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; |
1d7bb1d5 | 378 | struct list_head cq_overflow_list; |
7f1129d2 PB |
379 | struct xarray io_buffers; |
380 | struct xarray personalities; | |
381 | u32 pers_next; | |
382 | unsigned sq_thread_idle; | |
2b188cc1 JA |
383 | } ____cacheline_aligned_in_smp; |
384 | ||
d0acdee2 PB |
385 | /* IRQ completion list, under ->completion_lock */ |
386 | struct list_head locked_free_list; | |
387 | unsigned int locked_free_nr; | |
3c1a2ead | 388 | |
7c30f36a | 389 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
390 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
391 | ||
90554200 | 392 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 393 | struct list_head sqd_list; |
75b28aff | 394 | |
5ed7a37d PB |
395 | unsigned long check_cq_overflow; |
396 | ||
206aefde JA |
397 | struct { |
398 | unsigned cached_cq_tail; | |
399 | unsigned cq_entries; | |
0499e582 | 400 | struct eventfd_ctx *cq_ev_fd; |
311997b3 | 401 | struct wait_queue_head poll_wait; |
0499e582 PB |
402 | struct wait_queue_head cq_wait; |
403 | unsigned cq_extra; | |
404 | atomic_t cq_timeouts; | |
206aefde | 405 | struct fasync_struct *cq_fasync; |
0499e582 | 406 | unsigned cq_last_tm_flush; |
206aefde | 407 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 408 | |
2b188cc1 JA |
409 | struct { |
410 | spinlock_t completion_lock; | |
e94f141b | 411 | |
89850fce JA |
412 | spinlock_t timeout_lock; |
413 | ||
def596e9 | 414 | /* |
540e32a0 | 415 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
416 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
417 | * For SQPOLL, only the single threaded io_sq_thread() will | |
418 | * manipulate the list, hence no extra locking is needed there. | |
419 | */ | |
540e32a0 | 420 | struct list_head iopoll_list; |
78076bb6 JA |
421 | struct hlist_head *cancel_hash; |
422 | unsigned cancel_hash_bits; | |
915b3dde | 423 | bool poll_multi_queue; |
2b188cc1 | 424 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 425 | |
21b55dbc | 426 | struct io_restriction restrictions; |
3c1a2ead | 427 | |
b13a8918 PB |
428 | /* slow path rsrc auxilary data, used by update/register */ |
429 | struct { | |
430 | struct io_rsrc_node *rsrc_backup_node; | |
431 | struct io_mapped_ubuf *dummy_ubuf; | |
432 | struct io_rsrc_data *file_data; | |
433 | struct io_rsrc_data *buf_data; | |
434 | ||
435 | struct delayed_work rsrc_put_work; | |
436 | struct llist_head rsrc_put_llist; | |
437 | struct list_head rsrc_ref_list; | |
438 | spinlock_t rsrc_ref_lock; | |
439 | }; | |
440 | ||
3c1a2ead | 441 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
442 | struct { |
443 | #if defined(CONFIG_UNIX) | |
444 | struct socket *ring_sock; | |
445 | #endif | |
446 | /* hashed buffered write serialization */ | |
447 | struct io_wq_hash *hash_map; | |
448 | ||
449 | /* Only used for accounting purposes */ | |
450 | struct user_struct *user; | |
451 | struct mm_struct *mm_account; | |
452 | ||
453 | /* ctx exit and cancelation */ | |
9011bf9a PB |
454 | struct llist_head fallback_llist; |
455 | struct delayed_work fallback_work; | |
b986af7e PB |
456 | struct work_struct exit_work; |
457 | struct list_head tctx_list; | |
458 | struct completion ref_comp; | |
459 | }; | |
2b188cc1 JA |
460 | }; |
461 | ||
53e043b2 SM |
462 | struct io_uring_task { |
463 | /* submission side */ | |
09899b19 | 464 | int cached_refs; |
53e043b2 SM |
465 | struct xarray xa; |
466 | struct wait_queue_head wait; | |
ee53fb2b SM |
467 | const struct io_ring_ctx *last; |
468 | struct io_wq *io_wq; | |
53e043b2 | 469 | struct percpu_counter inflight; |
b303fe2e | 470 | atomic_t inflight_tracked; |
53e043b2 | 471 | atomic_t in_idle; |
53e043b2 SM |
472 | |
473 | spinlock_t task_lock; | |
474 | struct io_wq_work_list task_list; | |
53e043b2 | 475 | struct callback_head task_work; |
6294f368 | 476 | bool task_running; |
53e043b2 SM |
477 | }; |
478 | ||
09bb8394 JA |
479 | /* |
480 | * First field must be the file pointer in all the | |
481 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
482 | */ | |
221c5eb2 JA |
483 | struct io_poll_iocb { |
484 | struct file *file; | |
018043be | 485 | struct wait_queue_head *head; |
221c5eb2 | 486 | __poll_t events; |
8c838788 | 487 | bool done; |
221c5eb2 | 488 | bool canceled; |
392edb45 | 489 | struct wait_queue_entry wait; |
221c5eb2 JA |
490 | }; |
491 | ||
9d805892 | 492 | struct io_poll_update { |
018043be | 493 | struct file *file; |
9d805892 PB |
494 | u64 old_user_data; |
495 | u64 new_user_data; | |
496 | __poll_t events; | |
b69de288 JA |
497 | bool update_events; |
498 | bool update_user_data; | |
018043be PB |
499 | }; |
500 | ||
b5dba59e JA |
501 | struct io_close { |
502 | struct file *file; | |
b5dba59e JA |
503 | int fd; |
504 | }; | |
505 | ||
ad8a48ac JA |
506 | struct io_timeout_data { |
507 | struct io_kiocb *req; | |
508 | struct hrtimer timer; | |
509 | struct timespec64 ts; | |
510 | enum hrtimer_mode mode; | |
511 | }; | |
512 | ||
8ed8d3c3 JA |
513 | struct io_accept { |
514 | struct file *file; | |
515 | struct sockaddr __user *addr; | |
516 | int __user *addr_len; | |
517 | int flags; | |
09952e3e | 518 | unsigned long nofile; |
8ed8d3c3 JA |
519 | }; |
520 | ||
521 | struct io_sync { | |
522 | struct file *file; | |
523 | loff_t len; | |
524 | loff_t off; | |
525 | int flags; | |
d63d1b5e | 526 | int mode; |
8ed8d3c3 JA |
527 | }; |
528 | ||
fbf23849 JA |
529 | struct io_cancel { |
530 | struct file *file; | |
531 | u64 addr; | |
532 | }; | |
533 | ||
b29472ee JA |
534 | struct io_timeout { |
535 | struct file *file; | |
bfe68a22 PB |
536 | u32 off; |
537 | u32 target_seq; | |
135fcde8 | 538 | struct list_head list; |
90cd7e42 PB |
539 | /* head of the link, used by linked timeouts only */ |
540 | struct io_kiocb *head; | |
89b263f6 JA |
541 | /* for linked completions */ |
542 | struct io_kiocb *prev; | |
b29472ee JA |
543 | }; |
544 | ||
0bdf7a2d PB |
545 | struct io_timeout_rem { |
546 | struct file *file; | |
547 | u64 addr; | |
9c8e11b3 PB |
548 | |
549 | /* timeout update */ | |
550 | struct timespec64 ts; | |
551 | u32 flags; | |
0bdf7a2d PB |
552 | }; |
553 | ||
9adbd45d JA |
554 | struct io_rw { |
555 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
556 | struct kiocb kiocb; | |
557 | u64 addr; | |
558 | u64 len; | |
559 | }; | |
560 | ||
3fbb51c1 JA |
561 | struct io_connect { |
562 | struct file *file; | |
563 | struct sockaddr __user *addr; | |
564 | int addr_len; | |
565 | }; | |
566 | ||
e47293fd JA |
567 | struct io_sr_msg { |
568 | struct file *file; | |
fddaface | 569 | union { |
4af3417a PB |
570 | struct compat_msghdr __user *umsg_compat; |
571 | struct user_msghdr __user *umsg; | |
572 | void __user *buf; | |
fddaface | 573 | }; |
e47293fd | 574 | int msg_flags; |
bcda7baa | 575 | int bgid; |
fddaface | 576 | size_t len; |
bcda7baa | 577 | struct io_buffer *kbuf; |
e47293fd JA |
578 | }; |
579 | ||
15b71abe JA |
580 | struct io_open { |
581 | struct file *file; | |
582 | int dfd; | |
15b71abe | 583 | struct filename *filename; |
c12cedf2 | 584 | struct open_how how; |
4022e7af | 585 | unsigned long nofile; |
15b71abe JA |
586 | }; |
587 | ||
269bbe5f | 588 | struct io_rsrc_update { |
05f3fb3c JA |
589 | struct file *file; |
590 | u64 arg; | |
591 | u32 nr_args; | |
592 | u32 offset; | |
593 | }; | |
594 | ||
4840e418 JA |
595 | struct io_fadvise { |
596 | struct file *file; | |
597 | u64 offset; | |
598 | u32 len; | |
599 | u32 advice; | |
600 | }; | |
601 | ||
c1ca757b JA |
602 | struct io_madvise { |
603 | struct file *file; | |
604 | u64 addr; | |
605 | u32 len; | |
606 | u32 advice; | |
607 | }; | |
608 | ||
3e4827b0 JA |
609 | struct io_epoll { |
610 | struct file *file; | |
611 | int epfd; | |
612 | int op; | |
613 | int fd; | |
614 | struct epoll_event event; | |
e47293fd JA |
615 | }; |
616 | ||
7d67af2c PB |
617 | struct io_splice { |
618 | struct file *file_out; | |
619 | struct file *file_in; | |
620 | loff_t off_out; | |
621 | loff_t off_in; | |
622 | u64 len; | |
623 | unsigned int flags; | |
624 | }; | |
625 | ||
ddf0322d JA |
626 | struct io_provide_buf { |
627 | struct file *file; | |
628 | __u64 addr; | |
38134ada | 629 | __u32 len; |
ddf0322d JA |
630 | __u32 bgid; |
631 | __u16 nbufs; | |
632 | __u16 bid; | |
633 | }; | |
634 | ||
1d9e1288 BM |
635 | struct io_statx { |
636 | struct file *file; | |
637 | int dfd; | |
638 | unsigned int mask; | |
639 | unsigned int flags; | |
e62753e4 | 640 | const char __user *filename; |
1d9e1288 BM |
641 | struct statx __user *buffer; |
642 | }; | |
643 | ||
36f4fa68 JA |
644 | struct io_shutdown { |
645 | struct file *file; | |
646 | int how; | |
647 | }; | |
648 | ||
80a261fd JA |
649 | struct io_rename { |
650 | struct file *file; | |
651 | int old_dfd; | |
652 | int new_dfd; | |
653 | struct filename *oldpath; | |
654 | struct filename *newpath; | |
655 | int flags; | |
656 | }; | |
657 | ||
14a1143b JA |
658 | struct io_unlink { |
659 | struct file *file; | |
660 | int dfd; | |
661 | int flags; | |
662 | struct filename *filename; | |
663 | }; | |
664 | ||
3ca405eb PB |
665 | struct io_completion { |
666 | struct file *file; | |
8c3f9cd1 | 667 | u32 cflags; |
3ca405eb PB |
668 | }; |
669 | ||
f499a021 JA |
670 | struct io_async_connect { |
671 | struct sockaddr_storage address; | |
672 | }; | |
673 | ||
03b1230c JA |
674 | struct io_async_msghdr { |
675 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
676 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
677 | struct iovec *free_iov; | |
03b1230c JA |
678 | struct sockaddr __user *uaddr; |
679 | struct msghdr msg; | |
b537916c | 680 | struct sockaddr_storage addr; |
03b1230c JA |
681 | }; |
682 | ||
f67676d1 JA |
683 | struct io_async_rw { |
684 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
685 | const struct iovec *free_iovec; |
686 | struct iov_iter iter; | |
227c0c96 | 687 | size_t bytes_done; |
bcf5a063 | 688 | struct wait_page_queue wpq; |
f67676d1 JA |
689 | }; |
690 | ||
6b47ee6e PB |
691 | enum { |
692 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
693 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
694 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
695 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
696 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 697 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 698 | |
dddca226 | 699 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 700 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
701 | REQ_F_INFLIGHT_BIT, |
702 | REQ_F_CUR_POS_BIT, | |
703 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 704 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 705 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 706 | REQ_F_POLLED_BIT, |
bcda7baa | 707 | REQ_F_BUFFER_SELECTED_BIT, |
e342c807 | 708 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 709 | REQ_F_REISSUE_BIT, |
8c130827 | 710 | REQ_F_DONT_REISSUE_BIT, |
b8e64b53 | 711 | REQ_F_CREDS_BIT, |
20e60a38 | 712 | REQ_F_REFCOUNT_BIT, |
4d13d1a4 | 713 | REQ_F_ARM_LTIMEOUT_BIT, |
7b29f92d | 714 | /* keep async read/write and isreg together and in order */ |
b191e2df PB |
715 | REQ_F_NOWAIT_READ_BIT, |
716 | REQ_F_NOWAIT_WRITE_BIT, | |
7b29f92d | 717 | REQ_F_ISREG_BIT, |
84557871 JA |
718 | |
719 | /* not a real bit, just to check we're not overflowing the space */ | |
720 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
721 | }; |
722 | ||
723 | enum { | |
724 | /* ctx owns file */ | |
725 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
726 | /* drain existing IO first */ | |
727 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
728 | /* linked sqes */ | |
729 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
730 | /* doesn't sever on completion < 0 */ | |
731 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
732 | /* IOSQE_ASYNC */ | |
733 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
734 | /* IOSQE_BUFFER_SELECT */ |
735 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 736 | |
6b47ee6e | 737 | /* fail rest of links */ |
93d2bcd2 | 738 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 739 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
740 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
741 | /* read/write uses file position */ | |
742 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
743 | /* must not punt to workers */ | |
744 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 745 | /* has or had linked timeout */ |
6b47ee6e | 746 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
747 | /* needs cleanup */ |
748 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
749 | /* already went through poll handler */ |
750 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
751 | /* buffer already selected */ |
752 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
e342c807 PB |
753 | /* completion is deferred through io_comp_state */ |
754 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
755 | /* caller should reissue async */ |
756 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
757 | /* don't attempt request reissue, see io_rw_reissue() */ |
758 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d | 759 | /* supports async reads */ |
b191e2df | 760 | REQ_F_NOWAIT_READ = BIT(REQ_F_NOWAIT_READ_BIT), |
7b29f92d | 761 | /* supports async writes */ |
b191e2df | 762 | REQ_F_NOWAIT_WRITE = BIT(REQ_F_NOWAIT_WRITE_BIT), |
7b29f92d JA |
763 | /* regular file */ |
764 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
765 | /* has creds assigned */ |
766 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
20e60a38 PB |
767 | /* skip refcounting if not set */ |
768 | REQ_F_REFCOUNT = BIT(REQ_F_REFCOUNT_BIT), | |
4d13d1a4 PB |
769 | /* there is a linked timeout that has to be armed */ |
770 | REQ_F_ARM_LTIMEOUT = BIT(REQ_F_ARM_LTIMEOUT_BIT), | |
d7718a9d JA |
771 | }; |
772 | ||
773 | struct async_poll { | |
774 | struct io_poll_iocb poll; | |
807abcb0 | 775 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
776 | }; |
777 | ||
5b0a6acc PB |
778 | typedef void (*io_req_tw_func_t)(struct io_kiocb *req); |
779 | ||
7cbf1722 | 780 | struct io_task_work { |
5b0a6acc PB |
781 | union { |
782 | struct io_wq_work_node node; | |
783 | struct llist_node fallback_node; | |
784 | }; | |
785 | io_req_tw_func_t func; | |
7cbf1722 JA |
786 | }; |
787 | ||
992da01a PB |
788 | enum { |
789 | IORING_RSRC_FILE = 0, | |
790 | IORING_RSRC_BUFFER = 1, | |
791 | }; | |
792 | ||
09bb8394 JA |
793 | /* |
794 | * NOTE! Each of the iocb union members has the file pointer | |
795 | * as the first entry in their struct definition. So you can | |
796 | * access the file pointer through any of the sub-structs, | |
797 | * or directly as just 'ki_filp' in this struct. | |
798 | */ | |
2b188cc1 | 799 | struct io_kiocb { |
221c5eb2 | 800 | union { |
09bb8394 | 801 | struct file *file; |
9adbd45d | 802 | struct io_rw rw; |
221c5eb2 | 803 | struct io_poll_iocb poll; |
9d805892 | 804 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
805 | struct io_accept accept; |
806 | struct io_sync sync; | |
fbf23849 | 807 | struct io_cancel cancel; |
b29472ee | 808 | struct io_timeout timeout; |
0bdf7a2d | 809 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 810 | struct io_connect connect; |
e47293fd | 811 | struct io_sr_msg sr_msg; |
15b71abe | 812 | struct io_open open; |
b5dba59e | 813 | struct io_close close; |
269bbe5f | 814 | struct io_rsrc_update rsrc_update; |
4840e418 | 815 | struct io_fadvise fadvise; |
c1ca757b | 816 | struct io_madvise madvise; |
3e4827b0 | 817 | struct io_epoll epoll; |
7d67af2c | 818 | struct io_splice splice; |
ddf0322d | 819 | struct io_provide_buf pbuf; |
1d9e1288 | 820 | struct io_statx statx; |
36f4fa68 | 821 | struct io_shutdown shutdown; |
80a261fd | 822 | struct io_rename rename; |
14a1143b | 823 | struct io_unlink unlink; |
3ca405eb PB |
824 | /* use only after cleaning per-op data, see io_clean_op() */ |
825 | struct io_completion compl; | |
221c5eb2 | 826 | }; |
2b188cc1 | 827 | |
e8c2bc1f JA |
828 | /* opcode allocated if it needs to store data for async defer */ |
829 | void *async_data; | |
d625c6ee | 830 | u8 opcode; |
65a6543d XW |
831 | /* polled IO has completed */ |
832 | u8 iopoll_completed; | |
2b188cc1 | 833 | |
4f4eeba8 | 834 | u16 buf_index; |
9cf7c104 | 835 | u32 result; |
4f4eeba8 | 836 | |
010e8e6b PB |
837 | struct io_ring_ctx *ctx; |
838 | unsigned int flags; | |
abc54d63 | 839 | atomic_t refs; |
010e8e6b PB |
840 | struct task_struct *task; |
841 | u64 user_data; | |
d7718a9d | 842 | |
f2f87370 | 843 | struct io_kiocb *link; |
269bbe5f | 844 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 845 | |
b303fe2e | 846 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 847 | struct list_head inflight_entry; |
5b0a6acc | 848 | struct io_task_work io_task_work; |
010e8e6b PB |
849 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
850 | struct hlist_node hash_node; | |
851 | struct async_poll *apoll; | |
852 | struct io_wq_work work; | |
fe7e3257 | 853 | const struct cred *creds; |
c10d1f98 | 854 | |
eae071c9 PB |
855 | /* store used ubuf, so we can prevent reloading */ |
856 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 857 | }; |
05589553 | 858 | |
13bf43f5 PB |
859 | struct io_tctx_node { |
860 | struct list_head ctx_node; | |
861 | struct task_struct *task; | |
13bf43f5 PB |
862 | struct io_ring_ctx *ctx; |
863 | }; | |
864 | ||
27dc8338 PB |
865 | struct io_defer_entry { |
866 | struct list_head list; | |
867 | struct io_kiocb *req; | |
9cf7c104 | 868 | u32 seq; |
2b188cc1 JA |
869 | }; |
870 | ||
d3656344 | 871 | struct io_op_def { |
d3656344 JA |
872 | /* needs req->file assigned */ |
873 | unsigned needs_file : 1; | |
d3656344 JA |
874 | /* hash wq insertion if file is a regular file */ |
875 | unsigned hash_reg_file : 1; | |
876 | /* unbound wq insertion if file is a non-regular file */ | |
877 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
878 | /* opcode is not supported by this kernel */ |
879 | unsigned not_supported : 1; | |
8a72758c JA |
880 | /* set if opcode supports polled "wait" */ |
881 | unsigned pollin : 1; | |
882 | unsigned pollout : 1; | |
bcda7baa JA |
883 | /* op supports buffer selection */ |
884 | unsigned buffer_select : 1; | |
26f0505a PB |
885 | /* do prep async if is going to be punted */ |
886 | unsigned needs_async_setup : 1; | |
27926b68 JA |
887 | /* should block plug */ |
888 | unsigned plug : 1; | |
e8c2bc1f JA |
889 | /* size of async data needed, if any */ |
890 | unsigned short async_size; | |
d3656344 JA |
891 | }; |
892 | ||
0918682b | 893 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
894 | [IORING_OP_NOP] = {}, |
895 | [IORING_OP_READV] = { | |
d3656344 JA |
896 | .needs_file = 1, |
897 | .unbound_nonreg_file = 1, | |
8a72758c | 898 | .pollin = 1, |
4d954c25 | 899 | .buffer_select = 1, |
26f0505a | 900 | .needs_async_setup = 1, |
27926b68 | 901 | .plug = 1, |
e8c2bc1f | 902 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 903 | }, |
0463b6c5 | 904 | [IORING_OP_WRITEV] = { |
d3656344 JA |
905 | .needs_file = 1, |
906 | .hash_reg_file = 1, | |
907 | .unbound_nonreg_file = 1, | |
8a72758c | 908 | .pollout = 1, |
26f0505a | 909 | .needs_async_setup = 1, |
27926b68 | 910 | .plug = 1, |
e8c2bc1f | 911 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 912 | }, |
0463b6c5 | 913 | [IORING_OP_FSYNC] = { |
d3656344 JA |
914 | .needs_file = 1, |
915 | }, | |
0463b6c5 | 916 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
917 | .needs_file = 1, |
918 | .unbound_nonreg_file = 1, | |
8a72758c | 919 | .pollin = 1, |
27926b68 | 920 | .plug = 1, |
e8c2bc1f | 921 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 922 | }, |
0463b6c5 | 923 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
924 | .needs_file = 1, |
925 | .hash_reg_file = 1, | |
926 | .unbound_nonreg_file = 1, | |
8a72758c | 927 | .pollout = 1, |
27926b68 | 928 | .plug = 1, |
e8c2bc1f | 929 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 930 | }, |
0463b6c5 | 931 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
932 | .needs_file = 1, |
933 | .unbound_nonreg_file = 1, | |
934 | }, | |
0463b6c5 PB |
935 | [IORING_OP_POLL_REMOVE] = {}, |
936 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
937 | .needs_file = 1, |
938 | }, | |
0463b6c5 | 939 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
940 | .needs_file = 1, |
941 | .unbound_nonreg_file = 1, | |
8a72758c | 942 | .pollout = 1, |
26f0505a | 943 | .needs_async_setup = 1, |
e8c2bc1f | 944 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 945 | }, |
0463b6c5 | 946 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
947 | .needs_file = 1, |
948 | .unbound_nonreg_file = 1, | |
8a72758c | 949 | .pollin = 1, |
52de1fe1 | 950 | .buffer_select = 1, |
26f0505a | 951 | .needs_async_setup = 1, |
e8c2bc1f | 952 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 953 | }, |
0463b6c5 | 954 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 955 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 956 | }, |
9c8e11b3 PB |
957 | [IORING_OP_TIMEOUT_REMOVE] = { |
958 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 959 | }, |
0463b6c5 | 960 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
961 | .needs_file = 1, |
962 | .unbound_nonreg_file = 1, | |
8a72758c | 963 | .pollin = 1, |
d3656344 | 964 | }, |
0463b6c5 PB |
965 | [IORING_OP_ASYNC_CANCEL] = {}, |
966 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 967 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 968 | }, |
0463b6c5 | 969 | [IORING_OP_CONNECT] = { |
d3656344 JA |
970 | .needs_file = 1, |
971 | .unbound_nonreg_file = 1, | |
8a72758c | 972 | .pollout = 1, |
26f0505a | 973 | .needs_async_setup = 1, |
e8c2bc1f | 974 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 975 | }, |
0463b6c5 | 976 | [IORING_OP_FALLOCATE] = { |
d3656344 | 977 | .needs_file = 1, |
d3656344 | 978 | }, |
44526bed JA |
979 | [IORING_OP_OPENAT] = {}, |
980 | [IORING_OP_CLOSE] = {}, | |
981 | [IORING_OP_FILES_UPDATE] = {}, | |
982 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 983 | [IORING_OP_READ] = { |
3a6820f2 JA |
984 | .needs_file = 1, |
985 | .unbound_nonreg_file = 1, | |
8a72758c | 986 | .pollin = 1, |
bcda7baa | 987 | .buffer_select = 1, |
27926b68 | 988 | .plug = 1, |
e8c2bc1f | 989 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 990 | }, |
0463b6c5 | 991 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
992 | .needs_file = 1, |
993 | .unbound_nonreg_file = 1, | |
8a72758c | 994 | .pollout = 1, |
27926b68 | 995 | .plug = 1, |
e8c2bc1f | 996 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 997 | }, |
0463b6c5 | 998 | [IORING_OP_FADVISE] = { |
4840e418 | 999 | .needs_file = 1, |
c1ca757b | 1000 | }, |
44526bed | 1001 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1002 | [IORING_OP_SEND] = { |
fddaface JA |
1003 | .needs_file = 1, |
1004 | .unbound_nonreg_file = 1, | |
8a72758c | 1005 | .pollout = 1, |
fddaface | 1006 | }, |
0463b6c5 | 1007 | [IORING_OP_RECV] = { |
fddaface JA |
1008 | .needs_file = 1, |
1009 | .unbound_nonreg_file = 1, | |
8a72758c | 1010 | .pollin = 1, |
bcda7baa | 1011 | .buffer_select = 1, |
fddaface | 1012 | }, |
0463b6c5 | 1013 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1014 | }, |
3e4827b0 JA |
1015 | [IORING_OP_EPOLL_CTL] = { |
1016 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1017 | }, |
7d67af2c PB |
1018 | [IORING_OP_SPLICE] = { |
1019 | .needs_file = 1, | |
1020 | .hash_reg_file = 1, | |
1021 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1022 | }, |
1023 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1024 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1025 | [IORING_OP_TEE] = { |
1026 | .needs_file = 1, | |
1027 | .hash_reg_file = 1, | |
1028 | .unbound_nonreg_file = 1, | |
1029 | }, | |
36f4fa68 JA |
1030 | [IORING_OP_SHUTDOWN] = { |
1031 | .needs_file = 1, | |
1032 | }, | |
44526bed JA |
1033 | [IORING_OP_RENAMEAT] = {}, |
1034 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1035 | }; |
1036 | ||
0756a869 PB |
1037 | /* requests with any of those set should undergo io_disarm_next() */ |
1038 | #define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL) | |
1039 | ||
7a612350 | 1040 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1041 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1042 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1043 | struct task_struct *task, | |
3dd0c97a | 1044 | bool cancel_all); |
78cc687b | 1045 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
1ffc5422 | 1046 | |
d4d19c19 PB |
1047 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1048 | long res, unsigned int cflags); | |
ec9c02ad | 1049 | static void io_put_req(struct io_kiocb *req); |
91c2f697 | 1050 | static void io_put_req_deferred(struct io_kiocb *req); |
c7dae4ba | 1051 | static void io_dismantle_req(struct io_kiocb *req); |
94ae5e77 | 1052 | static void io_queue_linked_timeout(struct io_kiocb *req); |
fdecb662 | 1053 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1054 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1055 | unsigned nr_args); |
68fb8979 | 1056 | static void io_clean_op(struct io_kiocb *req); |
ac177053 | 1057 | static struct file *io_file_get(struct io_ring_ctx *ctx, |
8371adf5 | 1058 | struct io_kiocb *req, int fd, bool fixed); |
c5eef2b9 | 1059 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1060 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1061 | |
907d1df3 | 1062 | static void io_req_task_queue(struct io_kiocb *req); |
2a2758f2 | 1063 | static void io_submit_flush_completions(struct io_ring_ctx *ctx); |
179ae0d1 | 1064 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1065 | |
2b188cc1 JA |
1066 | static struct kmem_cache *req_cachep; |
1067 | ||
0918682b | 1068 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1069 | |
1070 | struct sock *io_uring_get_socket(struct file *file) | |
1071 | { | |
1072 | #if defined(CONFIG_UNIX) | |
1073 | if (file->f_op == &io_uring_fops) { | |
1074 | struct io_ring_ctx *ctx = file->private_data; | |
1075 | ||
1076 | return ctx->ring_sock->sk; | |
1077 | } | |
1078 | #endif | |
1079 | return NULL; | |
1080 | } | |
1081 | EXPORT_SYMBOL(io_uring_get_socket); | |
1082 | ||
f2f87370 PB |
1083 | #define io_for_each_link(pos, head) \ |
1084 | for (pos = (head); pos; pos = pos->link) | |
1085 | ||
21c843d5 PB |
1086 | /* |
1087 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1088 | * see commit f958d7b528b1 for details. | |
1089 | */ | |
1090 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1091 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1092 | ||
1093 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) | |
1094 | { | |
20e60a38 | 1095 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1096 | return atomic_inc_not_zero(&req->refs); |
1097 | } | |
1098 | ||
21c843d5 PB |
1099 | static inline bool req_ref_put_and_test(struct io_kiocb *req) |
1100 | { | |
20e60a38 PB |
1101 | if (likely(!(req->flags & REQ_F_REFCOUNT))) |
1102 | return true; | |
1103 | ||
21c843d5 PB |
1104 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1105 | return atomic_dec_and_test(&req->refs); | |
1106 | } | |
1107 | ||
1108 | static inline void req_ref_put(struct io_kiocb *req) | |
1109 | { | |
20e60a38 | 1110 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1111 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
1112 | } | |
1113 | ||
1114 | static inline void req_ref_get(struct io_kiocb *req) | |
1115 | { | |
20e60a38 | 1116 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1117 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1118 | atomic_inc(&req->refs); | |
1119 | } | |
1120 | ||
48dcd38d | 1121 | static inline void __io_req_set_refcount(struct io_kiocb *req, int nr) |
20e60a38 PB |
1122 | { |
1123 | if (!(req->flags & REQ_F_REFCOUNT)) { | |
1124 | req->flags |= REQ_F_REFCOUNT; | |
48dcd38d | 1125 | atomic_set(&req->refs, nr); |
20e60a38 PB |
1126 | } |
1127 | } | |
1128 | ||
48dcd38d PB |
1129 | static inline void io_req_set_refcount(struct io_kiocb *req) |
1130 | { | |
1131 | __io_req_set_refcount(req, 1); | |
1132 | } | |
1133 | ||
b895c9a6 | 1134 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1135 | { |
1136 | struct io_ring_ctx *ctx = req->ctx; | |
1137 | ||
269bbe5f | 1138 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1139 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1140 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1141 | } |
1142 | } | |
1143 | ||
f70865db PB |
1144 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1145 | { | |
1146 | bool got = percpu_ref_tryget(ref); | |
1147 | ||
1148 | /* already at zero, wait for ->release() */ | |
1149 | if (!got) | |
1150 | wait_for_completion(compl); | |
1151 | percpu_ref_resurrect(ref); | |
1152 | if (got) | |
1153 | percpu_ref_put(ref); | |
1154 | } | |
1155 | ||
3dd0c97a PB |
1156 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1157 | bool cancel_all) | |
08d23634 PB |
1158 | { |
1159 | struct io_kiocb *req; | |
1160 | ||
68207680 | 1161 | if (task && head->task != task) |
08d23634 | 1162 | return false; |
3dd0c97a | 1163 | if (cancel_all) |
08d23634 PB |
1164 | return true; |
1165 | ||
1166 | io_for_each_link(req, head) { | |
b05a1bcd | 1167 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1168 | return true; |
08d23634 PB |
1169 | } |
1170 | return false; | |
1171 | } | |
1172 | ||
93d2bcd2 | 1173 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1174 | { |
93d2bcd2 | 1175 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1176 | } |
4a38aed2 | 1177 | |
2b188cc1 JA |
1178 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1179 | { | |
1180 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1181 | ||
0f158b4c | 1182 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1183 | } |
1184 | ||
8eb7e2d0 PB |
1185 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1186 | { | |
1187 | return !req->timeout.off; | |
1188 | } | |
1189 | ||
f56165e6 PB |
1190 | static void io_fallback_req_func(struct work_struct *work) |
1191 | { | |
1192 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
1193 | fallback_work.work); | |
1194 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
1195 | struct io_kiocb *req, *tmp; | |
1196 | ||
1197 | percpu_ref_get(&ctx->refs); | |
1198 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) | |
1199 | req->io_task_work.func(req); | |
1200 | percpu_ref_put(&ctx->refs); | |
1201 | } | |
1202 | ||
2b188cc1 JA |
1203 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1204 | { | |
1205 | struct io_ring_ctx *ctx; | |
78076bb6 | 1206 | int hash_bits; |
2b188cc1 JA |
1207 | |
1208 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1209 | if (!ctx) | |
1210 | return NULL; | |
1211 | ||
78076bb6 JA |
1212 | /* |
1213 | * Use 5 bits less than the max cq entries, that should give us around | |
1214 | * 32 entries per hash list if totally full and uniformly spread. | |
1215 | */ | |
1216 | hash_bits = ilog2(p->cq_entries); | |
1217 | hash_bits -= 5; | |
1218 | if (hash_bits <= 0) | |
1219 | hash_bits = 1; | |
1220 | ctx->cancel_hash_bits = hash_bits; | |
1221 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1222 | GFP_KERNEL); | |
1223 | if (!ctx->cancel_hash) | |
1224 | goto err; | |
1225 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1226 | ||
6224843d PB |
1227 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1228 | if (!ctx->dummy_ubuf) | |
1229 | goto err; | |
1230 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1231 | ctx->dummy_ubuf->ubuf = -1UL; | |
1232 | ||
21482896 | 1233 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1234 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1235 | goto err; | |
2b188cc1 JA |
1236 | |
1237 | ctx->flags = p->flags; | |
90554200 | 1238 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1239 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1240 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1241 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1242 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1243 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1244 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1245 | mutex_init(&ctx->uring_lock); |
311997b3 | 1246 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1247 | spin_lock_init(&ctx->completion_lock); |
89850fce | 1248 | spin_lock_init(&ctx->timeout_lock); |
540e32a0 | 1249 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1250 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1251 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1252 | spin_lock_init(&ctx->rsrc_ref_lock); |
1253 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1254 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1255 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1256 | INIT_LIST_HEAD(&ctx->tctx_list); |
cd0ca2e0 | 1257 | INIT_LIST_HEAD(&ctx->submit_state.free_list); |
d0acdee2 | 1258 | INIT_LIST_HEAD(&ctx->locked_free_list); |
9011bf9a | 1259 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
2b188cc1 | 1260 | return ctx; |
206aefde | 1261 | err: |
6224843d | 1262 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1263 | kfree(ctx->cancel_hash); |
206aefde JA |
1264 | kfree(ctx); |
1265 | return NULL; | |
2b188cc1 JA |
1266 | } |
1267 | ||
8f6ed49a PB |
1268 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1269 | { | |
1270 | struct io_rings *r = ctx->rings; | |
1271 | ||
1272 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1273 | ctx->cq_extra--; | |
1274 | } | |
1275 | ||
9cf7c104 | 1276 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1277 | { |
2bc9930e JA |
1278 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1279 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1280 | |
8f6ed49a | 1281 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1282 | } |
de0617e4 | 1283 | |
9d858b21 | 1284 | return false; |
de0617e4 JA |
1285 | } |
1286 | ||
c97d8a0f PB |
1287 | #define FFS_ASYNC_READ 0x1UL |
1288 | #define FFS_ASYNC_WRITE 0x2UL | |
