | |
6808 | if (ret != -EAGAIN) | |
6809 | break; | |
6810 | cond_resched(); | |
6811 | } while (1); | |
6812 | } | |
31b51510 | 6813 | |
a3df7698 | 6814 | /* avoid locking problems by failing it from a clean context */ |
5d5901a3 | 6815 | if (ret) |
a3df7698 | 6816 | io_req_task_queue_fail(req, ret); |
2b188cc1 JA |
6817 | } |
6818 | ||
aeca241b | 6819 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
042b0d85 | 6820 | unsigned i) |
65e19f54 | 6821 | { |
042b0d85 | 6822 | return &table->files[i]; |
dafecf19 PB |
6823 | } |
6824 | ||
65e19f54 JA |
6825 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6826 | int index) | |
6827 | { | |
aeca241b | 6828 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6829 | |
a04b0ac0 | 6830 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6831 | } |
6832 | ||
a04b0ac0 | 6833 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6834 | { |
6835 | unsigned long file_ptr = (unsigned long) file; | |
6836 | ||
b191e2df | 6837 | if (__io_file_supports_nowait(file, READ)) |
9a321c98 | 6838 | file_ptr |= FFS_ASYNC_READ; |
b191e2df | 6839 | if (__io_file_supports_nowait(file, WRITE)) |
9a321c98 PB |
6840 | file_ptr |= FFS_ASYNC_WRITE; |
6841 | if (S_ISREG(file_inode(file)->i_mode)) | |
6842 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6843 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6844 | } |
6845 | ||
ac177053 PB |
6846 | static inline struct file *io_file_get_fixed(struct io_ring_ctx *ctx, |
6847 | struct io_kiocb *req, int fd) | |
09bb8394 | 6848 | { |
8da11c19 | 6849 | struct file *file; |
ac177053 | 6850 | unsigned long file_ptr; |
09bb8394 | 6851 | |
ac177053 PB |
6852 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
6853 | return NULL; | |
6854 | fd = array_index_nospec(fd, ctx->nr_user_files); | |
6855 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; | |
6856 | file = (struct file *) (file_ptr & FFS_MASK); | |
6857 | file_ptr &= ~FFS_MASK; | |
6858 | /* mask in overlapping REQ_F and FFS bits */ | |
b191e2df | 6859 | req->flags |= (file_ptr << REQ_F_NOWAIT_READ_BIT); |
ac177053 PB |
6860 | io_req_set_rsrc_node(req); |
6861 | return file; | |
6862 | } | |
d44f554e | 6863 | |
ac177053 | 6864 | static struct file *io_file_get_normal(struct io_ring_ctx *ctx, |
ac177053 PB |
6865 | struct io_kiocb *req, int fd) |
6866 | { | |
62906e89 | 6867 | struct file *file = fget(fd); |
ac177053 PB |
6868 | |
6869 | trace_io_uring_file_get(ctx, fd); | |
09bb8394 | 6870 | |
ac177053 PB |
6871 | /* we don't allow fixed io_uring files */ |
6872 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6873 | io_req_track_inflight(req); | |
8371adf5 | 6874 | return file; |
09bb8394 JA |
6875 | } |
6876 | ||
ac177053 | 6877 | static inline struct file *io_file_get(struct io_ring_ctx *ctx, |
ac177053 PB |
6878 | struct io_kiocb *req, int fd, bool fixed) |
6879 | { | |
6880 | if (fixed) | |
6881 | return io_file_get_fixed(ctx, req, fd); | |
6882 | else | |
62906e89 | 6883 | return io_file_get_normal(ctx, req, fd); |
ac177053 PB |
6884 | } |
6885 | ||
f237c30a | 6886 | static void io_req_task_link_timeout(struct io_kiocb *req, bool *locked) |
89b263f6 JA |
6887 | { |
6888 | struct io_kiocb *prev = req->timeout.prev; | |
8cb01fac | 6889 | int ret; |
89b263f6 JA |
6890 | |
6891 | if (prev) { | |
8cb01fac | 6892 | ret = io_try_cancel_userdata(req, prev->user_data); |
505657bc | 6893 | io_req_complete_post(req, ret ?: -ETIME, 0); |
89b263f6 | 6894 | io_put_req(prev); |
89b263f6 JA |
6895 | } else { |
6896 | io_req_complete_post(req, -ETIME, 0); | |
6897 | } | |
6898 | } | |
6899 | ||
2665abfd | 6900 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6901 | { |
ad8a48ac JA |
6902 | struct io_timeout_data *data = container_of(timer, |
6903 | struct io_timeout_data, timer); | |
90cd7e42 | 6904 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6905 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6906 | unsigned long flags; |
2665abfd | 6907 | |
89b263f6 | 6908 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
90cd7e42 PB |
6909 | prev = req->timeout.head; |
6910 | req->timeout.head = NULL; | |
2665abfd JA |
6911 | |
6912 | /* | |
6913 | * We don't expect the list to be empty, that will only happen if we | |
6914 | * race with the completion of the linked work. | |
6915 | */ | |
447c19f3 | 6916 | if (prev) { |
f2f87370 | 6917 | io_remove_next_linked(prev); |
447c19f3 PB |
6918 | if (!req_ref_inc_not_zero(prev)) |
6919 | prev = NULL; | |
6920 | } | |
ef9dd637 | 6921 | list_del(&req->timeout.list); |
89b263f6 JA |
6922 | req->timeout.prev = prev; |
6923 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); | |
2665abfd | 6924 | |
89b263f6 JA |
6925 | req->io_task_work.func = io_req_task_link_timeout; |
6926 | io_req_task_work_add(req); | |
2665abfd JA |
6927 | return HRTIMER_NORESTART; |
6928 | } | |
6929 | ||
de968c18 | 6930 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6931 | { |
de968c18 PB |
6932 | struct io_ring_ctx *ctx = req->ctx; |
6933 | ||
89b263f6 | 6934 | spin_lock_irq(&ctx->timeout_lock); |
76a46e06 | 6935 | /* |
f2f87370 PB |
6936 | * If the back reference is NULL, then our linked request finished |
6937 | * before we got a chance to setup the timer | |
76a46e06 | 6938 | */ |
90cd7e42 | 6939 | if (req->timeout.head) { |
e8c2bc1f | 6940 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6941 | |
ad8a48ac JA |
6942 | data->timer.function = io_link_timeout_fn; |
6943 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6944 | data->mode); | |
ef9dd637 | 6945 | list_add_tail(&req->timeout.list, &ctx->ltimeout_list); |
2665abfd | 6946 | } |
89b263f6 | 6947 | spin_unlock_irq(&ctx->timeout_lock); |
2665abfd | 6948 | /* drop submission reference */ |
76a46e06 JA |
6949 | io_put_req(req); |
6950 | } | |
2665abfd | 6951 | |
c5eef2b9 | 6952 | static void __io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 6953 | __must_hold(&req->ctx->uring_lock) |
2b188cc1 | 6954 | { |
906c6caa | 6955 | struct io_kiocb *linked_timeout; |
e0c5c576 | 6956 | int ret; |
2b188cc1 | 6957 | |
59b735ae | 6958 | issue_sqe: |
c5eef2b9 | 6959 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6960 | |
491381ce JA |
6961 | /* |
6962 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6963 | * doesn't support non-blocking read/write attempts | |
6964 | */ | |
1840038e | 6965 | if (likely(!ret)) { |
e342c807 | 6966 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 | 6967 | struct io_ring_ctx *ctx = req->ctx; |
cd0ca2e0 | 6968 | struct io_submit_state *state = &ctx->submit_state; |
e65ef56d | 6969 | |
cd0ca2e0 PB |
6970 | state->compl_reqs[state->compl_nr++] = req; |
6971 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2a2758f2 | 6972 | io_submit_flush_completions(ctx); |
906c6caa | 6973 | return; |
0d63c148 | 6974 | } |
906c6caa PB |
6975 | |
6976 | linked_timeout = io_prep_linked_timeout(req); | |
6977 | if (linked_timeout) | |
6978 | io_queue_linked_timeout(linked_timeout); | |
1840038e | 6979 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
906c6caa PB |
6980 | linked_timeout = io_prep_linked_timeout(req); |
6981 | ||
59b735ae OL |
6982 | switch (io_arm_poll_handler(req)) { |
6983 | case IO_APOLL_READY: | |
906c6caa PB |
6984 | if (linked_timeout) |
6985 | io_unprep_linked_timeout(req); | |
59b735ae OL |
6986 | goto issue_sqe; |
6987 | case IO_APOLL_ABORTED: | |
1840038e PB |
6988 | /* |
6989 | * Queued up for async execution, worker will release | |
6990 | * submit reference when the iocb is actually submitted. | |
6991 | */ | |
f237c30a | 6992 | io_queue_async_work(req, NULL); |
59b735ae | 6993 | break; |
1840038e | 6994 | } |
906c6caa PB |
6995 | |
6996 | if (linked_timeout) | |
6997 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 | 6998 | } else { |
f41db273 | 6999 | io_req_complete_failed(req, ret); |
9e645e11 | 7000 | } |
2b188cc1 JA |
7001 | } |
7002 | ||
441b8a78 | 7003 | static inline void io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 7004 | __must_hold(&req->ctx->uring_lock) |
4fe2c963 | 7005 | { |
10c66904 | 7006 | if (unlikely(req->ctx->drain_active) && io_drain_req(req)) |
76cc33d7 | 7007 | return; |
4fe2c963 | 7008 | |
a8295b98 | 7009 | if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)))) { |
c5eef2b9 | 7010 | __io_queue_sqe(req); |
a8295b98 | 7011 | } else if (req->flags & REQ_F_FAIL) { |
c6d3d9cb | 7012 | io_req_complete_fail_submit(req); |
76cc33d7 PB |
7013 | } else { |
7014 | int ret = io_req_prep_async(req); | |
7015 | ||
7016 | if (unlikely(ret)) | |
7017 | io_req_complete_failed(req, ret); | |
7018 | else | |
f237c30a | 7019 | io_queue_async_work(req, NULL); |
ce35a47a | 7020 | } |
4fe2c963 JL |
7021 | } |
7022 | ||
b16fed66 PB |
7023 | /* |
7024 | * Check SQE restrictions (opcode and flags). | |
7025 | * | |
7026 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
7027 | */ | |
7028 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
7029 | struct io_kiocb *req, | |
7030 | unsigned int sqe_flags) | |
4fe2c963 | 7031 | { |
4cfb25bf | 7032 | if (likely(!ctx->restricted)) |
b16fed66 PB |
7033 | return true; |
7034 | ||
7035 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
7036 | return false; | |
7037 | ||
7038 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
7039 | ctx->restrictions.sqe_flags_required) | |
7040 | return false; | |
7041 | ||
7042 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
7043 | ctx->restrictions.sqe_flags_required)) | |
7044 | return false; | |
7045 | ||
7046 | return true; | |
4fe2c963 JL |
7047 | } |
7048 | ||
b16fed66 PB |
7049 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
7050 | const struct io_uring_sqe *sqe) | |
282cdc86 | 7051 | __must_hold(&ctx->uring_lock) |
b16fed66 PB |
7052 | { |
7053 | struct io_submit_state *state; | |
7054 | unsigned int sqe_flags; | |
003e8dcc | 7055 | int personality, ret = 0; |
b16fed66 | 7056 | |
864ea921 | 7057 | /* req is partially pre-initialised, see io_preinit_req() */ |
b16fed66 PB |
7058 | req->opcode = READ_ONCE(sqe->opcode); |
7059 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
7060 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
7061 | req->user_data = READ_ONCE(sqe->user_data); | |
b16fed66 | 7062 | req->file = NULL; |
b16fed66 | 7063 | req->fixed_rsrc_refs = NULL; |
b16fed66 | 7064 | req->task = current; |
b16fed66 PB |
7065 | |
7066 | /* enforce forwards compatibility on users */ | |
dddca226 | 7067 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 7068 | return -EINVAL; |
b16fed66 PB |
7069 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
7070 | return -EINVAL; | |
4cfb25bf | 7071 | if (!io_check_restriction(ctx, req, sqe_flags)) |
b16fed66 PB |
7072 | return -EACCES; |
7073 | ||
7074 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
7075 | !io_op_defs[req->opcode].buffer_select) | |
7076 | return -EOPNOTSUPP; | |
3c19966d PB |
7077 | if (unlikely(sqe_flags & IOSQE_IO_DRAIN)) |
7078 | ctx->drain_active = true; | |
863e0560 | 7079 | |
003e8dcc JA |
7080 | personality = READ_ONCE(sqe->personality); |
7081 | if (personality) { | |
c10d1f98 PB |
7082 | req->creds = xa_load(&ctx->personalities, personality); |
7083 | if (!req->creds) | |
003e8dcc | 7084 | return -EINVAL; |
c10d1f98 | 7085 | get_cred(req->creds); |
b8e64b53 | 7086 | req->flags |= REQ_F_CREDS; |
003e8dcc | 7087 | } |
b16fed66 PB |
7088 | state = &ctx->submit_state; |
7089 | ||
7090 | /* | |
7091 | * Plug now if we have more than 1 IO left after this, and the target | |
7092 | * is potentially a read/write to block based storage. | |
7093 | */ | |
7094 | if (!state->plug_started && state->ios_left > 1 && | |
7095 | io_op_defs[req->opcode].plug) { | |
7096 | blk_start_plug(&state->plug); | |
7097 | state->plug_started = true; | |
7098 | } | |
7099 | ||
7100 | if (io_op_defs[req->opcode].needs_file) { | |
62906e89 | 7101 | req->file = io_file_get(ctx, req, READ_ONCE(sqe->fd), |
ac177053 | 7102 | (sqe_flags & IOSQE_FIXED_FILE)); |
b16fed66 PB |
7103 | if (unlikely(!req->file)) |
7104 | ret = -EBADF; | |
7105 | } | |
7106 | ||
7107 | state->ios_left--; | |
7108 | return ret; | |
7109 | } | |
7110 | ||
a6b8cadc | 7111 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 7112 | const struct io_uring_sqe *sqe) |
282cdc86 | 7113 | __must_hold(&ctx->uring_lock) |
9e645e11 | 7114 | { |
a1ab7b35 | 7115 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 7116 | int ret; |
9e645e11 | 7117 | |
a6b8cadc PB |
7118 | ret = io_init_req(ctx, req, sqe); |
7119 | if (unlikely(ret)) { | |
7120 | fail_req: | |
a8295b98 | 7121 | /* fail even hard links since we don't submit */ |
de59bc10 | 7122 | if (link->head) { |
a8295b98 HX |
7123 | /* |
7124 | * we can judge a link req is failed or cancelled by if | |
7125 | * REQ_F_FAIL is set, but the head is an exception since | |
7126 | * it may be set REQ_F_FAIL because of other req's failure | |
7127 | * so let's leverage req->result to distinguish if a head | |
7128 | * is set REQ_F_FAIL because of its failure or other req's | |
7129 | * failure so that we can set the correct ret code for it. | |
7130 | * init result here to avoid affecting the normal path. | |
7131 | */ | |
7132 | if (!(link->head->flags & REQ_F_FAIL)) | |
7133 | req_fail_link_node(link->head, -ECANCELED); | |
7134 | } else if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { | |
7135 | /* | |
7136 | * the current req is a normal req, we should return | |
7137 | * error and thus break the submittion loop. | |
7138 | */ | |
7139 | io_req_complete_failed(req, ret); | |
7140 | return ret; | |
de59bc10 | 7141 | } |
a8295b98 HX |
7142 | req_fail_link_node(req, ret); |
7143 | } else { | |
7144 | ret = io_req_prep(req, sqe); | |
7145 | if (unlikely(ret)) | |
7146 | goto fail_req; | |
a6b8cadc | 7147 | } |
441b8a78 | 7148 | |
be7053b7 | 7149 | /* don't need @sqe from now on */ |
236daeae OL |
7150 | trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, |
7151 | req->flags, true, | |
7152 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 7153 | |
9e645e11 JA |
7154 | /* |
7155 | * If we already have a head request, queue this one for async | |
7156 | * submittal once the head completes. If we don't have a head but | |
7157 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
7158 | * submitted sync once the chain is complete. If none of those | |
7159 | * conditions are true (normal request), then just queue it. | |
7160 | */ | |
863e0560 PB |
7161 | if (link->head) { |
7162 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 7163 | |
a8295b98 HX |
7164 | if (!(req->flags & REQ_F_FAIL)) { |
7165 | ret = io_req_prep_async(req); | |
7166 | if (unlikely(ret)) { | |
7167 | req_fail_link_node(req, ret); | |
7168 | if (!(head->flags & REQ_F_FAIL)) | |
7169 | req_fail_link_node(head, -ECANCELED); | |
7170 | } | |
7171 | } | |
9d76377f | 7172 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 7173 | link->last->link = req; |
863e0560 | 7174 | link->last = req; |
32fe525b PB |
7175 | |
7176 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 7177 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 7178 | link->head = NULL; |
5e159204 | 7179 | io_queue_sqe(head); |
32fe525b | 7180 | } |
9e645e11 | 7181 | } else { |
ef4ff581 | 7182 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
7183 | link->head = req; |
7184 | link->last = req; | |
711be031 | 7185 | } else { |
be7053b7 | 7186 | io_queue_sqe(req); |
711be031 | 7187 | } |
9e645e11 | 7188 | } |
2e6e1fde | 7189 | |
1d4240cc | 7190 | return 0; |
9e645e11 JA |
7191 | } |
7192 | ||
9a56a232 JA |
7193 | /* |
7194 | * Batched submission is done, ensure local IO is flushed out. | |
7195 | */ | |
ba88ff11 PB |
7196 | static void io_submit_state_end(struct io_submit_state *state, |
7197 | struct io_ring_ctx *ctx) | |
9a56a232 | 7198 | { |
a1ab7b35 | 7199 | if (state->link.head) |
de59bc10 | 7200 | io_queue_sqe(state->link.head); |
cd0ca2e0 | 7201 | if (state->compl_nr) |
2a2758f2 | 7202 | io_submit_flush_completions(ctx); |
27926b68 JA |
7203 | if (state->plug_started) |
7204 | blk_finish_plug(&state->plug); | |
9a56a232 JA |
7205 | } |
7206 | ||
7207 | /* | |
7208 | * Start submission side cache. | |
7209 | */ | |
7210 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 7211 | unsigned int max_ios) |
9a56a232 | 7212 | { |
27926b68 | 7213 | state->plug_started = false; |
9a56a232 | 7214 | state->ios_left = max_ios; |
a1ab7b35 PB |
7215 | /* set only head, no need to init link_last in advance */ |
7216 | state->link.head = NULL; | |
9a56a232 JA |
7217 | } |
7218 | ||
2b188cc1 JA |
7219 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
7220 | { | |
75b28aff | 7221 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 7222 | |
caf582c6 PB |
7223 | /* |
7224 | * Ensure any loads from the SQEs are done at this point, | |
7225 | * since once we write the new head, the application could | |
7226 | * write new data to them. | |
7227 | */ | |
7228 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
7229 | } |
7230 | ||
2b188cc1 | 7231 | /* |
dd9ae8a0 | 7232 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
7233 | * that is mapped by userspace. This means that care needs to be taken to |
7234 | * ensure that reads are stable, as we cannot rely on userspace always | |
7235 | * being a good citizen. If members of the sqe are validated and then later | |
7236 | * used, it's important that those reads are done through READ_ONCE() to | |
7237 | * prevent a re-load down the line. | |
7238 | */ | |
709b302f | 7239 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 7240 | { |
ea5ab3b5 | 7241 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 7242 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
7243 | |
7244 | /* | |
7245 | * The cached sq head (or cq tail) serves two purposes: | |
7246 | * | |
7247 | * 1) allows us to batch the cost of updating the user visible | |
7248 | * head updates. | |
7249 | * 2) allows the kernel side to track the head on its own, even | |
7250 | * though the application is the one updating it. | |
7251 | */ | |
17d3aeb3 | 7252 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
7253 | if (likely(head < ctx->sq_entries)) |
7254 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
7255 | |
7256 | /* drop invalid entries */ | |
15641e42 PB |
7257 | ctx->cq_extra--; |
7258 | WRITE_ONCE(ctx->rings->sq_dropped, | |
7259 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
7260 | return NULL; |
7261 | } | |
7262 | ||
0f212204 | 7263 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
282cdc86 | 7264 | __must_hold(&ctx->uring_lock) |
6c271ce2 | 7265 | { |
46c4e16a | 7266 | int submitted = 0; |
6c271ce2 | 7267 | |
ee7d46d9 PB |
7268 | /* make sure SQ entry isn't read before tail */ |
7269 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
7270 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
7271 | return -EAGAIN; | |
9a10867a | 7272 | io_get_task_refs(nr); |
6c271ce2 | 7273 | |
ba88ff11 | 7274 | io_submit_state_start(&ctx->submit_state, nr); |
46c4e16a | 7275 | while (submitted < nr) { |
3529d8c2 | 7276 | const struct io_uring_sqe *sqe; |
196be95c | 7277 | struct io_kiocb *req; |