1289 | #ifdef CONFIG_64BIT | |
1290 | #define FFS_ISREG 0x4UL | |
1291 | #else | |
1292 | #define FFS_ISREG 0x0UL | |
1293 | #endif | |
1294 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
1295 | ||
1296 | static inline bool io_req_ffs_set(struct io_kiocb *req) | |
1297 | { | |
1298 | return IS_ENABLED(CONFIG_64BIT) && (req->flags & REQ_F_FIXED_FILE); | |
1299 | } | |
1300 | ||
ce3d5aae PB |
1301 | static void io_req_track_inflight(struct io_kiocb *req) |
1302 | { | |
ce3d5aae | 1303 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1304 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1305 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1306 | } |
1307 | } | |
1308 | ||
906c6caa PB |
1309 | static inline void io_unprep_linked_timeout(struct io_kiocb *req) |
1310 | { | |
1311 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
1312 | } | |
1313 | ||
fd08e530 PB |
1314 | static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req) |
1315 | { | |
906c6caa PB |
1316 | if (WARN_ON_ONCE(!req->link)) |
1317 | return NULL; | |
1318 | ||
4d13d1a4 PB |
1319 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
1320 | req->flags |= REQ_F_LINK_TIMEOUT; | |
fd08e530 PB |
1321 | |
1322 | /* linked timeouts should have two refs once prep'ed */ | |
48dcd38d | 1323 | io_req_set_refcount(req); |
4d13d1a4 PB |
1324 | __io_req_set_refcount(req->link, 2); |
1325 | return req->link; | |
fd08e530 PB |
1326 | } |
1327 | ||
1328 | static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) | |
1329 | { | |
4d13d1a4 | 1330 | if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) |
fd08e530 PB |
1331 | return NULL; |
1332 | return __io_prep_linked_timeout(req); | |
1333 | } | |
1334 | ||
1e6fa521 JA |
1335 | static void io_prep_async_work(struct io_kiocb *req) |
1336 | { | |
1337 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1338 | struct io_ring_ctx *ctx = req->ctx; |
1339 | ||
b8e64b53 PB |
1340 | if (!(req->flags & REQ_F_CREDS)) { |
1341 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1342 | req->creds = get_current_cred(); |
b8e64b53 | 1343 | } |
003e8dcc | 1344 | |
e1d675df PB |
1345 | req->work.list.next = NULL; |
1346 | req->work.flags = 0; | |
feaadc4f PB |
1347 | if (req->flags & REQ_F_FORCE_ASYNC) |
1348 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1349 | ||
1e6fa521 JA |
1350 | if (req->flags & REQ_F_ISREG) { |
1351 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1352 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1353 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1354 | if (def->unbound_nonreg_file) |
1355 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1356 | } | |
e1d675df PB |
1357 | |
1358 | switch (req->opcode) { | |
1359 | case IORING_OP_SPLICE: | |
1360 | case IORING_OP_TEE: | |
e1d675df PB |
1361 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1362 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1363 | break; | |
1364 | } | |
561fb04a | 1365 | } |
cccf0ee8 | 1366 | |
cbdcb435 | 1367 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1368 | { |
cbdcb435 | 1369 | struct io_kiocb *cur; |
54a91f3b | 1370 | |
44eff40a PB |
1371 | if (req->flags & REQ_F_LINK_TIMEOUT) { |
1372 | struct io_ring_ctx *ctx = req->ctx; | |
1373 | ||
79ebeaee | 1374 | spin_lock(&ctx->completion_lock); |
44eff40a PB |
1375 | io_for_each_link(cur, req) |
1376 | io_prep_async_work(cur); | |
79ebeaee | 1377 | spin_unlock(&ctx->completion_lock); |
44eff40a PB |
1378 | } else { |
1379 | io_for_each_link(cur, req) | |
1380 | io_prep_async_work(cur); | |
1381 | } | |
561fb04a JA |
1382 | } |
1383 | ||
ebf93667 | 1384 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1385 | { |
a197f664 | 1386 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1387 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1388 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1389 | |
3bfe6106 JA |
1390 | BUG_ON(!tctx); |
1391 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1392 | |
cbdcb435 PB |
1393 | /* init ->work of the whole link before punting */ |
1394 | io_prep_async_link(req); | |
991468dc JA |
1395 | |
1396 | /* | |
1397 | * Not expected to happen, but if we do have a bug where this _can_ | |
1398 | * happen, catch it here and ensure the request is marked as | |
1399 | * canceled. That will make io-wq go through the usual work cancel | |
1400 | * procedure rather than attempt to run this request (or create a new | |
1401 | * worker for it). | |
1402 | */ | |
1403 | if (WARN_ON_ONCE(!same_thread_group(req->task, current))) | |
1404 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1405 | ||
d07f1e8a PB |
1406 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1407 | &req->work, req->flags); | |
ebf93667 | 1408 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1409 | if (link) |
1410 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1411 | } |
1412 | ||
1ee4160c | 1413 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1414 | __must_hold(&req->ctx->completion_lock) |
89850fce | 1415 | __must_hold(&req->ctx->timeout_lock) |
5262f567 | 1416 | { |
e8c2bc1f | 1417 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1418 | |
fd9c7bc5 | 1419 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1420 | atomic_set(&req->ctx->cq_timeouts, |
1421 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1422 | list_del_init(&req->timeout.list); |
d4d19c19 | 1423 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
91c2f697 | 1424 | io_put_req_deferred(req); |
5262f567 JA |
1425 | } |
1426 | } | |
1427 | ||
441b8a78 | 1428 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1429 | { |
441b8a78 | 1430 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1431 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1432 | struct io_defer_entry, list); | |
de0617e4 | 1433 | |
9cf7c104 | 1434 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1435 | break; |
27dc8338 | 1436 | list_del_init(&de->list); |
907d1df3 | 1437 | io_req_task_queue(de->req); |
27dc8338 | 1438 | kfree(de); |
441b8a78 | 1439 | } |
04518945 PB |
1440 | } |
1441 | ||
360428f8 | 1442 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
89850fce | 1443 | __must_hold(&ctx->completion_lock) |
de0617e4 | 1444 | { |
441b8a78 | 1445 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1446 | |
79ebeaee | 1447 | spin_lock_irq(&ctx->timeout_lock); |
f18ee4cf | 1448 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1449 | u32 events_needed, events_got; |
360428f8 | 1450 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1451 | struct io_kiocb, timeout.list); |
de0617e4 | 1452 | |
8eb7e2d0 | 1453 | if (io_is_timeout_noseq(req)) |
360428f8 | 1454 | break; |
f010505b MDG |
1455 | |
1456 | /* | |
1457 | * Since seq can easily wrap around over time, subtract | |
1458 | * the last seq at which timeouts were flushed before comparing. | |
1459 | * Assuming not more than 2^31-1 events have happened since, | |
1460 | * these subtractions won't have wrapped, so we can check if | |
1461 | * target is in [last_seq, current_seq] by comparing the two. | |
1462 | */ | |
1463 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1464 | events_got = seq - ctx->cq_last_tm_flush; | |
1465 | if (events_got < events_needed) | |
360428f8 | 1466 | break; |
bfe68a22 | 1467 | |
135fcde8 | 1468 | list_del_init(&req->timeout.list); |
1ee4160c | 1469 | io_kill_timeout(req, 0); |
f18ee4cf | 1470 | } |
f010505b | 1471 | ctx->cq_last_tm_flush = seq; |
79ebeaee | 1472 | spin_unlock_irq(&ctx->timeout_lock); |
360428f8 | 1473 | } |
5262f567 | 1474 | |
2335f6f5 | 1475 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1476 | { |
2335f6f5 PB |
1477 | if (ctx->off_timeout_used) |
1478 | io_flush_timeouts(ctx); | |
1479 | if (ctx->drain_active) | |
1480 | io_queue_deferred(ctx); | |
1481 | } | |
1482 | ||
1483 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) | |
1484 | { | |
1485 | if (unlikely(ctx->off_timeout_used || ctx->drain_active)) | |
1486 | __io_commit_cqring_flush(ctx); | |
ec30e04b PB |
1487 | /* order cqe stores with ring update */ |
1488 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1489 | } |
1490 | ||
90554200 JA |
1491 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1492 | { | |
1493 | struct io_rings *r = ctx->rings; | |
1494 | ||
a566c556 | 1495 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1496 | } |
1497 | ||
888aae2e PB |
1498 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1499 | { | |
1500 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1501 | } | |
1502 | ||
d068b506 | 1503 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1504 | { |
75b28aff | 1505 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1506 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1507 | |
115e12e5 SB |
1508 | /* |
1509 | * writes to the cq entry need to come after reading head; the | |
1510 | * control dependency is enough as we're using WRITE_ONCE to | |
1511 | * fill the cq entry | |
1512 | */ | |
a566c556 | 1513 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1514 | return NULL; |
1515 | ||
888aae2e | 1516 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1517 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1518 | } |
1519 | ||
f2842ab5 JA |
1520 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1521 | { | |
44c769de | 1522 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1523 | return false; |
7e55a19c SG |
1524 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1525 | return false; | |
44c769de | 1526 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1527 | } |
1528 | ||
2c5d763c JA |
1529 | /* |
1530 | * This should only get called when at least one event has been posted. | |
1531 | * Some applications rely on the eventfd notification count only changing | |
1532 | * IFF a new CQE has been added to the CQ ring. There's no depedency on | |
1533 | * 1:1 relationship between how many times this function is called (and | |
1534 | * hence the eventfd count) and number of CQEs posted to the CQ ring. | |
1535 | */ | |
b41e9852 | 1536 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1537 | { |
5fd46178 JA |
1538 | /* |
1539 | * wake_up_all() may seem excessive, but io_wake_function() and | |
1540 | * io_should_wake() handle the termination of the loop and only | |
1541 | * wake as many waiters as we need to. | |
1542 | */ | |
1543 | if (wq_has_sleeper(&ctx->cq_wait)) | |
1544 | wake_up_all(&ctx->cq_wait); | |
534ca6d6 JA |
1545 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1546 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1547 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1548 | eventfd_signal(ctx->cq_ev_fd, 1); |
311997b3 PB |
1549 | if (waitqueue_active(&ctx->poll_wait)) { |
1550 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1551 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1552 | } | |
1d7bb1d5 JA |
1553 | } |
1554 | ||
80c18e4a PB |
1555 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1556 | { | |
1557 | if (ctx->flags & IORING_SETUP_SQPOLL) { | |
5fd46178 JA |
1558 | if (wq_has_sleeper(&ctx->cq_wait)) |
1559 | wake_up_all(&ctx->cq_wait); | |
80c18e4a PB |
1560 | } |
1561 | if (io_should_trigger_evfd(ctx)) | |
1562 | eventfd_signal(ctx->cq_ev_fd, 1); | |
311997b3 PB |
1563 | if (waitqueue_active(&ctx->poll_wait)) { |
1564 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1565 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1566 | } | |
80c18e4a PB |
1567 | } |
1568 | ||
c4a2ed72 | 1569 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1570 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1571 | { |
b18032bb | 1572 | bool all_flushed, posted; |
1d7bb1d5 | 1573 | |
a566c556 | 1574 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1575 | return false; |
1d7bb1d5 | 1576 | |
b18032bb | 1577 | posted = false; |
79ebeaee | 1578 | spin_lock(&ctx->completion_lock); |
6c2450ae | 1579 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1580 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1581 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1582 | |
1d7bb1d5 JA |
1583 | if (!cqe && !force) |
1584 | break; | |
6c2450ae PB |
1585 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1586 | struct io_overflow_cqe, list); | |
1587 | if (cqe) | |
1588 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1589 | else | |
8f6ed49a PB |
1590 | io_account_cq_overflow(ctx); |
1591 | ||
b18032bb | 1592 | posted = true; |
6c2450ae PB |
1593 | list_del(&ocqe->list); |
1594 | kfree(ocqe); | |
1d7bb1d5 JA |
1595 | } |
1596 | ||
09e88404 PB |
1597 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1598 | if (all_flushed) { | |
5ed7a37d | 1599 | clear_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1600 | WRITE_ONCE(ctx->rings->sq_flags, |
1601 | ctx->rings->sq_flags & ~IORING_SQ_CQ_OVERFLOW); | |
09e88404 | 1602 | } |
46930143 | 1603 | |
b18032bb JA |
1604 | if (posted) |
1605 | io_commit_cqring(ctx); | |
79ebeaee | 1606 | spin_unlock(&ctx->completion_lock); |
b18032bb JA |
1607 | if (posted) |
1608 | io_cqring_ev_posted(ctx); | |
09e88404 | 1609 | return all_flushed; |
1d7bb1d5 JA |
1610 | } |
1611 | ||
90f67366 | 1612 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx) |
6c503150 | 1613 | { |
ca0a2651 JA |
1614 | bool ret = true; |
1615 | ||
5ed7a37d | 1616 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1617 | /* iopoll syncs against uring_lock, not completion_lock */ |
1618 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1619 | mutex_lock(&ctx->uring_lock); | |
90f67366 | 1620 | ret = __io_cqring_overflow_flush(ctx, false); |
6c503150 PB |
1621 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1622 | mutex_unlock(&ctx->uring_lock); | |
1623 | } | |
ca0a2651 JA |
1624 | |
1625 | return ret; | |
6c503150 PB |
1626 | } |
1627 | ||
6a290a14 PB |
1628 | /* must to be called somewhat shortly after putting a request */ |
1629 | static inline void io_put_task(struct task_struct *task, int nr) | |
1630 | { | |
1631 | struct io_uring_task *tctx = task->io_uring; | |
1632 | ||
e98e49b2 PB |
1633 | if (likely(task == current)) { |
1634 | tctx->cached_refs += nr; | |
1635 | } else { | |
1636 | percpu_counter_sub(&tctx->inflight, nr); | |
1637 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1638 | wake_up(&tctx->wait); | |
1639 | put_task_struct_many(task, nr); | |
1640 | } | |
6a290a14 PB |
1641 | } |
1642 | ||
d4d19c19 PB |
1643 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1644 | long res, unsigned int cflags) | |
2b188cc1 | 1645 | { |
cce4b8b0 | 1646 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1647 | |
cce4b8b0 PB |
1648 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1649 | if (!ocqe) { | |
1650 | /* | |
1651 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1652 | * or cannot allocate an overflow entry, then we need to drop it | |
1653 | * on the floor. | |
1654 | */ | |
8f6ed49a | 1655 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1656 | return false; |
2b188cc1 | 1657 | } |
cce4b8b0 | 1658 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1659 | set_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1660 | WRITE_ONCE(ctx->rings->sq_flags, |
1661 | ctx->rings->sq_flags | IORING_SQ_CQ_OVERFLOW); | |
1662 | ||
cce4b8b0 | 1663 | } |
d4d19c19 | 1664 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1665 | ocqe->cqe.res = res; |
1666 | ocqe->cqe.flags = cflags; | |
1667 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1668 | return true; | |
2b188cc1 JA |
1669 | } |
1670 | ||
d4d19c19 PB |
1671 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1672 | long res, unsigned int cflags) | |
2b188cc1 JA |
1673 | { |
1674 | struct io_uring_cqe *cqe; | |
1675 | ||
d4d19c19 | 1676 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1677 | |
2b188cc1 JA |
1678 | /* |
1679 | * If we can't get a cq entry, userspace overflowed the | |
1680 | * submission (by quite a lot). Increment the overflow count in | |
1681 | * the ring. | |
1682 | */ | |
d068b506 | 1683 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1684 | if (likely(cqe)) { |
d4d19c19 | 1685 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1686 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1687 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1688 | return true; |
2b188cc1 | 1689 | } |
d4d19c19 | 1690 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1691 | } |
1692 | ||
8d13326e | 1693 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1694 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1695 | long res, unsigned int cflags) | |
bcda7baa | 1696 | { |
d4d19c19 | 1697 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1698 | } |
1699 | ||
7a612350 PB |
1700 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1701 | unsigned int cflags) | |
2b188cc1 | 1702 | { |
78e19bbe | 1703 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 | 1704 | |
79ebeaee | 1705 | spin_lock(&ctx->completion_lock); |
d4d19c19 | 1706 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1707 | /* |
1708 | * If we're the last reference to this request, add to our locked | |
1709 | * free_list cache. | |
1710 | */ | |
de9b4cca | 1711 | if (req_ref_put_and_test(req)) { |
7a612350 | 1712 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
0756a869 | 1713 | if (req->flags & IO_DISARM_MASK) |
7a612350 PB |
1714 | io_disarm_next(req); |
1715 | if (req->link) { | |
1716 | io_req_task_queue(req->link); | |
1717 | req->link = NULL; | |
1718 | } | |
1719 | } | |
c7dae4ba JA |
1720 | io_dismantle_req(req); |
1721 | io_put_task(req->task, 1); | |
bb943b82 | 1722 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
d0acdee2 | 1723 | ctx->locked_free_nr++; |
180f829f PB |
1724 | } else { |
1725 | if (!percpu_ref_tryget(&ctx->refs)) | |
1726 | req = NULL; | |
1727 | } | |
7a612350 | 1728 | io_commit_cqring(ctx); |
79ebeaee | 1729 | spin_unlock(&ctx->completion_lock); |
7a612350 | 1730 | |
180f829f PB |
1731 | if (req) { |
1732 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1733 | percpu_ref_put(&ctx->refs); |
180f829f | 1734 | } |
229a7b63 JA |
1735 | } |
1736 | ||
4e3d9ff9 JA |
1737 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1738 | { | |
c854357b | 1739 | return req->flags & IO_REQ_CLEAN_FLAGS; |
4e3d9ff9 JA |
1740 | } |
1741 | ||
a38d68db | 1742 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1743 | unsigned int cflags) |
229a7b63 | 1744 | { |
4e3d9ff9 | 1745 | if (io_req_needs_clean(req)) |
68fb8979 | 1746 | io_clean_op(req); |
a38d68db PB |
1747 | req->result = res; |
1748 | req->compl.cflags = cflags; | |
e342c807 | 1749 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1750 | } |
1751 | ||
889fca73 PB |
1752 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1753 | long res, unsigned cflags) | |
bcda7baa | 1754 | { |
889fca73 PB |
1755 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1756 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1757 | else |
c7dae4ba | 1758 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1759 | } |
1760 | ||
a38d68db | 1761 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1762 | { |
889fca73 | 1763 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1764 | } |
1765 | ||
f41db273 PB |
1766 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1767 | { | |
93d2bcd2 | 1768 | req_set_fail(req); |
f41db273 PB |
1769 | io_req_complete_post(req, res, 0); |
1770 | } | |
1771 | ||
864ea921 PB |
1772 | /* |
1773 | * Don't initialise the fields below on every allocation, but do that in | |
1774 | * advance and keep them valid across allocations. | |
1775 | */ | |
1776 | static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
1777 | { | |
1778 | req->ctx = ctx; | |
1779 | req->link = NULL; | |
1780 | req->async_data = NULL; | |
1781 | /* not necessary, but safer to zero */ | |
1782 | req->result = 0; | |
1783 | } | |
1784 | ||
dac7a098 | 1785 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
cd0ca2e0 | 1786 | struct io_submit_state *state) |
dac7a098 | 1787 | { |
79ebeaee | 1788 | spin_lock(&ctx->completion_lock); |
cd0ca2e0 | 1789 | list_splice_init(&ctx->locked_free_list, &state->free_list); |
d0acdee2 | 1790 | ctx->locked_free_nr = 0; |
79ebeaee | 1791 | spin_unlock(&ctx->completion_lock); |
dac7a098 PB |
1792 | } |
1793 | ||
dd78f492 | 1794 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1795 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1796 | { |
c7dae4ba | 1797 | struct io_submit_state *state = &ctx->submit_state; |
dd78f492 | 1798 | int nr; |
0ddf92e8 | 1799 | |
c7dae4ba JA |
1800 | /* |
1801 | * If we have more than a batch's worth of requests in our IRQ side | |
1802 | * locked cache, grab the lock and move them over to our submission | |
1803 | * side cache. | |
1804 | */ | |
d0acdee2 | 1805 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
cd0ca2e0 | 1806 | io_flush_cached_locked_reqs(ctx, state); |
0ddf92e8 | 1807 | |
dd78f492 | 1808 | nr = state->free_reqs; |
cd0ca2e0 PB |
1809 | while (!list_empty(&state->free_list)) { |
1810 | struct io_kiocb *req = list_first_entry(&state->free_list, | |
bb943b82 | 1811 | struct io_kiocb, inflight_entry); |
dd78f492 | 1812 | |
bb943b82 | 1813 | list_del(&req->inflight_entry); |
dd78f492 PB |
1814 | state->reqs[nr++] = req; |
1815 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1816 | break; |
1b4c351f JA |
1817 | } |
1818 | ||
dd78f492 PB |
1819 | state->free_reqs = nr; |
1820 | return nr != 0; | |
0ddf92e8 JA |
1821 | } |
1822 | ||
5d5901a3 PB |
1823 | /* |
1824 | * A request might get retired back into the request caches even before opcode | |
1825 | * handlers and io_issue_sqe() are done with it, e.g. inline completion path. | |
1826 | * Because of that, io_alloc_req() should be called only under ->uring_lock | |
1827 | * and with extra caution to not get a request that is still worked on. | |
1828 | */ | |
e5d1bc0a | 1829 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
5d5901a3 | 1830 | __must_hold(&ctx->uring_lock) |
2b188cc1 | 1831 | { |
e5d1bc0a | 1832 | struct io_submit_state *state = &ctx->submit_state; |
864ea921 PB |
1833 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
1834 | int ret, i; | |
e5d1bc0a | 1835 | |
fe7e3257 | 1836 | BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); |
e5d1bc0a | 1837 | |
864ea921 PB |
1838 | if (likely(state->free_reqs || io_flush_cached_reqs(ctx))) |
1839 | goto got_req; | |
e5d1bc0a | 1840 | |
864ea921 PB |
1841 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1842 | state->reqs); | |
fd6fab2c | 1843 | |
864ea921 PB |
1844 | /* |
1845 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1846 | * retry single alloc to be on the safe side. | |
1847 | */ | |
1848 | if (unlikely(ret <= 0)) { | |
1849 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1850 | if (!state->reqs[0]) | |
1851 | return NULL; | |
1852 | ret = 1; | |
2b188cc1 | 1853 | } |
864ea921 PB |
1854 | |
1855 | for (i = 0; i < ret; i++) | |
1856 | io_preinit_req(state->reqs[i], ctx); | |
1857 | state->free_reqs = ret; | |
e5d1bc0a | 1858 | got_req: |
291b2821 PB |
1859 | state->free_reqs--; |
1860 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1861 | } |
1862 | ||
e1d767f0 | 1863 | static inline void io_put_file(struct file *file) |
8da11c19 | 1864 | { |
e1d767f0 | 1865 | if (file) |
8da11c19 PB |
1866 | fput(file); |
1867 | } | |
1868 | ||
4edf20f9 | 1869 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1870 | { |
094bae49 | 1871 | unsigned int flags = req->flags; |
929a3af9 | 1872 | |
3a0a6902 PB |
1873 | if (io_req_needs_clean(req)) |
1874 | io_clean_op(req); | |
e1d767f0 PB |
1875 | if (!(flags & REQ_F_FIXED_FILE)) |
1876 | io_put_file(req->file); | |
269bbe5f BM |
1877 | if (req->fixed_rsrc_refs) |
1878 | percpu_ref_put(req->fixed_rsrc_refs); | |
99ebe4ef | 1879 | if (req->async_data) { |
094bae49 | 1880 | kfree(req->async_data); |
99ebe4ef PB |
1881 | req->async_data = NULL; |
1882 | } | |
e65ef56d JA |
1883 | } |
1884 | ||
216578e5 | 1885 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1886 | { |
51a4cc11 | 1887 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1888 | |
216578e5 | 1889 | io_dismantle_req(req); |
7c660731 | 1890 | io_put_task(req->task, 1); |
c6ca97b3 | 1891 | |
79ebeaee | 1892 | spin_lock(&ctx->completion_lock); |
bb943b82 | 1893 | list_add(&req->inflight_entry, &ctx->locked_free_list); |
c34b025f | 1894 | ctx->locked_free_nr++; |
79ebeaee | 1895 | spin_unlock(&ctx->completion_lock); |
c34b025f | 1896 | |
ecfc5177 | 1897 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1898 | } |
1899 | ||
f2f87370 PB |
1900 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1901 | { | |
1902 | struct io_kiocb *nxt = req->link; | |
1903 | ||
1904 | req->link = nxt->link; | |
1905 | nxt->link = NULL; | |
1906 | } | |
1907 | ||
33cc89a9 PB |
1908 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1909 | __must_hold(&req->ctx->completion_lock) | |
89b263f6 | 1910 | __must_hold(&req->ctx->timeout_lock) |
2665abfd | 1911 | { |
33cc89a9 | 1912 | struct io_kiocb *link = req->link; |
f2f87370 | 1913 | |
b97e736a | 1914 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
c9abd7ad | 1915 | struct io_timeout_data *io = link->async_data; |
7c86ffee | 1916 | |
f2f87370 | 1917 | io_remove_next_linked(req); |
90cd7e42 | 1918 | link->timeout.head = NULL; |
fd9c7bc5 | 1919 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
d4d19c19 PB |
1920 | io_cqring_fill_event(link->ctx, link->user_data, |
1921 | -ECANCELED, 0); | |
91c2f697 | 1922 | io_put_req_deferred(link); |
d4729fbd | 1923 | return true; |
c9abd7ad PB |
1924 | } |
1925 | } | |
d4729fbd | 1926 | return false; |
7c86ffee PB |
1927 | } |
1928 | ||
d148ca4b | 1929 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1930 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1931 | { |
33cc89a9 | 1932 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1933 | |
f2f87370 | 1934 | req->link = NULL; |
f2f87370 PB |
1935 | while (link) { |
1936 | nxt = link->link; | |
1937 | link->link = NULL; | |
2665abfd | 1938 | |
f2f87370 | 1939 | trace_io_uring_fail_link(req, link); |
d4d19c19 | 1940 | io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0); |
91c2f697 | 1941 | io_put_req_deferred(link); |
f2f87370 | 1942 | link = nxt; |
9e645e11 | 1943 | } |
33cc89a9 | 1944 | } |
9e645e11 | 1945 | |
33cc89a9 PB |
1946 | static bool io_disarm_next(struct io_kiocb *req) |
1947 | __must_hold(&req->ctx->completion_lock) | |
1948 | { | |
1949 | bool posted = false; | |
1950 | ||
0756a869 PB |
1951 | if (req->flags & REQ_F_ARM_LTIMEOUT) { |
1952 | struct io_kiocb *link = req->link; | |
1953 | ||
906c6caa | 1954 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
0756a869 PB |
1955 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
1956 | io_remove_next_linked(req); | |
1957 | io_cqring_fill_event(link->ctx, link->user_data, | |
1958 | -ECANCELED, 0); | |
1959 | io_put_req_deferred(link); | |
1960 | posted = true; | |
1961 | } | |
1962 | } else if (req->flags & REQ_F_LINK_TIMEOUT) { | |
89b263f6 JA |
1963 | struct io_ring_ctx *ctx = req->ctx; |
1964 | ||
1965 | spin_lock_irq(&ctx->timeout_lock); | |
33cc89a9 | 1966 | posted = io_kill_linked_timeout(req); |
89b263f6 JA |
1967 | spin_unlock_irq(&ctx->timeout_lock); |
1968 | } | |
93d2bcd2 | 1969 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 1970 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
1971 | posted |= (req->link != NULL); |
1972 | io_fail_links(req); | |
1973 | } | |
1974 | return posted; | |
9e645e11 JA |
1975 | } |
1976 | ||
3fa5e0f3 | 1977 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1978 | { |
33cc89a9 | 1979 | struct io_kiocb *nxt; |
944e58bf | 1980 | |
9e645e11 JA |
1981 | /* |
1982 | * If LINK is set, we have dependent requests in this chain. If we | |
1983 | * didn't fail this request, queue the first one up, moving any other | |
1984 | * dependencies to the next request. In case of failure, fail the rest | |
1985 | * of the chain. | |
1986 | */ | |
0756a869 | 1987 | if (req->flags & IO_DISARM_MASK) { |
33cc89a9 | 1988 | struct io_ring_ctx *ctx = req->ctx; |
33cc89a9 PB |
1989 | bool posted; |
1990 | ||
79ebeaee | 1991 | spin_lock(&ctx->completion_lock); |
33cc89a9 PB |
1992 | posted = io_disarm_next(req); |
1993 | if (posted) | |
1994 | io_commit_cqring(req->ctx); | |
79ebeaee | 1995 | spin_unlock(&ctx->completion_lock); |
33cc89a9 PB |
1996 | if (posted) |
1997 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1998 | } |
33cc89a9 PB |
1999 | nxt = req->link; |
2000 | req->link = NULL; | |
2001 | return nxt; | |
4d7dd462 | 2002 | } |
9e645e11 | 2003 | |
f2f87370 | 2004 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2005 | { |
cdbff982 | 2006 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2007 | return NULL; |
2008 | return __io_req_find_next(req); | |
2009 | } | |
2010 | ||
2c32395d PB |
2011 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
2012 | { | |
2013 | if (!ctx) | |
2014 | return; | |
cd0ca2e0 | 2015 | if (ctx->submit_state.compl_nr) { |
2c32395d | 2016 | mutex_lock(&ctx->uring_lock); |
99c8bc52 HX |
2017 | if (ctx->submit_state.compl_nr) |
2018 | io_submit_flush_completions(ctx); | |
2c32395d PB |
2019 | mutex_unlock(&ctx->uring_lock); |
2020 | } | |
2021 | percpu_ref_put(&ctx->refs); | |
2022 | } | |
2023 | ||
7cbf1722 | 2024 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2025 | { |
ebd0df2e | 2026 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
2027 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
2028 | task_work); | |
c40f6379 | 2029 | |
16f72070 | 2030 | while (1) { |
3f18407d PB |
2031 | struct io_wq_work_node *node; |
2032 | ||
2033 | spin_lock_irq(&tctx->task_lock); | |
c6538be9 | 2034 | node = tctx->task_list.first; |
3f18407d | 2035 | INIT_WQ_LIST(&tctx->task_list); |
6294f368 PB |
2036 | if (!node) |
2037 | tctx->task_running = false; | |
3f18407d | 2038 | spin_unlock_irq(&tctx->task_lock); |
6294f368 PB |
2039 | if (!node) |
2040 | break; | |
3f18407d | 2041 | |
6294f368 | 2042 | do { |
3f18407d PB |
2043 | struct io_wq_work_node *next = node->next; |
2044 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2045 | io_task_work.node); | |
2046 | ||
2047 | if (req->ctx != ctx) { | |
2048 | ctx_flush_and_put(ctx); | |
2049 | ctx = req->ctx; | |
2050 | percpu_ref_get(&ctx->refs); | |
2051 | } | |
5b0a6acc | 2052 | req->io_task_work.func(req); |
3f18407d | 2053 | node = next; |
6294f368 PB |
2054 | } while (node); |
2055 | ||
7cbf1722 | 2056 | cond_resched(); |
3f18407d | 2057 | } |
ebd0df2e PB |
2058 | |
2059 | ctx_flush_and_put(ctx); | |
7cbf1722 JA |
2060 | } |
2061 | ||
e09ee510 | 2062 | static void io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 2063 | { |
c15b79de | 2064 | struct task_struct *tsk = req->task; |
7cbf1722 | 2065 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 2066 | enum task_work_notify_mode notify; |
e09ee510 | 2067 | struct io_wq_work_node *node; |
0b81e80c | 2068 | unsigned long flags; |
6294f368 | 2069 | bool running; |
7cbf1722 JA |
2070 | |
2071 | WARN_ON_ONCE(!tctx); | |
2072 | ||
0b81e80c | 2073 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2074 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
6294f368 PB |
2075 | running = tctx->task_running; |
2076 | if (!running) | |
2077 | tctx->task_running = true; | |
0b81e80c | 2078 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2079 | |
2080 | /* task_work already pending, we're done */ | |
6294f368 | 2081 | if (running) |
e09ee510 | 2082 | return; |
7cbf1722 | 2083 | |
c15b79de PB |
2084 | /* |
2085 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
2086 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2087 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2088 | * will do the job. | |
2089 | */ | |
2090 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
c15b79de PB |
2091 | if (!task_work_add(tsk, &tctx->task_work, notify)) { |
2092 | wake_up_process(tsk); | |
e09ee510 | 2093 | return; |
c15b79de | 2094 | } |
2215bed9 | 2095 | |
0b81e80c | 2096 | spin_lock_irqsave(&tctx->task_lock, flags); |
6294f368 | 2097 | tctx->task_running = false; |
e09ee510 PB |
2098 | node = tctx->task_list.first; |
2099 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2100 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 2101 | |
e09ee510 PB |
2102 | while (node) { |
2103 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
2104 | node = node->next; | |
2105 | if (llist_add(&req->io_task_work.fallback_node, | |
2106 | &req->ctx->fallback_llist)) | |
2107 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
2108 | } | |
eab30c4d PB |
2109 | } |
2110 | ||
5b0a6acc | 2111 | static void io_req_task_cancel(struct io_kiocb *req) |
c40f6379 | 2112 | { |
87ceb6a6 | 2113 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2114 | |
e83acd7d | 2115 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2116 | mutex_lock(&ctx->uring_lock); |
2593553a | 2117 | io_req_complete_failed(req, req->result); |
792bb6eb | 2118 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2119 | } |
2120 | ||
5b0a6acc | 2121 | static void io_req_task_submit(struct io_kiocb *req) |
c40f6379 JA |
2122 | { |
2123 | struct io_ring_ctx *ctx = req->ctx; | |
2124 | ||
04fc6c80 | 2125 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2126 | mutex_lock(&ctx->uring_lock); |
316319e8 | 2127 | /* req->task == current here, checking PF_EXITING is safe */ |
af066f31 | 2128 | if (likely(!(req->task->flags & PF_EXITING))) |
c5eef2b9 | 2129 | __io_queue_sqe(req); |
81b6d05c | 2130 | else |
2593553a | 2131 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2132 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2133 | } |
2134 | ||
2c4b8eb6 | 2135 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2136 | { |
2c4b8eb6 | 2137 | req->result = ret; |
5b0a6acc | 2138 | req->io_task_work.func = io_req_task_cancel; |
e09ee510 | 2139 | io_req_task_work_add(req); |
c40f6379 JA |
2140 | } |
2141 | ||
2c4b8eb6 | 2142 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2143 | { |
5b0a6acc | 2144 | req->io_task_work.func = io_req_task_submit; |
e09ee510 | 2145 | io_req_task_work_add(req); |
a3df7698 PB |
2146 | } |
2147 | ||
773af691 JA |
2148 | static void io_req_task_queue_reissue(struct io_kiocb *req) |
2149 | { | |
2150 | req->io_task_work.func = io_queue_async_work; | |
2151 | io_req_task_work_add(req); | |
2152 | } | |
2153 | ||
f2f87370 | 2154 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2155 | { |
9b5f7bd9 | 2156 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2157 | |
2158 | if (nxt) | |
906a8c3f | 2159 | io_req_task_queue(nxt); |
c69f8dbe JL |
2160 | } |
2161 | ||
c3524383 | 2162 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2163 | { |
c3524383 PB |
2164 | io_queue_next(req); |
2165 | __io_free_req(req); | |
2166 | } | |
8766dd51 | 2167 | |
2d6500d4 | 2168 | struct req_batch { |
5af1d13e PB |
2169 | struct task_struct *task; |
2170 | int task_refs; | |
1b4c351f | 2171 | int ctx_refs; |
2d6500d4 PB |
2172 | }; |
2173 | ||
5af1d13e PB |
2174 | static inline void io_init_req_batch(struct req_batch *rb) |
2175 | { | |
5af1d13e | 2176 | rb->task_refs = 0; |
9ae72463 | 2177 | rb->ctx_refs = 0; |
5af1d13e PB |
2178 | rb->task = NULL; |
2179 | } | |
2180 | ||
2d6500d4 PB |
2181 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2182 | struct req_batch *rb) | |
2183 | { | |
9ae72463 PB |
2184 | if (rb->ctx_refs) |
2185 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
e98e49b2 | 2186 | if (rb->task) |
e9dbe221 | 2187 | io_put_task(rb->task, rb->task_refs); |
2d6500d4 PB |
2188 | } |
2189 | ||
6ff119a6 PB |
2190 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2191 | struct io_submit_state *state) | |
2d6500d4 | 2192 | { |
f2f87370 | 2193 | io_queue_next(req); |
96670657 | 2194 | io_dismantle_req(req); |
2d6500d4 | 2195 | |
e3bc8e9d | 2196 | if (req->task != rb->task) { |
7c660731 PB |
2197 | if (rb->task) |
2198 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2199 | rb->task = req->task; |
2200 | rb->task_refs = 0; | |
5af1d13e | 2201 | } |
e3bc8e9d | 2202 | rb->task_refs++; |
9ae72463 | 2203 | rb->ctx_refs++; |
5af1d13e | 2204 | |
bd759045 | 2205 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2206 | state->reqs[state->free_reqs++] = req; |
bd759045 | 2207 | else |
cd0ca2e0 | 2208 | list_add(&req->inflight_entry, &state->free_list); |
7a743e22 PB |
2209 | } |
2210 | ||
2a2758f2 | 2211 | static void io_submit_flush_completions(struct io_ring_ctx *ctx) |
a141dd89 | 2212 | __must_hold(&ctx->uring_lock) |
905c172f | 2213 | { |
cd0ca2e0 PB |
2214 | struct io_submit_state *state = &ctx->submit_state; |
2215 | int i, nr = state->compl_nr; | |
905c172f PB |
2216 | struct req_batch rb; |
2217 | ||
79ebeaee | 2218 | spin_lock(&ctx->completion_lock); |
905c172f | 2219 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2220 | struct io_kiocb *req = state->compl_reqs[i]; |
5182ed2e | 2221 | |
d4d19c19 PB |
2222 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2223 | req->compl.cflags); | |
905c172f PB |
2224 | } |
2225 | io_commit_cqring(ctx); | |
79ebeaee | 2226 | spin_unlock(&ctx->completion_lock); |
905c172f | 2227 | io_cqring_ev_posted(ctx); |
5182ed2e PB |
2228 | |
2229 | io_init_req_batch(&rb); | |
905c172f | 2230 | for (i = 0; i < nr; i++) { |
cd0ca2e0 | 2231 | struct io_kiocb *req = state->compl_reqs[i]; |
905c172f | 2232 | |
91c2f697 | 2233 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2234 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2235 | } |
2236 | ||
2237 | io_req_free_batch_finish(ctx, &rb); | |
cd0ca2e0 | 2238 | state->compl_nr = 0; |
7a743e22 PB |
2239 | } |
2240 | ||
ba816ad6 JA |
2241 | /* |
2242 | * Drop reference to request, return next in chain (if there is one) if this | |
2243 | * was the last reference to this request. | |
2244 | */ | |
0d85035a | 2245 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2246 | { |
9b5f7bd9 PB |
2247 | struct io_kiocb *nxt = NULL; |
2248 | ||
de9b4cca | 2249 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2250 | nxt = io_req_find_next(req); |
4d7dd462 | 2251 | __io_free_req(req); |
2a44f467 | 2252 | } |
9b5f7bd9 | 2253 | return nxt; |
2b188cc1 JA |
2254 | } |
2255 | ||
0d85035a | 2256 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2257 | { |
de9b4cca | 2258 | if (req_ref_put_and_test(req)) |
e65ef56d | 2259 | io_free_req(req); |
2b188cc1 JA |
2260 | } |
2261 | ||
91c2f697 | 2262 | static inline void io_put_req_deferred(struct io_kiocb *req) |
216578e5 | 2263 | { |
91c2f697 | 2264 | if (req_ref_put_and_test(req)) { |
543af3a1 PB |
2265 | req->io_task_work.func = io_free_req; |
2266 | io_req_task_work_add(req); | |
2267 | } | |
216578e5 PB |
2268 | } |
2269 | ||
6c503150 | 2270 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2271 | { |
2272 | /* See comment at the top of this file */ | |
2273 | smp_rmb(); | |
e23de15f | 2274 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2275 | } |
2276 | ||
fb5ccc98 PB |
2277 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2278 | { | |
2279 | struct io_rings *rings = ctx->rings; | |
2280 | ||
2281 | /* make sure SQ entry isn't read before tail */ | |
2282 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2283 | } | |
2284 | ||
8ff069bf | 2285 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2286 | { |
8ff069bf | 2287 | unsigned int cflags; |
e94f141b | 2288 | |
bcda7baa JA |
2289 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2290 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2291 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2292 | kfree(kbuf); |
2293 | return cflags; | |
e94f141b JA |
2294 | } |
2295 | ||
8ff069bf | 2296 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2297 | { |
4d954c25 | 2298 | struct io_buffer *kbuf; |
bcda7baa | 2299 | |
ae421d93 PB |
2300 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) |
2301 | return 0; | |
4d954c25 | 2302 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2303 | return io_put_kbuf(req, kbuf); |
2304 | } | |
2305 | ||
4c6e277c JA |
2306 | static inline bool io_run_task_work(void) |
2307 | { | |
ef98eb04 | 2308 | if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) { |
4c6e277c | 2309 | __set_current_state(TASK_RUNNING); |
ef98eb04 | 2310 | tracehook_notify_signal(); |
4c6e277c JA |
2311 | return true; |
2312 | } | |
2313 | ||
2314 | return false; | |
bcda7baa JA |
2315 | } |
2316 | ||
def596e9 JA |
2317 | /* |
2318 | * Find and free completed poll iocbs | |
2319 | */ | |
2320 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
a8576af9 | 2321 | struct list_head *done) |
def596e9 | 2322 | { |
8237e045 | 2323 | struct req_batch rb; |
def596e9 | 2324 | struct io_kiocb *req; |
bbde017a XW |
2325 | |
2326 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2327 | smp_rmb(); | |
def596e9 | 2328 | |
5af1d13e | 2329 | io_init_req_batch(&rb); |
def596e9 | 2330 | while (!list_empty(done)) { |
d21ffe7e | 2331 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2332 | list_del(&req->inflight_entry); |
2333 | ||
a8576af9 | 2334 | if (READ_ONCE(req->result) == -EAGAIN && |
8c130827 | 2335 | !(req->flags & REQ_F_DONT_REISSUE)) { |
bbde017a | 2336 | req->iopoll_completed = 0; |
773af691 | 2337 | io_req_task_queue_reissue(req); |
8c130827 | 2338 | continue; |
bbde017a | 2339 | } |
def596e9 | 2340 | |
ae421d93 PB |
2341 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2342 | io_put_rw_kbuf(req)); | |
def596e9 JA |
2343 | (*nr_events)++; |
2344 | ||
de9b4cca | 2345 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2346 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2347 | } |
def596e9 | 2348 | |
09bb8394 | 2349 | io_commit_cqring(ctx); |
80c18e4a | 2350 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2351 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2352 | } |
2353 | ||
def596e9 | 2354 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
a8576af9 | 2355 | long min) |
def596e9 JA |
2356 | { |
2357 | struct io_kiocb *req, *tmp; | |
2358 | LIST_HEAD(done); | |
2359 | bool spin; | |
def596e9 JA |
2360 | |
2361 | /* | |
2362 | * Only spin for completions if we don't have multiple devices hanging | |
2363 | * off our complete list, and we're under the requested amount. | |
2364 | */ | |
915b3dde | 2365 | spin = !ctx->poll_multi_queue && *nr_events < min; |
def596e9 | 2366 | |
d21ffe7e | 2367 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2368 | struct kiocb *kiocb = &req->rw.kiocb; |
a2416e1e | 2369 | int ret; |
def596e9 JA |
2370 | |
2371 | /* | |
581f9810 BM |
2372 | * Move completed and retryable entries to our local lists. |
2373 | * If we find a request that requires polling, break out | |
2374 | * and complete those lists first, if we have entries there. | |
def596e9 | 2375 | */ |
65a6543d | 2376 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2377 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2378 | continue; |
2379 | } | |
2380 | if (!list_empty(&done)) | |
2381 | break; | |
2382 | ||
2383 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
a2416e1e PB |
2384 | if (unlikely(ret < 0)) |
2385 | return ret; | |
2386 | else if (ret) | |
2387 | spin = false; | |
def596e9 | 2388 | |
3aadc23e PB |
2389 | /* iopoll may have completed current req */ |
2390 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2391 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2392 | } |
2393 | ||
2394 | if (!list_empty(&done)) | |
a8576af9 | 2395 | io_iopoll_complete(ctx, nr_events, &done); |
def596e9 | 2396 | |
a2416e1e | 2397 | return 0; |
def596e9 JA |
2398 | } |
2399 | ||
def596e9 JA |
2400 | /* |
2401 | * We can't just wait for polled events to come to us, we have to actively | |
2402 | * find and complete them. | |
2403 | */ | |
b2edc0a7 | 2404 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2405 | { |
2406 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2407 | return; | |
2408 | ||
2409 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2410 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2411 | unsigned int nr_events = 0; |
2412 | ||
a8576af9 | 2413 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2414 | |
b2edc0a7 PB |
2415 | /* let it sleep and repeat later if can't complete a request */ |
2416 | if (nr_events == 0) | |
2417 | break; | |
08f5439f JA |
2418 | /* |
2419 | * Ensure we allow local-to-the-cpu processing to take place, | |
2420 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2421 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2422 | */ |
3fcee5a6 PB |
2423 | if (need_resched()) { |
2424 | mutex_unlock(&ctx->uring_lock); | |
2425 | cond_resched(); | |
2426 | mutex_lock(&ctx->uring_lock); | |
2427 | } | |
def596e9 JA |
2428 | } |
2429 | mutex_unlock(&ctx->uring_lock); | |
2430 | } | |
2431 | ||
7668b92a | 2432 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2433 | { |
7668b92a | 2434 | unsigned int nr_events = 0; |
e9979b36 | 2435 | int ret = 0; |
500f9fba | 2436 | |
c7849be9 XW |
2437 | /* |
2438 | * We disallow the app entering submit/complete with polling, but we | |
2439 | * still need to lock the ring to prevent racing with polled issue | |
2440 | * that got punted to a workqueue. | |
2441 | */ | |
2442 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2443 | /* |
2444 | * Don't enter poll loop if we already have events pending. | |
2445 | * If we do, we can potentially be spinning for commands that | |
2446 | * already triggered a CQE (eg in error). | |
2447 | */ | |
5ed7a37d | 2448 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2449 | __io_cqring_overflow_flush(ctx, false); |
2450 | if (io_cqring_events(ctx)) | |
2451 | goto out; | |
def596e9 | 2452 | do { |
500f9fba JA |
2453 | /* |
2454 | * If a submit got punted to a workqueue, we can have the | |
2455 | * application entering polling for a command before it gets | |
2456 | * issued. That app will hold the uring_lock for the duration | |
2457 | * of the poll right here, so we need to take a breather every | |
2458 | * now and then to ensure that the issue has a chance to add | |
2459 | * the poll to the issued list. Otherwise we can spin here | |
2460 | * forever, while the workqueue is stuck trying to acquire the | |
2461 | * very same mutex. | |
2462 | */ | |
e9979b36 | 2463 | if (list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
2464 | u32 tail = ctx->cached_cq_tail; |
2465 | ||
500f9fba | 2466 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2467 | io_run_task_work(); |
500f9fba | 2468 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2469 | |
8f487ef2 PB |
2470 | /* some requests don't go through iopoll_list */ |
2471 | if (tail != ctx->cached_cq_tail || | |
2472 | list_empty(&ctx->iopoll_list)) | |
e9979b36 | 2473 | break; |
500f9fba | 2474 | } |
a8576af9 | 2475 | ret = io_do_iopoll(ctx, &nr_events, min); |
f39c8a5b PB |
2476 | } while (!ret && nr_events < min && !need_resched()); |
2477 | out: | |
500f9fba | 2478 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2479 | return ret; |
2480 | } | |
2481 | ||
491381ce | 2482 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2483 | { |
491381ce JA |
2484 | /* |
2485 | * Tell lockdep we inherited freeze protection from submission | |
2486 | * thread. | |
2487 | */ | |
2488 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2489 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2490 | |
1c98679d PB |
2491 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2492 | sb_end_write(sb); | |
2b188cc1 JA |
2493 | } |
2494 | } | |
2495 | ||
b63534c4 | 2496 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2497 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2498 | { |
ab454438 | 2499 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2500 | |
ab454438 PB |
2501 | if (!rw) |
2502 | return !io_req_prep_async(req); | |
2503 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2504 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2505 | return true; | |
b63534c4 | 2506 | } |
b63534c4 | 2507 | |
3e6a0d3c | 2508 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2509 | { |
355afaeb | 2510 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2511 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2512 | |
355afaeb JA |
2513 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2514 | return false; | |
3e6a0d3c JA |
2515 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2516 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2517 | return false; |
7c977a58 JA |
2518 | /* |
2519 | * If ref is dying, we might be running poll reap from the exit work. | |
2520 | * Don't attempt to reissue from that path, just let it fail with | |
2521 | * -EAGAIN. | |
2522 | */ | |
3e6a0d3c JA |
2523 | if (percpu_ref_is_dying(&ctx->refs)) |
2524 | return false; | |
ef046888 JA |
2525 | /* |
2526 | * Play it safe and assume not safe to re-import and reissue if we're | |
2527 | * not in the original thread group (or in task context). | |
2528 | */ | |
2529 | if (!same_thread_group(req->task, current) || !in_task()) | |
2530 | return false; | |
3e6a0d3c JA |
2531 | return true; |
2532 | } | |
e82ad485 | 2533 | #else |
a1ff1e3f | 2534 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2535 | { |
2536 | return false; | |
2537 | } | |
e82ad485 | 2538 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2539 | { |
b63534c4 JA |
2540 | return false; |
2541 | } | |
3e6a0d3c | 2542 | #endif |
b63534c4 | 2543 | |
8ef12efe | 2544 | static bool __io_complete_rw_common(struct io_kiocb *req, long res) |
a1d7c393 | 2545 | { |
b65c128f PB |
2546 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2547 | kiocb_end_write(req); | |
9532b99b PB |
2548 | if (res != req->result) { |
2549 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2550 | io_rw_should_reissue(req)) { | |
2551 | req->flags |= REQ_F_REISSUE; | |
8ef12efe | 2552 | return true; |
9532b99b | 2553 | } |
93d2bcd2 | 2554 | req_set_fail(req); |
8ef12efe | 2555 | req->result = res; |
9532b99b | 2556 | } |
8ef12efe JA |
2557 | return false; |
2558 | } | |
2559 | ||
2560 | static void io_req_task_complete(struct io_kiocb *req) | |
2561 | { | |
ae421d93 | 2562 | __io_req_complete(req, 0, req->result, io_put_rw_kbuf(req)); |
8ef12efe JA |
2563 | } |
2564 | ||
2565 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, | |
2566 | unsigned int issue_flags) | |
2567 | { | |
2568 | if (__io_complete_rw_common(req, res)) | |
2569 | return; | |
2570 | io_req_task_complete(req); | |
ba816ad6 JA |
2571 | } |
2572 | ||
2573 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2574 | { | |
9adbd45d | 2575 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2576 | |
8ef12efe JA |
2577 | if (__io_complete_rw_common(req, res)) |
2578 | return; | |
2579 | req->result = res; | |
2580 | req->io_task_work.func = io_req_task_complete; | |
2581 | io_req_task_work_add(req); | |
2b188cc1 JA |
2582 | } |
2583 | ||
def596e9 JA |
2584 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2585 | { | |
9adbd45d | 2586 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2587 | |
491381ce JA |
2588 | if (kiocb->ki_flags & IOCB_WRITE) |
2589 | kiocb_end_write(req); | |
9532b99b | 2590 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2591 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2592 | io_resubmit_prep(req))) { | |
93d2bcd2 | 2593 | req_set_fail(req); |
9532b99b PB |
2594 | req->flags |= REQ_F_DONT_REISSUE; |
2595 | } | |
8c130827 | 2596 | } |
bbde017a XW |
2597 | |
2598 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2599 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2600 | smp_wmb(); |
2601 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2602 | } |
2603 | ||
2604 | /* | |
2605 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2606 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2607 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2608 | * accessing the kiocb cookie. |
2609 | */ | |
cb3d8972 | 2610 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2611 | { |
2612 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2613 | const bool in_async = io_wq_current_is_worker(); |
2614 | ||
2615 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2616 | if (unlikely(in_async)) | |
2617 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2618 | |
2619 | /* | |
2620 | * Track whether we have multiple files in our lists. This will impact | |
2621 | * how we do polling eventually, not spinning if we're on potentially | |
2622 | * different devices. | |
2623 | */ | |
540e32a0 | 2624 | if (list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
2625 | ctx->poll_multi_queue = false; |
2626 | } else if (!ctx->poll_multi_queue) { | |
def596e9 | 2627 | struct io_kiocb *list_req; |
915b3dde | 2628 | unsigned int queue_num0, queue_num1; |
def596e9 | 2629 | |
540e32a0 | 2630 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2631 | inflight_entry); |
915b3dde HX |
2632 | |
2633 | if (list_req->file != req->file) { | |
2634 | ctx->poll_multi_queue = true; | |
2635 | } else { | |
2636 | queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie); | |
2637 | queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie); | |
2638 | if (queue_num0 != queue_num1) | |
2639 | ctx->poll_multi_queue = true; | |
2640 | } | |
def596e9 JA |
2641 | } |
2642 | ||
2643 | /* | |
2644 | * For fast devices, IO may have already completed. If it has, add | |
2645 | * it to the front so we find it first. | |
2646 | */ | |
65a6543d | 2647 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2648 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2649 | else |
d21ffe7e | 2650 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2651 | |
cb3d8972 PB |
2652 | if (unlikely(in_async)) { |
2653 | /* | |
2654 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2655 | * in sq thread task context or in io worker task context. If | |
2656 | * current task context is sq thread, we don't need to check | |
2657 | * whether should wake up sq thread. | |
2658 | */ | |
2659 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2660 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2661 | wake_up(&ctx->sq_data->wait); | |
2662 | ||
2663 | mutex_unlock(&ctx->uring_lock); | |
2664 | } | |
def596e9 JA |
2665 | } |
2666 | ||
4503b767 JA |
2667 | static bool io_bdev_nowait(struct block_device *bdev) |
2668 | { | |
9ba0d0c8 | 2669 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2670 | } |
2671 | ||
2b188cc1 JA |
2672 | /* |
2673 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2674 | * any file. For now, just ensure that anything potentially problematic is done | |
2675 | * inline. | |
2676 | */ | |
b191e2df | 2677 | static bool __io_file_supports_nowait(struct file *file, int rw) |
2b188cc1 JA |
2678 | { |
2679 | umode_t mode = file_inode(file)->i_mode; | |
2680 | ||
4503b767 | 2681 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2682 | if (IS_ENABLED(CONFIG_BLOCK) && |
2683 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2684 | return true; |
2685 | return false; | |
2686 | } | |
976517f1 | 2687 | if (S_ISSOCK(mode)) |
2b188cc1 | 2688 | return true; |
4503b767 | 2689 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2690 | if (IS_ENABLED(CONFIG_BLOCK) && |
2691 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2692 | file->f_op != &io_uring_fops) |
2693 | return true; | |
2694 | return false; | |
2695 | } | |
2b188cc1 | 2696 | |
c5b85625 JA |
2697 | /* any ->read/write should understand O_NONBLOCK */ |
2698 | if (file->f_flags & O_NONBLOCK) | |
2699 | return true; | |
2700 | ||
af197f50 JA |
2701 | if (!(file->f_mode & FMODE_NOWAIT)) |
2702 | return false; | |
2703 | ||
2704 | if (rw == READ) | |
2705 | return file->f_op->read_iter != NULL; | |
2706 | ||
2707 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2708 | } |
2709 | ||
b191e2df | 2710 | static bool io_file_supports_nowait(struct io_kiocb *req, int rw) |
7b29f92d | 2711 | { |
b191e2df | 2712 | if (rw == READ && (req->flags & REQ_F_NOWAIT_READ)) |
7b29f92d | 2713 | return true; |
b191e2df | 2714 | else if (rw == WRITE && (req->flags & REQ_F_NOWAIT_WRITE)) |
7b29f92d JA |
2715 | return true; |
2716 | ||
b191e2df | 2717 | return __io_file_supports_nowait(req->file, rw); |
7b29f92d JA |
2718 | } |
2719 | ||
a88fc400 | 2720 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2721 | { |
def596e9 | 2722 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2723 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2724 | struct file *file = req->file; |
09bb8394 JA |
2725 | unsigned ioprio; |
2726 | int ret; | |
2b188cc1 | 2727 | |
c97d8a0f | 2728 | if (!io_req_ffs_set(req) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2729 | req->flags |= REQ_F_ISREG; |
2730 | ||
2b188cc1 | 2731 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2732 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2733 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2734 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2735 | } |
2b188cc1 | 2736 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2737 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2738 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2739 | if (unlikely(ret)) | |
2740 | return ret; | |
2b188cc1 | 2741 | |
75c668cd PB |
2742 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2743 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2744 | req->flags |= REQ_F_NOWAIT; | |
2745 | ||
2b188cc1 JA |
2746 | ioprio = READ_ONCE(sqe->ioprio); |
2747 | if (ioprio) { | |
2748 | ret = ioprio_check_cap(ioprio); | |
2749 | if (ret) | |
09bb8394 | 2750 | return ret; |
2b188cc1 JA |
2751 | |
2752 | kiocb->ki_ioprio = ioprio; | |
2753 | } else | |
2754 | kiocb->ki_ioprio = get_current_ioprio(); | |
2755 | ||
def596e9 | 2756 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2757 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2758 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2759 | return -EOPNOTSUPP; |
2b188cc1 | 2760 | |
def596e9 JA |
2761 | kiocb->ki_flags |= IOCB_HIPRI; |
2762 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2763 | req->iopoll_completed = 0; |
def596e9 | 2764 | } else { |
09bb8394 JA |
2765 | if (kiocb->ki_flags & IOCB_HIPRI) |
2766 | return -EINVAL; | |
def596e9 JA |
2767 | kiocb->ki_complete = io_complete_rw; |
2768 | } | |
9adbd45d | 2769 | |
eae071c9 PB |
2770 | if (req->opcode == IORING_OP_READ_FIXED || |
2771 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2772 | req->imu = NULL; | |
2773 | io_req_set_rsrc_node(req); | |
2774 | } | |
2775 | ||
3529d8c2 JA |
2776 | req->rw.addr = READ_ONCE(sqe->addr); |
2777 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2778 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2779 | return 0; |
2b188cc1 JA |
2780 | } |
2781 | ||
2782 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2783 | { | |
2784 | switch (ret) { | |
2785 | case -EIOCBQUEUED: | |
2786 | break; | |
2787 | case -ERESTARTSYS: | |
2788 | case -ERESTARTNOINTR: | |
2789 | case -ERESTARTNOHAND: | |
2790 | case -ERESTART_RESTARTBLOCK: | |
2791 | /* | |
2792 | * We can't just restart the syscall, since previously | |
2793 | * submitted sqes may already be in progress. Just fail this | |
2794 | * IO with EINTR. | |
2795 | */ | |
2796 | ret = -EINTR; | |
df561f66 | 2797 | fallthrough; |
2b188cc1 JA |
2798 | default: |
2799 | kiocb->ki_complete(kiocb, ret, 0); | |
2800 | } | |
2801 | } | |
2802 | ||
a1d7c393 | 2803 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2804 | unsigned int issue_flags) |
ba816ad6 | 2805 | { |
ba04291e | 2806 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2807 | struct io_async_rw *io = req->async_data; |
97284637 | 2808 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2809 | |
227c0c96 | 2810 | /* add previously done IO, if any */ |
e8c2bc1f | 2811 | if (io && io->bytes_done > 0) { |
227c0c96 | 2812 | if (ret < 0) |
e8c2bc1f | 2813 | ret = io->bytes_done; |
227c0c96 | 2814 | else |
e8c2bc1f | 2815 | ret += io->bytes_done; |
227c0c96 JA |
2816 | } |
2817 | ||
ba04291e JA |
2818 | if (req->flags & REQ_F_CUR_POS) |
2819 | req->file->f_pos = kiocb->ki_pos; | |
e149bd74 | 2820 | if (ret >= 0 && check_reissue) |
889fca73 | 2821 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2822 | else |
2823 | io_rw_done(kiocb, ret); | |
97284637 | 2824 | |
fe7e3257 | 2825 | if (check_reissue && (req->flags & REQ_F_REISSUE)) { |
97284637 | 2826 | req->flags &= ~REQ_F_REISSUE; |
a7be7c23 | 2827 | if (io_resubmit_prep(req)) { |
773af691 | 2828 | io_req_task_queue_reissue(req); |
8c130827 | 2829 | } else { |
93d2bcd2 | 2830 | req_set_fail(req); |
ae421d93 PB |
2831 | __io_req_complete(req, issue_flags, ret, |
2832 | io_put_rw_kbuf(req)); | |
97284637 PB |
2833 | } |
2834 | } | |
ba816ad6 JA |
2835 | } |
2836 | ||
eae071c9 PB |
2837 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2838 | struct io_mapped_ubuf *imu) | |
edafccee | 2839 | { |
9adbd45d | 2840 | size_t len = req->rw.len; |
75769e3f | 2841 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2842 | size_t offset; |
edafccee | 2843 | |
75769e3f | 2844 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2845 | return -EFAULT; |
2846 | /* not inside the mapped region */ | |
4751f53d | 2847 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2848 | return -EFAULT; |
2849 | ||
2850 | /* | |
2851 | * May not be a start of buffer, set size appropriately | |
2852 | * and advance us to the beginning. | |
2853 | */ | |
2854 | offset = buf_addr - imu->ubuf; | |
2855 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2856 | |
2857 | if (offset) { | |
2858 | /* | |
2859 | * Don't use iov_iter_advance() here, as it's really slow for | |
2860 | * using the latter parts of a big fixed buffer - it iterates | |
2861 | * over each segment manually. We can cheat a bit here, because | |
2862 | * we know that: | |
2863 | * | |
2864 | * 1) it's a BVEC iter, we set it up | |
2865 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2866 | * first and last bvec | |
2867 | * | |
2868 | * So just find our index, and adjust the iterator afterwards. | |
2869 | * If the offset is within the first bvec (or the whole first | |
2870 | * bvec, just use iov_iter_advance(). This makes it easier | |
2871 | * since we can just skip the first segment, which may not | |
2872 | * be PAGE_SIZE aligned. | |
2873 | */ | |
2874 | const struct bio_vec *bvec = imu->bvec; | |
2875 | ||
2876 | if (offset <= bvec->bv_len) { | |
2877 | iov_iter_advance(iter, offset); | |
2878 | } else { | |
2879 | unsigned long seg_skip; | |
2880 | ||
2881 | /* skip first vec */ | |
2882 | offset -= bvec->bv_len; | |
2883 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2884 | ||
2885 | iter->bvec = bvec + seg_skip; | |
2886 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2887 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2888 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2889 | } |
2890 | } | |
2891 | ||
847595de | 2892 | return 0; |
edafccee JA |
2893 | } |
2894 | ||
eae071c9 PB |
2895 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
2896 | { | |
2897 | struct io_ring_ctx *ctx = req->ctx; | |
2898 | struct io_mapped_ubuf *imu = req->imu; | |
2899 | u16 index, buf_index = req->buf_index; | |
2900 | ||
2901 | if (likely(!imu)) { | |
2902 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
2903 | return -EFAULT; | |
2904 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2905 | imu = READ_ONCE(ctx->user_bufs[index]); | |
2906 | req->imu = imu; | |
2907 | } | |
2908 | return __io_import_fixed(req, rw, iter, imu); | |
2909 | } | |
2910 | ||
bcda7baa JA |
2911 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2912 | { | |
2913 | if (needs_lock) | |
2914 | mutex_unlock(&ctx->uring_lock); | |
2915 | } | |
2916 | ||
2917 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2918 | { | |
2919 | /* | |
2920 | * "Normal" inline submissions always hold the uring_lock, since we | |
2921 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2922 | * The only exception is when we've detached the request and issue it | |
2923 | * from an async worker thread, grab the lock for that case. | |
2924 | */ | |
2925 | if (needs_lock) | |
2926 | mutex_lock(&ctx->uring_lock); | |
2927 | } | |
2928 | ||
2929 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2930 | int bgid, struct io_buffer *kbuf, | |
2931 | bool needs_lock) | |
2932 | { | |
2933 | struct io_buffer *head; | |
2934 | ||
2935 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2936 | return kbuf; | |
2937 | ||
2938 | io_ring_submit_lock(req->ctx, needs_lock); | |
2939 | ||
2940 | lockdep_assert_held(&req->ctx->uring_lock); | |
2941 | ||
9e15c3a0 | 2942 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2943 | if (head) { |
2944 | if (!list_empty(&head->list)) { | |
2945 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2946 | list); | |
2947 | list_del(&kbuf->list); | |
2948 | } else { | |
2949 | kbuf = head; | |
9e15c3a0 | 2950 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2951 | } |
2952 | if (*len > kbuf->len) | |
2953 | *len = kbuf->len; | |
2954 | } else { | |
2955 | kbuf = ERR_PTR(-ENOBUFS); | |
2956 | } | |
2957 | ||
2958 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2959 | ||
2960 | return kbuf; | |
2961 | } | |
2962 | ||
4d954c25 JA |
2963 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2964 | bool needs_lock) | |
2965 | { | |
2966 | struct io_buffer *kbuf; | |
4f4eeba8 | 2967 | u16 bgid; |
4d954c25 JA |
2968 | |
2969 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2970 | bgid = req->buf_index; |
4d954c25 JA |
2971 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2972 | if (IS_ERR(kbuf)) | |
2973 | return kbuf; | |
2974 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2975 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2976 | return u64_to_user_ptr(kbuf->addr); | |
2977 | } | |
2978 | ||
2979 | #ifdef CONFIG_COMPAT | |
2980 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2981 | bool needs_lock) | |
2982 | { | |
2983 | struct compat_iovec __user *uiov; | |
2984 | compat_ssize_t clen; | |
2985 | void __user *buf; | |
2986 | ssize_t len; | |
2987 | ||
2988 | uiov = u64_to_user_ptr(req->rw.addr); | |
2989 | if (!access_ok(uiov, sizeof(*uiov))) | |
2990 | return -EFAULT; | |
2991 | if (__get_user(clen, &uiov->iov_len)) | |
2992 | return -EFAULT; | |
2993 | if (clen < 0) | |
2994 | return -EINVAL; | |
2995 | ||
2996 | len = clen; | |
2997 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2998 | if (IS_ERR(buf)) | |
2999 | return PTR_ERR(buf); | |
3000 | iov[0].iov_base = buf; | |
3001 | iov[0].iov_len = (compat_size_t) len; | |
3002 | return 0; | |
3003 | } | |
3004 | #endif | |
3005 | ||
3006 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3007 | bool needs_lock) | |
3008 | { | |
3009 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3010 | void __user *buf; | |
3011 | ssize_t len; | |
3012 | ||
3013 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3014 | return -EFAULT; | |
3015 | ||
3016 | len = iov[0].iov_len; | |
3017 | if (len < 0) | |
3018 | return -EINVAL; | |
3019 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3020 | if (IS_ERR(buf)) | |
3021 | return PTR_ERR(buf); | |
3022 | iov[0].iov_base = buf; | |
3023 | iov[0].iov_len = len; | |
3024 | return 0; | |
3025 | } | |
3026 | ||
3027 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3028 | bool needs_lock) | |
3029 | { | |
dddb3e26 JA |
3030 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3031 | struct io_buffer *kbuf; | |
3032 | ||
3033 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3034 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3035 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3036 | return 0; |
dddb3e26 | 3037 | } |
dd201662 | 3038 | if (req->rw.len != 1) |
4d954c25 JA |
3039 | return -EINVAL; |
3040 | ||
3041 | #ifdef CONFIG_COMPAT | |
3042 | if (req->ctx->compat) | |
3043 | return io_compat_import(req, iov, needs_lock); | |
3044 | #endif | |
3045 | ||
3046 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3047 | } | |
3048 | ||
847595de PB |
3049 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3050 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3051 | { |