fb5ccc98 | 7278 | |
258b29a9 | 7279 | req = io_alloc_req(ctx); |
196be95c PB |
7280 | if (unlikely(!req)) { |
7281 | if (!submitted) | |
7282 | submitted = -EAGAIN; | |
fb5ccc98 | 7283 | break; |
196be95c | 7284 | } |
4fccfcbb PB |
7285 | sqe = io_get_sqe(ctx); |
7286 | if (unlikely(!sqe)) { | |
0c6e1d7f | 7287 | list_add(&req->inflight_entry, &ctx->submit_state.free_list); |
4fccfcbb PB |
7288 | break; |
7289 | } | |
d3656344 JA |
7290 | /* will complete beyond this point, count as submitted */ |
7291 | submitted++; | |
a1ab7b35 | 7292 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 7293 | break; |
6c271ce2 JA |
7294 | } |
7295 | ||
9466f437 PB |
7296 | if (unlikely(submitted != nr)) { |
7297 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 7298 | int unused = nr - ref_used; |
9466f437 | 7299 | |
09899b19 | 7300 | current->io_uring->cached_refs += unused; |
d8a6df10 | 7301 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 7302 | } |
6c271ce2 | 7303 | |
a1ab7b35 | 7304 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
7305 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
7306 | io_commit_sqring(ctx); | |
7307 | ||
6c271ce2 JA |
7308 | return submitted; |
7309 | } | |
7310 | ||
e4b6d902 PB |
7311 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
7312 | { | |
7313 | return READ_ONCE(sqd->state); | |
7314 | } | |
7315 | ||
23b3628e XW |
7316 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
7317 | { | |
7318 | /* Tell userspace we may need a wakeup call */ | |
79ebeaee | 7319 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7320 | WRITE_ONCE(ctx->rings->sq_flags, |
7321 | ctx->rings->sq_flags | IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7322 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7323 | } |
7324 | ||
7325 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
7326 | { | |
79ebeaee | 7327 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7328 | WRITE_ONCE(ctx->rings->sq_flags, |
7329 | ctx->rings->sq_flags & ~IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7330 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7331 | } |
7332 | ||
08369246 | 7333 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 7334 | { |
c8d1ba58 | 7335 | unsigned int to_submit; |
bdcd3eab | 7336 | int ret = 0; |
6c271ce2 | 7337 | |
c8d1ba58 | 7338 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 7339 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
7340 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
7341 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 7342 | |
906a3c6f | 7343 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 7344 | unsigned nr_events = 0; |
948e1947 PB |
7345 | const struct cred *creds = NULL; |
7346 | ||
7347 | if (ctx->sq_creds != current_cred()) | |
7348 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 7349 | |
c8d1ba58 | 7350 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 7351 | if (!list_empty(&ctx->iopoll_list)) |
a8576af9 | 7352 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 7353 | |
3b763ba1 PB |
7354 | /* |
7355 | * Don't submit if refs are dying, good for io_uring_register(), | |
7356 | * but also it is relied upon by io_ring_exit_work() | |
7357 | */ | |
0298ef96 PB |
7358 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
7359 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 7360 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 7361 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 7362 | |
acfb381d PB |
7363 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
7364 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
7365 | if (creds) |
7366 | revert_creds(creds); | |
acfb381d | 7367 | } |
6c271ce2 | 7368 | |
08369246 XW |
7369 | return ret; |
7370 | } | |
6c271ce2 | 7371 | |
08369246 XW |
7372 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7373 | { | |
7374 | struct io_ring_ctx *ctx; | |
7375 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7376 | |
c9dca27d PB |
7377 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7378 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 7379 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7380 | } |
6c271ce2 | 7381 | |
e4b6d902 PB |
7382 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
7383 | { | |
7384 | bool did_sig = false; | |
7385 | struct ksignal ksig; | |
7386 | ||
7387 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
7388 | signal_pending(current)) { | |
7389 | mutex_unlock(&sqd->lock); | |
7390 | if (signal_pending(current)) | |
7391 | did_sig = get_signal(&ksig); | |
7392 | cond_resched(); | |
7393 | mutex_lock(&sqd->lock); | |
7394 | } | |
e4b6d902 PB |
7395 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
7396 | } | |
7397 | ||
c8d1ba58 JA |
7398 | static int io_sq_thread(void *data) |
7399 | { | |
69fb2131 JA |
7400 | struct io_sq_data *sqd = data; |
7401 | struct io_ring_ctx *ctx; | |
a0d9205f | 7402 | unsigned long timeout = 0; |
37d1e2e3 | 7403 | char buf[TASK_COMM_LEN]; |
08369246 | 7404 | DEFINE_WAIT(wait); |
6c271ce2 | 7405 | |
696ee88a | 7406 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 7407 | set_task_comm(current, buf); |
37d1e2e3 JA |
7408 | |
7409 | if (sqd->sq_cpu != -1) | |
7410 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
7411 | else | |
7412 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
7413 | current->flags |= PF_NO_SETAFFINITY; | |
7414 | ||
09a6f4ef | 7415 | mutex_lock(&sqd->lock); |
e4b6d902 | 7416 | while (1) { |
1a924a80 | 7417 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 7418 | |
e4b6d902 PB |
7419 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
7420 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 7421 | break; |
08369246 XW |
7422 | timeout = jiffies + sqd->sq_thread_idle; |
7423 | } | |
e4b6d902 | 7424 | |
e95eee2d | 7425 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 7426 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 7427 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 7428 | |
08369246 XW |
7429 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
7430 | sqt_spin = true; | |
69fb2131 | 7431 | } |
dd432ea5 PB |
7432 | if (io_run_task_work()) |
7433 | sqt_spin = true; | |
6c271ce2 | 7434 | |
08369246 | 7435 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7436 | cond_resched(); |
08369246 XW |
7437 | if (sqt_spin) |
7438 | timeout = jiffies + sqd->sq_thread_idle; | |
7439 | continue; | |
7440 | } | |
7441 | ||
08369246 | 7442 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
dd432ea5 | 7443 | if (!io_sqd_events_pending(sqd) && !current->task_works) { |
1a924a80 PB |
7444 | bool needs_sched = true; |
7445 | ||
724cb4f9 | 7446 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
7447 | io_ring_set_wakeup_flag(ctx); |
7448 | ||
724cb4f9 HX |
7449 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
7450 | !list_empty_careful(&ctx->iopoll_list)) { | |
7451 | needs_sched = false; | |
7452 | break; | |
7453 | } | |
7454 | if (io_sqring_entries(ctx)) { | |
7455 | needs_sched = false; | |
7456 | break; | |
7457 | } | |
7458 | } | |
7459 | ||
7460 | if (needs_sched) { | |
7461 | mutex_unlock(&sqd->lock); | |
7462 | schedule(); | |
7463 | mutex_lock(&sqd->lock); | |
7464 | } | |
69fb2131 JA |
7465 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7466 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7467 | } |
08369246 XW |
7468 | |
7469 | finish_wait(&sqd->wait, &wait); | |
7470 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 7471 | } |
28cea78a | 7472 | |
78cc687b | 7473 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 7474 | sqd->thread = NULL; |
05962f95 | 7475 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 7476 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 7477 | io_run_task_work(); |
734551df PB |
7478 | mutex_unlock(&sqd->lock); |
7479 | ||
37d1e2e3 JA |
7480 | complete(&sqd->exited); |
7481 | do_exit(0); | |
6c271ce2 JA |
7482 | } |
7483 | ||
bda52162 JA |
7484 | struct io_wait_queue { |
7485 | struct wait_queue_entry wq; | |
7486 | struct io_ring_ctx *ctx; | |
5fd46178 | 7487 | unsigned cq_tail; |
bda52162 JA |
7488 | unsigned nr_timeouts; |
7489 | }; | |
7490 | ||
6c503150 | 7491 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7492 | { |
7493 | struct io_ring_ctx *ctx = iowq->ctx; | |
5fd46178 | 7494 | int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; |
bda52162 JA |
7495 | |
7496 | /* | |
d195a66e | 7497 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7498 | * started waiting. For timeouts, we always want to return to userspace, |
7499 | * regardless of event count. | |
7500 | */ | |
5fd46178 | 7501 | return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
bda52162 JA |
7502 | } |
7503 | ||
7504 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7505 | int wake_flags, void *key) | |
7506 | { | |
7507 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7508 | wq); | |
7509 | ||
6c503150 PB |
7510 | /* |
7511 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7512 | * the task, and the next invocation will do it. | |
7513 | */ | |
5ed7a37d | 7514 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
7515 | return autoremove_wake_function(curr, mode, wake_flags, key); |
7516 | return -1; | |
bda52162 JA |
7517 | } |
7518 | ||
af9c1a44 JA |
7519 | static int io_run_task_work_sig(void) |
7520 | { | |
7521 | if (io_run_task_work()) | |
7522 | return 1; | |
7523 | if (!signal_pending(current)) | |
7524 | return 0; | |
0b8cfa97 | 7525 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 7526 | return -ERESTARTSYS; |
af9c1a44 JA |
7527 | return -EINTR; |
7528 | } | |
7529 | ||
eeb60b9a PB |
7530 | /* when returns >0, the caller should retry */ |
7531 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7532 | struct io_wait_queue *iowq, | |
7533 | signed long *timeout) | |
7534 | { | |
7535 | int ret; | |
7536 | ||
7537 | /* make sure we run task_work before checking for signals */ | |
7538 | ret = io_run_task_work_sig(); | |
7539 | if (ret || io_should_wake(iowq)) | |
7540 | return ret; | |
7541 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 7542 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
7543 | return 1; |
7544 | ||
7545 | *timeout = schedule_timeout(*timeout); | |
7546 | return !*timeout ? -ETIME : 1; | |
7547 | } | |
7548 | ||
2b188cc1 JA |
7549 | /* |
7550 | * Wait until events become available, if we don't already have some. The | |
7551 | * application must reap them itself, as they reside on the shared cq ring. | |
7552 | */ | |
7553 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7554 | const sigset_t __user *sig, size_t sigsz, |
7555 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7556 | { |
90291099 | 7557 | struct io_wait_queue iowq; |
75b28aff | 7558 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7559 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7560 | int ret; | |
2b188cc1 | 7561 | |
b41e9852 | 7562 | do { |
90f67366 | 7563 | io_cqring_overflow_flush(ctx); |
6c503150 | 7564 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7565 | return 0; |
4c6e277c | 7566 | if (!io_run_task_work()) |
b41e9852 | 7567 | break; |
b41e9852 | 7568 | } while (1); |
2b188cc1 | 7569 | |
44df58d4 XW |
7570 | if (uts) { |
7571 | struct timespec64 ts; | |
7572 | ||
7573 | if (get_timespec64(&ts, uts)) | |
7574 | return -EFAULT; | |
7575 | timeout = timespec64_to_jiffies(&ts); | |
7576 | } | |
7577 | ||
2b188cc1 | 7578 | if (sig) { |
9e75ad5d AB |
7579 | #ifdef CONFIG_COMPAT |
7580 | if (in_compat_syscall()) | |
7581 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7582 | sigsz); |
9e75ad5d AB |
7583 | else |
7584 | #endif | |
b772434b | 7585 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7586 | |
2b188cc1 JA |
7587 | if (ret) |
7588 | return ret; | |
7589 | } | |
7590 | ||
90291099 PB |
7591 | init_waitqueue_func_entry(&iowq.wq, io_wake_function); |
7592 | iowq.wq.private = current; | |
7593 | INIT_LIST_HEAD(&iowq.wq.entry); | |
7594 | iowq.ctx = ctx; | |
bda52162 | 7595 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
5fd46178 | 7596 | iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; |
90291099 | 7597 | |
c826bd7a | 7598 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7599 | do { |
ca0a2651 | 7600 | /* if we can't even flush overflow, don't wait for more */ |
90f67366 | 7601 | if (!io_cqring_overflow_flush(ctx)) { |
ca0a2651 JA |
7602 | ret = -EBUSY; |
7603 | break; | |
7604 | } | |
311997b3 | 7605 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7606 | TASK_INTERRUPTIBLE); |
eeb60b9a | 7607 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
311997b3 | 7608 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7609 | cond_resched(); |
eeb60b9a | 7610 | } while (ret > 0); |
bda52162 | 7611 | |
b7db41c9 | 7612 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7613 | |
75b28aff | 7614 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7615 | } |
7616 | ||
9123c8ff | 7617 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7618 | { |
9123c8ff | 7619 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7620 | |
846a4ef2 | 7621 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7622 | kfree(table[i]); |
7623 | kfree(table); | |
7624 | } | |
7625 | ||
7626 | static void **io_alloc_page_table(size_t size) | |
7627 | { | |
7628 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7629 | size_t init_size = size; | |
7630 | void **table; | |
7631 | ||
0bea96f5 | 7632 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7633 | if (!table) |
7634 | return NULL; | |
7635 | ||
7636 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 7637 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff | 7638 | |
0bea96f5 | 7639 | table[i] = kzalloc(this_size, GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7640 | if (!table[i]) { |
7641 | io_free_page_table(table, init_size); | |
7642 | return NULL; | |
7643 | } | |
7644 | size -= this_size; | |
7645 | } | |
7646 | return table; | |
05f3fb3c JA |
7647 | } |
7648 | ||
28a9fe25 | 7649 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7650 | { |
28a9fe25 PB |
7651 | percpu_ref_exit(&ref_node->refs); |
7652 | kfree(ref_node); | |
1642b445 PB |
7653 | } |
7654 | ||
b9bd2bea PB |
7655 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
7656 | { | |
7657 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); | |
7658 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; | |
7659 | unsigned long flags; | |
7660 | bool first_add = false; | |
7661 | ||
7662 | spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); | |
7663 | node->done = true; | |
7664 | ||
7665 | while (!list_empty(&ctx->rsrc_ref_list)) { | |
7666 | node = list_first_entry(&ctx->rsrc_ref_list, | |
7667 | struct io_rsrc_node, node); | |
7668 | /* recycle ref nodes in order */ | |
7669 | if (!node->done) | |
7670 | break; | |
7671 | list_del(&node->node); | |
7672 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
7673 | } | |
7674 | spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); | |
7675 | ||
7676 | if (first_add) | |
7677 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
7678 | } | |
7679 | ||
7680 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) | |
7681 | { | |
7682 | struct io_rsrc_node *ref_node; | |
7683 | ||
7684 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); | |
7685 | if (!ref_node) | |
7686 | return NULL; | |
7687 | ||
7688 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, | |
7689 | 0, GFP_KERNEL)) { | |
7690 | kfree(ref_node); | |
7691 | return NULL; | |
7692 | } | |
7693 | INIT_LIST_HEAD(&ref_node->node); | |
7694 | INIT_LIST_HEAD(&ref_node->rsrc_list); | |
7695 | ref_node->done = false; | |
7696 | return ref_node; | |
7697 | } | |
7698 | ||
a7f0ed5a PB |
7699 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7700 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7701 | { |
a7f0ed5a PB |
7702 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7703 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7704 | |
a7f0ed5a PB |
7705 | if (data_to_kill) { |
7706 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7707 | |
a7f0ed5a | 7708 | rsrc_node->rsrc_data = data_to_kill; |
4956b9ea | 7709 | spin_lock_irq(&ctx->rsrc_ref_lock); |
a7f0ed5a | 7710 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); |
4956b9ea | 7711 | spin_unlock_irq(&ctx->rsrc_ref_lock); |
82fbcfa9 | 7712 | |
3e942498 | 7713 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7714 | percpu_ref_kill(&rsrc_node->refs); |
7715 | ctx->rsrc_node = NULL; | |
7716 | } | |
6b06314c | 7717 | |
a7f0ed5a PB |
7718 | if (!ctx->rsrc_node) { |
7719 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7720 | ctx->rsrc_backup_node = NULL; | |
7721 | } | |
8bad28d8 HX |
7722 | } |
7723 | ||
a7f0ed5a | 7724 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7725 | { |
7726 | if (ctx->rsrc_backup_node) | |
7727 | return 0; | |
b895c9a6 | 7728 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7729 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7730 | } |
7731 | ||
40ae0ff7 | 7732 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7733 | { |
7734 | int ret; | |
05589553 | 7735 | |
215c3902 | 7736 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7737 | if (data->quiesce) |
7738 | return -ENXIO; | |
05589553 | 7739 | |
8bad28d8 | 7740 | data->quiesce = true; |
1ffc5422 | 7741 | do { |
a7f0ed5a | 7742 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7743 | if (ret) |
f2303b1f | 7744 | break; |
a7f0ed5a | 7745 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7746 | |
3e942498 PB |
7747 | /* kill initial ref, already quiesced if zero */ |
7748 | if (atomic_dec_and_test(&data->refs)) | |
7749 | break; | |
c018db4a | 7750 | mutex_unlock(&ctx->uring_lock); |
8bad28d8 | 7751 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 | 7752 | ret = wait_for_completion_interruptible(&data->done); |
c018db4a JA |
7753 | if (!ret) { |
7754 | mutex_lock(&ctx->uring_lock); | |
1ffc5422 | 7755 | break; |
c018db4a | 7756 | } |
8bad28d8 | 7757 | |
3e942498 PB |
7758 | atomic_inc(&data->refs); |
7759 | /* wait for all works potentially completing data->done */ | |
7760 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7761 | reinit_completion(&data->done); |
8dd03afe | 7762 | |
1ffc5422 | 7763 | ret = io_run_task_work_sig(); |
8bad28d8 | 7764 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7765 | } while (ret >= 0); |
8bad28d8 | 7766 | data->quiesce = false; |
05f3fb3c | 7767 | |
8bad28d8 | 7768 | return ret; |
d7954b2b BM |
7769 | } |
7770 | ||
2d091d62 PB |
7771 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7772 | { | |
7773 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7774 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7775 | ||
7776 | return &data->tags[table_idx][off]; | |
7777 | } | |
7778 | ||
44b31f2f | 7779 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7780 | { |
2d091d62 PB |
7781 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7782 | ||
7783 | if (data->tags) | |
7784 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7785 | kfree(data); |