9adbd45d JA |
3052 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3053 | size_t sqe_len = req->rw.len; | |
847595de | 3054 | u8 opcode = req->opcode; |
4d954c25 | 3055 | ssize_t ret; |
edafccee | 3056 | |
7d009165 | 3057 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3058 | *iovec = NULL; |
9adbd45d | 3059 | return io_import_fixed(req, rw, iter); |
edafccee | 3060 | } |
2b188cc1 | 3061 | |
bcda7baa | 3062 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3063 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3064 | return -EINVAL; |
3065 | ||
3a6820f2 | 3066 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3067 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3068 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3069 | if (IS_ERR(buf)) |
4d954c25 | 3070 | return PTR_ERR(buf); |
3f9d6441 | 3071 | req->rw.len = sqe_len; |
bcda7baa JA |
3072 | } |
3073 | ||
3a6820f2 JA |
3074 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3075 | *iovec = NULL; | |
10fc72e4 | 3076 | return ret; |
3a6820f2 JA |
3077 | } |
3078 | ||
4d954c25 JA |
3079 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3080 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3081 | if (!ret) |
3082 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3083 | *iovec = NULL; |
3084 | return ret; | |
3085 | } | |
3086 | ||
89cd35c5 CH |
3087 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3088 | req->ctx->compat); | |
2b188cc1 JA |
3089 | } |
3090 | ||
0fef9483 JA |
3091 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3092 | { | |
5b09e37e | 3093 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3094 | } |
3095 | ||
31b51510 | 3096 | /* |
32960613 JA |
3097 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3098 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3099 | */ |
4017eb91 | 3100 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3101 | { |
4017eb91 JA |
3102 | struct kiocb *kiocb = &req->rw.kiocb; |
3103 | struct file *file = req->file; | |
32960613 JA |
3104 | ssize_t ret = 0; |
3105 | ||
3106 | /* | |
3107 | * Don't support polled IO through this interface, and we can't | |
3108 | * support non-blocking either. For the latter, this just causes | |
3109 | * the kiocb to be handled from an async context. | |
3110 | */ | |
3111 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3112 | return -EOPNOTSUPP; | |
3113 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3114 | return -EAGAIN; | |
3115 | ||
3116 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3117 | struct iovec iovec; |
32960613 JA |
3118 | ssize_t nr; |
3119 | ||
311ae9e1 PB |
3120 | if (!iov_iter_is_bvec(iter)) { |
3121 | iovec = iov_iter_iovec(iter); | |
3122 | } else { | |
4017eb91 JA |
3123 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3124 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3125 | } |
3126 | ||
32960613 JA |
3127 | if (rw == READ) { |
3128 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3129 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3130 | } else { |
3131 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3132 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3133 | } |
3134 | ||
3135 | if (nr < 0) { | |
3136 | if (!ret) | |
3137 | ret = nr; | |
3138 | break; | |
3139 | } | |
3140 | ret += nr; | |
3141 | if (nr != iovec.iov_len) | |
3142 | break; | |
4017eb91 JA |
3143 | req->rw.len -= nr; |
3144 | req->rw.addr += nr; | |
32960613 JA |
3145 | iov_iter_advance(iter, nr); |
3146 | } | |
3147 | ||
3148 | return ret; | |
3149 | } | |
3150 | ||
ff6165b2 JA |
3151 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3152 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3153 | { |
e8c2bc1f | 3154 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3155 | |
ff6165b2 | 3156 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3157 | rw->free_iovec = iovec; |
227c0c96 | 3158 | rw->bytes_done = 0; |
ff6165b2 | 3159 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3160 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3161 | return; |
b64e3444 | 3162 | if (!iovec) { |
ff6165b2 JA |
3163 | unsigned iov_off = 0; |
3164 | ||
3165 | rw->iter.iov = rw->fast_iov; | |
3166 | if (iter->iov != fast_iov) { | |
3167 | iov_off = iter->iov - fast_iov; | |
3168 | rw->iter.iov += iov_off; | |
3169 | } | |
3170 | if (rw->fast_iov != fast_iov) | |
3171 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3172 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3173 | } else { |
3174 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3175 | } |
3176 | } | |
3177 | ||
6cb78689 | 3178 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3179 | { |
e8c2bc1f JA |
3180 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3181 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3182 | return req->async_data == NULL; | |
3d9932a8 XW |
3183 | } |
3184 | ||
ff6165b2 JA |
3185 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3186 | const struct iovec *fast_iov, | |
227c0c96 | 3187 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3188 | { |
26f0505a | 3189 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3190 | return 0; |
e8c2bc1f | 3191 | if (!req->async_data) { |
6cb78689 | 3192 | if (io_alloc_async_data(req)) { |
6bf985dc | 3193 | kfree(iovec); |
5d204bcf | 3194 | return -ENOMEM; |
6bf985dc | 3195 | } |
b7bb4f7d | 3196 | |
ff6165b2 | 3197 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3198 | } |
b7bb4f7d | 3199 | return 0; |
f67676d1 JA |
3200 | } |
3201 | ||
73debe68 | 3202 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3203 | { |
e8c2bc1f | 3204 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3205 | struct iovec *iov = iorw->fast_iov; |
847595de | 3206 | int ret; |
c3e330a4 | 3207 | |
2846c481 | 3208 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3209 | if (unlikely(ret < 0)) |
3210 | return ret; | |
3211 | ||
ab0b196c PB |
3212 | iorw->bytes_done = 0; |
3213 | iorw->free_iovec = iov; | |
3214 | if (iov) | |
3215 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3216 | return 0; |
3217 | } | |
3218 | ||
73debe68 | 3219 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3220 | { |
3529d8c2 JA |
3221 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3222 | return -EBADF; | |
93642ef8 | 3223 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3224 | } |
3225 | ||
c1dd91d1 JA |
3226 | /* |
3227 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3228 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3229 | * This gets called when the page is unlocked, and we generally expect that to | |
3230 | * happen when the page IO is completed and the page is now uptodate. This will | |
3231 | * queue a task_work based retry of the operation, attempting to copy the data | |
3232 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3233 | * do a thread based blocking retry of the operation. That's the unexpected | |
3234 | * slow path. | |
3235 | */ | |
bcf5a063 JA |
3236 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3237 | int sync, void *arg) | |
3238 | { | |
3239 | struct wait_page_queue *wpq; | |
3240 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3241 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3242 | |
3243 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3244 | ||
cdc8fcb4 LT |
3245 | if (!wake_page_match(wpq, key)) |
3246 | return 0; | |
3247 | ||
c8d317aa | 3248 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 | 3249 | list_del_init(&wait->entry); |
921b9054 | 3250 | io_req_task_queue(req); |
bcf5a063 JA |
3251 | return 1; |
3252 | } | |
3253 | ||
c1dd91d1 JA |
3254 | /* |
3255 | * This controls whether a given IO request should be armed for async page | |
3256 | * based retry. If we return false here, the request is handed to the async | |
3257 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3258 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3259 | * will either succeed because the page is now uptodate and unlocked, or it | |
3260 | * will register a callback when the page is unlocked at IO completion. Through | |
3261 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3262 | * That retry will attempt the buffered read again. The retry will generally | |
3263 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3264 | * async worker threads for a blocking retry. | |
3265 | */ | |
227c0c96 | 3266 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3267 | { |
e8c2bc1f JA |
3268 | struct io_async_rw *rw = req->async_data; |
3269 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3270 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3271 | |
bcf5a063 JA |
3272 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3273 | if (req->flags & REQ_F_NOWAIT) | |
3274 | return false; | |
f67676d1 | 3275 | |
227c0c96 | 3276 | /* Only for buffered IO */ |
3b2a4439 | 3277 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3278 | return false; |
3b2a4439 | 3279 | |
bcf5a063 JA |
3280 | /* |
3281 | * just use poll if we can, and don't attempt if the fs doesn't | |
3282 | * support callback based unlocks | |
3283 | */ | |
3284 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3285 | return false; | |
f67676d1 | 3286 | |
3b2a4439 JA |
3287 | wait->wait.func = io_async_buf_func; |
3288 | wait->wait.private = req; | |
3289 | wait->wait.flags = 0; | |
3290 | INIT_LIST_HEAD(&wait->wait.entry); | |
3291 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3292 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3293 | kiocb->ki_waitq = wait; |
3b2a4439 | 3294 | return true; |
bcf5a063 JA |
3295 | } |
3296 | ||
aeab9506 | 3297 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3298 | { |
3299 | if (req->file->f_op->read_iter) | |
3300 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3301 | else if (req->file->f_op->read) |
4017eb91 | 3302 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3303 | else |
3304 | return -EINVAL; | |
f67676d1 JA |
3305 | } |
3306 | ||
889fca73 | 3307 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3308 | { |
3309 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3310 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3311 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3312 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3313 | ssize_t io_size, ret, ret2; |
45d189c6 | 3314 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3315 | |
2846c481 | 3316 | if (rw) { |
e8c2bc1f | 3317 | iter = &rw->iter; |
2846c481 PB |
3318 | iovec = NULL; |
3319 | } else { | |
3320 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3321 | if (ret < 0) | |
3322 | return ret; | |
3323 | } | |
632546c4 | 3324 | io_size = iov_iter_count(iter); |
fa15bafb | 3325 | req->result = io_size; |
2b188cc1 | 3326 | |
fd6c2e4c JA |
3327 | /* Ensure we clear previously set non-block flag */ |
3328 | if (!force_nonblock) | |
29de5f6a | 3329 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3330 | else |
3331 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3332 | ||
24c74678 | 3333 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3334 | if (force_nonblock && !io_file_supports_nowait(req, READ)) { |
6713e7a6 | 3335 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3336 | return ret ?: -EAGAIN; |
6713e7a6 | 3337 | } |
9e645e11 | 3338 | |
632546c4 | 3339 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3340 | if (unlikely(ret)) { |
3341 | kfree(iovec); | |
3342 | return ret; | |
3343 | } | |
2b188cc1 | 3344 | |
227c0c96 | 3345 | ret = io_iter_do_read(req, iter); |
32960613 | 3346 | |
230d50d4 | 3347 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3348 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3349 | /* IOPOLL retry should happen for io-wq threads */ |
3350 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3351 | goto done; |
75c668cd PB |
3352 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3353 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3354 | goto done; |
84216315 | 3355 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3356 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3357 | ret = 0; |
230d50d4 JA |
3358 | } else if (ret == -EIOCBQUEUED) { |
3359 | goto out_free; | |
7335e3bf | 3360 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3361 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3362 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3363 | goto done; |
227c0c96 JA |
3364 | } |
3365 | ||
227c0c96 | 3366 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3367 | if (ret2) |
3368 | return ret2; | |
3369 | ||
fe1cdd55 | 3370 | iovec = NULL; |
e8c2bc1f | 3371 | rw = req->async_data; |
227c0c96 | 3372 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3373 | iter = &rw->iter; |
227c0c96 | 3374 | |
b23df91b PB |
3375 | do { |
3376 | io_size -= ret; | |
3377 | rw->bytes_done += ret; | |
3378 | /* if we can retry, do so with the callbacks armed */ | |
3379 | if (!io_rw_should_retry(req)) { | |
3380 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3381 | return -EAGAIN; | |
3382 | } | |
3383 | ||
3384 | /* | |
3385 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3386 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3387 | * desired page gets unlocked. We can also get a partial read | |
3388 | * here, and if we do, then just retry at the new offset. | |
3389 | */ | |
3390 | ret = io_iter_do_read(req, iter); | |
3391 | if (ret == -EIOCBQUEUED) | |
3392 | return 0; | |
227c0c96 | 3393 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3394 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3395 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3396 | done: |
889fca73 | 3397 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3398 | out_free: |
3399 | /* it's faster to check here then delegate to kfree */ | |
3400 | if (iovec) | |
3401 | kfree(iovec); | |
5ea5dd45 | 3402 | return 0; |
2b188cc1 JA |
3403 | } |
3404 | ||
73debe68 | 3405 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3406 | { |
3529d8c2 JA |
3407 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3408 | return -EBADF; | |
93642ef8 | 3409 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3410 | } |
3411 | ||
889fca73 | 3412 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3413 | { |
3414 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3415 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3416 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3417 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3418 | ssize_t ret, ret2, io_size; |
45d189c6 | 3419 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3420 | |
2846c481 | 3421 | if (rw) { |
e8c2bc1f | 3422 | iter = &rw->iter; |
2846c481 PB |
3423 | iovec = NULL; |
3424 | } else { | |
3425 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3426 | if (ret < 0) | |
3427 | return ret; | |
3428 | } | |
632546c4 | 3429 | io_size = iov_iter_count(iter); |
fa15bafb | 3430 | req->result = io_size; |
2b188cc1 | 3431 | |
fd6c2e4c JA |
3432 | /* Ensure we clear previously set non-block flag */ |
3433 | if (!force_nonblock) | |
a88fc400 PB |
3434 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3435 | else | |
3436 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3437 | |
24c74678 | 3438 | /* If the file doesn't support async, just async punt */ |
b191e2df | 3439 | if (force_nonblock && !io_file_supports_nowait(req, WRITE)) |
f67676d1 | 3440 | goto copy_iov; |
31b51510 | 3441 | |
10d59345 JA |
3442 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3443 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3444 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3445 | goto copy_iov; |
31b51510 | 3446 | |
632546c4 | 3447 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3448 | if (unlikely(ret)) |
3449 | goto out_free; | |
4ed734b0 | 3450 | |
fa15bafb PB |
3451 | /* |
3452 | * Open-code file_start_write here to grab freeze protection, | |
3453 | * which will be released by another thread in | |
3454 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3455 | * released so that it doesn't complain about the held lock when | |
3456 | * we return to userspace. | |
3457 | */ | |
3458 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3459 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3460 | __sb_writers_release(file_inode(req->file)->i_sb, |
3461 | SB_FREEZE_WRITE); | |
3462 | } | |
3463 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3464 | |
fa15bafb | 3465 | if (req->file->f_op->write_iter) |
ff6165b2 | 3466 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3467 | else if (req->file->f_op->write) |
4017eb91 | 3468 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3469 | else |
3470 | ret2 = -EINVAL; | |
4ed734b0 | 3471 | |
6ad7f233 PB |
3472 | if (req->flags & REQ_F_REISSUE) { |
3473 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3474 | ret2 = -EAGAIN; |
6ad7f233 | 3475 | } |
230d50d4 | 3476 | |
fa15bafb PB |
3477 | /* |
3478 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3479 | * retry them without IOCB_NOWAIT. | |
3480 | */ | |
3481 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3482 | ret2 = -EAGAIN; | |
75c668cd PB |
3483 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3484 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3485 | goto done; |
fa15bafb | 3486 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3487 | /* IOPOLL retry should happen for io-wq threads */ |
3488 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3489 | goto copy_iov; | |
355afaeb | 3490 | done: |
889fca73 | 3491 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3492 | } else { |
f67676d1 | 3493 | copy_iov: |
84216315 | 3494 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3495 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3496 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3497 | return ret ?: -EAGAIN; |
2b188cc1 | 3498 | } |
31b51510 | 3499 | out_free: |
f261c168 | 3500 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3501 | if (iovec) |
6f2cc166 | 3502 | kfree(iovec); |
2b188cc1 JA |
3503 | return ret; |
3504 | } | |
3505 | ||
80a261fd JA |
3506 | static int io_renameat_prep(struct io_kiocb *req, |
3507 | const struct io_uring_sqe *sqe) | |
3508 | { | |
3509 | struct io_rename *ren = &req->rename; | |
3510 | const char __user *oldf, *newf; | |
3511 | ||
ed7eb259 JA |
3512 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3513 | return -EINVAL; | |
26578cda | 3514 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
ed7eb259 | 3515 | return -EINVAL; |
80a261fd JA |
3516 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3517 | return -EBADF; | |
3518 | ||
3519 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3520 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3521 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3522 | ren->new_dfd = READ_ONCE(sqe->len); | |
3523 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3524 | ||
3525 | ren->oldpath = getname(oldf); | |
3526 | if (IS_ERR(ren->oldpath)) | |
3527 | return PTR_ERR(ren->oldpath); | |
3528 | ||
3529 | ren->newpath = getname(newf); | |
3530 | if (IS_ERR(ren->newpath)) { | |
3531 | putname(ren->oldpath); | |
3532 | return PTR_ERR(ren->newpath); | |
3533 | } | |
3534 | ||
3535 | req->flags |= REQ_F_NEED_CLEANUP; | |
3536 | return 0; | |
3537 | } | |
3538 | ||
45d189c6 | 3539 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3540 | { |
3541 | struct io_rename *ren = &req->rename; | |
3542 | int ret; | |
3543 | ||
45d189c6 | 3544 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3545 | return -EAGAIN; |
3546 | ||
3547 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3548 | ren->newpath, ren->flags); | |
3549 | ||
3550 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3551 | if (ret < 0) | |
93d2bcd2 | 3552 | req_set_fail(req); |
80a261fd JA |
3553 | io_req_complete(req, ret); |
3554 | return 0; | |
3555 | } | |
3556 | ||
14a1143b JA |
3557 | static int io_unlinkat_prep(struct io_kiocb *req, |
3558 | const struct io_uring_sqe *sqe) | |
3559 | { | |
3560 | struct io_unlink *un = &req->unlink; | |
3561 | const char __user *fname; | |
3562 | ||
22634bc5 JA |
3563 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3564 | return -EINVAL; | |
26578cda PB |
3565 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || |
3566 | sqe->splice_fd_in) | |
22634bc5 | 3567 | return -EINVAL; |
14a1143b JA |
3568 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3569 | return -EBADF; | |
3570 | ||
3571 | un->dfd = READ_ONCE(sqe->fd); | |
3572 | ||
3573 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3574 | if (un->flags & ~AT_REMOVEDIR) | |
3575 | return -EINVAL; | |
3576 | ||
3577 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3578 | un->filename = getname(fname); | |
3579 | if (IS_ERR(un->filename)) | |
3580 | return PTR_ERR(un->filename); | |
3581 | ||
3582 | req->flags |= REQ_F_NEED_CLEANUP; | |
3583 | return 0; | |
3584 | } | |
3585 | ||
45d189c6 | 3586 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3587 | { |
3588 | struct io_unlink *un = &req->unlink; | |
3589 | int ret; | |
3590 | ||
45d189c6 | 3591 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3592 | return -EAGAIN; |
3593 | ||
3594 | if (un->flags & AT_REMOVEDIR) | |
3595 | ret = do_rmdir(un->dfd, un->filename); | |
3596 | else | |
3597 | ret = do_unlinkat(un->dfd, un->filename); | |
3598 | ||
3599 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3600 | if (ret < 0) | |
93d2bcd2 | 3601 | req_set_fail(req); |
14a1143b JA |
3602 | io_req_complete(req, ret); |
3603 | return 0; | |
3604 | } | |
3605 | ||
36f4fa68 JA |
3606 | static int io_shutdown_prep(struct io_kiocb *req, |
3607 | const struct io_uring_sqe *sqe) | |
3608 | { | |
3609 | #if defined(CONFIG_NET) | |
3610 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3611 | return -EINVAL; | |
26578cda PB |
3612 | if (unlikely(sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || |
3613 | sqe->buf_index || sqe->splice_fd_in)) | |
36f4fa68 JA |
3614 | return -EINVAL; |
3615 | ||
3616 | req->shutdown.how = READ_ONCE(sqe->len); | |
3617 | return 0; | |
3618 | #else | |
3619 | return -EOPNOTSUPP; | |
3620 | #endif | |
3621 | } | |
3622 | ||
45d189c6 | 3623 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3624 | { |
3625 | #if defined(CONFIG_NET) | |
3626 | struct socket *sock; | |
3627 | int ret; | |
3628 | ||
45d189c6 | 3629 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3630 | return -EAGAIN; |
3631 | ||
48aba79b | 3632 | sock = sock_from_file(req->file); |
36f4fa68 | 3633 | if (unlikely(!sock)) |
48aba79b | 3634 | return -ENOTSOCK; |
36f4fa68 JA |
3635 | |
3636 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3637 | if (ret < 0) |
93d2bcd2 | 3638 | req_set_fail(req); |
36f4fa68 JA |
3639 | io_req_complete(req, ret); |
3640 | return 0; | |
3641 | #else | |
3642 | return -EOPNOTSUPP; | |
3643 | #endif | |
3644 | } | |
3645 | ||
f2a8d5c7 PB |
3646 | static int __io_splice_prep(struct io_kiocb *req, |
3647 | const struct io_uring_sqe *sqe) | |
7d67af2c | 3648 | { |
fe7e3257 | 3649 | struct io_splice *sp = &req->splice; |
7d67af2c | 3650 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 3651 | |
3232dd02 PB |
3652 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3653 | return -EINVAL; | |
7d67af2c PB |
3654 | |
3655 | sp->file_in = NULL; | |
7d67af2c PB |
3656 | sp->len = READ_ONCE(sqe->len); |
3657 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3658 | ||
3659 | if (unlikely(sp->flags & ~valid_flags)) | |
3660 | return -EINVAL; | |
3661 | ||
62906e89 | 3662 | sp->file_in = io_file_get(req->ctx, req, READ_ONCE(sqe->splice_fd_in), |
8371adf5 PB |
3663 | (sp->flags & SPLICE_F_FD_IN_FIXED)); |
3664 | if (!sp->file_in) | |
3665 | return -EBADF; | |
7d67af2c | 3666 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3667 | return 0; |
3668 | } | |
3669 | ||
f2a8d5c7 PB |
3670 | static int io_tee_prep(struct io_kiocb *req, |
3671 | const struct io_uring_sqe *sqe) | |
3672 | { | |
3673 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3674 | return -EINVAL; | |
3675 | return __io_splice_prep(req, sqe); | |
3676 | } | |
3677 | ||
45d189c6 | 3678 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3679 | { |
3680 | struct io_splice *sp = &req->splice; | |
3681 | struct file *in = sp->file_in; | |
3682 | struct file *out = sp->file_out; | |
3683 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3684 | long ret = 0; | |
3685 | ||
45d189c6 | 3686 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3687 | return -EAGAIN; |
3688 | if (sp->len) | |
3689 | ret = do_tee(in, out, sp->len, flags); | |
3690 | ||
e1d767f0 PB |
3691 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3692 | io_put_file(in); | |
f2a8d5c7 PB |
3693 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3694 | ||
f2a8d5c7 | 3695 | if (ret != sp->len) |
93d2bcd2 | 3696 | req_set_fail(req); |
e1e16097 | 3697 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3698 | return 0; |
3699 | } | |
3700 | ||
3701 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3702 | { | |
fe7e3257 | 3703 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
3704 | |
3705 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3706 | sp->off_out = READ_ONCE(sqe->off); | |
3707 | return __io_splice_prep(req, sqe); | |
3708 | } | |
3709 | ||
45d189c6 | 3710 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3711 | { |
3712 | struct io_splice *sp = &req->splice; | |
3713 | struct file *in = sp->file_in; | |
3714 | struct file *out = sp->file_out; | |
3715 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3716 | loff_t *poff_in, *poff_out; | |
c9687426 | 3717 | long ret = 0; |
7d67af2c | 3718 | |
45d189c6 | 3719 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3720 | return -EAGAIN; |
7d67af2c PB |
3721 | |
3722 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3723 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3724 | |
948a7749 | 3725 | if (sp->len) |
c9687426 | 3726 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3727 | |
e1d767f0 PB |
3728 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3729 | io_put_file(in); | |
7d67af2c PB |
3730 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3731 | ||
7d67af2c | 3732 | if (ret != sp->len) |
93d2bcd2 | 3733 | req_set_fail(req); |
e1e16097 | 3734 | io_req_complete(req, ret); |
7d67af2c PB |
3735 | return 0; |
3736 | } | |
3737 | ||
2b188cc1 JA |
3738 | /* |
3739 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3740 | */ | |
889fca73 | 3741 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3742 | { |
3743 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3744 | |
def596e9 JA |
3745 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3746 | return -EINVAL; | |
3747 | ||
889fca73 | 3748 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3749 | return 0; |
3750 | } | |
3751 | ||
1155c76a | 3752 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3753 | { |
6b06314c | 3754 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3755 | |
09bb8394 JA |
3756 | if (!req->file) |
3757 | return -EBADF; | |
c992fe29 | 3758 | |
6b06314c | 3759 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3760 | return -EINVAL; |
26578cda PB |
3761 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
3762 | sqe->splice_fd_in)) | |
c992fe29 CH |
3763 | return -EINVAL; |
3764 | ||
8ed8d3c3 JA |
3765 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3766 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3767 | return -EINVAL; | |
3768 | ||
3769 | req->sync.off = READ_ONCE(sqe->off); | |
3770 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3771 | return 0; |
3772 | } | |
3773 | ||
45d189c6 | 3774 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3775 | { |
8ed8d3c3 | 3776 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3777 | int ret; |
3778 | ||
ac45abc0 | 3779 | /* fsync always requires a blocking context */ |
45d189c6 | 3780 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3781 | return -EAGAIN; |
3782 | ||
9adbd45d | 3783 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3784 | end > 0 ? end : LLONG_MAX, |
3785 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3786 | if (ret < 0) | |
93d2bcd2 | 3787 | req_set_fail(req); |
e1e16097 | 3788 | io_req_complete(req, ret); |
c992fe29 CH |
3789 | return 0; |
3790 | } | |
3791 | ||
d63d1b5e JA |
3792 | static int io_fallocate_prep(struct io_kiocb *req, |
3793 | const struct io_uring_sqe *sqe) | |
3794 | { | |
26578cda PB |
3795 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags || |
3796 | sqe->splice_fd_in) | |
d63d1b5e | 3797 | return -EINVAL; |
3232dd02 PB |
3798 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3799 | return -EINVAL; | |
d63d1b5e JA |
3800 | |
3801 | req->sync.off = READ_ONCE(sqe->off); | |
3802 | req->sync.len = READ_ONCE(sqe->addr); | |
3803 | req->sync.mode = READ_ONCE(sqe->len); | |
3804 | return 0; | |
3805 | } | |
3806 | ||
45d189c6 | 3807 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3808 | { |
ac45abc0 PB |
3809 | int ret; |
3810 | ||
d63d1b5e | 3811 | /* fallocate always requiring blocking context */ |
45d189c6 | 3812 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3813 | return -EAGAIN; |
ac45abc0 PB |
3814 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3815 | req->sync.len); | |
ac45abc0 | 3816 | if (ret < 0) |
93d2bcd2 | 3817 | req_set_fail(req); |
e1e16097 | 3818 | io_req_complete(req, ret); |
5d17b4a4 JA |
3819 | return 0; |
3820 | } | |
3821 | ||
ec65fea5 | 3822 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3823 | { |
f8748881 | 3824 | const char __user *fname; |
15b71abe | 3825 | int ret; |
b7bb4f7d | 3826 | |
d3fddf6d PB |
3827 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3828 | return -EINVAL; | |
26578cda | 3829 | if (unlikely(sqe->ioprio || sqe->buf_index || sqe->splice_fd_in)) |
15b71abe | 3830 | return -EINVAL; |
ec65fea5 | 3831 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3832 | return -EBADF; |
03b1230c | 3833 | |
ec65fea5 PB |
3834 | /* open.how should be already initialised */ |
3835 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3836 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3837 | |
25e72d10 PB |
3838 | req->open.dfd = READ_ONCE(sqe->fd); |
3839 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3840 | req->open.filename = getname(fname); |
15b71abe JA |
3841 | if (IS_ERR(req->open.filename)) { |
3842 | ret = PTR_ERR(req->open.filename); | |
3843 | req->open.filename = NULL; | |
3844 | return ret; | |
3845 | } | |
4022e7af | 3846 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3847 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3848 | return 0; |
03b1230c JA |
3849 | } |
3850 | ||
ec65fea5 PB |
3851 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3852 | { | |
d3fddf6d PB |
3853 | u64 mode = READ_ONCE(sqe->len); |
3854 | u64 flags = READ_ONCE(sqe->open_flags); | |
ec65fea5 | 3855 | |
ec65fea5 PB |
3856 | req->open.how = build_open_how(flags, mode); |
3857 | return __io_openat_prep(req, sqe); | |
3858 | } | |
3859 | ||
cebdb986 | 3860 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3861 | { |
cebdb986 | 3862 | struct open_how __user *how; |
cebdb986 | 3863 | size_t len; |
0fa03c62 JA |
3864 | int ret; |
3865 | ||
cebdb986 JA |
3866 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3867 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3868 | if (len < OPEN_HOW_SIZE_VER0) |
3869 | return -EINVAL; | |
3529d8c2 | 3870 | |
cebdb986 JA |
3871 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3872 | len); | |
3873 | if (ret) | |
3874 | return ret; | |
3529d8c2 | 3875 | |
ec65fea5 | 3876 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3877 | } |
3878 | ||
45d189c6 | 3879 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3880 | { |
3881 | struct open_flags op; | |
15b71abe | 3882 | struct file *file; |
3a81fd02 JA |
3883 | bool nonblock_set; |
3884 | bool resolve_nonblock; | |
15b71abe JA |
3885 | int ret; |
3886 | ||
cebdb986 | 3887 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3888 | if (ret) |
3889 | goto err; | |
3a81fd02 JA |
3890 | nonblock_set = op.open_flag & O_NONBLOCK; |
3891 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3892 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3893 | /* |
3894 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3895 | * it'll always -EAGAIN | |
3896 | */ | |
3897 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3898 | return -EAGAIN; | |
3899 | op.lookup_flags |= LOOKUP_CACHED; | |
3900 | op.open_flag |= O_NONBLOCK; | |
3901 | } | |
15b71abe | 3902 | |
4022e7af | 3903 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3904 | if (ret < 0) |
3905 | goto err; | |
3906 | ||
3907 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 3908 | if (IS_ERR(file)) { |
944d1444 | 3909 | /* |
12dcb58a PB |
3910 | * We could hang on to this 'fd' on retrying, but seems like |
3911 | * marginal gain for something that is now known to be a slower | |
3912 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 3913 | */ |
3a81fd02 | 3914 | put_unused_fd(ret); |
3a81fd02 | 3915 | |
15b71abe | 3916 | ret = PTR_ERR(file); |
12dcb58a PB |
3917 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
3918 | if (ret == -EAGAIN && | |
3919 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
3920 | return -EAGAIN; | |
3921 | goto err; | |
15b71abe | 3922 | } |
12dcb58a PB |
3923 | |
3924 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
3925 | file->f_flags &= ~O_NONBLOCK; | |
3926 | fsnotify_open(file); | |
3927 | fd_install(ret, file); | |
15b71abe JA |
3928 | err: |
3929 | putname(req->open.filename); | |
8fef80bf | 3930 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 3931 | if (ret < 0) |
93d2bcd2 | 3932 | req_set_fail(req); |
0bdf3398 | 3933 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3934 | return 0; |
3935 | } | |
3936 | ||
45d189c6 | 3937 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3938 | { |
e45cff58 | 3939 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3940 | } |
3941 | ||
067524e9 JA |