7786 | } | |
7787 | ||
d878c816 PB |
7788 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7789 | u64 __user *utags, unsigned nr, | |
7790 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7791 | { |
b895c9a6 | 7792 | struct io_rsrc_data *data; |
2d091d62 | 7793 | int ret = -ENOMEM; |
d878c816 | 7794 | unsigned i; |
1ad555c6 BM |
7795 | |
7796 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7797 | if (!data) | |
d878c816 | 7798 | return -ENOMEM; |
2d091d62 | 7799 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7800 | if (!data->tags) { |
1ad555c6 | 7801 | kfree(data); |
d878c816 PB |
7802 | return -ENOMEM; |
7803 | } | |
2d091d62 PB |
7804 | |
7805 | data->nr = nr; | |
7806 | data->ctx = ctx; | |
7807 | data->do_put = do_put; | |
d878c816 | 7808 | if (utags) { |
2d091d62 | 7809 | ret = -EFAULT; |
d878c816 | 7810 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7811 | u64 *tag_slot = io_get_tag_slot(data, i); |
7812 | ||
7813 | if (copy_from_user(tag_slot, &utags[i], | |
7814 | sizeof(*tag_slot))) | |
2d091d62 | 7815 | goto fail; |
d878c816 | 7816 | } |
1ad555c6 | 7817 | } |
b60c8dce | 7818 | |
3e942498 | 7819 | atomic_set(&data->refs, 1); |
1ad555c6 | 7820 | init_completion(&data->done); |
d878c816 PB |
7821 | *pdata = data; |
7822 | return 0; | |
2d091d62 PB |
7823 | fail: |
7824 | io_rsrc_data_free(data); | |
7825 | return ret; | |
1ad555c6 BM |
7826 | } |
7827 | ||
9123c8ff PB |
7828 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7829 | { | |
0bea96f5 PB |
7830 | table->files = kvcalloc(nr_files, sizeof(table->files[0]), |
7831 | GFP_KERNEL_ACCOUNT); | |
9123c8ff PB |
7832 | return !!table->files; |
7833 | } | |
7834 | ||
042b0d85 | 7835 | static void io_free_file_tables(struct io_file_table *table) |
9123c8ff | 7836 | { |
042b0d85 | 7837 | kvfree(table->files); |
9123c8ff PB |
7838 | table->files = NULL; |
7839 | } | |
7840 | ||
fff4db76 | 7841 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7842 | { |
fff4db76 PB |
7843 | #if defined(CONFIG_UNIX) |
7844 | if (ctx->ring_sock) { | |
7845 | struct sock *sock = ctx->ring_sock->sk; | |
7846 | struct sk_buff *skb; | |
7847 | ||
7848 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7849 | kfree_skb(skb); | |
7850 | } | |
7851 | #else | |
7852 | int i; | |
7853 | ||
7854 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7855 | struct file *file; | |
7856 | ||
7857 | file = io_file_from_index(ctx, i); | |
7858 | if (file) | |
7859 | fput(file); | |
7860 | } | |
7861 | #endif | |
042b0d85 | 7862 | io_free_file_tables(&ctx->file_table); |
44b31f2f | 7863 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7864 | ctx->file_data = NULL; |
7865 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7866 | } |
7867 | ||
d7954b2b BM |
7868 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7869 | { | |
d7954b2b BM |
7870 | int ret; |
7871 | ||
08480400 | 7872 | if (!ctx->file_data) |
d7954b2b | 7873 | return -ENXIO; |
08480400 PB |
7874 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7875 | if (!ret) | |
7876 | __io_sqe_files_unregister(ctx); | |
7877 | return ret; | |
6b06314c JA |
7878 | } |
7879 | ||
37d1e2e3 | 7880 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7881 | __releases(&sqd->lock) |
37d1e2e3 | 7882 | { |
521d6a73 PB |
7883 | WARN_ON_ONCE(sqd->thread == current); |
7884 | ||
9e138a48 PB |
7885 | /* |
7886 | * Do the dance but not conditional clear_bit() because it'd race with | |
7887 | * other threads incrementing park_pending and setting the bit. | |
7888 | */ | |
37d1e2e3 | 7889 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7890 | if (atomic_dec_return(&sqd->park_pending)) |
7891 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7892 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7893 | } |
7894 | ||
86e0d676 | 7895 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7896 | __acquires(&sqd->lock) |
37d1e2e3 | 7897 | { |
521d6a73 PB |
7898 | WARN_ON_ONCE(sqd->thread == current); |
7899 | ||
9e138a48 | 7900 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7901 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7902 | mutex_lock(&sqd->lock); |
05962f95 | 7903 | if (sqd->thread) |
86e0d676 | 7904 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7905 | } |
7906 | ||
7907 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7908 | { | |
521d6a73 | 7909 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7910 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7911 | |
05962f95 | 7912 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7913 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7914 | if (sqd->thread) |
7915 | wake_up_process(sqd->thread); | |
09a6f4ef | 7916 | mutex_unlock(&sqd->lock); |
05962f95 | 7917 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7918 | } |
7919 | ||
534ca6d6 | 7920 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7921 | { |
534ca6d6 | 7922 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7923 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7924 | ||
37d1e2e3 JA |
7925 | io_sq_thread_stop(sqd); |
7926 | kfree(sqd); | |
7927 | } | |
7928 | } | |
7929 | ||
7930 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7931 | { | |
7932 | struct io_sq_data *sqd = ctx->sq_data; | |
7933 | ||
7934 | if (sqd) { | |
05962f95 | 7935 | io_sq_thread_park(sqd); |
521d6a73 | 7936 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7937 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7938 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7939 | |
7940 | io_put_sq_data(sqd); | |
7941 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7942 | } |
7943 | } | |
7944 | ||
aa06165d JA |
7945 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7946 | { | |
7947 | struct io_ring_ctx *ctx_attach; | |
7948 | struct io_sq_data *sqd; | |
7949 | struct fd f; | |
7950 | ||
7951 | f = fdget(p->wq_fd); | |
7952 | if (!f.file) | |
7953 | return ERR_PTR(-ENXIO); | |
7954 | if (f.file->f_op != &io_uring_fops) { | |
7955 | fdput(f); | |
7956 | return ERR_PTR(-EINVAL); | |
7957 | } | |
7958 | ||
7959 | ctx_attach = f.file->private_data; | |
7960 | sqd = ctx_attach->sq_data; | |
7961 | if (!sqd) { | |
7962 | fdput(f); | |
7963 | return ERR_PTR(-EINVAL); | |
7964 | } | |
5c2469e0 JA |
7965 | if (sqd->task_tgid != current->tgid) { |
7966 | fdput(f); | |
7967 | return ERR_PTR(-EPERM); | |
7968 | } | |
aa06165d JA |
7969 | |
7970 | refcount_inc(&sqd->refs); | |
7971 | fdput(f); | |
7972 | return sqd; | |
7973 | } | |
7974 | ||
26984fbf PB |
7975 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7976 | bool *attached) | |
534ca6d6 JA |
7977 | { |
7978 | struct io_sq_data *sqd; | |
7979 | ||
26984fbf | 7980 | *attached = false; |
5c2469e0 JA |
7981 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7982 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7983 | if (!IS_ERR(sqd)) { |
7984 | *attached = true; | |
5c2469e0 | 7985 | return sqd; |
26984fbf | 7986 | } |
5c2469e0 JA |
7987 | /* fall through for EPERM case, setup new sqd/task */ |
7988 | if (PTR_ERR(sqd) != -EPERM) | |
7989 | return sqd; | |
7990 | } | |
aa06165d | 7991 | |
534ca6d6 JA |
7992 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7993 | if (!sqd) | |
7994 | return ERR_PTR(-ENOMEM); | |
7995 | ||
9e138a48 | 7996 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7997 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7998 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7999 | mutex_init(&sqd->lock); |
534ca6d6 | 8000 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 8001 | init_completion(&sqd->exited); |
534ca6d6 JA |
8002 | return sqd; |
8003 | } | |
8004 | ||
6b06314c | 8005 | #if defined(CONFIG_UNIX) |
6b06314c JA |
8006 | /* |
8007 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
8008 | * the io_uring can be safely unregistered on process exit, even if we have | |
8009 | * loops in the file referencing. | |
8010 | */ | |
8011 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
8012 | { | |
8013 | struct sock *sk = ctx->ring_sock->sk; | |
8014 | struct scm_fp_list *fpl; | |
8015 | struct sk_buff *skb; | |
08a45173 | 8016 | int i, nr_files; |
6b06314c | 8017 | |
6b06314c JA |
8018 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
8019 | if (!fpl) | |
8020 | return -ENOMEM; | |
8021 | ||
8022 | skb = alloc_skb(0, GFP_KERNEL); | |
8023 | if (!skb) { | |
8024 | kfree(fpl); | |
8025 | return -ENOMEM; | |
8026 | } | |
8027 | ||
8028 | skb->sk = sk; | |
6b06314c | 8029 | |
08a45173 | 8030 | nr_files = 0; |
62e398be | 8031 | fpl->user = get_uid(current_user()); |
6b06314c | 8032 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
8033 | struct file *file = io_file_from_index(ctx, i + offset); |
8034 | ||
8035 | if (!file) | |
08a45173 | 8036 | continue; |
65e19f54 | 8037 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
8038 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
8039 | nr_files++; | |
6b06314c JA |
8040 | } |
8041 | ||
08a45173 JA |
8042 | if (nr_files) { |
8043 | fpl->max = SCM_MAX_FD; | |
8044 | fpl->count = nr_files; | |
8045 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 8046 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
8047 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
8048 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 8049 | |
08a45173 JA |
8050 | for (i = 0; i < nr_files; i++) |
8051 | fput(fpl->fp[i]); | |
8052 | } else { | |
8053 | kfree_skb(skb); | |
8054 | kfree(fpl); | |
8055 | } | |
6b06314c JA |
8056 | |
8057 | return 0; | |
8058 | } | |
8059 | ||
8060 | /* | |
8061 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
8062 | * causes regular reference counting to break down. We rely on the UNIX | |
8063 | * garbage collection to take care of this problem for us. | |
8064 | */ | |
8065 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8066 | { | |
8067 | unsigned left, total; | |
8068 | int ret = 0; | |
8069 | ||
8070 | total = 0; | |
8071 | left = ctx->nr_user_files; | |
8072 | while (left) { | |
8073 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
8074 | |
8075 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
8076 | if (ret) | |
8077 | break; | |
8078 | left -= this_files; | |
8079 | total += this_files; | |
8080 | } | |
8081 | ||
8082 | if (!ret) | |
8083 | return 0; | |
8084 | ||
8085 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
8086 | struct file *file = io_file_from_index(ctx, total); |
8087 | ||
8088 | if (file) | |
8089 | fput(file); | |
6b06314c JA |
8090 | total++; |
8091 | } | |
8092 | ||
8093 | return ret; | |
8094 | } | |
8095 | #else | |
8096 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8097 | { | |
8098 | return 0; | |
8099 | } | |
8100 | #endif | |
8101 | ||
47e90392 | 8102 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 8103 | { |
50238531 | 8104 | struct file *file = prsrc->file; |
05f3fb3c JA |
8105 | #if defined(CONFIG_UNIX) |
8106 | struct sock *sock = ctx->ring_sock->sk; | |
8107 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
8108 | struct sk_buff *skb; | |
8109 | int i; | |
8110 | ||
8111 | __skb_queue_head_init(&list); | |
8112 | ||
8113 | /* | |
8114 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
8115 | * remove this entry and rearrange the file array. | |
8116 | */ | |
8117 | skb = skb_dequeue(head); | |
8118 | while (skb) { | |
8119 | struct scm_fp_list *fp; | |
8120 | ||
8121 | fp = UNIXCB(skb).fp; | |
8122 | for (i = 0; i < fp->count; i++) { | |
8123 | int left; | |
8124 | ||
8125 | if (fp->fp[i] != file) | |
8126 | continue; | |
8127 | ||
8128 | unix_notinflight(fp->user, fp->fp[i]); | |
8129 | left = fp->count - 1 - i; | |
8130 | if (left) { | |
8131 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
8132 | left * sizeof(struct file *)); | |
8133 | } | |
8134 | fp->count--; | |
8135 | if (!fp->count) { | |
8136 | kfree_skb(skb); | |
8137 | skb = NULL; | |
8138 | } else { | |
8139 | __skb_queue_tail(&list, skb); | |
8140 | } | |
8141 | fput(file); | |
8142 | file = NULL; | |
8143 | break; | |
8144 | } | |
8145 | ||
8146 | if (!file) | |
8147 | break; | |
8148 | ||
8149 | __skb_queue_tail(&list, skb); | |
8150 | ||
8151 | skb = skb_dequeue(head); | |
8152 | } | |
8153 | ||
8154 | if (skb_peek(&list)) { | |
8155 | spin_lock_irq(&head->lock); | |
8156 | while ((skb = __skb_dequeue(&list)) != NULL) | |
8157 | __skb_queue_tail(head, skb); | |
8158 | spin_unlock_irq(&head->lock); | |
8159 | } | |
8160 | #else | |
8161 | fput(file); | |
8162 | #endif | |
8163 | } | |
8164 | ||
b895c9a6 | 8165 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 8166 | { |
b895c9a6 | 8167 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
8168 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
8169 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 8170 | |
269bbe5f BM |
8171 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
8172 | list_del(&prsrc->list); | |
b60c8dce PB |
8173 | |
8174 | if (prsrc->tag) { | |
8175 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
8176 | |
8177 | io_ring_submit_lock(ctx, lock_ring); | |
79ebeaee | 8178 | spin_lock(&ctx->completion_lock); |
b60c8dce | 8179 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 8180 | ctx->cq_extra++; |
b60c8dce | 8181 | io_commit_cqring(ctx); |
79ebeaee | 8182 | spin_unlock(&ctx->completion_lock); |
b60c8dce PB |
8183 | io_cqring_ev_posted(ctx); |
8184 | io_ring_submit_unlock(ctx, lock_ring); | |
8185 | } | |
8186 | ||
40ae0ff7 | 8187 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 8188 | kfree(prsrc); |
65e19f54 | 8189 | } |
05589553 | 8190 | |
28a9fe25 | 8191 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
8192 | if (atomic_dec_and_test(&rsrc_data->refs)) |
8193 | complete(&rsrc_data->done); | |
2faf852d | 8194 | } |
65e19f54 | 8195 | |
269bbe5f | 8196 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
8197 | { |
8198 | struct io_ring_ctx *ctx; | |
8199 | struct llist_node *node; | |
8200 | ||
269bbe5f BM |
8201 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
8202 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
8203 | |
8204 | while (node) { | |
b895c9a6 | 8205 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
8206 | struct llist_node *next = node->next; |
8207 | ||
b895c9a6 | 8208 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 8209 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
8210 | node = next; |
8211 | } | |
8212 | } | |
8213 | ||
6b06314c | 8214 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 8215 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
8216 | { |
8217 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 8218 | struct file *file; |
f3baed39 | 8219 | int fd, ret; |
846a4ef2 | 8220 | unsigned i; |
6b06314c | 8221 | |
05f3fb3c | 8222 | if (ctx->file_data) |
6b06314c JA |
8223 | return -EBUSY; |
8224 | if (!nr_args) | |
8225 | return -EINVAL; | |
8226 | if (nr_args > IORING_MAX_FIXED_FILES) | |
8227 | return -EMFILE; | |
3a1b8a4e PB |
8228 | if (nr_args > rlimit(RLIMIT_NOFILE)) |
8229 | return -EMFILE; | |
a7f0ed5a | 8230 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
8231 | if (ret) |
8232 | return ret; | |
d878c816 PB |
8233 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
8234 | &ctx->file_data); | |
8235 | if (ret) | |
8236 | return ret; | |
6b06314c | 8237 | |
f3baed39 | 8238 | ret = -ENOMEM; |
aeca241b | 8239 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 8240 | goto out_free; |
65e19f54 | 8241 | |
08a45173 | 8242 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 8243 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
8244 | ret = -EFAULT; |
8245 | goto out_fput; | |
8246 | } | |
08a45173 | 8247 | /* allow sparse sets */ |
792e3582 PB |
8248 | if (fd == -1) { |
8249 | ret = -EINVAL; | |
2d091d62 | 8250 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 8251 | goto out_fput; |
08a45173 | 8252 | continue; |
792e3582 | 8253 | } |
6b06314c | 8254 | |
05f3fb3c | 8255 | file = fget(fd); |
6b06314c | 8256 | ret = -EBADF; |
792e3582 | 8257 | if (unlikely(!file)) |
600cf3f8 | 8258 | goto out_fput; |
05f3fb3c | 8259 | |
6b06314c JA |
8260 | /* |
8261 | * Don't allow io_uring instances to be registered. If UNIX | |
8262 | * isn't enabled, then this causes a reference cycle and this | |
8263 | * instance can never get freed. If UNIX is enabled we'll | |
8264 | * handle it just fine, but there's still no point in allowing | |
8265 | * a ring fd as it doesn't support regular read/write anyway. | |
8266 | */ | |
05f3fb3c JA |
8267 | if (file->f_op == &io_uring_fops) { |
8268 | fput(file); | |
600cf3f8 | 8269 | goto out_fput; |
6b06314c | 8270 | } |
aeca241b | 8271 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
8272 | } |
8273 | ||
6b06314c | 8274 | ret = io_sqe_files_scm(ctx); |
05589553 | 8275 | if (ret) { |
08480400 | 8276 | __io_sqe_files_unregister(ctx); |
05589553 XW |
8277 | return ret; |
8278 | } | |
6b06314c | 8279 | |
a7f0ed5a | 8280 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 8281 | return ret; |
600cf3f8 PB |
8282 | out_fput: |
8283 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8284 | file = io_file_from_index(ctx, i); | |
8285 | if (file) | |
8286 | fput(file); | |
8287 | } | |
042b0d85 | 8288 | io_free_file_tables(&ctx->file_table); |
600cf3f8 | 8289 | ctx->nr_user_files = 0; |
600cf3f8 | 8290 | out_free: |
44b31f2f | 8291 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 8292 | ctx->file_data = NULL; |
6b06314c JA |
8293 | return ret; |
8294 | } | |
8295 | ||
c3a31e60 JA |
8296 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
8297 | int index) | |
8298 | { | |
8299 | #if defined(CONFIG_UNIX) | |
8300 | struct sock *sock = ctx->ring_sock->sk; | |
8301 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
8302 | struct sk_buff *skb; | |
8303 | ||
8304 | /* | |
8305 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
8306 | * file set. If there's no room, fall back to allocating a new skb | |
8307 | * and filling it in. | |
8308 | */ | |
8309 | spin_lock_irq(&head->lock); | |
8310 | skb = skb_peek(head); | |
8311 | if (skb) { | |
8312 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
8313 | ||
8314 | if (fpl->count < SCM_MAX_FD) { | |
8315 | __skb_unlink(skb, head); | |
8316 | spin_unlock_irq(&head->lock); | |
8317 | fpl->fp[fpl->count] = get_file(file); | |
8318 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
8319 | fpl->count++; | |
8320 | spin_lock_irq(&head->lock); | |
8321 | __skb_queue_head(head, skb); | |
8322 | } else { | |
8323 | skb = NULL; | |
8324 | } | |
8325 | } | |
8326 | spin_unlock_irq(&head->lock); | |
8327 | ||
8328 | if (skb) { | |
8329 | fput(file); | |
8330 | return 0; | |
8331 | } | |
8332 | ||
8333 | return __io_sqe_files_scm(ctx, 1, index); | |
8334 | #else | |
8335 | return 0; | |
8336 | #endif | |
8337 | } | |
8338 | ||
9c7b0ba8 PB |
8339 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
8340 | struct io_rsrc_node *node, void *rsrc) | |
8341 | { | |
8342 | struct io_rsrc_put *prsrc; | |
8343 | ||
8344 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); | |
8345 | if (!prsrc) | |
8346 | return -ENOMEM; | |
8347 | ||
8348 | prsrc->tag = *io_get_tag_slot(data, idx); | |
8349 | prsrc->rsrc = rsrc; | |
8350 | list_add(&prsrc->list, &node->rsrc_list); | |
8351 | return 0; | |
8352 | } | |
8353 | ||
b9445598 PB |
8354 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
8355 | unsigned int issue_flags, u32 slot_index) | |
8356 | { | |
8357 | struct io_ring_ctx *ctx = req->ctx; | |
8358 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; | |
9c7b0ba8 | 8359 | bool needs_switch = false; |
b9445598 PB |
8360 | struct io_fixed_file *file_slot; |
8361 | int ret = -EBADF; | |
8362 | ||
8363 | io_ring_submit_lock(ctx, !force_nonblock); | |
8364 | if (file->f_op == &io_uring_fops) | |
8365 | goto err; | |
8366 | ret = -ENXIO; | |
8367 | if (!ctx->file_data) | |
8368 | goto err; | |
8369 | ret = -EINVAL; | |
8370 | if (slot_index >= ctx->nr_user_files) | |
8371 | goto err; | |
8372 | ||
8373 | slot_index = array_index_nospec(slot_index, ctx->nr_user_files); | |
8374 | file_slot = io_fixed_file_slot(&ctx->file_table, slot_index); | |
9c7b0ba8 PB |
8375 | |
8376 | if (file_slot->file_ptr) { | |
8377 | struct file *old_file; | |
8378 | ||
8379 | ret = io_rsrc_node_switch_start(ctx); | |
8380 | if (ret) | |
8381 | goto err; | |
8382 | ||
8383 | old_file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
8384 | ret = io_queue_rsrc_removal(ctx->file_data, slot_index, | |
8385 | ctx->rsrc_node, old_file); | |
8386 | if (ret) | |
8387 | goto err; | |
8388 | file_slot->file_ptr = 0; | |
8389 | needs_switch = true; | |
8390 | } | |
b9445598 PB |
8391 | |
8392 | *io_get_tag_slot(ctx->file_data, slot_index) = 0; | |
8393 | io_fixed_file_set(file_slot, file); | |
8394 | ret = io_sqe_file_register(ctx, file, slot_index); | |
8395 | if (ret) { | |
8396 | file_slot->file_ptr = 0; | |
8397 | goto err; | |
8398 | } | |
8399 | ||
8400 | ret = 0; | |
8401 | err: | |
9c7b0ba8 PB |
8402 | if (needs_switch) |
8403 | io_rsrc_node_switch(ctx, ctx->file_data); | |
b9445598 PB |
8404 | io_ring_submit_unlock(ctx, !force_nonblock); |
8405 | if (ret) | |
8406 | fput(file); | |
8407 | return ret; | |
8408 | } | |
8409 | ||
05f3fb3c | 8410 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
c3bdad02 | 8411 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
8412 | unsigned nr_args) |
8413 | { | |
c3bdad02 | 8414 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 8415 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 8416 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
8417 | struct io_fixed_file *file_slot; |
8418 | struct file *file; | |
98f0b3b4 PB |
8419 | int fd, i, err = 0; |
8420 | unsigned int done; | |
05589553 | 8421 | bool needs_switch = false; |
c3a31e60 | 8422 | |
98f0b3b4 PB |
8423 | if (!ctx->file_data) |
8424 | return -ENXIO; | |
8425 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
8426 | return -EINVAL; |
8427 | ||
67973b93 | 8428 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
8429 | u64 tag = 0; |
8430 | ||
8431 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
8432 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
8433 | err = -EFAULT; |
8434 | break; | |
8435 | } | |
c3bdad02 PB |
8436 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
8437 | err = -EINVAL; | |
8438 | break; | |
8439 | } | |
4e0377a1 | 8440 | if (fd == IORING_REGISTER_FILES_SKIP) |
8441 | continue; | |
8442 | ||
67973b93 | 8443 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 8444 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 8445 | |
a04b0ac0 PB |
8446 | if (file_slot->file_ptr) { |
8447 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
8448 | err = io_queue_rsrc_removal(data, up->offset + done, |
8449 | ctx->rsrc_node, file); | |
a5318d3c HD |
8450 | if (err) |
8451 | break; | |
a04b0ac0 | 8452 | file_slot->file_ptr = 0; |
05589553 | 8453 | needs_switch = true; |
c3a31e60 JA |
8454 | } |
8455 | if (fd != -1) { | |
c3a31e60 JA |
8456 | file = fget(fd); |
8457 | if (!file) { | |
8458 | err = -EBADF; | |
8459 | break; | |
8460 | } | |
8461 | /* | |
8462 | * Don't allow io_uring instances to be registered. If | |
8463 | * UNIX isn't enabled, then this causes a reference | |
8464 | * cycle and this instance can never get freed. If UNIX | |
8465 | * is enabled we'll handle it just fine, but there's | |
8466 | * still no point in allowing a ring fd as it doesn't | |
8467 | * support regular read/write anyway. | |
8468 | */ | |
8469 | if (file->f_op == &io_uring_fops) { | |
8470 | fput(file); | |
8471 | err = -EBADF; | |
8472 | break; | |
8473 | } | |
2d091d62 | 8474 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 8475 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 8476 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8477 | if (err) { |
a04b0ac0 | 8478 | file_slot->file_ptr = 0; |
f3bd9dae | 8479 | fput(file); |
c3a31e60 | 8480 | break; |
f3bd9dae | 8481 | } |
c3a31e60 | 8482 | } |
05f3fb3c JA |
8483 | } |
8484 | ||
a7f0ed5a PB |
8485 | if (needs_switch) |
8486 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
8487 | return done ? done : err; |
8488 | } | |
05589553 | 8489 | |
685fe7fe JA |
8490 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
8491 | struct task_struct *task) | |
24369c2e | 8492 | { |
e941894e | 8493 | struct io_wq_hash *hash; |
24369c2e | 8494 | struct io_wq_data data; |
24369c2e | 8495 | unsigned int concurrency; |
24369c2e | 8496 | |
362a9e65 | 8497 | mutex_lock(&ctx->uring_lock); |
e941894e JA |
8498 | hash = ctx->hash_map; |
8499 | if (!hash) { | |
8500 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
362a9e65 YY |
8501 | if (!hash) { |
8502 | mutex_unlock(&ctx->uring_lock); | |
e941894e | 8503 | return ERR_PTR(-ENOMEM); |
362a9e65 | 8504 | } |
e941894e JA |
8505 | refcount_set(&hash->refs, 1); |
8506 | init_waitqueue_head(&hash->wait); | |
8507 | ctx->hash_map = hash; | |
24369c2e | 8508 | } |
362a9e65 | 8509 | mutex_unlock(&ctx->uring_lock); |
24369c2e | 8510 | |
e941894e | 8511 | data.hash = hash; |
685fe7fe | 8512 | data.task = task; |
ebc11b6c | 8513 | data.free_work = io_wq_free_work; |
f5fa38c5 | 8514 | data.do_work = io_wq_submit_work; |
24369c2e | 8515 | |
d25e3a3d JA |
8516 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
8517 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 8518 | |
5aa75ed5 | 8519 | return io_wq_create(concurrency, &data); |
24369c2e PB |
8520 | } |
8521 | ||
5aa75ed5 JA |
8522 | static int io_uring_alloc_task_context(struct task_struct *task, |
8523 | struct io_ring_ctx *ctx) | |
0f212204 JA |
8524 | { |
8525 | struct io_uring_task *tctx; | |
d8a6df10 | 8526 | int ret; |
0f212204 | 8527 | |
09899b19 | 8528 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
8529 | if (unlikely(!tctx)) |
8530 | return -ENOMEM; | |
8531 | ||
d8a6df10 JA |
8532 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8533 | if (unlikely(ret)) { | |
8534 | kfree(tctx); | |
8535 | return ret; | |
8536 | } | |
8537 | ||
685fe7fe | 8538 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
8539 | if (IS_ERR(tctx->io_wq)) { |
8540 | ret = PTR_ERR(tctx->io_wq); | |
8541 | percpu_counter_destroy(&tctx->inflight); | |
8542 | kfree(tctx); | |
8543 | return ret; | |
8544 | } | |
8545 | ||
0f212204 JA |
8546 | xa_init(&tctx->xa); |
8547 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 8548 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 8549 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 8550 | task->io_uring = tctx; |
7cbf1722 JA |
8551 | spin_lock_init(&tctx->task_lock); |
8552 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 8553 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
8554 | return 0; |
8555 | } | |
8556 | ||
8557 | void __io_uring_free(struct task_struct *tsk) | |
8558 | { | |
8559 | struct io_uring_task *tctx = tsk->io_uring; | |
8560 | ||
8561 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 8562 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 8563 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 8564 | |
d8a6df10 | 8565 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8566 | kfree(tctx); |
8567 | tsk->io_uring = NULL; | |
8568 | } | |
8569 | ||
7e84e1c7 SG |
8570 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8571 | struct io_uring_params *p) | |
2b188cc1 JA |
8572 | { |
8573 | int ret; | |
8574 | ||
d25e3a3d JA |
8575 | /* Retain compatibility with failing for an invalid attach attempt */ |
8576 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
8577 | IORING_SETUP_ATTACH_WQ) { | |
8578 | struct fd f; | |
8579 | ||
8580 | f = fdget(p->wq_fd); | |
8581 | if (!f.file) | |
8582 | return -ENXIO; | |
0cc936f7 JA |
8583 | if (f.file->f_op != &io_uring_fops) { |
8584 | fdput(f); | |
f2a48dd0 | 8585 | return -EINVAL; |
0cc936f7 JA |
8586 | } |
8587 | fdput(f); | |
d25e3a3d | 8588 | } |
6c271ce2 | 8589 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 8590 | struct task_struct *tsk; |
534ca6d6 | 8591 | struct io_sq_data *sqd; |
26984fbf | 8592 | bool attached; |
534ca6d6 | 8593 | |
26984fbf | 8594 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8595 | if (IS_ERR(sqd)) { |
8596 | ret = PTR_ERR(sqd); | |
8597 | goto err; | |
8598 | } | |
69fb2131 | 8599 | |
7c30f36a | 8600 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8601 | ctx->sq_data = sqd; |
917257da JA |
8602 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8603 | if (!ctx->sq_thread_idle) | |
8604 | ctx->sq_thread_idle = HZ; | |
8605 | ||
78d7f6ba | 8606 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8607 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8608 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8609 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8610 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8611 | io_sq_thread_unpark(sqd); |
8612 | ||
de75a3d3 PB |
8613 | if (ret < 0) |
8614 | goto err; | |
8615 | if (attached) | |
5aa75ed5 | 8616 | return 0; |
aa06165d | 8617 | |
6c271ce2 | 8618 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8619 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8620 | |
917257da | 8621 | ret = -EINVAL; |
f2a48dd0 | 8622 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8623 | goto err_sqpoll; |
37d1e2e3 | 8624 | sqd->sq_cpu = cpu; |
6c271ce2 | 8625 | } else { |
37d1e2e3 | 8626 | sqd->sq_cpu = -1; |
6c271ce2 | 8627 | } |
37d1e2e3 JA |
8628 | |
8629 | sqd->task_pid = current->pid; | |
5c2469e0 | 8630 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8631 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8632 | if (IS_ERR(tsk)) { | |
8633 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8634 | goto err_sqpoll; |
6c271ce2 | 8635 | } |
97a73a0f | 8636 | |
46fe18b1 | 8637 | sqd->thread = tsk; |
97a73a0f | 8638 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8639 | wake_up_new_task(tsk); |
0f212204 JA |
8640 | if (ret) |
8641 | goto err; | |
6c271ce2 JA |
8642 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8643 | /* Can't have SQ_AFF without SQPOLL */ | |
8644 | ret = -EINVAL; | |
8645 | goto err; | |
8646 | } | |
8647 | ||
2b188cc1 | 8648 | return 0; |
f2a48dd0 PB |
8649 | err_sqpoll: |
8650 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8651 | err: |
37d1e2e3 | 8652 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8653 | return ret; |
8654 | } | |
8655 | ||
a087e2b5 BM |
8656 | static inline void __io_unaccount_mem(struct user_struct *user, |
8657 | unsigned long nr_pages) | |
2b188cc1 JA |
8658 | { |
8659 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8660 | } | |
8661 | ||
a087e2b5 BM |
8662 | static inline int __io_account_mem(struct user_struct *user, |
8663 | unsigned long nr_pages) | |
2b188cc1 JA |
8664 | { |
8665 | unsigned long page_limit, cur_pages, new_pages; | |
8666 | ||
8667 | /* Don't allow more pages than we can safely lock */ | |
8668 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8669 | ||
8670 | do { | |
8671 | cur_pages = atomic_long_read(&user->locked_vm); | |
8672 | new_pages = cur_pages + nr_pages; | |
8673 | if (new_pages > page_limit) | |
8674 | return -ENOMEM; | |
8675 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8676 | new_pages) != cur_pages); | |
8677 | ||
8678 | return 0; | |
8679 | } | |
8680 | ||
26bfa89e | 8681 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8682 | { |
62e398be | 8683 | if (ctx->user) |
a087e2b5 | 8684 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8685 | |
26bfa89e JA |
8686 | if (ctx->mm_account) |
8687 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8688 | } |
8689 | ||
26bfa89e | 8690 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8691 | { |
30975825 BM |
8692 | int ret; |
8693 | ||
62e398be | 8694 | if (ctx->user) { |
30975825 BM |
8695 | ret = __io_account_mem(ctx->user, nr_pages); |
8696 | if (ret) | |
8697 | return ret; | |
8698 | } | |
8699 | ||
26bfa89e JA |
8700 | if (ctx->mm_account) |
8701 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8702 | |
8703 | return 0; | |
8704 | } | |
8705 | ||
2b188cc1 JA |
8706 | static void io_mem_free(void *ptr) |
8707 | { | |
52e04ef4 MR |
8708 | struct page *page; |
8709 | ||
8710 | if (!ptr) | |
8711 | return; | |
2b188cc1 | 8712 | |
52e04ef4 | 8713 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8714 | if (put_page_testzero(page)) |
8715 | free_compound_page(page); | |
8716 | } | |
8717 | ||
8718 | static void *io_mem_alloc(size_t size) | |
8719 | { | |
8720 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8721 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8722 | |
8723 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8724 | } | |
8725 | ||
75b28aff HV |
8726 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8727 | size_t *sq_offset) | |
8728 | { | |
8729 | struct io_rings *rings; | |
8730 | size_t off, sq_array_size; | |
8731 | ||
8732 | off = struct_size(rings, cqes, cq_entries); | |
8733 | if (off == SIZE_MAX) | |
8734 | return SIZE_MAX; | |
8735 | ||
8736 | #ifdef CONFIG_SMP | |
8737 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8738 | if (off == 0) | |
8739 | return SIZE_MAX; | |
8740 | #endif | |
8741 | ||
b36200f5 DV |
8742 | if (sq_offset) |
8743 | *sq_offset = off; | |
8744 | ||
75b28aff HV |
8745 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8746 | if (sq_array_size == SIZE_MAX) | |
8747 | return SIZE_MAX; | |
8748 | ||
8749 | if (check_add_overflow(off, sq_array_size, &off)) | |
8750 | return SIZE_MAX; | |
8751 | ||
75b28aff HV |
8752 | return off; |
8753 | } | |
8754 | ||
41edf1a5 | 8755 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8756 | { |
41edf1a5 | 8757 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8758 | unsigned int i; |
8759 | ||
6224843d PB |
8760 | if (imu != ctx->dummy_ubuf) { |
8761 | for (i = 0; i < imu->nr_bvecs; i++) | |
8762 | unpin_user_page(imu->bvec[i].bv_page); | |
8763 | if (imu->acct_pages) | |
8764 | io_unaccount_mem(ctx, imu->acct_pages); | |
8765 | kvfree(imu); | |
8766 | } | |
41edf1a5 | 8767 | *slot = NULL; |
7f61a1e9 PB |
8768 | } |
8769 | ||
bd54b6fe | 8770 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8771 | { |
634d00df PB |
8772 | io_buffer_unmap(ctx, &prsrc->buf); |
8773 | prsrc->buf = NULL; | |
bd54b6fe | 8774 | } |
edafccee | 8775 | |
bd54b6fe BM |
8776 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8777 | { | |
8778 | unsigned int i; | |
edafccee | 8779 | |
7f61a1e9 PB |
8780 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8781 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8782 | kfree(ctx->user_bufs); |
bb6659cc | 8783 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8784 | ctx->user_bufs = NULL; |
bd54b6fe | 8785 | ctx->buf_data = NULL; |
edafccee | 8786 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8787 | } |
8788 | ||
0a96bbe4 | 8789 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8790 | { |
bd54b6fe | 8791 | int ret; |
edafccee | 8792 | |
bd54b6fe | 8793 | if (!ctx->buf_data) |
edafccee JA |
8794 | return -ENXIO; |
8795 | ||
bd54b6fe BM |
8796 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8797 | if (!ret) | |
8798 | __io_sqe_buffers_unregister(ctx); | |
8799 | return ret; | |
edafccee JA |
8800 | } |
8801 | ||
8802 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8803 | void __user *arg, unsigned index) | |
8804 | { | |
8805 | struct iovec __user *src; | |
8806 | ||
8807 | #ifdef CONFIG_COMPAT | |
8808 | if (ctx->compat) { | |
8809 | struct compat_iovec __user *ciovs; | |
8810 | struct compat_iovec ciov; | |
8811 | ||
8812 | ciovs = (struct compat_iovec __user *) arg; | |
8813 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8814 | return -EFAULT; | |
8815 | ||
d55e5f5b | 8816 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8817 | dst->iov_len = ciov.iov_len; |
8818 | return 0; | |
8819 | } | |
8820 | #endif | |
8821 | src = (struct iovec __user *) arg; | |
8822 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8823 | return -EFAULT; | |
8824 | return 0; | |
8825 | } | |
8826 | ||
de293938 JA |
8827 | /* |
8828 | * Not super efficient, but this is just a registration time. And we do cache | |
8829 | * the last compound head, so generally we'll only do a full search if we don't | |
8830 | * match that one. | |
8831 | * | |
8832 | * We check if the given compound head page has already been accounted, to | |
8833 | * avoid double accounting it. This allows us to account the full size of the | |
8834 | * page, not just the constituent pages of a huge page. | |
8835 | */ | |
8836 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8837 | int nr_pages, struct page *hpage) | |
8838 | { | |
8839 | int i, j; | |
8840 | ||
8841 | /* check current page array */ | |
8842 | for (i = 0; i < nr_pages; i++) { | |
8843 | if (!PageCompound(pages[i])) | |
8844 | continue; | |
8845 | if (compound_head(pages[i]) == hpage) | |
8846 | return true; | |
8847 | } | |
8848 | ||
8849 | /* check previously registered pages */ | |
8850 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8851 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8852 | |
8853 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8854 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8855 | continue; | |
8856 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8857 | return true; | |
8858 | } | |
8859 | } | |
8860 | ||
8861 | return false; | |
8862 | } | |
8863 | ||
8864 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8865 | int nr_pages, struct io_mapped_ubuf *imu, | |
8866 | struct page **last_hpage) | |
8867 | { | |
8868 | int i, ret; | |
8869 | ||
216e5835 | 8870 | imu->acct_pages = 0; |
de293938 JA |
8871 | for (i = 0; i < nr_pages; i++) { |
8872 | if (!PageCompound(pages[i])) { | |
8873 | imu->acct_pages++; | |
8874 | } else { | |
8875 | struct page *hpage; | |
8876 | ||
8877 | hpage = compound_head(pages[i]); | |
8878 | if (hpage == *last_hpage) | |
8879 | continue; | |
8880 | *last_hpage = hpage; | |
8881 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8882 | continue; | |
8883 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8884 | } | |
8885 | } | |
8886 | ||
8887 | if (!imu->acct_pages) | |
8888 | return 0; | |
8889 | ||