3942 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3943 | const struct io_uring_sqe *sqe) | |
3944 | { | |
3945 | struct io_provide_buf *p = &req->pbuf; | |
3946 | u64 tmp; | |
3947 | ||
26578cda PB |
3948 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off || |
3949 | sqe->splice_fd_in) | |
067524e9 JA |
3950 | return -EINVAL; |
3951 | ||
3952 | tmp = READ_ONCE(sqe->fd); | |
3953 | if (!tmp || tmp > USHRT_MAX) | |
3954 | return -EINVAL; | |
3955 | ||
3956 | memset(p, 0, sizeof(*p)); | |
3957 | p->nbufs = tmp; | |
3958 | p->bgid = READ_ONCE(sqe->buf_group); | |
3959 | return 0; | |
3960 | } | |
3961 | ||
3962 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3963 | int bgid, unsigned nbufs) | |
3964 | { | |
3965 | unsigned i = 0; | |
3966 | ||
3967 | /* shouldn't happen */ | |
3968 | if (!nbufs) | |
3969 | return 0; | |
3970 | ||
3971 | /* the head kbuf is the list itself */ | |
3972 | while (!list_empty(&buf->list)) { | |
3973 | struct io_buffer *nxt; | |
3974 | ||
3975 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3976 | list_del(&nxt->list); | |
3977 | kfree(nxt); | |
3978 | if (++i == nbufs) | |
3979 | return i; | |
3980 | } | |
3981 | i++; | |
3982 | kfree(buf); | |
9e15c3a0 | 3983 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3984 | |
3985 | return i; | |
3986 | } | |
3987 | ||
889fca73 | 3988 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3989 | { |
3990 | struct io_provide_buf *p = &req->pbuf; | |
3991 | struct io_ring_ctx *ctx = req->ctx; | |
3992 | struct io_buffer *head; | |
3993 | int ret = 0; | |
45d189c6 | 3994 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3995 | |
3996 | io_ring_submit_lock(ctx, !force_nonblock); | |
3997 | ||
3998 | lockdep_assert_held(&ctx->uring_lock); | |
3999 | ||
4000 | ret = -ENOENT; | |
9e15c3a0 | 4001 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
4002 | if (head) |
4003 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 4004 | if (ret < 0) |
93d2bcd2 | 4005 | req_set_fail(req); |
067524e9 | 4006 | |
9fb8cb49 PB |
4007 | /* complete before unlock, IOPOLL may need the lock */ |
4008 | __io_req_complete(req, issue_flags, ret, 0); | |
4009 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
4010 | return 0; |
4011 | } | |
4012 | ||
ddf0322d JA |
4013 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4014 | const struct io_uring_sqe *sqe) | |
4015 | { | |
38134ada | 4016 | unsigned long size, tmp_check; |
ddf0322d JA |
4017 | struct io_provide_buf *p = &req->pbuf; |
4018 | u64 tmp; | |
4019 | ||
26578cda | 4020 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
ddf0322d JA |
4021 | return -EINVAL; |
4022 | ||
4023 | tmp = READ_ONCE(sqe->fd); | |
4024 | if (!tmp || tmp > USHRT_MAX) | |
4025 | return -E2BIG; | |
4026 | p->nbufs = tmp; | |
4027 | p->addr = READ_ONCE(sqe->addr); | |
4028 | p->len = READ_ONCE(sqe->len); | |
4029 | ||
38134ada PB |
4030 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
4031 | &size)) | |
4032 | return -EOVERFLOW; | |
4033 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
4034 | return -EOVERFLOW; | |
4035 | ||
d81269fe PB |
4036 | size = (unsigned long)p->len * p->nbufs; |
4037 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
4038 | return -EFAULT; |
4039 | ||
4040 | p->bgid = READ_ONCE(sqe->buf_group); | |
4041 | tmp = READ_ONCE(sqe->off); | |
4042 | if (tmp > USHRT_MAX) | |
4043 | return -E2BIG; | |
4044 | p->bid = tmp; | |
4045 | return 0; | |
4046 | } | |
4047 | ||
4048 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4049 | { | |
4050 | struct io_buffer *buf; | |
4051 | u64 addr = pbuf->addr; | |
4052 | int i, bid = pbuf->bid; | |
4053 | ||
4054 | for (i = 0; i < pbuf->nbufs; i++) { | |
4055 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4056 | if (!buf) | |
4057 | break; | |
4058 | ||
4059 | buf->addr = addr; | |
d1f82808 | 4060 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
4061 | buf->bid = bid; |
4062 | addr += pbuf->len; | |
4063 | bid++; | |
4064 | if (!*head) { | |
4065 | INIT_LIST_HEAD(&buf->list); | |
4066 | *head = buf; | |
4067 | } else { | |
4068 | list_add_tail(&buf->list, &(*head)->list); | |
4069 | } | |
4070 | } | |
4071 | ||
4072 | return i ? i : -ENOMEM; | |
4073 | } | |
4074 | ||
889fca73 | 4075 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4076 | { |
4077 | struct io_provide_buf *p = &req->pbuf; | |
4078 | struct io_ring_ctx *ctx = req->ctx; | |
4079 | struct io_buffer *head, *list; | |
4080 | int ret = 0; | |
45d189c6 | 4081 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4082 | |
4083 | io_ring_submit_lock(ctx, !force_nonblock); | |
4084 | ||
4085 | lockdep_assert_held(&ctx->uring_lock); | |
4086 | ||
9e15c3a0 | 4087 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4088 | |
4089 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4090 | if (ret >= 0 && !list) { |
4091 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4092 | if (ret < 0) | |
067524e9 | 4093 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4094 | } |
ddf0322d | 4095 | if (ret < 0) |
93d2bcd2 | 4096 | req_set_fail(req); |
9fb8cb49 PB |
4097 | /* complete before unlock, IOPOLL may need the lock */ |
4098 | __io_req_complete(req, issue_flags, ret, 0); | |
4099 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4100 | return 0; |
cebdb986 JA |
4101 | } |
4102 | ||
3e4827b0 JA |
4103 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4104 | const struct io_uring_sqe *sqe) | |
4105 | { | |
4106 | #if defined(CONFIG_EPOLL) | |
26578cda | 4107 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
3e4827b0 | 4108 | return -EINVAL; |
2d74d042 | 4109 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4110 | return -EINVAL; |
3e4827b0 JA |
4111 | |
4112 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4113 | req->epoll.op = READ_ONCE(sqe->len); | |
4114 | req->epoll.fd = READ_ONCE(sqe->off); | |
4115 | ||
4116 | if (ep_op_has_event(req->epoll.op)) { | |
4117 | struct epoll_event __user *ev; | |
4118 | ||
4119 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4120 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4121 | return -EFAULT; | |
4122 | } | |
4123 | ||
4124 | return 0; | |
4125 | #else | |
4126 | return -EOPNOTSUPP; | |
4127 | #endif | |
4128 | } | |
4129 | ||
889fca73 | 4130 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4131 | { |
4132 | #if defined(CONFIG_EPOLL) | |
4133 | struct io_epoll *ie = &req->epoll; | |
4134 | int ret; | |
45d189c6 | 4135 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4136 | |
4137 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4138 | if (force_nonblock && ret == -EAGAIN) | |
4139 | return -EAGAIN; | |
4140 | ||
4141 | if (ret < 0) | |
93d2bcd2 | 4142 | req_set_fail(req); |
889fca73 | 4143 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4144 | return 0; |
4145 | #else | |
4146 | return -EOPNOTSUPP; | |
4147 | #endif | |
4148 | } | |
4149 | ||
c1ca757b JA |
4150 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4151 | { | |
4152 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
26578cda | 4153 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->splice_fd_in) |
c1ca757b | 4154 | return -EINVAL; |
3232dd02 PB |
4155 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4156 | return -EINVAL; | |
c1ca757b JA |
4157 | |
4158 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4159 | req->madvise.len = READ_ONCE(sqe->len); | |
4160 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4161 | return 0; | |
4162 | #else | |
4163 | return -EOPNOTSUPP; | |
4164 | #endif | |
4165 | } | |
4166 | ||
45d189c6 | 4167 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4168 | { |
4169 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4170 | struct io_madvise *ma = &req->madvise; | |
4171 | int ret; | |
4172 | ||
45d189c6 | 4173 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4174 | return -EAGAIN; |
4175 | ||
0726b01e | 4176 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4177 | if (ret < 0) |
93d2bcd2 | 4178 | req_set_fail(req); |
e1e16097 | 4179 | io_req_complete(req, ret); |
c1ca757b JA |
4180 | return 0; |
4181 | #else | |
4182 | return -EOPNOTSUPP; | |
4183 | #endif | |
4184 | } | |
4185 | ||
4840e418 JA |
4186 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4187 | { | |
26578cda | 4188 | if (sqe->ioprio || sqe->buf_index || sqe->addr || sqe->splice_fd_in) |
4840e418 | 4189 | return -EINVAL; |
3232dd02 PB |
4190 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4191 | return -EINVAL; | |
4840e418 JA |
4192 | |
4193 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4194 | req->fadvise.len = READ_ONCE(sqe->len); | |
4195 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4196 | return 0; | |
4197 | } | |
4198 | ||
45d189c6 | 4199 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4200 | { |
4201 | struct io_fadvise *fa = &req->fadvise; | |
4202 | int ret; | |
4203 | ||
45d189c6 | 4204 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4205 | switch (fa->advice) { |
4206 | case POSIX_FADV_NORMAL: | |
4207 | case POSIX_FADV_RANDOM: | |
4208 | case POSIX_FADV_SEQUENTIAL: | |
4209 | break; | |
4210 | default: | |
4211 | return -EAGAIN; | |
4212 | } | |
4213 | } | |
4840e418 JA |
4214 | |
4215 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4216 | if (ret < 0) | |
93d2bcd2 | 4217 | req_set_fail(req); |
0bdf3398 | 4218 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4219 | return 0; |
4220 | } | |
4221 | ||
eddc7ef5 JA |
4222 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4223 | { | |
2d74d042 | 4224 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4225 | return -EINVAL; |
26578cda | 4226 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
eddc7ef5 | 4227 | return -EINVAL; |
9c280f90 | 4228 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4229 | return -EBADF; |
eddc7ef5 | 4230 | |
1d9e1288 BM |
4231 | req->statx.dfd = READ_ONCE(sqe->fd); |
4232 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4233 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4234 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4235 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4236 | |
4237 | return 0; | |
4238 | } | |
4239 | ||
45d189c6 | 4240 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4241 | { |
1d9e1288 | 4242 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4243 | int ret; |
4244 | ||
59d70013 | 4245 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4246 | return -EAGAIN; |
4247 | ||
e62753e4 BM |
4248 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4249 | ctx->buffer); | |
eddc7ef5 | 4250 | |
eddc7ef5 | 4251 | if (ret < 0) |
93d2bcd2 | 4252 | req_set_fail(req); |
e1e16097 | 4253 | io_req_complete(req, ret); |
eddc7ef5 JA |
4254 | return 0; |
4255 | } | |
4256 | ||
b5dba59e JA |
4257 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4258 | { | |
14587a46 | 4259 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4260 | return -EINVAL; |
b5dba59e | 4261 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
26578cda | 4262 | sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) |
b5dba59e | 4263 | return -EINVAL; |
9c280f90 | 4264 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4265 | return -EBADF; |
b5dba59e JA |
4266 | |
4267 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4268 | return 0; |
b5dba59e JA |
4269 | } |
4270 | ||
889fca73 | 4271 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4272 | { |
9eac1904 | 4273 | struct files_struct *files = current->files; |
3af73b28 | 4274 | struct io_close *close = &req->close; |
9eac1904 | 4275 | struct fdtable *fdt; |
a1fde923 PB |
4276 | struct file *file = NULL; |
4277 | int ret = -EBADF; | |
b5dba59e | 4278 | |
9eac1904 JA |
4279 | spin_lock(&files->file_lock); |
4280 | fdt = files_fdtable(files); | |
4281 | if (close->fd >= fdt->max_fds) { | |
4282 | spin_unlock(&files->file_lock); | |
4283 | goto err; | |
4284 | } | |
4285 | file = fdt->fd[close->fd]; | |
a1fde923 | 4286 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4287 | spin_unlock(&files->file_lock); |
4288 | file = NULL; | |
4289 | goto err; | |
3af73b28 | 4290 | } |
b5dba59e JA |
4291 | |
4292 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4293 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4294 | spin_unlock(&files->file_lock); |
0bf0eefd | 4295 | return -EAGAIN; |
a2100672 | 4296 | } |
b5dba59e | 4297 | |
9eac1904 JA |
4298 | ret = __close_fd_get_file(close->fd, &file); |
4299 | spin_unlock(&files->file_lock); | |
4300 | if (ret < 0) { | |
4301 | if (ret == -ENOENT) | |
4302 | ret = -EBADF; | |
4303 | goto err; | |
4304 | } | |
4305 | ||
3af73b28 | 4306 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4307 | ret = filp_close(file, current->files); |
4308 | err: | |
3af73b28 | 4309 | if (ret < 0) |
93d2bcd2 | 4310 | req_set_fail(req); |
9eac1904 JA |
4311 | if (file) |
4312 | fput(file); | |
889fca73 | 4313 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4314 | return 0; |
b5dba59e JA |
4315 | } |
4316 | ||
1155c76a | 4317 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4318 | { |
4319 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4320 | |
5d17b4a4 JA |
4321 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4322 | return -EINVAL; | |
26578cda PB |
4323 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4324 | sqe->splice_fd_in)) | |
5d17b4a4 JA |
4325 | return -EINVAL; |
4326 | ||
8ed8d3c3 JA |
4327 | req->sync.off = READ_ONCE(sqe->off); |
4328 | req->sync.len = READ_ONCE(sqe->len); | |
4329 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4330 | return 0; |
4331 | } | |
4332 | ||
45d189c6 | 4333 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4334 | { |
8ed8d3c3 JA |
4335 | int ret; |
4336 | ||
ac45abc0 | 4337 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4338 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4339 | return -EAGAIN; |
4340 | ||
9adbd45d | 4341 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4342 | req->sync.flags); |
4343 | if (ret < 0) | |
93d2bcd2 | 4344 | req_set_fail(req); |
e1e16097 | 4345 | io_req_complete(req, ret); |
5d17b4a4 JA |
4346 | return 0; |
4347 | } | |
4348 | ||
469956e8 | 4349 | #if defined(CONFIG_NET) |
02d27d89 PB |
4350 | static int io_setup_async_msg(struct io_kiocb *req, |
4351 | struct io_async_msghdr *kmsg) | |
4352 | { | |
e8c2bc1f JA |
4353 | struct io_async_msghdr *async_msg = req->async_data; |
4354 | ||
4355 | if (async_msg) | |
02d27d89 | 4356 | return -EAGAIN; |
e8c2bc1f | 4357 | if (io_alloc_async_data(req)) { |
257e84a5 | 4358 | kfree(kmsg->free_iov); |
02d27d89 PB |
4359 | return -ENOMEM; |
4360 | } | |
e8c2bc1f | 4361 | async_msg = req->async_data; |
02d27d89 | 4362 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4363 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4364 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4365 | /* if were using fast_iov, set it to the new one */ |
4366 | if (!async_msg->free_iov) | |
4367 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4368 | ||
02d27d89 PB |
4369 | return -EAGAIN; |
4370 | } | |
4371 | ||
2ae523ed PB |
4372 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4373 | struct io_async_msghdr *iomsg) | |
4374 | { | |
2ae523ed | 4375 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4376 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4377 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4378 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4379 | } |
4380 | ||
93642ef8 PB |
4381 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4382 | { | |
4383 | int ret; | |
4384 | ||
93642ef8 PB |
4385 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4386 | if (!ret) | |
4387 | req->flags |= REQ_F_NEED_CLEANUP; | |
4388 | return ret; | |
4389 | } | |
4390 | ||
3529d8c2 | 4391 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4392 | { |
e47293fd | 4393 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4394 | |
d2b6f48b PB |
4395 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4396 | return -EINVAL; | |
4397 | ||
270a5940 | 4398 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4399 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4400 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4401 | if (sr->msg_flags & MSG_DONTWAIT) | |
4402 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4403 | |
d8768362 JA |
4404 | #ifdef CONFIG_COMPAT |
4405 | if (req->ctx->compat) | |
4406 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4407 | #endif | |
93642ef8 | 4408 | return 0; |
03b1230c JA |
4409 | } |
4410 | ||
889fca73 | 4411 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4412 | { |
6b754c8b | 4413 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4414 | struct socket *sock; |
7a7cacba | 4415 | unsigned flags; |
0031275d | 4416 | int min_ret = 0; |
0fa03c62 JA |
4417 | int ret; |
4418 | ||
dba4a925 | 4419 | sock = sock_from_file(req->file); |
7a7cacba | 4420 | if (unlikely(!sock)) |
dba4a925 | 4421 | return -ENOTSOCK; |
3529d8c2 | 4422 | |
257e84a5 PB |
4423 | kmsg = req->async_data; |
4424 | if (!kmsg) { | |
7a7cacba PB |
4425 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4426 | if (ret) | |
4427 | return ret; | |
4428 | kmsg = &iomsg; | |
0fa03c62 | 4429 | } |
0fa03c62 | 4430 | |
04411806 PB |
4431 | flags = req->sr_msg.msg_flags; |
4432 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4433 | flags |= MSG_DONTWAIT; |
0031275d SM |
4434 | if (flags & MSG_WAITALL) |
4435 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4436 | ||
7a7cacba | 4437 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4438 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4439 | return io_setup_async_msg(req, kmsg); |
4440 | if (ret == -ERESTARTSYS) | |
4441 | ret = -EINTR; | |
0fa03c62 | 4442 | |
257e84a5 PB |
4443 | /* fast path, check for non-NULL to avoid function call */ |
4444 | if (kmsg->free_iov) | |
4445 | kfree(kmsg->free_iov); | |
99bc4c38 | 4446 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4447 | if (ret < min_ret) |
93d2bcd2 | 4448 | req_set_fail(req); |
889fca73 | 4449 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4450 | return 0; |
03b1230c | 4451 | } |
aa1fa28f | 4452 | |
889fca73 | 4453 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4454 | { |
7a7cacba PB |
4455 | struct io_sr_msg *sr = &req->sr_msg; |
4456 | struct msghdr msg; | |
4457 | struct iovec iov; | |
fddaface | 4458 | struct socket *sock; |
7a7cacba | 4459 | unsigned flags; |
0031275d | 4460 | int min_ret = 0; |
fddaface JA |
4461 | int ret; |
4462 | ||
dba4a925 | 4463 | sock = sock_from_file(req->file); |
7a7cacba | 4464 | if (unlikely(!sock)) |
dba4a925 | 4465 | return -ENOTSOCK; |
fddaface | 4466 | |
7a7cacba PB |
4467 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4468 | if (unlikely(ret)) | |
14db8411 | 4469 | return ret; |
fddaface | 4470 | |
7a7cacba PB |
4471 | msg.msg_name = NULL; |
4472 | msg.msg_control = NULL; | |
4473 | msg.msg_controllen = 0; | |
4474 | msg.msg_namelen = 0; | |
fddaface | 4475 | |
04411806 PB |
4476 | flags = req->sr_msg.msg_flags; |
4477 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4478 | flags |= MSG_DONTWAIT; |
0031275d SM |
4479 | if (flags & MSG_WAITALL) |
4480 | min_ret = iov_iter_count(&msg.msg_iter); | |
4481 | ||
7a7cacba PB |
4482 | msg.msg_flags = flags; |
4483 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4484 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4485 | return -EAGAIN; |
4486 | if (ret == -ERESTARTSYS) | |
4487 | ret = -EINTR; | |
fddaface | 4488 | |
0031275d | 4489 | if (ret < min_ret) |
93d2bcd2 | 4490 | req_set_fail(req); |
889fca73 | 4491 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4492 | return 0; |
fddaface JA |
4493 | } |
4494 | ||
1400e697 PB |
4495 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4496 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4497 | { |
4498 | struct io_sr_msg *sr = &req->sr_msg; | |
4499 | struct iovec __user *uiov; | |
4500 | size_t iov_len; | |
4501 | int ret; | |
4502 | ||
1400e697 PB |
4503 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4504 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4505 | if (ret) |
4506 | return ret; | |
4507 | ||
4508 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4509 | if (iov_len > 1) | |
4510 | return -EINVAL; | |
5476dfed | 4511 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4512 | return -EFAULT; |
5476dfed | 4513 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4514 | iomsg->free_iov = NULL; |
52de1fe1 | 4515 | } else { |
257e84a5 | 4516 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4517 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4518 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4519 | false); |
52de1fe1 JA |
4520 | if (ret > 0) |
4521 | ret = 0; | |
4522 | } | |
4523 | ||
4524 | return ret; | |
4525 | } | |
4526 | ||
4527 | #ifdef CONFIG_COMPAT | |
4528 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4529 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4530 | { |
52de1fe1 JA |
4531 | struct io_sr_msg *sr = &req->sr_msg; |
4532 | struct compat_iovec __user *uiov; | |
4533 | compat_uptr_t ptr; | |
4534 | compat_size_t len; | |
4535 | int ret; | |
4536 | ||
4af3417a PB |
4537 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4538 | &ptr, &len); | |
52de1fe1 JA |
4539 | if (ret) |
4540 | return ret; | |
4541 | ||
4542 | uiov = compat_ptr(ptr); | |
4543 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4544 | compat_ssize_t clen; | |
4545 | ||
4546 | if (len > 1) | |
4547 | return -EINVAL; | |
4548 | if (!access_ok(uiov, sizeof(*uiov))) | |
4549 | return -EFAULT; | |
4550 | if (__get_user(clen, &uiov->iov_len)) | |
4551 | return -EFAULT; | |
4552 | if (clen < 0) | |
4553 | return -EINVAL; | |
2d280bc8 | 4554 | sr->len = clen; |
257e84a5 | 4555 | iomsg->free_iov = NULL; |
52de1fe1 | 4556 | } else { |
257e84a5 | 4557 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4558 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4559 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4560 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4561 | if (ret < 0) |
4562 | return ret; | |
4563 | } | |
4564 | ||
4565 | return 0; | |
4566 | } | |
4567 | #endif | |
4568 | ||
1400e697 PB |
4569 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4570 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4571 | { |
1400e697 | 4572 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4573 | |
4574 | #ifdef CONFIG_COMPAT | |
4575 | if (req->ctx->compat) | |
1400e697 | 4576 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4577 | #endif |
52de1fe1 | 4578 | |
1400e697 | 4579 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4580 | } |
4581 | ||
bcda7baa | 4582 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4583 | bool needs_lock) |
bcda7baa JA |
4584 | { |
4585 | struct io_sr_msg *sr = &req->sr_msg; | |
4586 | struct io_buffer *kbuf; | |
4587 | ||
bcda7baa JA |
4588 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4589 | if (IS_ERR(kbuf)) | |
4590 | return kbuf; | |
4591 | ||
4592 | sr->kbuf = kbuf; | |
4593 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4594 | return kbuf; |
fddaface JA |
4595 | } |
4596 | ||
7fbb1b54 PB |
4597 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4598 | { | |
4599 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4600 | } | |
4601 | ||
93642ef8 | 4602 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4603 | { |
99bc4c38 | 4604 | int ret; |
3529d8c2 | 4605 | |
93642ef8 PB |
4606 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4607 | if (!ret) | |
4608 | req->flags |= REQ_F_NEED_CLEANUP; | |
4609 | return ret; | |
4610 | } | |
4611 | ||
4612 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4613 | { | |
4614 | struct io_sr_msg *sr = &req->sr_msg; | |
4615 | ||
d2b6f48b PB |
4616 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4617 | return -EINVAL; | |
4618 | ||
270a5940 | 4619 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4620 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4621 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4622 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4623 | if (sr->msg_flags & MSG_DONTWAIT) | |
4624 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4625 | |
d8768362 JA |
4626 | #ifdef CONFIG_COMPAT |
4627 | if (req->ctx->compat) | |
4628 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4629 | #endif | |
93642ef8 | 4630 | return 0; |
aa1fa28f JA |
4631 | } |
4632 | ||
889fca73 | 4633 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4634 | { |
6b754c8b | 4635 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4636 | struct socket *sock; |
7fbb1b54 | 4637 | struct io_buffer *kbuf; |
7a7cacba | 4638 | unsigned flags; |
0031275d | 4639 | int min_ret = 0; |
52de1fe1 | 4640 | int ret, cflags = 0; |
45d189c6 | 4641 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4642 | |
dba4a925 | 4643 | sock = sock_from_file(req->file); |
7a7cacba | 4644 | if (unlikely(!sock)) |
dba4a925 | 4645 | return -ENOTSOCK; |
3529d8c2 | 4646 | |
257e84a5 PB |
4647 | kmsg = req->async_data; |
4648 | if (!kmsg) { | |
7a7cacba PB |
4649 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4650 | if (ret) | |
681fda8d | 4651 | return ret; |
7a7cacba PB |
4652 | kmsg = &iomsg; |
4653 | } | |
03b1230c | 4654 | |
bc02ef33 | 4655 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4656 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4657 | if (IS_ERR(kbuf)) |
52de1fe1 | 4658 | return PTR_ERR(kbuf); |
7a7cacba | 4659 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4660 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4661 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4662 | 1, req->sr_msg.len); |
4663 | } | |
52de1fe1 | 4664 | |
04411806 PB |
4665 | flags = req->sr_msg.msg_flags; |
4666 | if (force_nonblock) | |
7a7cacba | 4667 | flags |= MSG_DONTWAIT; |
0031275d SM |
4668 | if (flags & MSG_WAITALL) |
4669 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4670 | ||
7a7cacba PB |
4671 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4672 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4673 | if (force_nonblock && ret == -EAGAIN) |
4674 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4675 | if (ret == -ERESTARTSYS) |
4676 | ret = -EINTR; | |
03b1230c | 4677 | |
7fbb1b54 PB |
4678 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4679 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4680 | /* fast path, check for non-NULL to avoid function call */ |
4681 | if (kmsg->free_iov) | |
4682 | kfree(kmsg->free_iov); | |
99bc4c38 | 4683 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4684 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4685 | req_set_fail(req); |
889fca73 | 4686 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4687 | return 0; |
0fa03c62 | 4688 | } |
5d17b4a4 | 4689 | |
889fca73 | 4690 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4691 | { |
6b754c8b | 4692 | struct io_buffer *kbuf; |
7a7cacba PB |
4693 | struct io_sr_msg *sr = &req->sr_msg; |
4694 | struct msghdr msg; | |
4695 | void __user *buf = sr->buf; | |
fddaface | 4696 | struct socket *sock; |
7a7cacba PB |
4697 | struct iovec iov; |
4698 | unsigned flags; | |
0031275d | 4699 | int min_ret = 0; |
bcda7baa | 4700 | int ret, cflags = 0; |
45d189c6 | 4701 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4702 | |
dba4a925 | 4703 | sock = sock_from_file(req->file); |
7a7cacba | 4704 | if (unlikely(!sock)) |
dba4a925 | 4705 | return -ENOTSOCK; |
fddaface | 4706 | |
bc02ef33 | 4707 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4708 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4709 | if (IS_ERR(kbuf)) |
4710 | return PTR_ERR(kbuf); | |
7a7cacba | 4711 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4712 | } |
bcda7baa | 4713 | |
7a7cacba | 4714 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4715 | if (unlikely(ret)) |
4716 | goto out_free; | |
fddaface | 4717 | |
7a7cacba PB |
4718 | msg.msg_name = NULL; |
4719 | msg.msg_control = NULL; | |
4720 | msg.msg_controllen = 0; | |
4721 | msg.msg_namelen = 0; | |
4722 | msg.msg_iocb = NULL; | |
4723 | msg.msg_flags = 0; | |
fddaface | 4724 | |
04411806 PB |
4725 | flags = req->sr_msg.msg_flags; |
4726 | if (force_nonblock) | |
7a7cacba | 4727 | flags |= MSG_DONTWAIT; |
0031275d SM |
4728 | if (flags & MSG_WAITALL) |
4729 | min_ret = iov_iter_count(&msg.msg_iter); | |
4730 | ||
7a7cacba PB |
4731 | ret = sock_recvmsg(sock, &msg, flags); |
4732 | if (force_nonblock && ret == -EAGAIN) | |
4733 | return -EAGAIN; | |
4734 | if (ret == -ERESTARTSYS) | |
4735 | ret = -EINTR; | |
14c32eee | 4736 | out_free: |
7fbb1b54 PB |
4737 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4738 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4739 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4740 | req_set_fail(req); |
889fca73 | 4741 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4742 | return 0; |
fddaface JA |
4743 | } |
4744 | ||
3529d8c2 | 4745 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4746 | { |
8ed8d3c3 JA |
4747 | struct io_accept *accept = &req->accept; |
4748 | ||
14587a46 | 4749 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4750 | return -EINVAL; |
26578cda | 4751 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->splice_fd_in) |
17f2fe35 JA |
4752 | return -EINVAL; |
4753 | ||
d55e5f5b JA |
4754 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4755 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4756 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4757 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4758 | return 0; |
8ed8d3c3 | 4759 | } |
17f2fe35 | 4760 | |
889fca73 | 4761 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4762 | { |
4763 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4764 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4765 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4766 | int ret; |
4767 | ||
e697deed JX |
4768 | if (req->file->f_flags & O_NONBLOCK) |
4769 | req->flags |= REQ_F_NOWAIT; | |
4770 | ||
8ed8d3c3 | 4771 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4772 | accept->addr_len, accept->flags, |
4773 | accept->nofile); | |
8ed8d3c3 | 4774 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4775 | return -EAGAIN; |
ac45abc0 PB |
4776 | if (ret < 0) { |
4777 | if (ret == -ERESTARTSYS) | |
4778 | ret = -EINTR; | |
93d2bcd2 | 4779 | req_set_fail(req); |
ac45abc0 | 4780 | } |
889fca73 | 4781 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4782 | return 0; |
8ed8d3c3 JA |
4783 | } |
4784 | ||
93642ef8 PB |
4785 | static int io_connect_prep_async(struct io_kiocb *req) |
4786 | { | |
4787 | struct io_async_connect *io = req->async_data; | |
4788 | struct io_connect *conn = &req->connect; | |
4789 | ||
4790 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4791 | } | |
4792 | ||
3529d8c2 | 4793 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4794 | { |
3529d8c2 | 4795 | struct io_connect *conn = &req->connect; |
f499a021 | 4796 | |
14587a46 | 4797 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 | 4798 | return -EINVAL; |
26578cda PB |
4799 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags || |
4800 | sqe->splice_fd_in) | |
3fbb51c1 JA |
4801 | return -EINVAL; |
4802 | ||
3529d8c2 JA |
4803 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4804 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4805 | return 0; |
f499a021 JA |
4806 | } |
4807 | ||
889fca73 | 4808 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4809 | { |
e8c2bc1f | 4810 | struct io_async_connect __io, *io; |
f8e85cf2 | 4811 | unsigned file_flags; |
3fbb51c1 | 4812 | int ret; |
45d189c6 | 4813 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4814 | |
e8c2bc1f JA |
4815 | if (req->async_data) { |
4816 | io = req->async_data; | |
f499a021 | 4817 | } else { |
3529d8c2 JA |
4818 | ret = move_addr_to_kernel(req->connect.addr, |
4819 | req->connect.addr_len, | |
e8c2bc1f | 4820 | &__io.address); |
f499a021 JA |
4821 | if (ret) |
4822 | goto out; | |
4823 | io = &__io; | |
4824 | } | |
4825 | ||
3fbb51c1 JA |
4826 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4827 | ||
e8c2bc1f | 4828 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4829 | req->connect.addr_len, file_flags); |
87f80d62 | 4830 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4831 | if (req->async_data) |
b7bb4f7d | 4832 | return -EAGAIN; |
e8c2bc1f | 4833 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4834 | ret = -ENOMEM; |
4835 | goto out; | |
4836 | } | |
e8c2bc1f | 4837 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4838 | return -EAGAIN; |
f499a021 | 4839 | } |
f8e85cf2 JA |
4840 | if (ret == -ERESTARTSYS) |
4841 | ret = -EINTR; | |
f499a021 | 4842 | out: |
4e88d6e7 | 4843 | if (ret < 0) |
93d2bcd2 | 4844 | req_set_fail(req); |
889fca73 | 4845 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4846 | return 0; |
469956e8 Y |
4847 | } |
4848 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4849 | #define IO_NETOP_FN(op) \ |
4850 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4851 | { \ | |
4852 | return -EOPNOTSUPP; \ | |
4853 | } | |
4854 | ||
4855 | #define IO_NETOP_PREP(op) \ | |
4856 | IO_NETOP_FN(op) \ | |
4857 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4858 | { \ | |
4859 | return -EOPNOTSUPP; \ | |
4860 | } \ | |
4861 | ||
4862 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4863 | IO_NETOP_PREP(op) \ | |
4864 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4865 | { \ | |
4866 | return -EOPNOTSUPP; \ | |
4867 | } | |
4868 | ||
4869 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4870 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4871 | IO_NETOP_PREP_ASYNC(connect); | |
4872 | IO_NETOP_PREP(accept); | |
4873 | IO_NETOP_FN(send); | |
4874 | IO_NETOP_FN(recv); | |
469956e8 | 4875 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4876 | |
d7718a9d JA |
4877 | struct io_poll_table { |
4878 | struct poll_table_struct pt; | |
4879 | struct io_kiocb *req; | |
68b11e8b | 4880 | int nr_entries; |