26bfa89e | 8890 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8891 | if (ret) |
8892 | imu->acct_pages = 0; | |
8893 | return ret; | |
8894 | } | |
8895 | ||
0a96bbe4 | 8896 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8897 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8898 | struct page **last_hpage) |
edafccee | 8899 | { |
41edf1a5 | 8900 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8901 | struct vm_area_struct **vmas = NULL; |
8902 | struct page **pages = NULL; | |
0a96bbe4 BM |
8903 | unsigned long off, start, end, ubuf; |
8904 | size_t size; | |
8905 | int ret, pret, nr_pages, i; | |
8906 | ||
6224843d PB |
8907 | if (!iov->iov_base) { |
8908 | *pimu = ctx->dummy_ubuf; | |
8909 | return 0; | |
8910 | } | |
8911 | ||
0a96bbe4 BM |
8912 | ubuf = (unsigned long) iov->iov_base; |
8913 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8914 | start = ubuf >> PAGE_SHIFT; | |
8915 | nr_pages = end - start; | |
8916 | ||
41edf1a5 | 8917 | *pimu = NULL; |
0a96bbe4 BM |
8918 | ret = -ENOMEM; |
8919 | ||
8920 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8921 | if (!pages) | |
8922 | goto done; | |
8923 | ||
8924 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8925 | GFP_KERNEL); | |
8926 | if (!vmas) | |
8927 | goto done; | |
edafccee | 8928 | |
41edf1a5 | 8929 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8930 | if (!imu) |
0a96bbe4 BM |
8931 | goto done; |
8932 | ||
8933 | ret = 0; | |
8934 | mmap_read_lock(current->mm); | |
8935 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8936 | pages, vmas); | |
8937 | if (pret == nr_pages) { | |
8938 | /* don't support file backed memory */ | |
8939 | for (i = 0; i < nr_pages; i++) { | |
8940 | struct vm_area_struct *vma = vmas[i]; | |
8941 | ||
40dad765 PB |
8942 | if (vma_is_shmem(vma)) |
8943 | continue; | |
0a96bbe4 BM |
8944 | if (vma->vm_file && |
8945 | !is_file_hugepages(vma->vm_file)) { | |
8946 | ret = -EOPNOTSUPP; | |
8947 | break; | |
8948 | } | |
8949 | } | |
8950 | } else { | |
8951 | ret = pret < 0 ? pret : -EFAULT; | |
8952 | } | |
8953 | mmap_read_unlock(current->mm); | |
8954 | if (ret) { | |
8955 | /* | |
8956 | * if we did partial map, or found file backed vmas, | |
8957 | * release any pages we did get | |
8958 | */ | |
8959 | if (pret > 0) | |
8960 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8961 | goto done; |
8962 | } | |
8963 | ||
8964 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8965 | if (ret) { | |
8966 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8967 | goto done; |
8968 | } | |
8969 | ||
8970 | off = ubuf & ~PAGE_MASK; | |
8971 | size = iov->iov_len; | |
8972 | for (i = 0; i < nr_pages; i++) { | |
8973 | size_t vec_len; | |
8974 | ||
8975 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8976 | imu->bvec[i].bv_page = pages[i]; | |
8977 | imu->bvec[i].bv_len = vec_len; | |
8978 | imu->bvec[i].bv_offset = off; | |
8979 | off = 0; | |
8980 | size -= vec_len; | |
8981 | } | |
8982 | /* store original address for later verification */ | |
8983 | imu->ubuf = ubuf; | |
4751f53d | 8984 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8985 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8986 | *pimu = imu; |
0a96bbe4 BM |
8987 | ret = 0; |
8988 | done: | |
41edf1a5 PB |
8989 | if (ret) |
8990 | kvfree(imu); | |
0a96bbe4 BM |
8991 | kvfree(pages); |
8992 | kvfree(vmas); | |
8993 | return ret; | |
8994 | } | |
8995 | ||
2b358604 | 8996 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8997 | { |
87094465 PB |
8998 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8999 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 9000 | } |
edafccee | 9001 | |
2b358604 BM |
9002 | static int io_buffer_validate(struct iovec *iov) |
9003 | { | |
50e96989 PB |
9004 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
9005 | ||
2b358604 BM |
9006 | /* |
9007 | * Don't impose further limits on the size and buffer | |
9008 | * constraints here, we'll -EINVAL later when IO is | |
9009 | * submitted if they are wrong. | |
9010 | */ | |
6224843d PB |
9011 | if (!iov->iov_base) |
9012 | return iov->iov_len ? -EFAULT : 0; | |
9013 | if (!iov->iov_len) | |
2b358604 | 9014 | return -EFAULT; |
edafccee | 9015 | |
2b358604 BM |
9016 | /* arbitrary limit, but we need something */ |
9017 | if (iov->iov_len > SZ_1G) | |
9018 | return -EFAULT; | |
edafccee | 9019 | |
50e96989 PB |
9020 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
9021 | return -EOVERFLOW; | |
9022 | ||
2b358604 BM |
9023 | return 0; |
9024 | } | |
edafccee | 9025 | |
2b358604 | 9026 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 9027 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 9028 | { |
bd54b6fe BM |
9029 | struct page *last_hpage = NULL; |
9030 | struct io_rsrc_data *data; | |
2b358604 BM |
9031 | int i, ret; |
9032 | struct iovec iov; | |
edafccee | 9033 | |
87094465 PB |
9034 | if (ctx->user_bufs) |
9035 | return -EBUSY; | |
489809e2 | 9036 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 9037 | return -EINVAL; |
bd54b6fe | 9038 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
9039 | if (ret) |
9040 | return ret; | |
d878c816 PB |
9041 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
9042 | if (ret) | |
9043 | return ret; | |
bd54b6fe BM |
9044 | ret = io_buffers_map_alloc(ctx, nr_args); |
9045 | if (ret) { | |
bb6659cc | 9046 | io_rsrc_data_free(data); |
bd54b6fe BM |
9047 | return ret; |
9048 | } | |
edafccee | 9049 | |
87094465 | 9050 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
9051 | ret = io_copy_iov(ctx, &iov, arg, i); |
9052 | if (ret) | |
0a96bbe4 | 9053 | break; |
2b358604 BM |
9054 | ret = io_buffer_validate(&iov); |
9055 | if (ret) | |
0a96bbe4 | 9056 | break; |
2d091d62 | 9057 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
9058 | ret = -EINVAL; |
9059 | break; | |
9060 | } | |
edafccee | 9061 | |
41edf1a5 PB |
9062 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
9063 | &last_hpage); | |
0a96bbe4 BM |
9064 | if (ret) |
9065 | break; | |
edafccee | 9066 | } |
0a96bbe4 | 9067 | |
bd54b6fe | 9068 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 9069 | |
bd54b6fe BM |
9070 | ctx->buf_data = data; |
9071 | if (ret) | |
9072 | __io_sqe_buffers_unregister(ctx); | |
9073 | else | |
9074 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
9075 | return ret; |
9076 | } | |
9077 | ||
634d00df PB |
9078 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
9079 | struct io_uring_rsrc_update2 *up, | |
9080 | unsigned int nr_args) | |
9081 | { | |
9082 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
9083 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
9084 | struct page *last_hpage = NULL; |
9085 | bool needs_switch = false; | |
9086 | __u32 done; | |
9087 | int i, err; | |
9088 | ||
9089 | if (!ctx->buf_data) | |
9090 | return -ENXIO; | |
9091 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
9092 | return -EINVAL; | |
9093 | ||
9094 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
9095 | struct io_mapped_ubuf *imu; |
9096 | int offset = up->offset + done; | |
634d00df PB |
9097 | u64 tag = 0; |
9098 | ||
9099 | err = io_copy_iov(ctx, &iov, iovs, done); | |
9100 | if (err) | |
9101 | break; | |
9102 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
9103 | err = -EFAULT; | |
9104 | break; | |
9105 | } | |
0b8c0e7c PB |
9106 | err = io_buffer_validate(&iov); |
9107 | if (err) | |
9108 | break; | |
cf3770e7 CIK |
9109 | if (!iov.iov_base && tag) { |
9110 | err = -EINVAL; | |
9111 | break; | |
9112 | } | |
0b8c0e7c PB |
9113 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
9114 | if (err) | |
9115 | break; | |
634d00df | 9116 | |
0b8c0e7c | 9117 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 9118 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
9119 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
9120 | ctx->rsrc_node, ctx->user_bufs[i]); | |
9121 | if (unlikely(err)) { | |
9122 | io_buffer_unmap(ctx, &imu); | |
634d00df | 9123 | break; |
0b8c0e7c | 9124 | } |
634d00df PB |
9125 | ctx->user_bufs[i] = NULL; |
9126 | needs_switch = true; | |
9127 | } | |
9128 | ||
0b8c0e7c | 9129 | ctx->user_bufs[i] = imu; |
2d091d62 | 9130 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
9131 | } |
9132 | ||
9133 | if (needs_switch) | |
9134 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
9135 | return done ? done : err; | |
9136 | } | |
9137 | ||
9b402849 JA |
9138 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
9139 | { | |
9140 | __s32 __user *fds = arg; | |
9141 | int fd; | |
9142 | ||
9143 | if (ctx->cq_ev_fd) | |
9144 | return -EBUSY; | |
9145 | ||
9146 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
9147 | return -EFAULT; | |
9148 | ||
9149 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
9150 | if (IS_ERR(ctx->cq_ev_fd)) { | |
9151 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
fe7e3257 | 9152 | |
9b402849 JA |
9153 | ctx->cq_ev_fd = NULL; |
9154 | return ret; | |
9155 | } | |
9156 | ||
9157 | return 0; | |
9158 | } | |
9159 | ||
9160 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
9161 | { | |
9162 | if (ctx->cq_ev_fd) { | |
9163 | eventfd_ctx_put(ctx->cq_ev_fd); | |
9164 | ctx->cq_ev_fd = NULL; | |
9165 | return 0; | |
9166 | } | |
9167 | ||
9168 | return -ENXIO; | |
9169 | } | |
9170 | ||
5a2e745d JA |
9171 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
9172 | { | |
9e15c3a0 JA |
9173 | struct io_buffer *buf; |
9174 | unsigned long index; | |
9175 | ||
8bab4c09 | 9176 | xa_for_each(&ctx->io_buffers, index, buf) { |
9e15c3a0 | 9177 | __io_remove_buffers(ctx, buf, index, -1U); |
8bab4c09 JA |
9178 | cond_resched(); |
9179 | } | |
5a2e745d JA |
9180 | } |
9181 | ||
7255834e | 9182 | static void io_req_cache_free(struct list_head *list) |
1b4c351f | 9183 | { |
68e68ee6 | 9184 | struct io_kiocb *req, *nxt; |
1b4c351f | 9185 | |
bb943b82 PB |
9186 | list_for_each_entry_safe(req, nxt, list, inflight_entry) { |
9187 | list_del(&req->inflight_entry); | |
1b4c351f JA |
9188 | kmem_cache_free(req_cachep, req); |
9189 | } | |
9190 | } | |
9191 | ||
4010fec4 | 9192 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 9193 | { |
cd0ca2e0 | 9194 | struct io_submit_state *state = &ctx->submit_state; |
bf019da7 | 9195 | |
9a4fdbd8 JA |
9196 | mutex_lock(&ctx->uring_lock); |
9197 | ||
cd0ca2e0 PB |
9198 | if (state->free_reqs) { |
9199 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); | |
9200 | state->free_reqs = 0; | |
8e5c66c4 | 9201 | } |
9a4fdbd8 | 9202 | |
cd0ca2e0 PB |
9203 | io_flush_cached_locked_reqs(ctx, state); |
9204 | io_req_cache_free(&state->free_list); | |
9a4fdbd8 JA |
9205 | mutex_unlock(&ctx->uring_lock); |
9206 | } | |
9207 | ||
43597aac | 9208 | static void io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 9209 | { |
43597aac | 9210 | if (data && !atomic_dec_and_test(&data->refs)) |
bd54b6fe | 9211 | wait_for_completion(&data->done); |
bd54b6fe | 9212 | } |
04fc6c80 | 9213 | |
2b188cc1 JA |
9214 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
9215 | { | |
37d1e2e3 | 9216 | io_sq_thread_finish(ctx); |
2aede0e4 | 9217 | |
37d1e2e3 | 9218 | if (ctx->mm_account) { |
2aede0e4 JA |
9219 | mmdrop(ctx->mm_account); |
9220 | ctx->mm_account = NULL; | |
30975825 | 9221 | } |
def596e9 | 9222 | |
43597aac PB |
9223 | /* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */ |
9224 | io_wait_rsrc_data(ctx->buf_data); | |
9225 | io_wait_rsrc_data(ctx->file_data); | |
9226 | ||
8bad28d8 | 9227 | mutex_lock(&ctx->uring_lock); |
43597aac | 9228 | if (ctx->buf_data) |
bd54b6fe | 9229 | __io_sqe_buffers_unregister(ctx); |
43597aac | 9230 | if (ctx->file_data) |
08480400 | 9231 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
9232 | if (ctx->rings) |
9233 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 9234 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 9235 | io_eventfd_unregister(ctx); |
5a2e745d | 9236 | io_destroy_buffers(ctx); |
07db298a PB |
9237 | if (ctx->sq_creds) |
9238 | put_cred(ctx->sq_creds); | |
def596e9 | 9239 | |
a7f0ed5a PB |
9240 | /* there are no registered resources left, nobody uses it */ |
9241 | if (ctx->rsrc_node) | |
9242 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 9243 | if (ctx->rsrc_backup_node) |
b895c9a6 | 9244 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
9245 | flush_delayed_work(&ctx->rsrc_put_work); |
9246 | ||
9247 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
9248 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 9249 | |
2b188cc1 | 9250 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
9251 | if (ctx->ring_sock) { |
9252 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 9253 | sock_release(ctx->ring_sock); |
355e8d26 | 9254 | } |
2b188cc1 | 9255 | #endif |
ef9dd637 | 9256 | WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list)); |
2b188cc1 | 9257 | |
75b28aff | 9258 | io_mem_free(ctx->rings); |
2b188cc1 | 9259 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
9260 | |
9261 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 9262 | free_uid(ctx->user); |
4010fec4 | 9263 | io_req_caches_free(ctx); |
e941894e JA |
9264 | if (ctx->hash_map) |
9265 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 9266 | kfree(ctx->cancel_hash); |
6224843d | 9267 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
9268 | kfree(ctx); |
9269 | } | |
9270 | ||
9271 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
9272 | { | |
9273 | struct io_ring_ctx *ctx = file->private_data; | |
9274 | __poll_t mask = 0; | |
9275 | ||
311997b3 | 9276 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
9277 | /* |
9278 | * synchronizes with barrier from wq_has_sleeper call in | |
9279 | * io_commit_cqring | |
9280 | */ | |
2b188cc1 | 9281 | smp_rmb(); |
90554200 | 9282 | if (!io_sqring_full(ctx)) |
2b188cc1 | 9283 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
9284 | |
9285 | /* | |
9286 | * Don't flush cqring overflow list here, just do a simple check. | |
9287 | * Otherwise there could possible be ABBA deadlock: | |
9288 | * CPU0 CPU1 | |
9289 | * ---- ---- | |
9290 | * lock(&ctx->uring_lock); | |
9291 | * lock(&ep->mtx); | |
9292 | * lock(&ctx->uring_lock); | |
9293 | * lock(&ep->mtx); | |
9294 | * | |
9295 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
9296 | * pushs them to do the flush. | |
9297 | */ | |
5ed7a37d | 9298 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
9299 | mask |= EPOLLIN | EPOLLRDNORM; |
9300 | ||
9301 | return mask; | |
9302 | } | |
9303 | ||
9304 | static int io_uring_fasync(int fd, struct file *file, int on) | |
9305 | { | |
9306 | struct io_ring_ctx *ctx = file->private_data; | |
9307 | ||
9308 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
9309 | } | |
9310 | ||
0bead8cd | 9311 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 9312 | { |
4379bf8b | 9313 | const struct cred *creds; |
071698e1 | 9314 | |
61cf9370 | 9315 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
9316 | if (creds) { |
9317 | put_cred(creds); | |
0bead8cd | 9318 | return 0; |
1e6fa521 | 9319 | } |
0bead8cd YD |
9320 | |
9321 | return -EINVAL; | |
9322 | } | |
9323 | ||
d56d938b PB |
9324 | struct io_tctx_exit { |
9325 | struct callback_head task_work; | |
9326 | struct completion completion; | |
baf186c4 | 9327 | struct io_ring_ctx *ctx; |
d56d938b PB |
9328 | }; |
9329 | ||
9330 | static void io_tctx_exit_cb(struct callback_head *cb) | |
9331 | { | |
9332 | struct io_uring_task *tctx = current->io_uring; | |
9333 | struct io_tctx_exit *work; | |
9334 | ||
9335 | work = container_of(cb, struct io_tctx_exit, task_work); | |
9336 | /* | |
9337 | * When @in_idle, we're in cancellation and it's racy to remove the | |
9338 | * node. It'll be removed by the end of cancellation, just ignore it. | |
9339 | */ | |
9340 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 9341 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
9342 | complete(&work->completion); |
9343 | } | |
9344 | ||
28090c13 PB |
9345 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
9346 | { | |
9347 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
9348 | ||
9349 | return req->ctx == data; | |
9350 | } | |
9351 | ||
85faa7b8 JA |
9352 | static void io_ring_exit_work(struct work_struct *work) |
9353 | { | |
d56d938b | 9354 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 9355 | unsigned long timeout = jiffies + HZ * 60 * 5; |
58d3be2c | 9356 | unsigned long interval = HZ / 20; |
d56d938b PB |
9357 | struct io_tctx_exit exit; |
9358 | struct io_tctx_node *node; | |
9359 | int ret; | |
85faa7b8 | 9360 | |
56952e91 JA |
9361 | /* |
9362 | * If we're doing polled IO and end up having requests being | |
9363 | * submitted async (out-of-line), then completions can come in while | |
9364 | * we're waiting for refs to drop. We need to reap these manually, | |
9365 | * as nobody else will be looking for them. | |
9366 | */ | |
b2edc0a7 | 9367 | do { |
3dd0c97a | 9368 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
9369 | if (ctx->sq_data) { |
9370 | struct io_sq_data *sqd = ctx->sq_data; | |
9371 | struct task_struct *tsk; | |
9372 | ||
9373 | io_sq_thread_park(sqd); | |
9374 | tsk = sqd->thread; | |
9375 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
9376 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
9377 | io_cancel_ctx_cb, ctx, true); | |
9378 | io_sq_thread_unpark(sqd); | |
9379 | } | |
b5bb3a24 | 9380 | |
58d3be2c PB |
9381 | if (WARN_ON_ONCE(time_after(jiffies, timeout))) { |
9382 | /* there is little hope left, don't run it too often */ | |
9383 | interval = HZ * 60; | |
9384 | } | |
9385 | } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); | |
d56d938b | 9386 | |
7f00651a PB |
9387 | init_completion(&exit.completion); |
9388 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
9389 | exit.ctx = ctx; | |
89b5066e PB |
9390 | /* |
9391 | * Some may use context even when all refs and requests have been put, | |
9392 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 9393 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
9394 | * this lock/unlock section also waits them to finish. |
9395 | */ | |
d56d938b PB |
9396 | mutex_lock(&ctx->uring_lock); |
9397 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
9398 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
9399 | ||
d56d938b PB |
9400 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
9401 | ctx_node); | |
7f00651a PB |
9402 | /* don't spin on a single task if cancellation failed */ |
9403 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
9404 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
9405 | if (WARN_ON_ONCE(ret)) | |
9406 | continue; | |
9407 | wake_up_process(node->task); | |
9408 | ||
9409 | mutex_unlock(&ctx->uring_lock); | |
9410 | wait_for_completion(&exit.completion); | |
d56d938b PB |
9411 | mutex_lock(&ctx->uring_lock); |
9412 | } | |
9413 | mutex_unlock(&ctx->uring_lock); | |
79ebeaee JA |
9414 | spin_lock(&ctx->completion_lock); |
9415 | spin_unlock(&ctx->completion_lock); | |
d56d938b | 9416 | |
85faa7b8 JA |
9417 | io_ring_ctx_free(ctx); |
9418 | } | |
9419 | ||
80c4cbdb PB |
9420 | /* Returns true if we found and killed one or more timeouts */ |
9421 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 9422 | bool cancel_all) |
80c4cbdb PB |
9423 | { |
9424 | struct io_kiocb *req, *tmp; | |
9425 | int canceled = 0; | |
9426 | ||
79ebeaee JA |
9427 | spin_lock(&ctx->completion_lock); |
9428 | spin_lock_irq(&ctx->timeout_lock); | |
80c4cbdb | 9429 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
3dd0c97a | 9430 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
9431 | io_kill_timeout(req, -ECANCELED); |
9432 | canceled++; | |
9433 | } | |
9434 | } | |
79ebeaee | 9435 | spin_unlock_irq(&ctx->timeout_lock); |
51520426 PB |
9436 | if (canceled != 0) |
9437 | io_commit_cqring(ctx); | |
79ebeaee | 9438 | spin_unlock(&ctx->completion_lock); |
80c4cbdb PB |
9439 | if (canceled != 0) |
9440 | io_cqring_ev_posted(ctx); | |
9441 | return canceled != 0; | |
9442 | } | |
9443 | ||
2b188cc1 JA |
9444 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
9445 | { | |
61cf9370 MWO |
9446 | unsigned long index; |
9447 | struct creds *creds; | |
9448 | ||
2b188cc1 JA |
9449 | mutex_lock(&ctx->uring_lock); |
9450 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 9451 | if (ctx->rings) |
6c2450ae | 9452 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
9453 | xa_for_each(&ctx->personalities, index, creds) |
9454 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
9455 | mutex_unlock(&ctx->uring_lock); |
9456 | ||
3dd0c97a PB |
9457 | io_kill_timeouts(ctx, NULL, true); |
9458 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 9459 | |
15dff286 | 9460 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 9461 | io_iopoll_try_reap_events(ctx); |
309fc03a | 9462 | |
85faa7b8 | 9463 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
9464 | /* |
9465 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
9466 | * if we're exiting a ton of rings at the same time. It just adds | |
9467 | * noise and overhead, there's no discernable change in runtime | |
9468 | * over using system_wq. | |
9469 | */ | |
9470 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
9471 | } |
9472 | ||
9473 | static int io_uring_release(struct inode *inode, struct file *file) | |
9474 | { | |
9475 | struct io_ring_ctx *ctx = file->private_data; | |
9476 | ||
9477 | file->private_data = NULL; | |
9478 | io_ring_ctx_wait_and_kill(ctx); | |
9479 | return 0; | |
9480 | } | |
9481 | ||
f6edbabb PB |
9482 | struct io_task_cancel { |
9483 | struct task_struct *task; | |
3dd0c97a | 9484 | bool all; |
f6edbabb | 9485 | }; |
f254ac04 | 9486 | |
f6edbabb | 9487 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 9488 | { |
9a472ef7 | 9489 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 9490 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
9491 | bool ret; |
9492 | ||
3dd0c97a | 9493 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
9494 | struct io_ring_ctx *ctx = req->ctx; |
9495 | ||
9496 | /* protect against races with linked timeouts */ | |
79ebeaee | 9497 | spin_lock(&ctx->completion_lock); |
3dd0c97a | 9498 | ret = io_match_task(req, cancel->task, cancel->all); |
79ebeaee | 9499 | spin_unlock(&ctx->completion_lock); |
9a472ef7 | 9500 | } else { |
3dd0c97a | 9501 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
9502 | } |
9503 | return ret; | |
b711d4ea JA |
9504 | } |
9505 | ||
e1915f76 | 9506 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 9507 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 9508 | { |
e1915f76 | 9509 | struct io_defer_entry *de; |
b7ddce3c PB |
9510 | LIST_HEAD(list); |
9511 | ||
79ebeaee | 9512 | spin_lock(&ctx->completion_lock); |
b7ddce3c | 9513 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { |
3dd0c97a | 9514 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
9515 | list_cut_position(&list, &ctx->defer_list, &de->list); |
9516 | break; | |
9517 | } | |
9518 | } | |
79ebeaee | 9519 | spin_unlock(&ctx->completion_lock); |
e1915f76 PB |
9520 | if (list_empty(&list)) |
9521 | return false; | |
b7ddce3c PB |
9522 | |
9523 | while (!list_empty(&list)) { | |
9524 | de = list_first_entry(&list, struct io_defer_entry, list); | |
9525 | list_del_init(&de->list); | |
f41db273 | 9526 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
9527 | kfree(de); |
9528 | } | |
e1915f76 | 9529 | return true; |
b7ddce3c PB |
9530 | } |
9531 | ||
1b00764f PB |
9532 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
9533 | { | |
9534 | struct io_tctx_node *node; | |
9535 | enum io_wq_cancel cret; | |
9536 | bool ret = false; | |
9537 | ||
9538 | mutex_lock(&ctx->uring_lock); | |
9539 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
9540 | struct io_uring_task *tctx = node->task->io_uring; | |
9541 | ||
9542 | /* | |
9543 | * io_wq will stay alive while we hold uring_lock, because it's | |
9544 | * killed after ctx nodes, which requires to take the lock. | |
9545 | */ | |
9546 | if (!tctx || !tctx->io_wq) | |
9547 | continue; | |
9548 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
9549 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9550 | } | |
9551 | mutex_unlock(&ctx->uring_lock); | |
9552 | ||
9553 | return ret; | |
9554 | } | |
9555 | ||
9936c7c2 PB |
9556 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
9557 | struct task_struct *task, | |
3dd0c97a | 9558 | bool cancel_all) |
9936c7c2 | 9559 | { |
3dd0c97a | 9560 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 9561 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
9562 | |
9563 | while (1) { | |
9564 | enum io_wq_cancel cret; | |
9565 | bool ret = false; | |
9566 | ||
1b00764f PB |
9567 | if (!task) { |
9568 | ret |= io_uring_try_cancel_iowq(ctx); | |
9569 | } else if (tctx && tctx->io_wq) { | |
9570 | /* | |
9571 | * Cancels requests of all rings, not only @ctx, but | |
9572 | * it's fine as the task is in exit/exec. | |
9573 | */ | |
5aa75ed5 | 9574 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
9575 | &cancel, true); |
9576 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9577 | } | |
9578 | ||
9579 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 9580 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 9581 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
9582 | while (!list_empty_careful(&ctx->iopoll_list)) { |
9583 | io_iopoll_try_reap_events(ctx); | |
9584 | ret = true; | |
9585 | } | |
9586 | } | |
9587 | ||
3dd0c97a PB |
9588 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
9589 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
9590 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
9591 | if (task) |
9592 | ret |= io_run_task_work(); | |
9936c7c2 PB |
9593 | if (!ret) |
9594 | break; | |
9595 | cond_resched(); | |
9596 | } | |
9597 | } | |
9598 | ||
eef51daa | 9599 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 9600 | { |
236434c3 | 9601 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 9602 | struct io_tctx_node *node; |
a528b04e | 9603 | int ret; |
236434c3 MWO |
9604 | |
9605 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9606 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9607 | if (unlikely(ret)) |
9608 | return ret; | |
236434c3 | 9609 | tctx = current->io_uring; |
0f212204 | 9610 | } |
cf27f3b1 PB |
9611 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9612 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9613 | if (!node) | |
9614 | return -ENOMEM; | |
9615 | node->ctx = ctx; | |
9616 | node->task = current; | |
13bf43f5 | 9617 | |
cf27f3b1 PB |
9618 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9619 | node, GFP_KERNEL)); | |
9620 | if (ret) { | |
9621 | kfree(node); | |
9622 | return ret; | |
0f212204 | 9623 | } |
cf27f3b1 PB |
9624 | |
9625 | mutex_lock(&ctx->uring_lock); | |
9626 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9627 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9628 | } |
cf27f3b1 | 9629 | tctx->last = ctx; |
0f212204 JA |
9630 | return 0; |
9631 | } | |
9632 | ||
cf27f3b1 PB |
9633 | /* |
9634 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9635 | */ | |
eef51daa | 9636 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9637 | { |
9638 | struct io_uring_task *tctx = current->io_uring; | |
9639 | ||
9640 | if (likely(tctx && tctx->last == ctx)) | |
9641 | return 0; | |
eef51daa | 9642 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9643 | } |
9644 | ||
0f212204 JA |
9645 | /* |
9646 | * Remove this io_uring_file -> task mapping. | |
9647 | */ | |
eef51daa | 9648 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9649 | { |
9650 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9651 | struct io_tctx_node *node; |
2941267b | 9652 | |
eebd2e37 PB |
9653 | if (!tctx) |
9654 | return; | |
13bf43f5 PB |
9655 | node = xa_erase(&tctx->xa, index); |
9656 | if (!node) | |
2941267b | 9657 | return; |
0f212204 | 9658 | |
13bf43f5 PB |
9659 | WARN_ON_ONCE(current != node->task); |
9660 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9661 | ||
9662 | mutex_lock(&node->ctx->uring_lock); | |
9663 | list_del(&node->ctx_node); | |
9664 | mutex_unlock(&node->ctx->uring_lock); | |
9665 | ||
baf186c4 | 9666 | if (tctx->last == node->ctx) |
0f212204 | 9667 | tctx->last = NULL; |
13bf43f5 | 9668 | kfree(node); |
0f212204 JA |
9669 | } |
9670 | ||
8452d4a6 | 9671 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9672 | { |
ba5ef6dc | 9673 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9674 | struct io_tctx_node *node; |
de7f1d9e PB |
9675 | unsigned long index; |
9676 | ||
8bab4c09 | 9677 | xa_for_each(&tctx->xa, index, node) { |
eef51daa | 9678 | io_uring_del_tctx_node(index); |
8bab4c09 JA |
9679 | cond_resched(); |
9680 | } | |
b16ef427 ME |
9681 | if (wq) { |
9682 | /* | |
9683 | * Must be after io_uring_del_task_file() (removes nodes under | |
9684 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9685 | */ | |
ba5ef6dc | 9686 | io_wq_put_and_exit(wq); |
dadebc35 | 9687 | tctx->io_wq = NULL; |
b16ef427 | 9688 | } |
de7f1d9e PB |
9689 | } |
9690 | ||
3f48cf18 | 9691 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9692 | { |
3f48cf18 PB |
9693 | if (tracked) |
9694 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9695 | return percpu_counter_sum(&tctx->inflight); |
9696 | } | |
9697 | ||
09899b19 PB |
9698 | static void io_uring_drop_tctx_refs(struct task_struct *task) |
9699 | { | |
9700 | struct io_uring_task *tctx = task->io_uring; | |
9701 | unsigned int refs = tctx->cached_refs; | |
9702 | ||
e9dbe221 PB |
9703 | if (refs) { |
9704 | tctx->cached_refs = 0; | |
9705 | percpu_counter_sub(&tctx->inflight, refs); | |
9706 | put_task_struct_many(task, refs); | |
9707 | } | |
09899b19 PB |
9708 | } |
9709 | ||
78cc687b PB |
9710 | /* |
9711 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
9712 | * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation. | |
9713 | */ | |
9714 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9715 | { |
521d6a73 | 9716 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9717 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9718 | s64 inflight; |
9719 | DEFINE_WAIT(wait); | |
fdaf083c | 9720 | |
78cc687b PB |
9721 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9722 | ||
6d042ffb PO |
9723 | if (!current->io_uring) |
9724 | return; | |
17a91051 PB |
9725 | if (tctx->io_wq) |
9726 | io_wq_exit_start(tctx->io_wq); | |
9727 | ||
0e9ddb39 PB |
9728 | atomic_inc(&tctx->in_idle); |
9729 | do { | |
e9dbe221 | 9730 | io_uring_drop_tctx_refs(current); |
0e9ddb39 | 9731 | /* read completions before cancelations */ |
78cc687b | 9732 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9733 | if (!inflight) |
9734 | break; | |
fdaf083c | 9735 | |
78cc687b PB |
9736 | if (!sqd) { |
9737 | struct io_tctx_node *node; | |
9738 | unsigned long index; | |
0f212204 | 9739 | |
78cc687b PB |
9740 | xa_for_each(&tctx->xa, index, node) { |
9741 | /* sqpoll task will cancel all its requests */ | |
9742 | if (node->ctx->sq_data) | |
9743 | continue; | |
9744 | io_uring_try_cancel_requests(node->ctx, current, | |
9745 | cancel_all); | |
9746 | } | |
9747 | } else { | |
9748 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9749 | io_uring_try_cancel_requests(ctx, current, | |
9750 | cancel_all); | |
9751 | } | |
17a91051 | 9752 | |
0f212204 | 9753 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
e9dbe221 | 9754 | io_uring_drop_tctx_refs(current); |
0f212204 | 9755 | /* |
a1bb3cd5 PB |
9756 | * If we've seen completions, retry without waiting. This |
9757 | * avoids a race where a completion comes in before we did | |
9758 | * prepare_to_wait(). | |
0f212204 | 9759 | */ |
3dd0c97a | 9760 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9761 | schedule(); |
f57555ed | 9762 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9763 | } while (1); |
fdaf083c | 9764 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9765 | |
8452d4a6 | 9766 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9767 | if (cancel_all) { |
3f48cf18 PB |
9768 | /* for exec all current's requests should be gone, kill tctx */ |
9769 | __io_uring_free(current); | |
9770 | } | |
44e728b8 PB |
9771 | } |
9772 | ||
f552a27a | 9773 | void __io_uring_cancel(bool cancel_all) |
78cc687b | 9774 | { |
f552a27a | 9775 | io_uring_cancel_generic(cancel_all, NULL); |
78cc687b PB |
9776 | } |
9777 | ||
6c5c240e RP |
9778 | static void *io_uring_validate_mmap_request(struct file *file, |
9779 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9780 | { |
2b188cc1 | 9781 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9782 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9783 | struct page *page; |
9784 | void *ptr; | |
9785 | ||
9786 | switch (offset) { | |
9787 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9788 | case IORING_OFF_CQ_RING: |
9789 | ptr = ctx->rings; | |
2b188cc1 JA |
9790 | break; |
9791 | case IORING_OFF_SQES: | |
9792 | ptr = ctx->sq_sqes; | |
9793 | break; | |
2b188cc1 | 9794 | default: |
6c5c240e | 9795 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9796 | } |
9797 | ||
9798 | page = virt_to_head_page(ptr); | |
a50b854e | 9799 | if (sz > page_size(page)) |
6c5c240e RP |
9800 | return ERR_PTR(-EINVAL); |
9801 | ||
9802 | return ptr; | |
9803 | } | |
9804 | ||
9805 | #ifdef CONFIG_MMU | |
9806 | ||
9807 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9808 | { | |
9809 | size_t sz = vma->vm_end - vma->vm_start; | |
9810 | unsigned long pfn; | |
9811 | void *ptr; | |
9812 | ||
9813 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9814 | if (IS_ERR(ptr)) | |
9815 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9816 | |
9817 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9818 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9819 | } | |
9820 | ||
6c5c240e RP |
9821 | #else /* !CONFIG_MMU */ |
9822 | ||
9823 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9824 | { | |
9825 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9826 | } | |
9827 | ||
9828 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9829 | { | |
9830 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9831 | } | |
9832 | ||
9833 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9834 | unsigned long addr, unsigned long len, | |
9835 | unsigned long pgoff, unsigned long flags) | |
9836 | { | |
9837 | void *ptr; | |
9838 | ||
9839 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9840 | if (IS_ERR(ptr)) | |
9841 | return PTR_ERR(ptr); | |
9842 | ||
9843 | return (unsigned long) ptr; | |
9844 | } | |
9845 | ||
9846 | #endif /* !CONFIG_MMU */ | |
9847 | ||
d9d05217 | 9848 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9849 | { |
9850 | DEFINE_WAIT(wait); | |
9851 | ||
9852 | do { | |
9853 | if (!io_sqring_full(ctx)) | |
9854 | break; | |
90554200 JA |
9855 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9856 | ||
9857 | if (!io_sqring_full(ctx)) | |
9858 | break; | |
90554200 JA |
9859 | schedule(); |
9860 | } while (!signal_pending(current)); | |
9861 | ||
9862 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9863 | return 0; |
90554200 JA |
9864 | } |
9865 | ||
c73ebb68 HX |
9866 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9867 | struct __kernel_timespec __user **ts, | |
9868 | const sigset_t __user **sig) | |
9869 | { | |
9870 | struct io_uring_getevents_arg arg; | |
9871 | ||
9872 | /* | |
9873 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9874 | * is just a pointer to the sigset_t. | |
9875 | */ | |
9876 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9877 | *sig = (const sigset_t __user *) argp; | |
9878 | *ts = NULL; | |
9879 | return 0; | |
9880 | } | |
9881 | ||
9882 | /* | |
9883 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9884 | * timespec and sigset_t pointers if good. | |
9885 | */ | |
9886 | if (*argsz != sizeof(arg)) | |
9887 | return -EINVAL; | |
9888 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9889 | return -EFAULT; | |
9890 | *sig = u64_to_user_ptr(arg.sigmask); | |
9891 | *argsz = arg.sigmask_sz; | |
9892 | *ts = u64_to_user_ptr(arg.ts); | |
9893 | return 0; | |
9894 | } | |
9895 | ||
2b188cc1 | 9896 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9897 | u32, min_complete, u32, flags, const void __user *, argp, |
9898 | size_t, argsz) | |
2b188cc1 JA |
9899 | { |
9900 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9901 | int submitted = 0; |
9902 | struct fd f; | |
33f993da | 9903 | long ret; |
2b188cc1 | 9904 | |
4c6e277c | 9905 | io_run_task_work(); |
b41e9852 | 9906 | |
33f993da PB |
9907 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9908 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9909 | return -EINVAL; |
9910 | ||
9911 | f = fdget(fd); | |
33f993da | 9912 | if (unlikely(!f.file)) |
2b188cc1 JA |
9913 | return -EBADF; |
9914 | ||
9915 | ret = -EOPNOTSUPP; | |
33f993da | 9916 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9917 | goto out_fput; |
9918 | ||
9919 | ret = -ENXIO; | |
9920 | ctx = f.file->private_data; | |
33f993da | 9921 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9922 | goto out_fput; |
9923 | ||
7e84e1c7 | 9924 | ret = -EBADFD; |
33f993da | 9925 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9926 | goto out; |
9927 | ||
6c271ce2 JA |
9928 | /* |
9929 | * For SQ polling, the thread will do all submissions and completions. | |
9930 | * Just return the requested submit count, and wake the thread if | |
9931 | * we were asked to. | |
9932 | */ | |
b2a9eada | 9933 | ret = 0; |
6c271ce2 | 9934 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
90f67366 | 9935 | io_cqring_overflow_flush(ctx); |
89448c47 | 9936 | |
21f96522 JA |
9937 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9938 | ret = -EOWNERDEAD; | |
04147488 | 9939 | goto out; |
21f96522 | 9940 | } |
6c271ce2 | 9941 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9942 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9943 | if (flags & IORING_ENTER_SQ_WAIT) { |
9944 | ret = io_sqpoll_wait_sq(ctx); | |
9945 | if (ret) | |
9946 | goto out; | |
9947 | } | |
6c271ce2 | 9948 | submitted = to_submit; |
b2a9eada | 9949 | } else if (to_submit) { |