d7718a9d JA |
4881 | int error; |
4882 | }; | |
ce593a6c | 4883 | |
d7718a9d | 4884 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5b0a6acc | 4885 | __poll_t mask, io_req_tw_func_t func) |
d7718a9d | 4886 | { |
d7718a9d JA |
4887 | /* for instances that support it check for an event match first: */ |
4888 | if (mask && !(mask & poll->events)) | |
4889 | return 0; | |
4890 | ||
4891 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4892 | ||
4893 | list_del_init(&poll->wait.entry); | |
4894 | ||
d7718a9d | 4895 | req->result = mask; |
5b0a6acc | 4896 | req->io_task_work.func = func; |
6d816e08 | 4897 | |
d7718a9d | 4898 | /* |
e3aabf95 JA |
4899 | * If this fails, then the task is exiting. When a task exits, the |
4900 | * work gets canceled, so just cancel this request as well instead | |
4901 | * of executing it. We can't safely execute it anyway, as we may not | |
4902 | * have the needed state needed for it anyway. | |
d7718a9d | 4903 | */ |
e09ee510 | 4904 | io_req_task_work_add(req); |
d7718a9d JA |
4905 | return 1; |
4906 | } | |
4907 | ||
74ce6ce4 JA |
4908 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4909 | __acquires(&req->ctx->completion_lock) | |
4910 | { | |
4911 | struct io_ring_ctx *ctx = req->ctx; | |
4912 | ||
316319e8 | 4913 | /* req->task == current here, checking PF_EXITING is safe */ |
e09ee510 PB |
4914 | if (unlikely(req->task->flags & PF_EXITING)) |
4915 | WRITE_ONCE(poll->canceled, true); | |
4916 | ||
74ce6ce4 JA |
4917 | if (!req->result && !READ_ONCE(poll->canceled)) { |
4918 | struct poll_table_struct pt = { ._key = poll->events }; | |
4919 | ||
4920 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4921 | } | |
4922 | ||
79ebeaee | 4923 | spin_lock(&ctx->completion_lock); |
74ce6ce4 JA |
4924 | if (!req->result && !READ_ONCE(poll->canceled)) { |
4925 | add_wait_queue(poll->head, &poll->wait); | |
4926 | return true; | |
4927 | } | |
4928 | ||
4929 | return false; | |
4930 | } | |
4931 | ||
d4e7cd36 | 4932 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4933 | { |
e8c2bc1f | 4934 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4935 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4936 | return req->async_data; |
d4e7cd36 JA |
4937 | return req->apoll->double_poll; |
4938 | } | |
4939 | ||
4940 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4941 | { | |
4942 | if (req->opcode == IORING_OP_POLL_ADD) | |
4943 | return &req->poll; | |
4944 | return &req->apoll->poll; | |
4945 | } | |
4946 | ||
4947 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4948 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4949 | { |
4950 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4951 | |
4952 | lockdep_assert_held(&req->ctx->completion_lock); | |
4953 | ||
4954 | if (poll && poll->head) { | |
4955 | struct wait_queue_head *head = poll->head; | |
4956 | ||
79ebeaee | 4957 | spin_lock_irq(&head->lock); |
18bceab1 JA |
4958 | list_del_init(&poll->wait.entry); |
4959 | if (poll->wait.private) | |
de9b4cca | 4960 | req_ref_put(req); |
18bceab1 | 4961 | poll->head = NULL; |
79ebeaee | 4962 | spin_unlock_irq(&head->lock); |
18bceab1 JA |
4963 | } |
4964 | } | |
4965 | ||
e27414be | 4966 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4967 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4968 | { |
4969 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4970 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4971 | int error; |
18bceab1 | 4972 | |
e27414be | 4973 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4974 | error = -ECANCELED; |
88e41cf9 | 4975 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4976 | } else { |
5082620f | 4977 | error = mangle_poll(mask); |
e27414be | 4978 | } |
b69de288 JA |
4979 | if (req->poll.events & EPOLLONESHOT) |
4980 | flags = 0; | |
d4d19c19 | 4981 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
88e41cf9 JA |
4982 | req->poll.done = true; |
4983 | flags = 0; | |
4984 | } | |
7b289c38 HX |
4985 | if (flags & IORING_CQE_F_MORE) |
4986 | ctx->cq_extra++; | |
18bceab1 | 4987 | |
18bceab1 | 4988 | io_commit_cqring(ctx); |
88e41cf9 | 4989 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4990 | } |
4991 | ||
5b0a6acc | 4992 | static void io_poll_task_func(struct io_kiocb *req) |
18bceab1 JA |
4993 | { |
4994 | struct io_ring_ctx *ctx = req->ctx; | |
dd221f46 | 4995 | struct io_kiocb *nxt; |
18bceab1 JA |
4996 | |
4997 | if (io_poll_rewait(req, &req->poll)) { | |
79ebeaee | 4998 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 4999 | } else { |
f40b964a | 5000 | bool done; |
18bceab1 | 5001 | |
e27414be | 5002 | done = io_poll_complete(req, req->result); |
88e41cf9 | 5003 | if (done) { |
a890d01e | 5004 | io_poll_remove_double(req); |
88e41cf9 | 5005 | hash_del(&req->hash_node); |
f40b964a | 5006 | } else { |
88e41cf9 JA |
5007 | req->result = 0; |
5008 | add_wait_queue(req->poll.head, &req->poll.wait); | |
5009 | } | |
79ebeaee | 5010 | spin_unlock(&ctx->completion_lock); |
dd221f46 | 5011 | io_cqring_ev_posted(ctx); |
18bceab1 | 5012 | |
88e41cf9 JA |
5013 | if (done) { |
5014 | nxt = io_put_req_find_next(req); | |
5015 | if (nxt) | |
5b0a6acc | 5016 | io_req_task_submit(nxt); |
88e41cf9 | 5017 | } |
dd221f46 | 5018 | } |
18bceab1 JA |
5019 | } |
5020 | ||
5021 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5022 | int sync, void *key) | |
5023 | { | |
5024 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5025 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 | 5026 | __poll_t mask = key_to_poll(key); |
79ebeaee | 5027 | unsigned long flags; |
18bceab1 JA |
5028 | |
5029 | /* for instances that support it check for an event match first: */ | |
5030 | if (mask && !(mask & poll->events)) | |
5031 | return 0; | |
88e41cf9 JA |
5032 | if (!(poll->events & EPOLLONESHOT)) |
5033 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 5034 | |
8706e04e JA |
5035 | list_del_init(&wait->entry); |
5036 | ||
9ce85ef2 | 5037 | if (poll->head) { |
18bceab1 JA |
5038 | bool done; |
5039 | ||
79ebeaee | 5040 | spin_lock_irqsave(&poll->head->lock, flags); |
807abcb0 | 5041 | done = list_empty(&poll->wait.entry); |
18bceab1 | 5042 | if (!done) |
807abcb0 | 5043 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5044 | /* make sure double remove sees this as being gone */ |
5045 | wait->private = NULL; | |
79ebeaee | 5046 | spin_unlock_irqrestore(&poll->head->lock, flags); |
c8b5e260 JA |
5047 | if (!done) { |
5048 | /* use wait func handler, so it matches the rq type */ | |
5049 | poll->wait.func(&poll->wait, mode, sync, key); | |
5050 | } | |
18bceab1 | 5051 | } |
de9b4cca | 5052 | req_ref_put(req); |
18bceab1 JA |
5053 | return 1; |
5054 | } | |
5055 | ||
5056 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5057 | wait_queue_func_t wake_func) | |
5058 | { | |
5059 | poll->head = NULL; | |
5060 | poll->done = false; | |
5061 | poll->canceled = false; | |
464dca61 JA |
5062 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5063 | /* mask in events that we always want/need */ | |
5064 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5065 | INIT_LIST_HEAD(&poll->wait.entry); |
5066 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5067 | } | |
5068 | ||
5069 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5070 | struct wait_queue_head *head, |
5071 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5072 | { |
5073 | struct io_kiocb *req = pt->req; | |
5074 | ||
5075 | /* | |
68b11e8b PB |
5076 | * The file being polled uses multiple waitqueues for poll handling |
5077 | * (e.g. one for read, one for write). Setup a separate io_poll_iocb | |
5078 | * if this happens. | |
18bceab1 | 5079 | */ |
68b11e8b | 5080 | if (unlikely(pt->nr_entries)) { |
58852d4d PB |
5081 | struct io_poll_iocb *poll_one = poll; |
5082 | ||
23a65db8 PB |
5083 | /* double add on the same waitqueue head, ignore */ |
5084 | if (poll_one->head == head) | |
5085 | return; | |
18bceab1 | 5086 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5087 | if (*poll_ptr) { |
23a65db8 PB |
5088 | if ((*poll_ptr)->head == head) |
5089 | return; | |
18bceab1 JA |
5090 | pt->error = -EINVAL; |
5091 | return; | |
5092 | } | |
ea6a693d JA |
5093 | /* |
5094 | * Can't handle multishot for double wait for now, turn it | |
5095 | * into one-shot mode. | |
5096 | */ | |
7a274727 PB |
5097 | if (!(poll_one->events & EPOLLONESHOT)) |
5098 | poll_one->events |= EPOLLONESHOT; | |
18bceab1 JA |
5099 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5100 | if (!poll) { | |
5101 | pt->error = -ENOMEM; | |
5102 | return; | |
5103 | } | |
58852d4d | 5104 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5105 | req_ref_get(req); |
18bceab1 | 5106 | poll->wait.private = req; |
807abcb0 | 5107 | *poll_ptr = poll; |
18bceab1 JA |
5108 | } |
5109 | ||
68b11e8b | 5110 | pt->nr_entries++; |
18bceab1 | 5111 | poll->head = head; |
a31eb4a2 JX |
5112 | |
5113 | if (poll->events & EPOLLEXCLUSIVE) | |
5114 | add_wait_queue_exclusive(head, &poll->wait); | |
5115 | else | |
5116 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5117 | } |
5118 | ||
5119 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5120 | struct poll_table_struct *p) | |
5121 | { | |
5122 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5123 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5124 | |
807abcb0 | 5125 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5126 | } |
5127 | ||
5b0a6acc | 5128 | static void io_async_task_func(struct io_kiocb *req) |
d7718a9d | 5129 | { |
d7718a9d JA |
5130 | struct async_poll *apoll = req->apoll; |
5131 | struct io_ring_ctx *ctx = req->ctx; | |
5132 | ||
236daeae | 5133 | trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data); |
d7718a9d | 5134 | |
74ce6ce4 | 5135 | if (io_poll_rewait(req, &apoll->poll)) { |
79ebeaee | 5136 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5137 | return; |
d7718a9d JA |
5138 | } |
5139 | ||
0ea13b44 | 5140 | hash_del(&req->hash_node); |
d4e7cd36 | 5141 | io_poll_remove_double(req); |
79ebeaee | 5142 | spin_unlock(&ctx->completion_lock); |
74ce6ce4 | 5143 | |
0be0b0e3 | 5144 | if (!READ_ONCE(apoll->poll.canceled)) |
5b0a6acc | 5145 | io_req_task_submit(req); |
0be0b0e3 | 5146 | else |
2593553a | 5147 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5148 | } |
5149 | ||
5150 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5151 | void *key) | |
5152 | { | |
5153 | struct io_kiocb *req = wait->private; | |
5154 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5155 | ||
5156 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5157 | key_to_poll(key)); | |
5158 | ||
5159 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5160 | } | |
5161 | ||
5162 | static void io_poll_req_insert(struct io_kiocb *req) | |
5163 | { | |
5164 | struct io_ring_ctx *ctx = req->ctx; | |
5165 | struct hlist_head *list; | |
5166 | ||
5167 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5168 | hlist_add_head(&req->hash_node, list); | |
5169 | } | |
5170 | ||
5171 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5172 | struct io_poll_iocb *poll, | |
5173 | struct io_poll_table *ipt, __poll_t mask, | |
5174 | wait_queue_func_t wake_func) | |
5175 | __acquires(&ctx->completion_lock) | |
5176 | { | |
5177 | struct io_ring_ctx *ctx = req->ctx; | |
5178 | bool cancel = false; | |
5179 | ||
4d52f338 | 5180 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5181 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5182 | poll->file = req->file; |
18bceab1 | 5183 | poll->wait.private = req; |
d7718a9d JA |
5184 | |
5185 | ipt->pt._key = mask; | |
5186 | ipt->req = req; | |
68b11e8b PB |
5187 | ipt->error = 0; |
5188 | ipt->nr_entries = 0; | |
d7718a9d | 5189 | |
d7718a9d | 5190 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
68b11e8b PB |
5191 | if (unlikely(!ipt->nr_entries) && !ipt->error) |
5192 | ipt->error = -EINVAL; | |
d7718a9d | 5193 | |
79ebeaee | 5194 | spin_lock(&ctx->completion_lock); |
a890d01e | 5195 | if (ipt->error || (mask && (poll->events & EPOLLONESHOT))) |
46fee9ab | 5196 | io_poll_remove_double(req); |
d7718a9d | 5197 | if (likely(poll->head)) { |
79ebeaee | 5198 | spin_lock_irq(&poll->head->lock); |
d7718a9d JA |
5199 | if (unlikely(list_empty(&poll->wait.entry))) { |
5200 | if (ipt->error) | |
5201 | cancel = true; | |
5202 | ipt->error = 0; | |
5203 | mask = 0; | |
5204 | } | |
88e41cf9 | 5205 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5206 | list_del_init(&poll->wait.entry); |
5207 | else if (cancel) | |
5208 | WRITE_ONCE(poll->canceled, true); | |
5209 | else if (!poll->done) /* actually waiting for an event */ | |
5210 | io_poll_req_insert(req); | |
79ebeaee | 5211 | spin_unlock_irq(&poll->head->lock); |
d7718a9d JA |
5212 | } |
5213 | ||
5214 | return mask; | |
5215 | } | |
5216 | ||
59b735ae OL |
5217 | enum { |
5218 | IO_APOLL_OK, | |
5219 | IO_APOLL_ABORTED, | |
5220 | IO_APOLL_READY | |
5221 | }; | |
5222 | ||
5223 | static int io_arm_poll_handler(struct io_kiocb *req) | |
d7718a9d JA |
5224 | { |
5225 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5226 | struct io_ring_ctx *ctx = req->ctx; | |
5227 | struct async_poll *apoll; | |
5228 | struct io_poll_table ipt; | |
b2d9c3da | 5229 | __poll_t ret, mask = EPOLLONESHOT | POLLERR | POLLPRI; |
9dab14b8 | 5230 | int rw; |
d7718a9d JA |
5231 | |
5232 | if (!req->file || !file_can_poll(req->file)) | |
59b735ae | 5233 | return IO_APOLL_ABORTED; |
24c74678 | 5234 | if (req->flags & REQ_F_POLLED) |
59b735ae | 5235 | return IO_APOLL_ABORTED; |
b2d9c3da PB |
5236 | if (!def->pollin && !def->pollout) |
5237 | return IO_APOLL_ABORTED; | |
5238 | ||
5239 | if (def->pollin) { | |
9dab14b8 | 5240 | rw = READ; |
b2d9c3da PB |
5241 | mask |= POLLIN | POLLRDNORM; |
5242 | ||
5243 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5244 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5245 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5246 | mask &= ~POLLIN; | |
5247 | } else { | |
9dab14b8 | 5248 | rw = WRITE; |
b2d9c3da PB |
5249 | mask |= POLLOUT | POLLWRNORM; |
5250 | } | |
5251 | ||
9dab14b8 | 5252 | /* if we can't nonblock try, then no point in arming a poll handler */ |
b191e2df | 5253 | if (!io_file_supports_nowait(req, rw)) |
59b735ae | 5254 | return IO_APOLL_ABORTED; |
d7718a9d JA |
5255 | |
5256 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5257 | if (unlikely(!apoll)) | |
59b735ae | 5258 | return IO_APOLL_ABORTED; |
807abcb0 | 5259 | apoll->double_poll = NULL; |
d7718a9d | 5260 | req->apoll = apoll; |
b2d9c3da | 5261 | req->flags |= REQ_F_POLLED; |
d7718a9d | 5262 | ipt.pt._qproc = io_async_queue_proc; |
48dcd38d | 5263 | io_req_set_refcount(req); |
d7718a9d JA |
5264 | |
5265 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5266 | io_async_wake); | |
79ebeaee | 5267 | spin_unlock(&ctx->completion_lock); |
41a5169c HX |
5268 | if (ret || ipt.error) |
5269 | return ret ? IO_APOLL_READY : IO_APOLL_ABORTED; | |
5270 | ||
236daeae OL |
5271 | trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, |
5272 | mask, apoll->poll.events); | |
59b735ae | 5273 | return IO_APOLL_OK; |
d7718a9d JA |
5274 | } |
5275 | ||
5276 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5277 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5278 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5279 | { |
b41e9852 | 5280 | bool do_complete = false; |
221c5eb2 | 5281 | |
5082620f JA |
5282 | if (!poll->head) |
5283 | return false; | |
79ebeaee | 5284 | spin_lock_irq(&poll->head->lock); |
b2e720ac JA |
5285 | if (do_cancel) |
5286 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5287 | if (!list_empty(&poll->wait.entry)) { |
5288 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5289 | do_complete = true; |
221c5eb2 | 5290 | } |
79ebeaee | 5291 | spin_unlock_irq(&poll->head->lock); |
3bfa5bcb | 5292 | hash_del(&req->hash_node); |
d7718a9d JA |
5293 | return do_complete; |
5294 | } | |
5295 | ||
5d709043 | 5296 | static bool io_poll_remove_one(struct io_kiocb *req) |
e07785b0 | 5297 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5298 | { |
5299 | bool do_complete; | |
5300 | ||
d4e7cd36 | 5301 | io_poll_remove_double(req); |
e31001a3 | 5302 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5303 | |
b41e9852 | 5304 | if (do_complete) { |
d4d19c19 | 5305 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5306 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5307 | req_set_fail(req); |
91c2f697 | 5308 | io_put_req_deferred(req); |
5d709043 | 5309 | } |
b41e9852 | 5310 | return do_complete; |
221c5eb2 JA |
5311 | } |
5312 | ||
76e1b642 JA |
5313 | /* |
5314 | * Returns true if we found and killed one or more poll requests | |
5315 | */ | |
6b81928d | 5316 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5317 | bool cancel_all) |
221c5eb2 | 5318 | { |
78076bb6 | 5319 | struct hlist_node *tmp; |
221c5eb2 | 5320 | struct io_kiocb *req; |
8e2e1faf | 5321 | int posted = 0, i; |
221c5eb2 | 5322 | |
79ebeaee | 5323 | spin_lock(&ctx->completion_lock); |
78076bb6 JA |
5324 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5325 | struct hlist_head *list; | |
5326 | ||
5327 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5328 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5329 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5330 | posted += io_poll_remove_one(req); |
5331 | } | |
221c5eb2 | 5332 | } |
79ebeaee | 5333 | spin_unlock(&ctx->completion_lock); |
b41e9852 | 5334 | |
8e2e1faf JA |
5335 | if (posted) |
5336 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5337 | |
5338 | return posted != 0; | |
221c5eb2 JA |
5339 | } |
5340 | ||
9ba5fac8 PB |
5341 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5342 | bool poll_only) | |
e07785b0 | 5343 | __must_hold(&ctx->completion_lock) |
47f46768 | 5344 | { |
78076bb6 | 5345 | struct hlist_head *list; |
47f46768 JA |
5346 | struct io_kiocb *req; |
5347 | ||
78076bb6 JA |
5348 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5349 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5350 | if (sqe_addr != req->user_data) |
5351 | continue; | |
9ba5fac8 PB |
5352 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5353 | continue; | |
b2cb805f | 5354 | return req; |
47f46768 | 5355 | } |
b2cb805f JA |
5356 | return NULL; |
5357 | } | |
5358 | ||
9ba5fac8 PB |
5359 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5360 | bool poll_only) | |
e07785b0 | 5361 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5362 | { |
5363 | struct io_kiocb *req; | |
5364 | ||
9ba5fac8 | 5365 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5366 | if (!req) |
5367 | return -ENOENT; | |
5368 | if (io_poll_remove_one(req)) | |
5369 | return 0; | |
5370 | ||
5371 | return -EALREADY; | |
47f46768 JA |
5372 | } |
5373 | ||
9096af3e PB |
5374 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5375 | unsigned int flags) | |
5376 | { | |
5377 | u32 events; | |
47f46768 | 5378 | |
9096af3e PB |
5379 | events = READ_ONCE(sqe->poll32_events); |
5380 | #ifdef __BIG_ENDIAN | |
5381 | events = swahw32(events); | |
5382 | #endif | |
5383 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5384 | events |= EPOLLONESHOT; | |
5385 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5386 | } |
5387 | ||
c5de0036 | 5388 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5389 | const struct io_uring_sqe *sqe) |
0969e783 | 5390 | { |
c5de0036 PB |
5391 | struct io_poll_update *upd = &req->poll_update; |
5392 | u32 flags; | |
5393 | ||
0969e783 JA |
5394 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5395 | return -EINVAL; | |
26578cda | 5396 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
c5de0036 PB |
5397 | return -EINVAL; |
5398 | flags = READ_ONCE(sqe->len); | |
5399 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5400 | IORING_POLL_ADD_MULTI)) | |
5401 | return -EINVAL; | |
5402 | /* meaningless without update */ | |
5403 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5404 | return -EINVAL; |
5405 | ||
c5de0036 PB |
5406 | upd->old_user_data = READ_ONCE(sqe->addr); |
5407 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5408 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5409 | |
c5de0036 PB |
5410 | upd->new_user_data = READ_ONCE(sqe->off); |
5411 | if (!upd->update_user_data && upd->new_user_data) | |
5412 | return -EINVAL; | |
5413 | if (upd->update_events) | |
5414 | upd->events = io_poll_parse_events(sqe, flags); | |
5415 | else if (sqe->poll32_events) | |
5416 | return -EINVAL; | |
221c5eb2 | 5417 | |
221c5eb2 JA |
5418 | return 0; |
5419 | } | |
5420 | ||
221c5eb2 JA |
5421 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5422 | void *key) | |
5423 | { | |
c2f2eb7d JA |
5424 | struct io_kiocb *req = wait->private; |
5425 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5426 | |
d7718a9d | 5427 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5428 | } |
5429 | ||
221c5eb2 JA |
5430 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5431 | struct poll_table_struct *p) | |
5432 | { | |
5433 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5434 | ||
e8c2bc1f | 5435 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5436 | } |
5437 | ||
3529d8c2 | 5438 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5439 | { |
5440 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5441 | u32 flags; |
221c5eb2 JA |
5442 | |
5443 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5444 | return -EINVAL; | |
c5de0036 | 5445 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5446 | return -EINVAL; |
5447 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5448 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5449 | return -EINVAL; |
5450 | ||
48dcd38d | 5451 | io_req_set_refcount(req); |
c5de0036 | 5452 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5453 | return 0; |
5454 | } | |
5455 | ||
61e98203 | 5456 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5457 | { |
5458 | struct io_poll_iocb *poll = &req->poll; | |
5459 | struct io_ring_ctx *ctx = req->ctx; | |
5460 | struct io_poll_table ipt; | |
0969e783 | 5461 | __poll_t mask; |
0969e783 | 5462 | |
d7718a9d | 5463 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5464 | |
d7718a9d JA |
5465 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5466 | io_poll_wake); | |
221c5eb2 | 5467 | |
8c838788 | 5468 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5469 | ipt.error = 0; |
e27414be | 5470 | io_poll_complete(req, mask); |
221c5eb2 | 5471 | } |
79ebeaee | 5472 | spin_unlock(&ctx->completion_lock); |
221c5eb2 | 5473 | |
8c838788 JA |
5474 | if (mask) { |
5475 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5476 | if (poll->events & EPOLLONESHOT) |
5477 | io_put_req(req); | |
221c5eb2 | 5478 | } |
8c838788 | 5479 | return ipt.error; |
221c5eb2 JA |
5480 | } |
5481 | ||
c5de0036 | 5482 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5483 | { |
5484 | struct io_ring_ctx *ctx = req->ctx; | |
5485 | struct io_kiocb *preq; | |
cb3b200e | 5486 | bool completing; |
b69de288 JA |
5487 | int ret; |
5488 | ||
79ebeaee | 5489 | spin_lock(&ctx->completion_lock); |
9ba5fac8 | 5490 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5491 | if (!preq) { |
5492 | ret = -ENOENT; | |
5493 | goto err; | |
b69de288 | 5494 | } |
cb3b200e | 5495 | |
c5de0036 PB |
5496 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5497 | completing = true; | |
5498 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5499 | goto err; | |
5500 | } | |
5501 | ||
cb3b200e JA |
5502 | /* |
5503 | * Don't allow racy completion with singleshot, as we cannot safely | |
5504 | * update those. For multishot, if we're racing with completion, just | |
5505 | * let completion re-add it. | |
5506 | */ | |
5507 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5508 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5509 | ret = -EALREADY; | |
5510 | goto err; | |
b69de288 JA |
5511 | } |
5512 | /* we now have a detached poll request. reissue. */ | |
5513 | ret = 0; | |
5514 | err: | |
b69de288 | 5515 | if (ret < 0) { |
79ebeaee | 5516 | spin_unlock(&ctx->completion_lock); |
93d2bcd2 | 5517 | req_set_fail(req); |
b69de288 JA |
5518 | io_req_complete(req, ret); |
5519 | return 0; | |
5520 | } | |
5521 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5522 | if (req->poll_update.update_events) { |
b69de288 | 5523 | preq->poll.events &= ~0xffff; |
9d805892 | 5524 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5525 | preq->poll.events |= IO_POLL_UNMASK; |
5526 | } | |
9d805892 PB |
5527 | if (req->poll_update.update_user_data) |
5528 | preq->user_data = req->poll_update.new_user_data; | |
79ebeaee | 5529 | spin_unlock(&ctx->completion_lock); |
cb3b200e | 5530 | |
b69de288 JA |
5531 | /* complete update request, we're done with it */ |
5532 | io_req_complete(req, ret); | |
5533 | ||
cb3b200e | 5534 | if (!completing) { |
c5de0036 | 5535 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5536 | if (ret < 0) { |
93d2bcd2 | 5537 | req_set_fail(preq); |
cb3b200e JA |
5538 | io_req_complete(preq, ret); |
5539 | } | |
b69de288 JA |
5540 | } |
5541 | return 0; | |
5542 | } | |
5543 | ||
89850fce JA |
5544 | static void io_req_task_timeout(struct io_kiocb *req) |
5545 | { | |
89850fce | 5546 | req_set_fail(req); |
505657bc | 5547 | io_req_complete_post(req, -ETIME, 0); |
89850fce JA |
5548 | } |
5549 | ||
5262f567 JA |
5550 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5551 | { | |
ad8a48ac JA |
5552 | struct io_timeout_data *data = container_of(timer, |
5553 | struct io_timeout_data, timer); | |
5554 | struct io_kiocb *req = data->req; | |
5555 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5556 | unsigned long flags; |
5557 | ||
89850fce | 5558 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
a71976f3 | 5559 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5560 | atomic_set(&req->ctx->cq_timeouts, |
5561 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
89850fce | 5562 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); |
01cec8c1 | 5563 | |
89850fce JA |
5564 | req->io_task_work.func = io_req_task_timeout; |
5565 | io_req_task_work_add(req); | |
5262f567 JA |
5566 | return HRTIMER_NORESTART; |
5567 | } | |
5568 | ||
fbd15848 PB |
5569 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5570 | __u64 user_data) | |
89850fce | 5571 | __must_hold(&ctx->timeout_lock) |
f254ac04 | 5572 | { |
fbd15848 | 5573 | struct io_timeout_data *io; |
47f46768 | 5574 | struct io_kiocb *req; |
fd9c7bc5 | 5575 | bool found = false; |
f254ac04 | 5576 | |
135fcde8 | 5577 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5578 | found = user_data == req->user_data; |
5579 | if (found) | |
47f46768 | 5580 | break; |
47f46768 | 5581 | } |
fd9c7bc5 PB |
5582 | if (!found) |
5583 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5584 | |
5585 | io = req->async_data; | |
fd9c7bc5 | 5586 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5587 | return ERR_PTR(-EALREADY); |
a71976f3 | 5588 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5589 | return req; |
5590 | } | |
47f46768 | 5591 | |
fbd15848 | 5592 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
ec3c3d0f | 5593 | __must_hold(&ctx->completion_lock) |
89850fce | 5594 | __must_hold(&ctx->timeout_lock) |
fbd15848 PB |
5595 | { |
5596 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5597 | ||
5598 | if (IS_ERR(req)) | |
5599 | return PTR_ERR(req); | |
f254ac04 | 5600 | |
93d2bcd2 | 5601 | req_set_fail(req); |
d4d19c19 | 5602 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
91c2f697 | 5603 | io_put_req_deferred(req); |
f254ac04 JA |
5604 | return 0; |
5605 | } | |
5606 | ||
9c8e11b3 PB |
5607 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5608 | struct timespec64 *ts, enum hrtimer_mode mode) | |
89850fce | 5609 | __must_hold(&ctx->timeout_lock) |
47f46768 | 5610 | { |
9c8e11b3 PB |
5611 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5612 | struct io_timeout_data *data; | |
47f46768 | 5613 | |
9c8e11b3 PB |
5614 | if (IS_ERR(req)) |
5615 | return PTR_ERR(req); | |
47f46768 | 5616 | |
9c8e11b3 PB |
5617 | req->timeout.off = 0; /* noseq */ |
5618 | data = req->async_data; | |
5619 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5620 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5621 | data->timer.function = io_timeout_fn; | |
5622 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5623 | return 0; | |
47f46768 JA |
5624 | } |
5625 | ||
3529d8c2 JA |
5626 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5627 | const struct io_uring_sqe *sqe) | |
b29472ee | 5628 | { |
9c8e11b3 PB |
5629 | struct io_timeout_rem *tr = &req->timeout_rem; |
5630 | ||
b29472ee JA |
5631 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5632 | return -EINVAL; | |
61710e43 DA |
5633 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5634 | return -EINVAL; | |
26578cda | 5635 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->splice_fd_in) |
b29472ee JA |
5636 | return -EINVAL; |
5637 | ||
9c8e11b3 PB |
5638 | tr->addr = READ_ONCE(sqe->addr); |
5639 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5640 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5641 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5642 | return -EINVAL; | |
5643 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5644 | return -EFAULT; | |
5645 | } else if (tr->flags) { | |
5646 | /* timeout removal doesn't support flags */ | |
b29472ee | 5647 | return -EINVAL; |
9c8e11b3 | 5648 | } |
b29472ee | 5649 | |
b29472ee JA |
5650 | return 0; |
5651 | } | |
5652 | ||
8662daec PB |
5653 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5654 | { | |
5655 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5656 | : HRTIMER_MODE_REL; | |
5657 | } | |
5658 | ||
11365043 JA |
5659 | /* |
5660 | * Remove or update an existing timeout command | |
5661 | */ | |
61e98203 | 5662 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5663 | { |
9c8e11b3 | 5664 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5665 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5666 | int ret; |
11365043 | 5667 | |
ec3c3d0f PB |
5668 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { |
5669 | spin_lock(&ctx->completion_lock); | |
5670 | spin_lock_irq(&ctx->timeout_lock); | |
9c8e11b3 | 5671 | ret = io_timeout_cancel(ctx, tr->addr); |
ec3c3d0f PB |
5672 | spin_unlock_irq(&ctx->timeout_lock); |
5673 | spin_unlock(&ctx->completion_lock); | |
5674 | } else { | |
5675 | spin_lock_irq(&ctx->timeout_lock); | |
8662daec PB |
5676 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, |
5677 | io_translate_timeout_mode(tr->flags)); | |
ec3c3d0f PB |
5678 | spin_unlock_irq(&ctx->timeout_lock); |
5679 | } | |
11365043 | 5680 | |
4e88d6e7 | 5681 | if (ret < 0) |
93d2bcd2 | 5682 | req_set_fail(req); |
505657bc | 5683 | io_req_complete_post(req, ret, 0); |
11365043 | 5684 | return 0; |
5262f567 JA |
5685 | } |
5686 | ||
3529d8c2 | 5687 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5688 | bool is_timeout_link) |
5262f567 | 5689 | { |
ad8a48ac | 5690 | struct io_timeout_data *data; |
a41525ab | 5691 | unsigned flags; |
56080b02 | 5692 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5693 | |
ad8a48ac | 5694 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5695 | return -EINVAL; |
26578cda PB |
5696 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1 || |
5697 | sqe->splice_fd_in) | |
a41525ab | 5698 | return -EINVAL; |
56080b02 | 5699 | if (off && is_timeout_link) |
2d28390a | 5700 | return -EINVAL; |
a41525ab JA |
5701 | flags = READ_ONCE(sqe->timeout_flags); |
5702 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5703 | return -EINVAL; |
bdf20073 | 5704 | |
bfe68a22 | 5705 | req->timeout.off = off; |
f18ee4cf PB |
5706 | if (unlikely(off && !req->ctx->off_timeout_used)) |
5707 | req->ctx->off_timeout_used = true; | |
26a61679 | 5708 | |
e8c2bc1f | 5709 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5710 | return -ENOMEM; |
5711 | ||
e8c2bc1f | 5712 | data = req->async_data; |
ad8a48ac | 5713 | data->req = req; |
ad8a48ac JA |
5714 | |
5715 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5716 | return -EFAULT; |
5717 | ||
8662daec | 5718 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5719 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
b97e736a PB |
5720 | |
5721 | if (is_timeout_link) { | |
5722 | struct io_submit_link *link = &req->ctx->submit_state.link; | |
5723 | ||
5724 | if (!link->head) | |
5725 | return -EINVAL; | |
5726 | if (link->last->opcode == IORING_OP_LINK_TIMEOUT) | |
5727 | return -EINVAL; | |
4d13d1a4 PB |
5728 | req->timeout.head = link->last; |
5729 | link->last->flags |= REQ_F_ARM_LTIMEOUT; | |
b97e736a | 5730 | } |
ad8a48ac JA |
5731 | return 0; |
5732 | } | |
5733 | ||
61e98203 | 5734 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5735 | { |
ad8a48ac | 5736 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5737 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5738 | struct list_head *entry; |
bfe68a22 | 5739 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5740 | |
89850fce | 5741 | spin_lock_irq(&ctx->timeout_lock); |
93bd25bb | 5742 | |
5262f567 JA |
5743 | /* |
5744 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5745 | * timeout event to be satisfied. If it isn't set, then this is |
5746 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5747 | */ |
8eb7e2d0 | 5748 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5749 | entry = ctx->timeout_list.prev; |
5750 | goto add; | |
5751 | } | |
5262f567 | 5752 | |
bfe68a22 PB |
5753 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5754 | req->timeout.target_seq = tail + off; | |
5262f567 | 5755 | |
f010505b MDG |
5756 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5757 | * This is safe because ->completion_lock is held, and submissions | |
5758 | * and completions are never mixed in the same ->completion_lock section. | |
5759 | */ | |
5760 | ctx->cq_last_tm_flush = tail; | |
5761 | ||
5262f567 JA |
5762 | /* |
5763 | * Insertion sort, ensuring the first entry in the list is always | |