eef51daa | 9950 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
9951 | if (unlikely(ret)) |
9952 | goto out; | |
2b188cc1 | 9953 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9954 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9955 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9956 | |
9957 | if (submitted != to_submit) | |
9958 | goto out; | |
2b188cc1 JA |
9959 | } |
9960 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9961 | const sigset_t __user *sig; |
9962 | struct __kernel_timespec __user *ts; | |
9963 | ||
9964 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9965 | if (unlikely(ret)) | |
9966 | goto out; | |
9967 | ||
2b188cc1 JA |
9968 | min_complete = min(min_complete, ctx->cq_entries); |
9969 | ||
32b2244a XW |
9970 | /* |
9971 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9972 | * space applications don't need to do io completion events | |
9973 | * polling again, they can rely on io_sq_thread to do polling | |
9974 | * work, which can reduce cpu usage and uring_lock contention. | |
9975 | */ | |
9976 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9977 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9978 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9979 | } else { |
c73ebb68 | 9980 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9981 | } |
2b188cc1 JA |
9982 | } |
9983 | ||
7c504e65 | 9984 | out: |
6805b32e | 9985 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9986 | out_fput: |
9987 | fdput(f); | |
9988 | return submitted ? submitted : ret; | |
9989 | } | |
9990 | ||
bebdb65e | 9991 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9992 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9993 | const struct cred *cred) | |
87ce955b | 9994 | { |
87ce955b JA |
9995 | struct user_namespace *uns = seq_user_ns(m); |
9996 | struct group_info *gi; | |
9997 | kernel_cap_t cap; | |
9998 | unsigned __capi; | |
9999 | int g; | |
10000 | ||
10001 | seq_printf(m, "%5d\n", id); | |
10002 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
10003 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
10004 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
10005 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
10006 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
10007 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
10008 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
10009 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
10010 | seq_puts(m, "\n\tGroups:\t"); | |
10011 | gi = cred->group_info; | |
10012 | for (g = 0; g < gi->ngroups; g++) { | |
10013 | seq_put_decimal_ull(m, g ? " " : "", | |
10014 | from_kgid_munged(uns, gi->gid[g])); | |
10015 | } | |
10016 | seq_puts(m, "\n\tCapEff:\t"); | |
10017 | cap = cred->cap_effective; | |
10018 | CAP_FOR_EACH_U32(__capi) | |
10019 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
10020 | seq_putc(m, '\n'); | |
10021 | return 0; | |
10022 | } | |
10023 | ||
10024 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
10025 | { | |
dbbe9c64 | 10026 | struct io_sq_data *sq = NULL; |
fad8e0de | 10027 | bool has_lock; |
87ce955b JA |
10028 | int i; |
10029 | ||
fad8e0de JA |
10030 | /* |
10031 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
10032 | * since fdinfo case grabs it in the opposite direction of normal use | |
10033 | * cases. If we fail to get the lock, we just don't iterate any | |
10034 | * structures that could be going away outside the io_uring mutex. | |
10035 | */ | |
10036 | has_lock = mutex_trylock(&ctx->uring_lock); | |
10037 | ||
5f3f26f9 | 10038 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 10039 | sq = ctx->sq_data; |
5f3f26f9 JA |
10040 | if (!sq->thread) |
10041 | sq = NULL; | |
10042 | } | |
dbbe9c64 JQ |
10043 | |
10044 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
10045 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 10046 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 10047 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 10048 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 10049 | |
87ce955b JA |
10050 | if (f) |
10051 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
10052 | else | |
10053 | seq_printf(m, "%5u: <none>\n", i); | |
10054 | } | |
10055 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 10056 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 10057 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 10058 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 10059 | |
4751f53d | 10060 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 10061 | } |
61cf9370 MWO |
10062 | if (has_lock && !xa_empty(&ctx->personalities)) { |
10063 | unsigned long index; | |
10064 | const struct cred *cred; | |
10065 | ||
87ce955b | 10066 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
10067 | xa_for_each(&ctx->personalities, index, cred) |
10068 | io_uring_show_cred(m, index, cred); | |
87ce955b | 10069 | } |
d7718a9d | 10070 | seq_printf(m, "PollList:\n"); |
79ebeaee | 10071 | spin_lock(&ctx->completion_lock); |
d7718a9d JA |
10072 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
10073 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
10074 | struct io_kiocb *req; | |
10075 | ||
10076 | hlist_for_each_entry(req, list, hash_node) | |
10077 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
10078 | req->task->task_works != NULL); | |
10079 | } | |
79ebeaee | 10080 | spin_unlock(&ctx->completion_lock); |
fad8e0de JA |
10081 | if (has_lock) |
10082 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
10083 | } |
10084 | ||
10085 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
10086 | { | |
10087 | struct io_ring_ctx *ctx = f->private_data; | |
10088 | ||
10089 | if (percpu_ref_tryget(&ctx->refs)) { | |
10090 | __io_uring_show_fdinfo(ctx, m); | |
10091 | percpu_ref_put(&ctx->refs); | |
10092 | } | |
10093 | } | |
bebdb65e | 10094 | #endif |
87ce955b | 10095 | |
2b188cc1 JA |
10096 | static const struct file_operations io_uring_fops = { |
10097 | .release = io_uring_release, | |
10098 | .mmap = io_uring_mmap, | |
6c5c240e RP |
10099 | #ifndef CONFIG_MMU |
10100 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
10101 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
10102 | #endif | |
2b188cc1 JA |
10103 | .poll = io_uring_poll, |
10104 | .fasync = io_uring_fasync, | |
bebdb65e | 10105 | #ifdef CONFIG_PROC_FS |
87ce955b | 10106 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 10107 | #endif |
2b188cc1 JA |
10108 | }; |
10109 | ||
10110 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
10111 | struct io_uring_params *p) | |
10112 | { | |
75b28aff HV |
10113 | struct io_rings *rings; |
10114 | size_t size, sq_array_offset; | |
2b188cc1 | 10115 | |
bd740481 JA |
10116 | /* make sure these are sane, as we already accounted them */ |
10117 | ctx->sq_entries = p->sq_entries; | |
10118 | ctx->cq_entries = p->cq_entries; | |
10119 | ||
75b28aff HV |
10120 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
10121 | if (size == SIZE_MAX) | |
10122 | return -EOVERFLOW; | |
10123 | ||
10124 | rings = io_mem_alloc(size); | |
10125 | if (!rings) | |
2b188cc1 JA |
10126 | return -ENOMEM; |
10127 | ||
75b28aff HV |
10128 | ctx->rings = rings; |
10129 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
10130 | rings->sq_ring_mask = p->sq_entries - 1; | |
10131 | rings->cq_ring_mask = p->cq_entries - 1; | |
10132 | rings->sq_ring_entries = p->sq_entries; | |
10133 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
10134 | |
10135 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
10136 | if (size == SIZE_MAX) { |
10137 | io_mem_free(ctx->rings); | |
10138 | ctx->rings = NULL; | |
2b188cc1 | 10139 | return -EOVERFLOW; |
eb065d30 | 10140 | } |
2b188cc1 JA |
10141 | |
10142 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
10143 | if (!ctx->sq_sqes) { |
10144 | io_mem_free(ctx->rings); | |
10145 | ctx->rings = NULL; | |
2b188cc1 | 10146 | return -ENOMEM; |
eb065d30 | 10147 | } |
2b188cc1 | 10148 | |
2b188cc1 JA |
10149 | return 0; |
10150 | } | |
10151 | ||
9faadcc8 PB |
10152 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
10153 | { | |
10154 | int ret, fd; | |
10155 | ||
10156 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
10157 | if (fd < 0) | |
10158 | return fd; | |
10159 | ||
eef51daa | 10160 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
10161 | if (ret) { |
10162 | put_unused_fd(fd); | |
10163 | return ret; | |
10164 | } | |
10165 | fd_install(fd, file); | |
10166 | return fd; | |
10167 | } | |
10168 | ||
2b188cc1 JA |
10169 | /* |
10170 | * Allocate an anonymous fd, this is what constitutes the application | |
10171 | * visible backing of an io_uring instance. The application mmaps this | |
10172 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
10173 | * we have to tie this fd to a socket for file garbage collection purposes. | |
10174 | */ | |
9faadcc8 | 10175 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
10176 | { |
10177 | struct file *file; | |
9faadcc8 | 10178 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
10179 | int ret; |
10180 | ||
2b188cc1 JA |
10181 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
10182 | &ctx->ring_sock); | |
10183 | if (ret) | |
9faadcc8 | 10184 | return ERR_PTR(ret); |
2b188cc1 JA |
10185 | #endif |
10186 | ||
2b188cc1 JA |
10187 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
10188 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 10189 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
10190 | if (IS_ERR(file)) { |
10191 | sock_release(ctx->ring_sock); | |
10192 | ctx->ring_sock = NULL; | |
10193 | } else { | |
10194 | ctx->ring_sock->file = file; | |
0f212204 | 10195 | } |
2b188cc1 | 10196 | #endif |
9faadcc8 | 10197 | return file; |
2b188cc1 JA |
10198 | } |
10199 | ||
7f13657d XW |
10200 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
10201 | struct io_uring_params __user *params) | |
2b188cc1 | 10202 | { |
2b188cc1 | 10203 | struct io_ring_ctx *ctx; |
9faadcc8 | 10204 | struct file *file; |
2b188cc1 JA |
10205 | int ret; |
10206 | ||
8110c1a6 | 10207 | if (!entries) |
2b188cc1 | 10208 | return -EINVAL; |
8110c1a6 JA |
10209 | if (entries > IORING_MAX_ENTRIES) { |
10210 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
10211 | return -EINVAL; | |
10212 | entries = IORING_MAX_ENTRIES; | |
10213 | } | |
2b188cc1 JA |
10214 | |
10215 | /* | |
10216 | * Use twice as many entries for the CQ ring. It's possible for the | |
10217 | * application to drive a higher depth than the size of the SQ ring, | |
10218 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
10219 | * some flexibility in overcommitting a bit. If the application has |
10220 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
10221 | * of CQ ring entries manually. | |
2b188cc1 JA |
10222 | */ |
10223 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
10224 | if (p->flags & IORING_SETUP_CQSIZE) { |
10225 | /* | |
10226 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
10227 | * to a power-of-two, if it isn't already. We do NOT impose | |
10228 | * any cq vs sq ring sizing. | |
10229 | */ | |
eb2667b3 | 10230 | if (!p->cq_entries) |
33a107f0 | 10231 | return -EINVAL; |
8110c1a6 JA |
10232 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
10233 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
10234 | return -EINVAL; | |
10235 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
10236 | } | |
eb2667b3 JQ |
10237 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
10238 | if (p->cq_entries < p->sq_entries) | |
10239 | return -EINVAL; | |
33a107f0 JA |
10240 | } else { |
10241 | p->cq_entries = 2 * p->sq_entries; | |
10242 | } | |
2b188cc1 | 10243 | |
2b188cc1 | 10244 | ctx = io_ring_ctx_alloc(p); |
62e398be | 10245 | if (!ctx) |
2b188cc1 | 10246 | return -ENOMEM; |
2b188cc1 | 10247 | ctx->compat = in_compat_syscall(); |
62e398be JA |
10248 | if (!capable(CAP_IPC_LOCK)) |
10249 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
10250 | |
10251 | /* | |
10252 | * This is just grabbed for accounting purposes. When a process exits, | |
10253 | * the mm is exited and dropped before the files, hence we need to hang | |
10254 | * on to this mm purely for the purposes of being able to unaccount | |
10255 | * memory (locked/pinned vm). It's not used for anything else. | |
10256 | */ | |
6b7898eb | 10257 | mmgrab(current->mm); |
2aede0e4 | 10258 | ctx->mm_account = current->mm; |
6b7898eb | 10259 | |
2b188cc1 JA |
10260 | ret = io_allocate_scq_urings(ctx, p); |
10261 | if (ret) | |
10262 | goto err; | |
10263 | ||
7e84e1c7 | 10264 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
10265 | if (ret) |
10266 | goto err; | |
eae071c9 | 10267 | /* always set a rsrc node */ |
47b228ce PB |
10268 | ret = io_rsrc_node_switch_start(ctx); |
10269 | if (ret) | |
10270 | goto err; | |
eae071c9 | 10271 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 10272 | |
2b188cc1 | 10273 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
10274 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
10275 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
10276 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
10277 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
10278 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
10279 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
10280 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
10281 | |
10282 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
10283 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
10284 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
10285 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
10286 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
10287 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
10288 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 10289 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 10290 | |
7f13657d XW |
10291 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
10292 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 10293 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 10294 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
10295 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
10296 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
10297 | |
10298 | if (copy_to_user(params, p, sizeof(*p))) { | |
10299 | ret = -EFAULT; | |
10300 | goto err; | |
10301 | } | |
d1719f70 | 10302 | |
9faadcc8 PB |
10303 | file = io_uring_get_file(ctx); |
10304 | if (IS_ERR(file)) { | |
10305 | ret = PTR_ERR(file); | |
10306 | goto err; | |
10307 | } | |
10308 | ||
044c1ab3 JA |
10309 | /* |
10310 | * Install ring fd as the very last thing, so we don't risk someone | |
10311 | * having closed it before we finish setup | |
10312 | */ | |
9faadcc8 PB |
10313 | ret = io_uring_install_fd(ctx, file); |
10314 | if (ret < 0) { | |
10315 | /* fput will clean it up */ | |
10316 | fput(file); | |
10317 | return ret; | |
10318 | } | |
044c1ab3 | 10319 | |
c826bd7a | 10320 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
10321 | return ret; |
10322 | err: | |
10323 | io_ring_ctx_wait_and_kill(ctx); | |
10324 | return ret; | |
10325 | } | |
10326 | ||
10327 | /* | |
10328 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
10329 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
10330 | * params structure passed in. | |
10331 | */ | |
10332 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
10333 | { | |
10334 | struct io_uring_params p; | |
2b188cc1 JA |
10335 | int i; |
10336 | ||
10337 | if (copy_from_user(&p, params, sizeof(p))) | |
10338 | return -EFAULT; | |
10339 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
10340 | if (p.resv[i]) | |
10341 | return -EINVAL; | |
10342 | } | |
10343 | ||
6c271ce2 | 10344 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 10345 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
10346 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
10347 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
10348 | return -EINVAL; |
10349 | ||
7f13657d | 10350 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
10351 | } |
10352 | ||
10353 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
10354 | struct io_uring_params __user *, params) | |
10355 | { | |
10356 | return io_uring_setup(entries, params); | |
10357 | } | |
10358 | ||
66f4af93 JA |
10359 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
10360 | { | |
10361 | struct io_uring_probe *p; | |
10362 | size_t size; | |
10363 | int i, ret; | |
10364 | ||
10365 | size = struct_size(p, ops, nr_args); | |
10366 | if (size == SIZE_MAX) | |
10367 | return -EOVERFLOW; | |
10368 | p = kzalloc(size, GFP_KERNEL); | |
10369 | if (!p) | |
10370 | return -ENOMEM; | |
10371 | ||
10372 | ret = -EFAULT; | |
10373 | if (copy_from_user(p, arg, size)) | |
10374 | goto out; | |
10375 | ret = -EINVAL; | |
10376 | if (memchr_inv(p, 0, size)) | |
10377 | goto out; | |
10378 | ||
10379 | p->last_op = IORING_OP_LAST - 1; | |
10380 | if (nr_args > IORING_OP_LAST) | |
10381 | nr_args = IORING_OP_LAST; | |
10382 | ||
10383 | for (i = 0; i < nr_args; i++) { | |
10384 | p->ops[i].op = i; | |
10385 | if (!io_op_defs[i].not_supported) | |
10386 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
10387 | } | |
10388 | p->ops_len = i; | |
10389 | ||
10390 | ret = 0; | |
10391 | if (copy_to_user(arg, p, size)) | |
10392 | ret = -EFAULT; | |
10393 | out: | |
10394 | kfree(p); | |
10395 | return ret; | |
10396 | } | |
10397 | ||
071698e1 JA |
10398 | static int io_register_personality(struct io_ring_ctx *ctx) |
10399 | { | |
4379bf8b | 10400 | const struct cred *creds; |
61cf9370 | 10401 | u32 id; |
1e6fa521 | 10402 | int ret; |
071698e1 | 10403 | |
4379bf8b | 10404 | creds = get_current_cred(); |
1e6fa521 | 10405 | |
61cf9370 MWO |
10406 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
10407 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
a30f895a JA |
10408 | if (ret < 0) { |
10409 | put_cred(creds); | |
10410 | return ret; | |
10411 | } | |
10412 | return id; | |
071698e1 JA |
10413 | } |
10414 | ||
21b55dbc SG |
10415 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
10416 | unsigned int nr_args) | |
10417 | { | |
10418 | struct io_uring_restriction *res; | |
10419 | size_t size; | |
10420 | int i, ret; | |
10421 | ||
7e84e1c7 SG |
10422 | /* Restrictions allowed only if rings started disabled */ |
10423 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10424 | return -EBADFD; | |
10425 | ||
21b55dbc | 10426 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 10427 | if (ctx->restrictions.registered) |
21b55dbc SG |
10428 | return -EBUSY; |
10429 | ||
10430 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
10431 | return -EINVAL; | |
10432 | ||
10433 | size = array_size(nr_args, sizeof(*res)); | |
10434 | if (size == SIZE_MAX) | |
10435 | return -EOVERFLOW; | |
10436 | ||
10437 | res = memdup_user(arg, size); | |
10438 | if (IS_ERR(res)) | |
10439 | return PTR_ERR(res); | |
10440 | ||
10441 | ret = 0; | |
10442 | ||
10443 | for (i = 0; i < nr_args; i++) { | |