5764 | * the one we need first. | |
5765 | */ | |
5262f567 | 5766 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5767 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5768 | timeout.list); | |
5262f567 | 5769 | |
8eb7e2d0 | 5770 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5771 | continue; |
bfe68a22 PB |
5772 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5773 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5774 | break; |
5775 | } | |
93bd25bb | 5776 | add: |
135fcde8 | 5777 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5778 | data->timer.function = io_timeout_fn; |
5779 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
89850fce | 5780 | spin_unlock_irq(&ctx->timeout_lock); |
5262f567 JA |
5781 | return 0; |
5782 | } | |
5262f567 | 5783 | |
f458dd84 PB |
5784 | struct io_cancel_data { |
5785 | struct io_ring_ctx *ctx; | |
5786 | u64 user_data; | |
5787 | }; | |
5788 | ||
62755e35 JA |
5789 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5790 | { | |
5791 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5792 | struct io_cancel_data *cd = data; |
62755e35 | 5793 | |
f458dd84 | 5794 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5795 | } |
5796 | ||
f458dd84 PB |
5797 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5798 | struct io_ring_ctx *ctx) | |
62755e35 | 5799 | { |
f458dd84 | 5800 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5801 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5802 | int ret = 0; |
5803 | ||
f458dd84 | 5804 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5805 | return -ENOENT; |
5806 | ||
f458dd84 | 5807 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5808 | switch (cancel_ret) { |
5809 | case IO_WQ_CANCEL_OK: | |
5810 | ret = 0; | |
5811 | break; | |
5812 | case IO_WQ_CANCEL_RUNNING: | |
5813 | ret = -EALREADY; | |
5814 | break; | |
5815 | case IO_WQ_CANCEL_NOTFOUND: | |
5816 | ret = -ENOENT; | |
5817 | break; | |
5818 | } | |
5819 | ||
e977d6d3 JA |
5820 | return ret; |
5821 | } | |
5822 | ||
8cb01fac | 5823 | static int io_try_cancel_userdata(struct io_kiocb *req, u64 sqe_addr) |
47f46768 | 5824 | { |
8cb01fac | 5825 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 JA |
5826 | int ret; |
5827 | ||
8cb01fac PB |
5828 | WARN_ON_ONCE(req->task != current); |
5829 | ||
f458dd84 | 5830 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
df9727af | 5831 | if (ret != -ENOENT) |
8cb01fac | 5832 | return ret; |
505657bc PB |
5833 | |
5834 | spin_lock(&ctx->completion_lock); | |
79ebeaee | 5835 | spin_lock_irq(&ctx->timeout_lock); |
47f46768 | 5836 | ret = io_timeout_cancel(ctx, sqe_addr); |
79ebeaee | 5837 | spin_unlock_irq(&ctx->timeout_lock); |
47f46768 | 5838 | if (ret != -ENOENT) |
505657bc PB |
5839 | goto out; |
5840 | ret = io_poll_cancel(ctx, sqe_addr, false); | |
5841 | out: | |
5842 | spin_unlock(&ctx->completion_lock); | |
5843 | return ret; | |
47f46768 JA |
5844 | } |
5845 | ||
3529d8c2 JA |
5846 | static int io_async_cancel_prep(struct io_kiocb *req, |
5847 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5848 | { |
fbf23849 | 5849 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5850 | return -EINVAL; |
61710e43 DA |
5851 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5852 | return -EINVAL; | |
26578cda PB |
5853 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags || |
5854 | sqe->splice_fd_in) | |
e977d6d3 JA |
5855 | return -EINVAL; |
5856 | ||
fbf23849 JA |
5857 | req->cancel.addr = READ_ONCE(sqe->addr); |
5858 | return 0; | |
5859 | } | |
5860 | ||
61e98203 | 5861 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5862 | { |
5863 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5864 | u64 sqe_addr = req->cancel.addr; |
5865 | struct io_tctx_node *node; | |
5866 | int ret; | |
5867 | ||
8cb01fac | 5868 | ret = io_try_cancel_userdata(req, sqe_addr); |
58f99373 PB |
5869 | if (ret != -ENOENT) |
5870 | goto done; | |
58f99373 PB |
5871 | |
5872 | /* slow path, try all io-wq's */ | |
5873 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5874 | ret = -ENOENT; | |
5875 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5876 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5877 | |
58f99373 PB |
5878 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5879 | if (ret != -ENOENT) | |
5880 | break; | |
5881 | } | |
5882 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
58f99373 | 5883 | done: |
58f99373 | 5884 | if (ret < 0) |
93d2bcd2 | 5885 | req_set_fail(req); |
505657bc | 5886 | io_req_complete_post(req, ret, 0); |
5262f567 JA |
5887 | return 0; |
5888 | } | |
5889 | ||
269bbe5f | 5890 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5891 | const struct io_uring_sqe *sqe) |
5892 | { | |
61710e43 DA |
5893 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5894 | return -EINVAL; | |
26578cda | 5895 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
05f3fb3c JA |
5896 | return -EINVAL; |
5897 | ||
269bbe5f BM |
5898 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5899 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5900 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5901 | return -EINVAL; |
269bbe5f | 5902 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5903 | return 0; |
5904 | } | |
5905 | ||
889fca73 | 5906 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5907 | { |
5908 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 5909 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 5910 | int ret; |
fbf23849 | 5911 | |
45d189c6 | 5912 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5913 | return -EAGAIN; |
05f3fb3c | 5914 | |
269bbe5f BM |
5915 | up.offset = req->rsrc_update.offset; |
5916 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
5917 | up.nr = 0; |
5918 | up.tags = 0; | |
615cee49 | 5919 | up.resv = 0; |
05f3fb3c JA |
5920 | |
5921 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 5922 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 5923 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5924 | mutex_unlock(&ctx->uring_lock); |
5925 | ||
5926 | if (ret < 0) | |
93d2bcd2 | 5927 | req_set_fail(req); |
889fca73 | 5928 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5929 | return 0; |
5930 | } | |
5931 | ||
bfe76559 | 5932 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5933 | { |
d625c6ee | 5934 | switch (req->opcode) { |
e781573e | 5935 | case IORING_OP_NOP: |
bfe76559 | 5936 | return 0; |
f67676d1 JA |
5937 | case IORING_OP_READV: |
5938 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5939 | case IORING_OP_READ: |
bfe76559 | 5940 | return io_read_prep(req, sqe); |
f67676d1 JA |
5941 | case IORING_OP_WRITEV: |
5942 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5943 | case IORING_OP_WRITE: |
bfe76559 | 5944 | return io_write_prep(req, sqe); |
0969e783 | 5945 | case IORING_OP_POLL_ADD: |
bfe76559 | 5946 | return io_poll_add_prep(req, sqe); |
0969e783 | 5947 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5948 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5949 | case IORING_OP_FSYNC: |
1155c76a | 5950 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5951 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5952 | return io_sfr_prep(req, sqe); |
03b1230c | 5953 | case IORING_OP_SENDMSG: |
fddaface | 5954 | case IORING_OP_SEND: |
bfe76559 | 5955 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5956 | case IORING_OP_RECVMSG: |
fddaface | 5957 | case IORING_OP_RECV: |
bfe76559 | 5958 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5959 | case IORING_OP_CONNECT: |
bfe76559 | 5960 | return io_connect_prep(req, sqe); |
2d28390a | 5961 | case IORING_OP_TIMEOUT: |
bfe76559 | 5962 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5963 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5964 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5965 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5966 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5967 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5968 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5969 | case IORING_OP_ACCEPT: |
bfe76559 | 5970 | return io_accept_prep(req, sqe); |
d63d1b5e | 5971 | case IORING_OP_FALLOCATE: |
bfe76559 | 5972 | return io_fallocate_prep(req, sqe); |
15b71abe | 5973 | case IORING_OP_OPENAT: |
bfe76559 | 5974 | return io_openat_prep(req, sqe); |
b5dba59e | 5975 | case IORING_OP_CLOSE: |
bfe76559 | 5976 | return io_close_prep(req, sqe); |
05f3fb3c | 5977 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5978 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5979 | case IORING_OP_STATX: |
bfe76559 | 5980 | return io_statx_prep(req, sqe); |
4840e418 | 5981 | case IORING_OP_FADVISE: |
bfe76559 | 5982 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5983 | case IORING_OP_MADVISE: |
bfe76559 | 5984 | return io_madvise_prep(req, sqe); |
cebdb986 | 5985 | case IORING_OP_OPENAT2: |
bfe76559 | 5986 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5987 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5988 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5989 | case IORING_OP_SPLICE: |
bfe76559 | 5990 | return io_splice_prep(req, sqe); |
ddf0322d | 5991 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5992 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5993 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5994 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5995 | case IORING_OP_TEE: |
bfe76559 | 5996 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5997 | case IORING_OP_SHUTDOWN: |
5998 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5999 | case IORING_OP_RENAMEAT: |
6000 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6001 | case IORING_OP_UNLINKAT: |
6002 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
6003 | } |
6004 | ||
bfe76559 PB |
6005 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6006 | req->opcode); | |
bd54b6fe | 6007 | return -EINVAL; |
bfe76559 PB |
6008 | } |
6009 | ||
93642ef8 | 6010 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 6011 | { |
b7e298d2 PB |
6012 | if (!io_op_defs[req->opcode].needs_async_setup) |
6013 | return 0; | |
6014 | if (WARN_ON_ONCE(req->async_data)) | |
6015 | return -EFAULT; | |
6016 | if (io_alloc_async_data(req)) | |
6017 | return -EAGAIN; | |
6018 | ||
93642ef8 PB |
6019 | switch (req->opcode) { |
6020 | case IORING_OP_READV: | |
93642ef8 PB |
6021 | return io_rw_prep_async(req, READ); |
6022 | case IORING_OP_WRITEV: | |
93642ef8 PB |
6023 | return io_rw_prep_async(req, WRITE); |
6024 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
6025 | return io_sendmsg_prep_async(req); |
6026 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
6027 | return io_recvmsg_prep_async(req); |
6028 | case IORING_OP_CONNECT: | |
6029 | return io_connect_prep_async(req); | |
6030 | } | |
b7e298d2 PB |
6031 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
6032 | req->opcode); | |
6033 | return -EFAULT; | |
f67676d1 JA |
6034 | } |
6035 | ||
9cf7c104 PB |
6036 | static u32 io_get_sequence(struct io_kiocb *req) |
6037 | { | |
a3dbdf54 | 6038 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 6039 | |
a3dbdf54 PB |
6040 | /* need original cached_sq_head, but it was increased for each req */ |
6041 | io_for_each_link(req, req) | |
6042 | seq--; | |
6043 | return seq; | |
9cf7c104 PB |
6044 | } |
6045 | ||
76cc33d7 | 6046 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 6047 | { |
3c19966d | 6048 | struct io_kiocb *pos; |
a197f664 | 6049 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6050 | struct io_defer_entry *de; |
f67676d1 | 6051 | int ret; |
9cf7c104 | 6052 | u32 seq; |
de0617e4 | 6053 | |
3c19966d PB |
6054 | /* |
6055 | * If we need to drain a request in the middle of a link, drain the | |
6056 | * head request and the next request/link after the current link. | |
6057 | * Considering sequential execution of links, IOSQE_IO_DRAIN will be | |
6058 | * maintained for every request of our link. | |
6059 | */ | |
6060 | if (ctx->drain_next) { | |
6061 | req->flags |= REQ_F_IO_DRAIN; | |
6062 | ctx->drain_next = false; | |
6063 | } | |
6064 | /* not interested in head, start from the first linked */ | |
6065 | io_for_each_link(pos, req->link) { | |
6066 | if (pos->flags & REQ_F_IO_DRAIN) { | |
6067 | ctx->drain_next = true; | |
6068 | req->flags |= REQ_F_IO_DRAIN; | |
6069 | break; | |
6070 | } | |
6071 | } | |
6072 | ||
9d858b21 | 6073 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 | 6074 | if (likely(list_empty_careful(&ctx->defer_list) && |
10c66904 PB |
6075 | !(req->flags & REQ_F_IO_DRAIN))) { |
6076 | ctx->drain_active = false; | |
76cc33d7 | 6077 | return false; |
10c66904 | 6078 | } |
9cf7c104 PB |
6079 | |
6080 | seq = io_get_sequence(req); | |
6081 | /* Still a chance to pass the sequence check */ | |
6082 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6083 | return false; |
de0617e4 | 6084 | |
b7e298d2 | 6085 | ret = io_req_prep_async(req); |
be7053b7 | 6086 | if (ret) |
1b48773f | 6087 | goto fail; |
cbdcb435 | 6088 | io_prep_async_link(req); |
27dc8338 | 6089 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 6090 | if (!de) { |
1b48773f PB |
6091 | ret = -ENOMEM; |
6092 | fail: | |
6093 | io_req_complete_failed(req, ret); | |
76cc33d7 PB |
6094 | return true; |
6095 | } | |
2d28390a | 6096 | |
79ebeaee | 6097 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 6098 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
79ebeaee | 6099 | spin_unlock(&ctx->completion_lock); |
27dc8338 | 6100 | kfree(de); |
ae34817b | 6101 | io_queue_async_work(req); |
76cc33d7 | 6102 | return true; |
de0617e4 JA |
6103 | } |
6104 | ||
915967f6 | 6105 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6106 | de->req = req; |
9cf7c104 | 6107 | de->seq = seq; |
27dc8338 | 6108 | list_add_tail(&de->list, &ctx->defer_list); |
79ebeaee | 6109 | spin_unlock(&ctx->completion_lock); |
76cc33d7 | 6110 | return true; |
de0617e4 JA |
6111 | } |
6112 | ||
68fb8979 | 6113 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6114 | { |
0e1b6fe3 PB |
6115 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6116 | switch (req->opcode) { | |
6117 | case IORING_OP_READV: | |
6118 | case IORING_OP_READ_FIXED: | |
6119 | case IORING_OP_READ: | |
bcda7baa | 6120 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6121 | break; |
6122 | case IORING_OP_RECVMSG: | |
6123 | case IORING_OP_RECV: | |
bcda7baa | 6124 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6125 | break; |
6126 | } | |
99bc4c38 PB |
6127 | } |
6128 | ||
0e1b6fe3 PB |
6129 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6130 | switch (req->opcode) { | |
6131 | case IORING_OP_READV: | |
6132 | case IORING_OP_READ_FIXED: | |
6133 | case IORING_OP_READ: | |
6134 | case IORING_OP_WRITEV: | |
6135 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6136 | case IORING_OP_WRITE: { |
6137 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
6138 | |
6139 | kfree(io->free_iovec); | |
0e1b6fe3 | 6140 | break; |
e8c2bc1f | 6141 | } |
0e1b6fe3 | 6142 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6143 | case IORING_OP_SENDMSG: { |
6144 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6145 | |
6146 | kfree(io->free_iov); | |
0e1b6fe3 | 6147 | break; |
e8c2bc1f | 6148 | } |
0e1b6fe3 PB |
6149 | case IORING_OP_SPLICE: |
6150 | case IORING_OP_TEE: | |
e1d767f0 PB |
6151 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6152 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6153 | break; |
f3cd4850 JA |
6154 | case IORING_OP_OPENAT: |
6155 | case IORING_OP_OPENAT2: | |
6156 | if (req->open.filename) | |
6157 | putname(req->open.filename); | |
6158 | break; | |
80a261fd JA |
6159 | case IORING_OP_RENAMEAT: |
6160 | putname(req->rename.oldpath); | |
6161 | putname(req->rename.newpath); | |
6162 | break; | |
14a1143b JA |
6163 | case IORING_OP_UNLINKAT: |
6164 | putname(req->unlink.filename); | |
6165 | break; | |
0e1b6fe3 | 6166 | } |
99bc4c38 | 6167 | } |
75652a30 JA |
6168 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6169 | kfree(req->apoll->double_poll); | |
6170 | kfree(req->apoll); | |
6171 | req->apoll = NULL; | |
6172 | } | |
3a0a6902 PB |
6173 | if (req->flags & REQ_F_INFLIGHT) { |
6174 | struct io_uring_task *tctx = req->task->io_uring; | |
6175 | ||
6176 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 6177 | } |
c854357b | 6178 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 6179 | put_cred(req->creds); |
c854357b PB |
6180 | |
6181 | req->flags &= ~IO_REQ_CLEAN_FLAGS; | |
99bc4c38 PB |
6182 | } |
6183 | ||
889fca73 | 6184 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6185 | { |
a197f664 | 6186 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6187 | const struct cred *creds = NULL; |
d625c6ee | 6188 | int ret; |
2b188cc1 | 6189 | |
b8e64b53 | 6190 | if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) |
c10d1f98 | 6191 | creds = override_creds(req->creds); |
5730b27e | 6192 | |
d625c6ee | 6193 | switch (req->opcode) { |
2b188cc1 | 6194 | case IORING_OP_NOP: |
889fca73 | 6195 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6196 | break; |
6197 | case IORING_OP_READV: | |
edafccee | 6198 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6199 | case IORING_OP_READ: |
889fca73 | 6200 | ret = io_read(req, issue_flags); |
edafccee | 6201 | break; |
3529d8c2 | 6202 | case IORING_OP_WRITEV: |
edafccee | 6203 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6204 | case IORING_OP_WRITE: |
889fca73 | 6205 | ret = io_write(req, issue_flags); |
2b188cc1 | 6206 | break; |
c992fe29 | 6207 | case IORING_OP_FSYNC: |
45d189c6 | 6208 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6209 | break; |
221c5eb2 | 6210 | case IORING_OP_POLL_ADD: |
61e98203 | 6211 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6212 | break; |
6213 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6214 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6215 | break; |
5d17b4a4 | 6216 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6217 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6218 | break; |
0fa03c62 | 6219 | case IORING_OP_SENDMSG: |
889fca73 | 6220 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6221 | break; |
fddaface | 6222 | case IORING_OP_SEND: |
889fca73 | 6223 | ret = io_send(req, issue_flags); |
0fa03c62 | 6224 | break; |
aa1fa28f | 6225 | case IORING_OP_RECVMSG: |
889fca73 | 6226 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6227 | break; |
fddaface | 6228 | case IORING_OP_RECV: |
889fca73 | 6229 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6230 | break; |
5262f567 | 6231 | case IORING_OP_TIMEOUT: |
61e98203 | 6232 | ret = io_timeout(req, issue_flags); |
5262f567 | 6233 | break; |
11365043 | 6234 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6235 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6236 | break; |
17f2fe35 | 6237 | case IORING_OP_ACCEPT: |
889fca73 | 6238 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6239 | break; |
f8e85cf2 | 6240 | case IORING_OP_CONNECT: |
889fca73 | 6241 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6242 | break; |
62755e35 | 6243 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6244 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6245 | break; |
d63d1b5e | 6246 | case IORING_OP_FALLOCATE: |
45d189c6 | 6247 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6248 | break; |
15b71abe | 6249 | case IORING_OP_OPENAT: |
45d189c6 | 6250 | ret = io_openat(req, issue_flags); |
15b71abe | 6251 | break; |
b5dba59e | 6252 | case IORING_OP_CLOSE: |
889fca73 | 6253 | ret = io_close(req, issue_flags); |
b5dba59e | 6254 | break; |
05f3fb3c | 6255 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6256 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6257 | break; |
eddc7ef5 | 6258 | case IORING_OP_STATX: |
45d189c6 | 6259 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6260 | break; |
4840e418 | 6261 | case IORING_OP_FADVISE: |
45d189c6 | 6262 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6263 | break; |
c1ca757b | 6264 | case IORING_OP_MADVISE: |
45d189c6 | 6265 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6266 | break; |
cebdb986 | 6267 | case IORING_OP_OPENAT2: |
45d189c6 | 6268 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6269 | break; |
3e4827b0 | 6270 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6271 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6272 | break; |
7d67af2c | 6273 | case IORING_OP_SPLICE: |
45d189c6 | 6274 | ret = io_splice(req, issue_flags); |
7d67af2c | 6275 | break; |
ddf0322d | 6276 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6277 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6278 | break; |
067524e9 | 6279 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6280 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6281 | break; |
f2a8d5c7 | 6282 | case IORING_OP_TEE: |
45d189c6 | 6283 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6284 | break; |
36f4fa68 | 6285 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6286 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6287 | break; |
80a261fd | 6288 | case IORING_OP_RENAMEAT: |
45d189c6 | 6289 | ret = io_renameat(req, issue_flags); |
80a261fd | 6290 | break; |
14a1143b | 6291 | case IORING_OP_UNLINKAT: |
45d189c6 | 6292 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6293 | break; |
2b188cc1 JA |
6294 | default: |
6295 | ret = -EINVAL; | |
6296 | break; | |
6297 | } | |
6298 | ||
5730b27e JA |
6299 | if (creds) |
6300 | revert_creds(creds); | |
def596e9 JA |
6301 | if (ret) |
6302 | return ret; | |
b532576e | 6303 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6304 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6305 | io_iopoll_req_issued(req); | |
def596e9 JA |
6306 | |
6307 | return 0; | |
2b188cc1 JA |
6308 | } |
6309 | ||
ebc11b6c PB |
6310 | static struct io_wq_work *io_wq_free_work(struct io_wq_work *work) |
6311 | { | |
6312 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6313 | ||
6314 | req = io_put_req_find_next(req); | |
6315 | return req ? &req->work : NULL; | |
6316 | } | |
6317 | ||
5280f7e5 | 6318 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6319 | { |
6320 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6321 | struct io_kiocb *timeout; |
561fb04a | 6322 | int ret = 0; |
2b188cc1 | 6323 | |
48dcd38d PB |
6324 | /* one will be dropped by ->io_free_work() after returning to io-wq */ |
6325 | if (!(req->flags & REQ_F_REFCOUNT)) | |
6326 | __io_req_set_refcount(req, 2); | |
6327 | else | |
6328 | req_ref_get(req); | |
5d5901a3 | 6329 | |
6df1db6b PB |
6330 | timeout = io_prep_linked_timeout(req); |
6331 | if (timeout) | |
6332 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6333 | |
4014d943 | 6334 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6335 | ret = -ECANCELED; |
31b51510 | 6336 | |
561fb04a | 6337 | if (!ret) { |
561fb04a | 6338 | do { |
889fca73 | 6339 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6340 | /* |
6341 | * We can get EAGAIN for polled IO even though we're | |
6342 | * forcing a sync submission from here, since we can't | |
6343 | * wait for request slots on the block side. | |
6344 | */ | |
6345 | if (ret != -EAGAIN) | |
6346 | break; | |
6347 | cond_resched(); | |
6348 | } while (1); | |
6349 | } | |
31b51510 | 6350 | |
a3df7698 | 6351 | /* avoid locking problems by failing it from a clean context */ |
5d5901a3 | 6352 | if (ret) |
a3df7698 | 6353 | io_req_task_queue_fail(req, ret); |
2b188cc1 JA |
6354 | } |
6355 | ||
aeca241b | 6356 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
042b0d85 | 6357 | unsigned i) |
65e19f54 | 6358 | { |
042b0d85 | 6359 | return &table->files[i]; |
dafecf19 PB |
6360 | } |
6361 | ||
65e19f54 JA |
6362 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6363 | int index) | |
6364 | { | |
aeca241b | 6365 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6366 | |
a04b0ac0 | 6367 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6368 | } |
6369 | ||
a04b0ac0 | 6370 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6371 | { |
6372 | unsigned long file_ptr = (unsigned long) file; | |
6373 | ||
b191e2df | 6374 | if (__io_file_supports_nowait(file, READ)) |
9a321c98 | 6375 | file_ptr |= FFS_ASYNC_READ; |
b191e2df | 6376 | if (__io_file_supports_nowait(file, WRITE)) |
9a321c98 PB |
6377 | file_ptr |= FFS_ASYNC_WRITE; |
6378 | if (S_ISREG(file_inode(file)->i_mode)) | |
6379 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6380 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6381 | } |
6382 | ||
ac177053 PB |
6383 | static inline struct file *io_file_get_fixed(struct io_ring_ctx *ctx, |
6384 | struct io_kiocb *req, int fd) | |
09bb8394 | 6385 | { |
8da11c19 | 6386 | struct file *file; |
ac177053 | 6387 | unsigned long file_ptr; |
09bb8394 | 6388 | |
ac177053 PB |
6389 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
6390 | return NULL; | |
6391 | fd = array_index_nospec(fd, ctx->nr_user_files); | |
6392 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; | |
6393 | file = (struct file *) (file_ptr & FFS_MASK); | |
6394 | file_ptr &= ~FFS_MASK; | |
6395 | /* mask in overlapping REQ_F and FFS bits */ | |
b191e2df | 6396 | req->flags |= (file_ptr << REQ_F_NOWAIT_READ_BIT); |
ac177053 PB |
6397 | io_req_set_rsrc_node(req); |
6398 | return file; | |
6399 | } | |
d44f554e | 6400 | |
ac177053 | 6401 | static struct file *io_file_get_normal(struct io_ring_ctx *ctx, |
ac177053 PB |
6402 | struct io_kiocb *req, int fd) |
6403 | { | |
62906e89 | 6404 | struct file *file = fget(fd); |
ac177053 PB |
6405 | |
6406 | trace_io_uring_file_get(ctx, fd); | |
09bb8394 | 6407 | |
ac177053 PB |
6408 | /* we don't allow fixed io_uring files */ |
6409 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6410 | io_req_track_inflight(req); | |
8371adf5 | 6411 | return file; |
09bb8394 JA |
6412 | } |
6413 | ||
ac177053 | 6414 | static inline struct file *io_file_get(struct io_ring_ctx *ctx, |
ac177053 PB |
6415 | struct io_kiocb *req, int fd, bool fixed) |
6416 | { | |
6417 | if (fixed) | |
6418 | return io_file_get_fixed(ctx, req, fd); | |
6419 | else | |
62906e89 | 6420 | return io_file_get_normal(ctx, req, fd); |
ac177053 PB |
6421 | } |
6422 | ||
89b263f6 JA |
6423 | static void io_req_task_link_timeout(struct io_kiocb *req) |
6424 | { | |
6425 | struct io_kiocb *prev = req->timeout.prev; | |
8cb01fac | 6426 | int ret; |
89b263f6 JA |
6427 | |
6428 | if (prev) { | |
8cb01fac | 6429 | ret = io_try_cancel_userdata(req, prev->user_data); |
505657bc | 6430 | io_req_complete_post(req, ret ?: -ETIME, 0); |
89b263f6 | 6431 | io_put_req(prev); |
89b263f6 JA |
6432 | } else { |
6433 | io_req_complete_post(req, -ETIME, 0); | |
6434 | } | |
6435 | } | |
6436 | ||
2665abfd | 6437 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6438 | { |
ad8a48ac JA |
6439 | struct io_timeout_data *data = container_of(timer, |
6440 | struct io_timeout_data, timer); | |
90cd7e42 | 6441 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6442 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6443 | unsigned long flags; |
2665abfd | 6444 | |
89b263f6 | 6445 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
90cd7e42 PB |
6446 | prev = req->timeout.head; |
6447 | req->timeout.head = NULL; | |
2665abfd JA |
6448 | |
6449 | /* | |
6450 | * We don't expect the list to be empty, that will only happen if we | |
6451 | * race with the completion of the linked work. | |
6452 | */ | |
447c19f3 | 6453 | if (prev) { |
f2f87370 | 6454 | io_remove_next_linked(prev); |
447c19f3 PB |
6455 | if (!req_ref_inc_not_zero(prev)) |
6456 | prev = NULL; | |
6457 | } | |
89b263f6 JA |
6458 | req->timeout.prev = prev; |
6459 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); | |
2665abfd | 6460 | |
89b263f6 JA |
6461 | req->io_task_work.func = io_req_task_link_timeout; |
6462 | io_req_task_work_add(req); | |
2665abfd JA |
6463 | return HRTIMER_NORESTART; |
6464 | } | |
6465 | ||
de968c18 | 6466 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6467 | { |
de968c18 PB |
6468 | struct io_ring_ctx *ctx = req->ctx; |
6469 | ||
89b263f6 | 6470 | spin_lock_irq(&ctx->timeout_lock); |
76a46e06 | 6471 | /* |
f2f87370 PB |
6472 | * If the back reference is NULL, then our linked request finished |
6473 | * before we got a chance to setup the timer | |
76a46e06 | 6474 | */ |
90cd7e42 | 6475 | if (req->timeout.head) { |
e8c2bc1f | 6476 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6477 | |
ad8a48ac JA |
6478 | data->timer.function = io_link_timeout_fn; |
6479 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6480 | data->mode); | |
2665abfd | 6481 | } |
89b263f6 | 6482 | spin_unlock_irq(&ctx->timeout_lock); |
2665abfd | 6483 | /* drop submission reference */ |
76a46e06 JA |
6484 | io_put_req(req); |
6485 | } | |
2665abfd | 6486 | |
c5eef2b9 | 6487 | static void __io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 6488 | __must_hold(&req->ctx->uring_lock) |
2b188cc1 | 6489 | { |
906c6caa | 6490 | struct io_kiocb *linked_timeout; |
e0c5c576 | 6491 | int ret; |
2b188cc1 | 6492 | |
59b735ae | 6493 | issue_sqe: |
c5eef2b9 | 6494 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6495 | |
491381ce JA |
6496 | /* |
6497 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6498 | * doesn't support non-blocking read/write attempts | |
6499 | */ | |
1840038e | 6500 | if (likely(!ret)) { |
e342c807 | 6501 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 | 6502 | struct io_ring_ctx *ctx = req->ctx; |
cd0ca2e0 | 6503 | struct io_submit_state *state = &ctx->submit_state; |
e65ef56d | 6504 | |
cd0ca2e0 PB |
6505 | state->compl_reqs[state->compl_nr++] = req; |
6506 | if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) | |
2a2758f2 | 6507 | io_submit_flush_completions(ctx); |
906c6caa | 6508 | return; |
0d63c148 | 6509 | } |
906c6caa PB |
6510 | |
6511 | linked_timeout = io_prep_linked_timeout(req); | |
6512 | if (linked_timeout) | |
6513 | io_queue_linked_timeout(linked_timeout); | |
1840038e | 6514 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
906c6caa PB |
6515 | linked_timeout = io_prep_linked_timeout(req); |
6516 | ||
59b735ae OL |
6517 | switch (io_arm_poll_handler(req)) { |
6518 | case IO_APOLL_READY: | |
906c6caa PB |
6519 | if (linked_timeout) |
6520 | io_unprep_linked_timeout(req); | |
59b735ae OL |
6521 | goto issue_sqe; |
6522 | case IO_APOLL_ABORTED: | |
1840038e PB |
6523 | /* |
6524 | * Queued up for async execution, worker will release | |
6525 | * submit reference when the iocb is actually submitted. | |
6526 | */ | |
6527 | io_queue_async_work(req); | |
59b735ae | 6528 | break; |
1840038e | 6529 | } |
906c6caa PB |
6530 | |
6531 | if (linked_timeout) | |
6532 | io_queue_linked_timeout(linked_timeout); | |
0d63c148 | 6533 | } else { |
f41db273 | 6534 | io_req_complete_failed(req, ret); |
9e645e11 | 6535 | } |
2b188cc1 JA |
6536 | } |
6537 | ||
441b8a78 | 6538 | static inline void io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 6539 | __must_hold(&req->ctx->uring_lock) |
4fe2c963 | 6540 | { |
10c66904 | 6541 | if (unlikely(req->ctx->drain_active) && io_drain_req(req)) |
76cc33d7 | 6542 | return; |
4fe2c963 | 6543 | |
76cc33d7 | 6544 | if (likely(!(req->flags & REQ_F_FORCE_ASYNC))) { |
c5eef2b9 | 6545 | __io_queue_sqe(req); |
76cc33d7 PB |
6546 | } else { |
6547 | int ret = io_req_prep_async(req); | |
6548 | ||
6549 | if (unlikely(ret)) | |
6550 | io_req_complete_failed(req, ret); | |
6551 | else | |
6552 | io_queue_async_work(req); | |
ce35a47a | 6553 | } |
4fe2c963 JL |
6554 | } |
6555 | ||
b16fed66 PB |
6556 | /* |
6557 | * Check SQE restrictions (opcode and flags). | |
6558 | * | |
6559 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6560 | */ | |
6561 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6562 | struct io_kiocb *req, | |
6563 | unsigned int sqe_flags) | |
4fe2c963 | 6564 | { |
4cfb25bf | 6565 | if (likely(!ctx->restricted)) |
b16fed66 PB |
6566 | return true; |
6567 | ||
6568 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6569 | return false; | |
6570 | ||
6571 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6572 | ctx->restrictions.sqe_flags_required) | |
6573 | return false; | |
6574 | ||
6575 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6576 | ctx->restrictions.sqe_flags_required)) | |
6577 | return false; | |
6578 | ||
6579 | return true; | |
4fe2c963 JL |
6580 | } |
6581 | ||
b16fed66 PB |
6582 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6583 | const struct io_uring_sqe *sqe) | |
282cdc86 | 6584 | __must_hold(&ctx->uring_lock) |
b16fed66 PB |
6585 | { |
6586 | struct io_submit_state *state; | |
6587 | unsigned int sqe_flags; | |
003e8dcc | 6588 | int personality, ret = 0; |
b16fed66 | 6589 | |
864ea921 | 6590 | /* req is partially pre-initialised, see io_preinit_req() */ |
b16fed66 PB |
6591 | req->opcode = READ_ONCE(sqe->opcode); |
6592 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6593 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6594 | req->user_data = READ_ONCE(sqe->user_data); | |
b16fed66 | 6595 | req->file = NULL; |
b16fed66 | 6596 | req->fixed_rsrc_refs = NULL; |
b16fed66 | 6597 | req->task = current; |
b16fed66 PB |
6598 | |
6599 | /* enforce forwards compatibility on users */ | |
dddca226 | 6600 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 6601 | return -EINVAL; |
b16fed66 PB |
6602 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
6603 | return -EINVAL; | |
4cfb25bf | 6604 | if (!io_check_restriction(ctx, req, sqe_flags)) |
b16fed66 PB |
6605 | return -EACCES; |
6606 | ||
6607 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6608 | !io_op_defs[req->opcode].buffer_select) | |
6609 | return -EOPNOTSUPP; | |
3c19966d PB |
6610 | if (unlikely(sqe_flags & IOSQE_IO_DRAIN)) |
6611 | ctx->drain_active = true; | |
863e0560 | 6612 | |
003e8dcc JA |
6613 | personality = READ_ONCE(sqe->personality); |
6614 | if (personality) { | |
c10d1f98 PB |
6615 | req->creds = xa_load(&ctx->personalities, personality); |
6616 | if (!req->creds) | |