10444 | switch (res[i].opcode) { | |
10445 | case IORING_RESTRICTION_REGISTER_OP: | |
10446 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
10447 | ret = -EINVAL; | |
10448 | goto out; | |
10449 | } | |
10450 | ||
10451 | __set_bit(res[i].register_op, | |
10452 | ctx->restrictions.register_op); | |
10453 | break; | |
10454 | case IORING_RESTRICTION_SQE_OP: | |
10455 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
10456 | ret = -EINVAL; | |
10457 | goto out; | |
10458 | } | |
10459 | ||
10460 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
10461 | break; | |
10462 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
10463 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
10464 | break; | |
10465 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
10466 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
10467 | break; | |
10468 | default: | |
10469 | ret = -EINVAL; | |
10470 | goto out; | |
10471 | } | |
10472 | } | |
10473 | ||
10474 | out: | |
10475 | /* Reset all restrictions if an error happened */ | |
10476 | if (ret != 0) | |
10477 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
10478 | else | |
7e84e1c7 | 10479 | ctx->restrictions.registered = true; |
21b55dbc SG |
10480 | |
10481 | kfree(res); | |
10482 | return ret; | |
10483 | } | |
10484 | ||
7e84e1c7 SG |
10485 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10486 | { | |
10487 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10488 | return -EBADFD; | |
10489 | ||
10490 | if (ctx->restrictions.registered) | |
10491 | ctx->restricted = 1; | |
10492 | ||
0298ef96 PB |
10493 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
10494 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
10495 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
10496 | return 0; |
10497 | } | |
10498 | ||
fdecb662 | 10499 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 10500 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
10501 | unsigned nr_args) |
10502 | { | |
10503 | __u32 tmp; | |
10504 | int err; | |
10505 | ||
c3bdad02 PB |
10506 | if (up->resv) |
10507 | return -EINVAL; | |
98f0b3b4 PB |
10508 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
10509 | return -EOVERFLOW; | |
10510 | err = io_rsrc_node_switch_start(ctx); | |
10511 | if (err) | |
10512 | return err; | |
10513 | ||
fdecb662 PB |
10514 | switch (type) { |
10515 | case IORING_RSRC_FILE: | |
98f0b3b4 | 10516 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
10517 | case IORING_RSRC_BUFFER: |
10518 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
10519 | } |
10520 | return -EINVAL; | |
10521 | } | |
10522 | ||
c3bdad02 PB |
10523 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
10524 | unsigned nr_args) | |
98f0b3b4 | 10525 | { |
c3bdad02 | 10526 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
10527 | |
10528 | if (!nr_args) | |
10529 | return -EINVAL; | |
c3bdad02 PB |
10530 | memset(&up, 0, sizeof(up)); |
10531 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
10532 | return -EFAULT; | |
10533 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
10534 | } | |
10535 | ||
10536 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 10537 | unsigned size, unsigned type) |
c3bdad02 PB |
10538 | { |
10539 | struct io_uring_rsrc_update2 up; | |
10540 | ||
10541 | if (size != sizeof(up)) | |
10542 | return -EINVAL; | |
98f0b3b4 PB |
10543 | if (copy_from_user(&up, arg, sizeof(up))) |
10544 | return -EFAULT; | |
992da01a | 10545 | if (!up.nr || up.resv) |
98f0b3b4 | 10546 | return -EINVAL; |
992da01a | 10547 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
10548 | } |
10549 | ||
792e3582 | 10550 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 10551 | unsigned int size, unsigned int type) |
792e3582 PB |
10552 | { |
10553 | struct io_uring_rsrc_register rr; | |
10554 | ||
10555 | /* keep it extendible */ | |
10556 | if (size != sizeof(rr)) | |
10557 | return -EINVAL; | |
10558 | ||
10559 | memset(&rr, 0, sizeof(rr)); | |
10560 | if (copy_from_user(&rr, arg, size)) | |
10561 | return -EFAULT; | |
992da01a | 10562 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
10563 | return -EINVAL; |
10564 | ||
992da01a | 10565 | switch (type) { |
792e3582 PB |
10566 | case IORING_RSRC_FILE: |
10567 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
10568 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
10569 | case IORING_RSRC_BUFFER: |
10570 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
10571 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
10572 | } |
10573 | return -EINVAL; | |
10574 | } | |
10575 | ||
fe76421d JA |
10576 | static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg, |
10577 | unsigned len) | |
10578 | { | |
10579 | struct io_uring_task *tctx = current->io_uring; | |
10580 | cpumask_var_t new_mask; | |
10581 | int ret; | |
10582 | ||
10583 | if (!tctx || !tctx->io_wq) | |
10584 | return -EINVAL; | |
10585 | ||
10586 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
10587 | return -ENOMEM; | |
10588 | ||
10589 | cpumask_clear(new_mask); | |
10590 | if (len > cpumask_size()) | |
10591 | len = cpumask_size(); | |
10592 | ||
10593 | if (copy_from_user(new_mask, arg, len)) { | |
10594 | free_cpumask_var(new_mask); | |
10595 | return -EFAULT; | |
10596 | } | |
10597 | ||
10598 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
10599 | free_cpumask_var(new_mask); | |
10600 | return ret; | |
10601 | } | |
10602 | ||
10603 | static int io_unregister_iowq_aff(struct io_ring_ctx *ctx) | |
10604 | { | |
10605 | struct io_uring_task *tctx = current->io_uring; | |
10606 | ||
10607 | if (!tctx || !tctx->io_wq) | |
10608 | return -EINVAL; | |
10609 | ||
10610 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
10611 | } | |
10612 | ||
2e480058 JA |
10613 | static int io_register_iowq_max_workers(struct io_ring_ctx *ctx, |
10614 | void __user *arg) | |
10615 | { | |
fa84693b JA |
10616 | struct io_uring_task *tctx = NULL; |
10617 | struct io_sq_data *sqd = NULL; | |
2e480058 JA |
10618 | __u32 new_count[2]; |
10619 | int i, ret; | |
10620 | ||
2e480058 JA |
10621 | if (copy_from_user(new_count, arg, sizeof(new_count))) |
10622 | return -EFAULT; | |
10623 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
10624 | if (new_count[i] > INT_MAX) | |
10625 | return -EINVAL; | |
10626 | ||
fa84693b JA |
10627 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
10628 | sqd = ctx->sq_data; | |
10629 | if (sqd) { | |
009ad9f0 JA |
10630 | /* |
10631 | * Observe the correct sqd->lock -> ctx->uring_lock | |
10632 | * ordering. Fine to drop uring_lock here, we hold | |
10633 | * a ref to the ctx. | |
10634 | */ | |
41d3a6bd | 10635 | refcount_inc(&sqd->refs); |
009ad9f0 | 10636 | mutex_unlock(&ctx->uring_lock); |
fa84693b | 10637 | mutex_lock(&sqd->lock); |
009ad9f0 | 10638 | mutex_lock(&ctx->uring_lock); |
41d3a6bd JA |
10639 | if (sqd->thread) |
10640 | tctx = sqd->thread->io_uring; | |
fa84693b JA |
10641 | } |
10642 | } else { | |
10643 | tctx = current->io_uring; | |
10644 | } | |
10645 | ||
10646 | ret = -EINVAL; | |
10647 | if (!tctx || !tctx->io_wq) | |
10648 | goto err; | |
10649 | ||
2e480058 JA |
10650 | ret = io_wq_max_workers(tctx->io_wq, new_count); |
10651 | if (ret) | |
fa84693b JA |
10652 | goto err; |
10653 | ||
41d3a6bd | 10654 | if (sqd) { |
fa84693b | 10655 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
10656 | io_put_sq_data(sqd); |
10657 | } | |
2e480058 JA |
10658 | |
10659 | if (copy_to_user(arg, new_count, sizeof(new_count))) | |
10660 | return -EFAULT; | |
10661 | ||
10662 | return 0; | |
fa84693b | 10663 | err: |
41d3a6bd | 10664 | if (sqd) { |
fa84693b | 10665 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
10666 | io_put_sq_data(sqd); |
10667 | } | |
fa84693b | 10668 | return ret; |
2e480058 JA |
10669 | } |
10670 | ||
071698e1 JA |
10671 | static bool io_register_op_must_quiesce(int op) |
10672 | { | |
10673 | switch (op) { | |
bd54b6fe BM |
10674 | case IORING_REGISTER_BUFFERS: |
10675 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 10676 | case IORING_REGISTER_FILES: |
071698e1 JA |
10677 | case IORING_UNREGISTER_FILES: |
10678 | case IORING_REGISTER_FILES_UPDATE: | |
10679 | case IORING_REGISTER_PROBE: | |
10680 | case IORING_REGISTER_PERSONALITY: | |
10681 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
10682 | case IORING_REGISTER_FILES2: |
10683 | case IORING_REGISTER_FILES_UPDATE2: | |
10684 | case IORING_REGISTER_BUFFERS2: | |
10685 | case IORING_REGISTER_BUFFERS_UPDATE: | |
fe76421d JA |
10686 | case IORING_REGISTER_IOWQ_AFF: |
10687 | case IORING_UNREGISTER_IOWQ_AFF: | |
2e480058 | 10688 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
071698e1 JA |
10689 | return false; |
10690 | default: | |
10691 | return true; | |
10692 | } | |
10693 | } | |
10694 | ||
e73c5c7c PB |
10695 | static int io_ctx_quiesce(struct io_ring_ctx *ctx) |
10696 | { | |
10697 | long ret; | |
10698 | ||
10699 | percpu_ref_kill(&ctx->refs); | |
10700 | ||
10701 | /* | |
10702 | * Drop uring mutex before waiting for references to exit. If another | |
10703 | * thread is currently inside io_uring_enter() it might need to grab the | |
10704 | * uring_lock to make progress. If we hold it here across the drain | |
10705 | * wait, then we can deadlock. It's safe to drop the mutex here, since | |
10706 | * no new references will come in after we've killed the percpu ref. | |
10707 | */ | |
10708 | mutex_unlock(&ctx->uring_lock); | |
10709 | do { | |
10710 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10711 | if (!ret) | |
10712 | break; | |
10713 | ret = io_run_task_work_sig(); | |
10714 | } while (ret >= 0); | |
10715 | mutex_lock(&ctx->uring_lock); | |
10716 | ||
10717 | if (ret) | |
10718 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); | |
10719 | return ret; | |
10720 | } | |
10721 | ||
edafccee JA |
10722 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10723 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10724 | __releases(ctx->uring_lock) |
10725 | __acquires(ctx->uring_lock) | |
edafccee JA |
10726 | { |
10727 | int ret; | |
10728 | ||
35fa71a0 JA |
10729 | /* |
10730 | * We're inside the ring mutex, if the ref is already dying, then | |
10731 | * someone else killed the ctx or is already going through | |
10732 | * io_uring_register(). | |
10733 | */ | |
10734 | if (percpu_ref_is_dying(&ctx->refs)) | |
10735 | return -ENXIO; | |
10736 | ||
75c4021a PB |
10737 | if (ctx->restricted) { |
10738 | if (opcode >= IORING_REGISTER_LAST) | |
10739 | return -EINVAL; | |
10740 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10741 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10742 | return -EACCES; | |
10743 | } | |
10744 | ||
071698e1 | 10745 | if (io_register_op_must_quiesce(opcode)) { |
e73c5c7c PB |
10746 | ret = io_ctx_quiesce(ctx); |
10747 | if (ret) | |
f70865db | 10748 | return ret; |
05f3fb3c | 10749 | } |
edafccee JA |
10750 | |
10751 | switch (opcode) { | |
10752 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10753 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10754 | break; |
10755 | case IORING_UNREGISTER_BUFFERS: | |
10756 | ret = -EINVAL; | |
10757 | if (arg || nr_args) | |
10758 | break; | |
0a96bbe4 | 10759 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10760 | break; |
6b06314c | 10761 | case IORING_REGISTER_FILES: |
792e3582 | 10762 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10763 | break; |
10764 | case IORING_UNREGISTER_FILES: | |
10765 | ret = -EINVAL; | |
10766 | if (arg || nr_args) | |
10767 | break; | |
10768 | ret = io_sqe_files_unregister(ctx); | |
10769 | break; | |
c3a31e60 | 10770 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10771 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10772 | break; |
9b402849 | 10773 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10774 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10775 | ret = -EINVAL; |
10776 | if (nr_args != 1) | |
10777 | break; | |
10778 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10779 | if (ret) |
10780 | break; | |
10781 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10782 | ctx->eventfd_async = 1; | |
10783 | else | |
10784 | ctx->eventfd_async = 0; | |
9b402849 JA |
10785 | break; |
10786 | case IORING_UNREGISTER_EVENTFD: | |
10787 | ret = -EINVAL; | |
10788 | if (arg || nr_args) | |
10789 | break; | |
10790 | ret = io_eventfd_unregister(ctx); | |
10791 | break; | |
66f4af93 JA |
10792 | case IORING_REGISTER_PROBE: |
10793 | ret = -EINVAL; | |
10794 | if (!arg || nr_args > 256) | |
10795 | break; | |
10796 | ret = io_probe(ctx, arg, nr_args); | |
10797 | break; | |
071698e1 JA |
10798 | case IORING_REGISTER_PERSONALITY: |
10799 | ret = -EINVAL; | |
10800 | if (arg || nr_args) | |
10801 | break; | |
10802 | ret = io_register_personality(ctx); | |
10803 | break; | |
10804 | case IORING_UNREGISTER_PERSONALITY: | |
10805 | ret = -EINVAL; | |
10806 | if (arg) | |
10807 | break; | |
10808 | ret = io_unregister_personality(ctx, nr_args); | |
10809 | break; | |
7e84e1c7 SG |
10810 | case IORING_REGISTER_ENABLE_RINGS: |
10811 | ret = -EINVAL; | |
10812 | if (arg || nr_args) | |
10813 | break; | |
10814 | ret = io_register_enable_rings(ctx); | |
10815 | break; | |
21b55dbc SG |
10816 | case IORING_REGISTER_RESTRICTIONS: |
10817 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10818 | break; | |
992da01a PB |
10819 | case IORING_REGISTER_FILES2: |
10820 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10821 | break; | |
10822 | case IORING_REGISTER_FILES_UPDATE2: | |
10823 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10824 | IORING_RSRC_FILE); | |
10825 | break; | |
10826 | case IORING_REGISTER_BUFFERS2: | |
10827 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10828 | break; |
992da01a PB |
10829 | case IORING_REGISTER_BUFFERS_UPDATE: |
10830 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10831 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10832 | break; |
fe76421d JA |
10833 | case IORING_REGISTER_IOWQ_AFF: |
10834 | ret = -EINVAL; | |
10835 | if (!arg || !nr_args) | |
10836 | break; | |
10837 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
10838 | break; | |
10839 | case IORING_UNREGISTER_IOWQ_AFF: | |
10840 | ret = -EINVAL; | |
10841 | if (arg || nr_args) | |
10842 | break; | |
10843 | ret = io_unregister_iowq_aff(ctx); | |
10844 | break; | |
2e480058 JA |
10845 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
10846 | ret = -EINVAL; | |
10847 | if (!arg || nr_args != 2) | |
10848 | break; | |
10849 | ret = io_register_iowq_max_workers(ctx, arg); | |
10850 | break; | |
edafccee JA |
10851 | default: |
10852 | ret = -EINVAL; | |
10853 | break; | |
10854 | } | |
10855 | ||
071698e1 | 10856 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10857 | /* bring the ctx back to life */ |
05f3fb3c | 10858 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10859 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10860 | } |
edafccee JA |
10861 | return ret; |
10862 | } | |
10863 | ||
10864 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10865 | void __user *, arg, unsigned int, nr_args) | |
10866 | { | |
10867 | struct io_ring_ctx *ctx; | |
10868 | long ret = -EBADF; | |
10869 | struct fd f; | |
10870 | ||
10871 | f = fdget(fd); | |
10872 | if (!f.file) | |
10873 | return -EBADF; | |
10874 | ||
10875 | ret = -EOPNOTSUPP; | |
10876 | if (f.file->f_op != &io_uring_fops) | |
10877 | goto out_fput; | |
10878 | ||
10879 | ctx = f.file->private_data; | |
10880 | ||
b6c23dd5 PB |
10881 | io_run_task_work(); |
10882 | ||
edafccee JA |
10883 | mutex_lock(&ctx->uring_lock); |
10884 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10885 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10886 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10887 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10888 | out_fput: |
10889 | fdput(f); | |
10890 | return ret; | |
10891 | } | |
10892 | ||
2b188cc1 JA |
10893 | static int __init io_uring_init(void) |
10894 | { | |
d7f62e82 SM |
10895 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10896 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10897 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10898 | } while (0) | |
10899 | ||
10900 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10901 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10902 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10903 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10904 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10905 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10906 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10907 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10908 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10909 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10910 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10911 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10912 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10913 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10914 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10915 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10916 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10917 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10918 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10919 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10920 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10921 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10922 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10923 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10924 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10925 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10926 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10927 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10928 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 10929 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 10930 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 10931 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
b9445598 | 10932 | BUILD_BUG_SQE_ELEM(44, __u32, file_index); |
d7f62e82 | 10933 | |
b0d658ec PB |
10934 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10935 | sizeof(struct io_uring_rsrc_update)); | |
10936 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10937 | sizeof(struct io_uring_rsrc_update2)); | |
90499ad0 PB |
10938 | |
10939 | /* ->buf_index is u16 */ | |
10940 | BUILD_BUG_ON(IORING_MAX_REG_BUFFERS >= (1u << 16)); | |
10941 | ||
b0d658ec PB |
10942 | /* should fit into one byte */ |
10943 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10944 | ||
d3656344 | 10945 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
32c2d33e | 10946 | BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int)); |
16340eab | 10947 | |
91f245d5 JA |
10948 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10949 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10950 | return 0; |
10951 | }; | |
10952 | __initcall(io_uring_init); |