003e8dcc | 6617 | return -EINVAL; |
c10d1f98 | 6618 | get_cred(req->creds); |
b8e64b53 | 6619 | req->flags |= REQ_F_CREDS; |
003e8dcc | 6620 | } |
b16fed66 PB |
6621 | state = &ctx->submit_state; |
6622 | ||
6623 | /* | |
6624 | * Plug now if we have more than 1 IO left after this, and the target | |
6625 | * is potentially a read/write to block based storage. | |
6626 | */ | |
6627 | if (!state->plug_started && state->ios_left > 1 && | |
6628 | io_op_defs[req->opcode].plug) { | |
6629 | blk_start_plug(&state->plug); | |
6630 | state->plug_started = true; | |
6631 | } | |
6632 | ||
6633 | if (io_op_defs[req->opcode].needs_file) { | |
62906e89 | 6634 | req->file = io_file_get(ctx, req, READ_ONCE(sqe->fd), |
ac177053 | 6635 | (sqe_flags & IOSQE_FIXED_FILE)); |
b16fed66 PB |
6636 | if (unlikely(!req->file)) |
6637 | ret = -EBADF; | |
6638 | } | |
6639 | ||
6640 | state->ios_left--; | |
6641 | return ret; | |
6642 | } | |
6643 | ||
a6b8cadc | 6644 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6645 | const struct io_uring_sqe *sqe) |
282cdc86 | 6646 | __must_hold(&ctx->uring_lock) |
9e645e11 | 6647 | { |
a1ab7b35 | 6648 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6649 | int ret; |
9e645e11 | 6650 | |
a6b8cadc PB |
6651 | ret = io_init_req(ctx, req, sqe); |
6652 | if (unlikely(ret)) { | |
6653 | fail_req: | |
de59bc10 PB |
6654 | if (link->head) { |
6655 | /* fail even hard links since we don't submit */ | |
93d2bcd2 | 6656 | req_set_fail(link->head); |
f41db273 | 6657 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6658 | link->head = NULL; |
6659 | } | |
f41db273 | 6660 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6661 | return ret; |
6662 | } | |
441b8a78 | 6663 | |
be7053b7 PB |
6664 | ret = io_req_prep(req, sqe); |
6665 | if (unlikely(ret)) | |
6666 | goto fail_req; | |
a6b8cadc | 6667 | |
be7053b7 | 6668 | /* don't need @sqe from now on */ |
236daeae OL |
6669 | trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, |
6670 | req->flags, true, | |
6671 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 6672 | |
9e645e11 JA |
6673 | /* |
6674 | * If we already have a head request, queue this one for async | |
6675 | * submittal once the head completes. If we don't have a head but | |
6676 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6677 | * submitted sync once the chain is complete. If none of those | |
6678 | * conditions are true (normal request), then just queue it. | |
6679 | */ | |
863e0560 PB |
6680 | if (link->head) { |
6681 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6682 | |
b7e298d2 | 6683 | ret = io_req_prep_async(req); |
cf109604 | 6684 | if (unlikely(ret)) |
a6b8cadc | 6685 | goto fail_req; |
9d76377f | 6686 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6687 | link->last->link = req; |
863e0560 | 6688 | link->last = req; |
32fe525b PB |
6689 | |
6690 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6691 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 6692 | link->head = NULL; |
5e159204 | 6693 | io_queue_sqe(head); |
32fe525b | 6694 | } |
9e645e11 | 6695 | } else { |
ef4ff581 | 6696 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6697 | link->head = req; |
6698 | link->last = req; | |
711be031 | 6699 | } else { |
be7053b7 | 6700 | io_queue_sqe(req); |
711be031 | 6701 | } |
9e645e11 | 6702 | } |
2e6e1fde | 6703 | |
1d4240cc | 6704 | return 0; |
9e645e11 JA |
6705 | } |
6706 | ||
9a56a232 JA |
6707 | /* |
6708 | * Batched submission is done, ensure local IO is flushed out. | |
6709 | */ | |
ba88ff11 PB |
6710 | static void io_submit_state_end(struct io_submit_state *state, |
6711 | struct io_ring_ctx *ctx) | |
9a56a232 | 6712 | { |
a1ab7b35 | 6713 | if (state->link.head) |
de59bc10 | 6714 | io_queue_sqe(state->link.head); |
cd0ca2e0 | 6715 | if (state->compl_nr) |
2a2758f2 | 6716 | io_submit_flush_completions(ctx); |
27926b68 JA |
6717 | if (state->plug_started) |
6718 | blk_finish_plug(&state->plug); | |
9a56a232 JA |
6719 | } |
6720 | ||
6721 | /* | |
6722 | * Start submission side cache. | |
6723 | */ | |
6724 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6725 | unsigned int max_ios) |
9a56a232 | 6726 | { |
27926b68 | 6727 | state->plug_started = false; |
9a56a232 | 6728 | state->ios_left = max_ios; |
a1ab7b35 PB |
6729 | /* set only head, no need to init link_last in advance */ |
6730 | state->link.head = NULL; | |
9a56a232 JA |
6731 | } |
6732 | ||
2b188cc1 JA |
6733 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6734 | { | |
75b28aff | 6735 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6736 | |
caf582c6 PB |
6737 | /* |
6738 | * Ensure any loads from the SQEs are done at this point, | |
6739 | * since once we write the new head, the application could | |
6740 | * write new data to them. | |
6741 | */ | |
6742 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6743 | } |
6744 | ||
2b188cc1 | 6745 | /* |
dd9ae8a0 | 6746 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
6747 | * that is mapped by userspace. This means that care needs to be taken to |
6748 | * ensure that reads are stable, as we cannot rely on userspace always | |
6749 | * being a good citizen. If members of the sqe are validated and then later | |
6750 | * used, it's important that those reads are done through READ_ONCE() to | |
6751 | * prevent a re-load down the line. | |
6752 | */ | |
709b302f | 6753 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6754 | { |
ea5ab3b5 | 6755 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 6756 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
6757 | |
6758 | /* | |
6759 | * The cached sq head (or cq tail) serves two purposes: | |
6760 | * | |
6761 | * 1) allows us to batch the cost of updating the user visible | |
6762 | * head updates. | |
6763 | * 2) allows the kernel side to track the head on its own, even | |
6764 | * though the application is the one updating it. | |
6765 | */ | |
17d3aeb3 | 6766 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
6767 | if (likely(head < ctx->sq_entries)) |
6768 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6769 | |
6770 | /* drop invalid entries */ | |
15641e42 PB |
6771 | ctx->cq_extra--; |
6772 | WRITE_ONCE(ctx->rings->sq_dropped, | |
6773 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
6774 | return NULL; |
6775 | } | |
6776 | ||
0f212204 | 6777 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
282cdc86 | 6778 | __must_hold(&ctx->uring_lock) |
6c271ce2 | 6779 | { |
09899b19 | 6780 | struct io_uring_task *tctx; |
46c4e16a | 6781 | int submitted = 0; |
6c271ce2 | 6782 | |
ee7d46d9 PB |
6783 | /* make sure SQ entry isn't read before tail */ |
6784 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
6785 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6786 | return -EAGAIN; | |
6c271ce2 | 6787 | |
09899b19 PB |
6788 | tctx = current->io_uring; |
6789 | tctx->cached_refs -= nr; | |
6790 | if (unlikely(tctx->cached_refs < 0)) { | |
6791 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
6792 | ||
6793 | percpu_counter_add(&tctx->inflight, refill); | |
6794 | refcount_add(refill, ¤t->usage); | |
6795 | tctx->cached_refs += refill; | |
6796 | } | |
ba88ff11 | 6797 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6798 | |
46c4e16a | 6799 | while (submitted < nr) { |
3529d8c2 | 6800 | const struct io_uring_sqe *sqe; |
196be95c | 6801 | struct io_kiocb *req; |
fb5ccc98 | 6802 | |
258b29a9 | 6803 | req = io_alloc_req(ctx); |
196be95c PB |
6804 | if (unlikely(!req)) { |
6805 | if (!submitted) | |
6806 | submitted = -EAGAIN; | |
fb5ccc98 | 6807 | break; |
196be95c | 6808 | } |
4fccfcbb PB |
6809 | sqe = io_get_sqe(ctx); |
6810 | if (unlikely(!sqe)) { | |
6811 | kmem_cache_free(req_cachep, req); | |
6812 | break; | |
6813 | } | |
d3656344 JA |
6814 | /* will complete beyond this point, count as submitted */ |
6815 | submitted++; | |
a1ab7b35 | 6816 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6817 | break; |
6c271ce2 JA |
6818 | } |
6819 | ||
9466f437 PB |
6820 | if (unlikely(submitted != nr)) { |
6821 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 6822 | int unused = nr - ref_used; |
9466f437 | 6823 | |
09899b19 | 6824 | current->io_uring->cached_refs += unused; |
d8a6df10 | 6825 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 6826 | } |
6c271ce2 | 6827 | |
a1ab7b35 | 6828 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6829 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6830 | io_commit_sqring(ctx); | |
6831 | ||
6c271ce2 JA |
6832 | return submitted; |
6833 | } | |
6834 | ||
e4b6d902 PB |
6835 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
6836 | { | |
6837 | return READ_ONCE(sqd->state); | |
6838 | } | |
6839 | ||
23b3628e XW |
6840 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6841 | { | |
6842 | /* Tell userspace we may need a wakeup call */ | |
79ebeaee | 6843 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
6844 | WRITE_ONCE(ctx->rings->sq_flags, |
6845 | ctx->rings->sq_flags | IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 6846 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
6847 | } |
6848 | ||
6849 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6850 | { | |
79ebeaee | 6851 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
6852 | WRITE_ONCE(ctx->rings->sq_flags, |
6853 | ctx->rings->sq_flags & ~IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 6854 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
6855 | } |
6856 | ||
08369246 | 6857 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6858 | { |
c8d1ba58 | 6859 | unsigned int to_submit; |
bdcd3eab | 6860 | int ret = 0; |
6c271ce2 | 6861 | |
c8d1ba58 | 6862 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 6863 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
6864 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
6865 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 6866 | |
906a3c6f | 6867 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6868 | unsigned nr_events = 0; |
948e1947 PB |
6869 | const struct cred *creds = NULL; |
6870 | ||
6871 | if (ctx->sq_creds != current_cred()) | |
6872 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 6873 | |
c8d1ba58 | 6874 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6875 | if (!list_empty(&ctx->iopoll_list)) |
a8576af9 | 6876 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6877 | |
3b763ba1 PB |
6878 | /* |
6879 | * Don't submit if refs are dying, good for io_uring_register(), | |
6880 | * but also it is relied upon by io_ring_exit_work() | |
6881 | */ | |
0298ef96 PB |
6882 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6883 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6884 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 6885 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 6886 | |
acfb381d PB |
6887 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6888 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
6889 | if (creds) |
6890 | revert_creds(creds); | |
acfb381d | 6891 | } |
6c271ce2 | 6892 | |
08369246 XW |
6893 | return ret; |
6894 | } | |
6c271ce2 | 6895 | |
08369246 XW |
6896 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6897 | { | |
6898 | struct io_ring_ctx *ctx; | |
6899 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6900 | |
c9dca27d PB |
6901 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6902 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6903 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6904 | } |
6c271ce2 | 6905 | |
e4b6d902 PB |
6906 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
6907 | { | |
6908 | bool did_sig = false; | |
6909 | struct ksignal ksig; | |
6910 | ||
6911 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
6912 | signal_pending(current)) { | |
6913 | mutex_unlock(&sqd->lock); | |
6914 | if (signal_pending(current)) | |
6915 | did_sig = get_signal(&ksig); | |
6916 | cond_resched(); | |
6917 | mutex_lock(&sqd->lock); | |
6918 | } | |
e4b6d902 PB |
6919 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
6920 | } | |
6921 | ||
c8d1ba58 JA |
6922 | static int io_sq_thread(void *data) |
6923 | { | |
69fb2131 JA |
6924 | struct io_sq_data *sqd = data; |
6925 | struct io_ring_ctx *ctx; | |
a0d9205f | 6926 | unsigned long timeout = 0; |
37d1e2e3 | 6927 | char buf[TASK_COMM_LEN]; |
08369246 | 6928 | DEFINE_WAIT(wait); |
6c271ce2 | 6929 | |
696ee88a | 6930 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6931 | set_task_comm(current, buf); |
37d1e2e3 JA |
6932 | |
6933 | if (sqd->sq_cpu != -1) | |
6934 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6935 | else | |
6936 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6937 | current->flags |= PF_NO_SETAFFINITY; | |
6938 | ||
09a6f4ef | 6939 | mutex_lock(&sqd->lock); |
e4b6d902 | 6940 | while (1) { |
1a924a80 | 6941 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 6942 | |
e4b6d902 PB |
6943 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
6944 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 6945 | break; |
08369246 XW |
6946 | timeout = jiffies + sqd->sq_thread_idle; |
6947 | } | |
e4b6d902 | 6948 | |
e95eee2d | 6949 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6950 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 6951 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 6952 | |
08369246 XW |
6953 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6954 | sqt_spin = true; | |
69fb2131 | 6955 | } |
dd432ea5 PB |
6956 | if (io_run_task_work()) |
6957 | sqt_spin = true; | |
6c271ce2 | 6958 | |
08369246 | 6959 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 6960 | cond_resched(); |
08369246 XW |
6961 | if (sqt_spin) |
6962 | timeout = jiffies + sqd->sq_thread_idle; | |
6963 | continue; | |
6964 | } | |
6965 | ||
08369246 | 6966 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
dd432ea5 | 6967 | if (!io_sqd_events_pending(sqd) && !current->task_works) { |
1a924a80 PB |
6968 | bool needs_sched = true; |
6969 | ||
724cb4f9 | 6970 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
6971 | io_ring_set_wakeup_flag(ctx); |
6972 | ||
724cb4f9 HX |
6973 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
6974 | !list_empty_careful(&ctx->iopoll_list)) { | |
6975 | needs_sched = false; | |
6976 | break; | |
6977 | } | |
6978 | if (io_sqring_entries(ctx)) { | |
6979 | needs_sched = false; | |
6980 | break; | |
6981 | } | |
6982 | } | |
6983 | ||
6984 | if (needs_sched) { | |
6985 | mutex_unlock(&sqd->lock); | |
6986 | schedule(); | |
6987 | mutex_lock(&sqd->lock); | |
6988 | } | |
69fb2131 JA |
6989 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6990 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6991 | } |
08369246 XW |
6992 | |
6993 | finish_wait(&sqd->wait, &wait); | |
6994 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6995 | } |
28cea78a | 6996 | |
78cc687b | 6997 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 6998 | sqd->thread = NULL; |
05962f95 | 6999 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 7000 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 7001 | io_run_task_work(); |
734551df PB |
7002 | mutex_unlock(&sqd->lock); |
7003 | ||
37d1e2e3 JA |
7004 | complete(&sqd->exited); |
7005 | do_exit(0); | |
6c271ce2 JA |
7006 | } |
7007 | ||
bda52162 JA |
7008 | struct io_wait_queue { |
7009 | struct wait_queue_entry wq; | |
7010 | struct io_ring_ctx *ctx; | |
5fd46178 | 7011 | unsigned cq_tail; |
bda52162 JA |
7012 | unsigned nr_timeouts; |
7013 | }; | |
7014 | ||
6c503150 | 7015 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7016 | { |
7017 | struct io_ring_ctx *ctx = iowq->ctx; | |
5fd46178 | 7018 | int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; |
bda52162 JA |
7019 | |
7020 | /* | |
d195a66e | 7021 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7022 | * started waiting. For timeouts, we always want to return to userspace, |
7023 | * regardless of event count. | |
7024 | */ | |
5fd46178 | 7025 | return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
bda52162 JA |
7026 | } |
7027 | ||
7028 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7029 | int wake_flags, void *key) | |
7030 | { | |
7031 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7032 | wq); | |
7033 | ||
6c503150 PB |
7034 | /* |
7035 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7036 | * the task, and the next invocation will do it. | |
7037 | */ | |
5ed7a37d | 7038 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
7039 | return autoremove_wake_function(curr, mode, wake_flags, key); |
7040 | return -1; | |
bda52162 JA |
7041 | } |
7042 | ||
af9c1a44 JA |
7043 | static int io_run_task_work_sig(void) |
7044 | { | |
7045 | if (io_run_task_work()) | |
7046 | return 1; | |
7047 | if (!signal_pending(current)) | |
7048 | return 0; | |
0b8cfa97 | 7049 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 7050 | return -ERESTARTSYS; |
af9c1a44 JA |
7051 | return -EINTR; |
7052 | } | |
7053 | ||
eeb60b9a PB |
7054 | /* when returns >0, the caller should retry */ |
7055 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7056 | struct io_wait_queue *iowq, | |
7057 | signed long *timeout) | |
7058 | { | |
7059 | int ret; | |
7060 | ||
7061 | /* make sure we run task_work before checking for signals */ | |
7062 | ret = io_run_task_work_sig(); | |
7063 | if (ret || io_should_wake(iowq)) | |
7064 | return ret; | |
7065 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 7066 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
7067 | return 1; |
7068 | ||
7069 | *timeout = schedule_timeout(*timeout); | |
7070 | return !*timeout ? -ETIME : 1; | |
7071 | } | |
7072 | ||
2b188cc1 JA |
7073 | /* |
7074 | * Wait until events become available, if we don't already have some. The | |
7075 | * application must reap them itself, as they reside on the shared cq ring. | |
7076 | */ | |
7077 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7078 | const sigset_t __user *sig, size_t sigsz, |
7079 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7080 | { |
90291099 | 7081 | struct io_wait_queue iowq; |
75b28aff | 7082 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7083 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7084 | int ret; | |
2b188cc1 | 7085 | |
b41e9852 | 7086 | do { |
90f67366 | 7087 | io_cqring_overflow_flush(ctx); |
6c503150 | 7088 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7089 | return 0; |
4c6e277c | 7090 | if (!io_run_task_work()) |
b41e9852 | 7091 | break; |
b41e9852 | 7092 | } while (1); |
2b188cc1 JA |
7093 | |
7094 | if (sig) { | |
9e75ad5d AB |
7095 | #ifdef CONFIG_COMPAT |
7096 | if (in_compat_syscall()) | |
7097 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7098 | sigsz); |
9e75ad5d AB |
7099 | else |
7100 | #endif | |
b772434b | 7101 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7102 | |
2b188cc1 JA |
7103 | if (ret) |
7104 | return ret; | |
7105 | } | |
7106 | ||
c73ebb68 | 7107 | if (uts) { |
c1d5a224 PB |
7108 | struct timespec64 ts; |
7109 | ||
c73ebb68 HX |
7110 | if (get_timespec64(&ts, uts)) |
7111 | return -EFAULT; | |
7112 | timeout = timespec64_to_jiffies(&ts); | |
7113 | } | |
7114 | ||
90291099 PB |
7115 | init_waitqueue_func_entry(&iowq.wq, io_wake_function); |
7116 | iowq.wq.private = current; | |
7117 | INIT_LIST_HEAD(&iowq.wq.entry); | |
7118 | iowq.ctx = ctx; | |
bda52162 | 7119 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
5fd46178 | 7120 | iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; |
90291099 | 7121 | |
c826bd7a | 7122 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7123 | do { |
ca0a2651 | 7124 | /* if we can't even flush overflow, don't wait for more */ |
90f67366 | 7125 | if (!io_cqring_overflow_flush(ctx)) { |
ca0a2651 JA |
7126 | ret = -EBUSY; |
7127 | break; | |
7128 | } | |
311997b3 | 7129 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7130 | TASK_INTERRUPTIBLE); |
eeb60b9a | 7131 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
311997b3 | 7132 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7133 | cond_resched(); |
eeb60b9a | 7134 | } while (ret > 0); |
bda52162 | 7135 | |
b7db41c9 | 7136 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7137 | |
75b28aff | 7138 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7139 | } |
7140 | ||
9123c8ff | 7141 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7142 | { |
9123c8ff | 7143 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7144 | |
846a4ef2 | 7145 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7146 | kfree(table[i]); |
7147 | kfree(table); | |
7148 | } | |
7149 | ||
7150 | static void **io_alloc_page_table(size_t size) | |
7151 | { | |
7152 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7153 | size_t init_size = size; | |
7154 | void **table; | |
7155 | ||
0bea96f5 | 7156 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7157 | if (!table) |
7158 | return NULL; | |
7159 | ||
7160 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 7161 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff | 7162 | |
0bea96f5 | 7163 | table[i] = kzalloc(this_size, GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
7164 | if (!table[i]) { |
7165 | io_free_page_table(table, init_size); | |
7166 | return NULL; | |
7167 | } | |
7168 | size -= this_size; | |
7169 | } | |
7170 | return table; | |
05f3fb3c JA |
7171 | } |
7172 | ||
28a9fe25 | 7173 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7174 | { |
28a9fe25 PB |
7175 | percpu_ref_exit(&ref_node->refs); |
7176 | kfree(ref_node); | |
1642b445 PB |
7177 | } |
7178 | ||
b9bd2bea PB |
7179 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
7180 | { | |
7181 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); | |
7182 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; | |
7183 | unsigned long flags; | |
7184 | bool first_add = false; | |
7185 | ||
7186 | spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); | |
7187 | node->done = true; | |
7188 | ||
7189 | while (!list_empty(&ctx->rsrc_ref_list)) { | |
7190 | node = list_first_entry(&ctx->rsrc_ref_list, | |
7191 | struct io_rsrc_node, node); | |
7192 | /* recycle ref nodes in order */ | |
7193 | if (!node->done) | |
7194 | break; | |
7195 | list_del(&node->node); | |
7196 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
7197 | } | |
7198 | spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); | |
7199 | ||
7200 | if (first_add) | |
7201 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
7202 | } | |
7203 | ||
7204 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) | |
7205 | { | |
7206 | struct io_rsrc_node *ref_node; | |
7207 | ||
7208 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); | |
7209 | if (!ref_node) | |
7210 | return NULL; | |
7211 | ||
7212 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, | |
7213 | 0, GFP_KERNEL)) { | |
7214 | kfree(ref_node); | |
7215 | return NULL; | |
7216 | } | |
7217 | INIT_LIST_HEAD(&ref_node->node); | |
7218 | INIT_LIST_HEAD(&ref_node->rsrc_list); | |
7219 | ref_node->done = false; | |
7220 | return ref_node; | |
7221 | } | |
7222 | ||
a7f0ed5a PB |
7223 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7224 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7225 | { |
a7f0ed5a PB |
7226 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7227 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7228 | |
a7f0ed5a PB |
7229 | if (data_to_kill) { |
7230 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7231 | |
a7f0ed5a | 7232 | rsrc_node->rsrc_data = data_to_kill; |
4956b9ea | 7233 | spin_lock_irq(&ctx->rsrc_ref_lock); |
a7f0ed5a | 7234 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); |
4956b9ea | 7235 | spin_unlock_irq(&ctx->rsrc_ref_lock); |
82fbcfa9 | 7236 | |
3e942498 | 7237 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7238 | percpu_ref_kill(&rsrc_node->refs); |
7239 | ctx->rsrc_node = NULL; | |
7240 | } | |
6b06314c | 7241 | |
a7f0ed5a PB |
7242 | if (!ctx->rsrc_node) { |
7243 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7244 | ctx->rsrc_backup_node = NULL; | |
7245 | } | |
8bad28d8 HX |
7246 | } |
7247 | ||
a7f0ed5a | 7248 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7249 | { |
7250 | if (ctx->rsrc_backup_node) | |
7251 | return 0; | |
b895c9a6 | 7252 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7253 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7254 | } |
7255 | ||
40ae0ff7 | 7256 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7257 | { |
7258 | int ret; | |
05589553 | 7259 | |
215c3902 | 7260 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7261 | if (data->quiesce) |
7262 | return -ENXIO; | |
05589553 | 7263 | |
8bad28d8 | 7264 | data->quiesce = true; |
1ffc5422 | 7265 | do { |
a7f0ed5a | 7266 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7267 | if (ret) |
f2303b1f | 7268 | break; |
a7f0ed5a | 7269 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7270 | |
3e942498 PB |
7271 | /* kill initial ref, already quiesced if zero */ |
7272 | if (atomic_dec_and_test(&data->refs)) | |
7273 | break; | |
c018db4a | 7274 | mutex_unlock(&ctx->uring_lock); |
8bad28d8 | 7275 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 | 7276 | ret = wait_for_completion_interruptible(&data->done); |
c018db4a JA |
7277 | if (!ret) { |
7278 | mutex_lock(&ctx->uring_lock); | |
1ffc5422 | 7279 | break; |
c018db4a | 7280 | } |
8bad28d8 | 7281 | |
3e942498 PB |
7282 | atomic_inc(&data->refs); |
7283 | /* wait for all works potentially completing data->done */ | |
7284 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7285 | reinit_completion(&data->done); |
8dd03afe | 7286 | |
1ffc5422 | 7287 | ret = io_run_task_work_sig(); |
8bad28d8 | 7288 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7289 | } while (ret >= 0); |
8bad28d8 | 7290 | data->quiesce = false; |
05f3fb3c | 7291 | |
8bad28d8 | 7292 | return ret; |
d7954b2b BM |
7293 | } |
7294 | ||
2d091d62 PB |
7295 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7296 | { | |
7297 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7298 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7299 | ||
7300 | return &data->tags[table_idx][off]; | |
7301 | } | |
7302 | ||
44b31f2f | 7303 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7304 | { |
2d091d62 PB |
7305 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7306 | ||
7307 | if (data->tags) | |
7308 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7309 | kfree(data); |
7310 | } | |
7311 | ||
d878c816 PB |
7312 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7313 | u64 __user *utags, unsigned nr, | |
7314 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7315 | { |
b895c9a6 | 7316 | struct io_rsrc_data *data; |
2d091d62 | 7317 | int ret = -ENOMEM; |
d878c816 | 7318 | unsigned i; |
1ad555c6 BM |
7319 | |
7320 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7321 | if (!data) | |
d878c816 | 7322 | return -ENOMEM; |
2d091d62 | 7323 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7324 | if (!data->tags) { |
1ad555c6 | 7325 | kfree(data); |
d878c816 PB |
7326 | return -ENOMEM; |
7327 | } | |
2d091d62 PB |
7328 | |
7329 | data->nr = nr; | |
7330 | data->ctx = ctx; | |
7331 | data->do_put = do_put; | |
d878c816 | 7332 | if (utags) { |
2d091d62 | 7333 | ret = -EFAULT; |
d878c816 | 7334 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7335 | u64 *tag_slot = io_get_tag_slot(data, i); |
7336 | ||
7337 | if (copy_from_user(tag_slot, &utags[i], | |
7338 | sizeof(*tag_slot))) | |
2d091d62 | 7339 | goto fail; |
d878c816 | 7340 | } |
1ad555c6 | 7341 | } |
b60c8dce | 7342 | |
3e942498 | 7343 | atomic_set(&data->refs, 1); |
1ad555c6 | 7344 | init_completion(&data->done); |
d878c816 PB |
7345 | *pdata = data; |
7346 | return 0; | |
2d091d62 PB |
7347 | fail: |
7348 | io_rsrc_data_free(data); | |
7349 | return ret; | |
1ad555c6 BM |
7350 | } |
7351 | ||
9123c8ff PB |
7352 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7353 | { | |
0bea96f5 PB |
7354 | table->files = kvcalloc(nr_files, sizeof(table->files[0]), |
7355 | GFP_KERNEL_ACCOUNT); | |
9123c8ff PB |
7356 | return !!table->files; |
7357 | } | |
7358 | ||
042b0d85 | 7359 | static void io_free_file_tables(struct io_file_table *table) |
9123c8ff | 7360 | { |
042b0d85 | 7361 | kvfree(table->files); |
9123c8ff PB |
7362 | table->files = NULL; |
7363 | } | |
7364 | ||
fff4db76 | 7365 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7366 | { |
fff4db76 PB |
7367 | #if defined(CONFIG_UNIX) |
7368 | if (ctx->ring_sock) { | |
7369 | struct sock *sock = ctx->ring_sock->sk; | |
7370 | struct sk_buff *skb; | |
7371 | ||
7372 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7373 | kfree_skb(skb); | |
7374 | } | |
7375 | #else | |
7376 | int i; | |
7377 | ||
7378 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7379 | struct file *file; | |
7380 | ||
7381 | file = io_file_from_index(ctx, i); | |
7382 | if (file) | |
7383 | fput(file); | |
7384 | } | |
7385 | #endif | |
042b0d85 | 7386 | io_free_file_tables(&ctx->file_table); |
44b31f2f | 7387 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7388 | ctx->file_data = NULL; |
7389 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7390 | } |
7391 | ||
d7954b2b BM |
7392 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7393 | { | |
d7954b2b BM |
7394 | int ret; |
7395 | ||
08480400 | 7396 | if (!ctx->file_data) |
d7954b2b | 7397 | return -ENXIO; |
08480400 PB |
7398 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7399 | if (!ret) | |
7400 | __io_sqe_files_unregister(ctx); | |
7401 | return ret; | |
6b06314c JA |
7402 | } |
7403 | ||
37d1e2e3 | 7404 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7405 | __releases(&sqd->lock) |
37d1e2e3 | 7406 | { |
521d6a73 PB |
7407 | WARN_ON_ONCE(sqd->thread == current); |
7408 | ||
9e138a48 PB |
7409 | /* |
7410 | * Do the dance but not conditional clear_bit() because it'd race with | |
7411 | * other threads incrementing park_pending and setting the bit. | |
7412 | */ | |
37d1e2e3 | 7413 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7414 | if (atomic_dec_return(&sqd->park_pending)) |
7415 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7416 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7417 | } |
7418 | ||
86e0d676 | 7419 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7420 | __acquires(&sqd->lock) |
37d1e2e3 | 7421 | { |
521d6a73 PB |
7422 | WARN_ON_ONCE(sqd->thread == current); |
7423 | ||
9e138a48 | 7424 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7425 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7426 | mutex_lock(&sqd->lock); |
05962f95 | 7427 | if (sqd->thread) |
86e0d676 | 7428 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7429 | } |
7430 | ||
7431 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7432 | { | |
521d6a73 | 7433 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7434 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7435 | |
05962f95 | 7436 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7437 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7438 | if (sqd->thread) |
7439 | wake_up_process(sqd->thread); | |
09a6f4ef | 7440 | mutex_unlock(&sqd->lock); |
05962f95 | 7441 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7442 | } |
7443 | ||
534ca6d6 | 7444 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7445 | { |
534ca6d6 | 7446 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7447 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7448 | ||
37d1e2e3 JA |
7449 | io_sq_thread_stop(sqd); |
7450 | kfree(sqd); | |
7451 | } | |
7452 | } | |
7453 | ||
7454 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7455 | { | |
7456 | struct io_sq_data *sqd = ctx->sq_data; | |
7457 | ||
7458 | if (sqd) { | |
05962f95 | 7459 | io_sq_thread_park(sqd); |
521d6a73 | 7460 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7461 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7462 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7463 | |
7464 | io_put_sq_data(sqd); | |
7465 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7466 | } |
7467 | } | |
7468 | ||
aa06165d JA |
7469 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7470 | { | |
7471 | struct io_ring_ctx *ctx_attach; | |
7472 | struct io_sq_data *sqd; | |
7473 | struct fd f; | |
7474 | ||
7475 | f = fdget(p->wq_fd); | |
7476 | if (!f.file) | |
7477 | return ERR_PTR(-ENXIO); | |
7478 | if (f.file->f_op != &io_uring_fops) { | |
7479 | fdput(f); | |
7480 | return ERR_PTR(-EINVAL); | |
7481 | } | |
7482 | ||
7483 | ctx_attach = f.file->private_data; | |
7484 | sqd = ctx_attach->sq_data; | |
7485 | if (!sqd) { | |
7486 | fdput(f); | |
7487 | return ERR_PTR(-EINVAL); | |
7488 | } | |
5c2469e0 JA |
7489 | if (sqd->task_tgid != current->tgid) { |
7490 | fdput(f); | |
7491 | return ERR_PTR(-EPERM); | |
7492 | } | |
aa06165d JA |
7493 | |
7494 | refcount_inc(&sqd->refs); | |
7495 | fdput(f); | |
7496 | return sqd; | |
7497 | } | |
7498 | ||
26984fbf PB |
7499 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7500 | bool *attached) | |
534ca6d6 JA |
7501 | { |
7502 | struct io_sq_data *sqd; | |
7503 | ||
26984fbf | 7504 | *attached = false; |
5c2469e0 JA |
7505 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7506 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7507 | if (!IS_ERR(sqd)) { |
7508 | *attached = true; | |
5c2469e0 | 7509 | return sqd; |
26984fbf | 7510 | } |
5c2469e0 JA |
7511 | /* fall through for EPERM case, setup new sqd/task */ |
7512 | if (PTR_ERR(sqd) != -EPERM) | |
7513 | return sqd; | |
7514 | } | |
aa06165d | 7515 | |
534ca6d6 JA |
7516 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7517 | if (!sqd) | |
7518 | return ERR_PTR(-ENOMEM); | |
7519 | ||
9e138a48 | 7520 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7521 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7522 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7523 | mutex_init(&sqd->lock); |
534ca6d6 | 7524 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7525 | init_completion(&sqd->exited); |
534ca6d6 JA |
7526 | return sqd; |
7527 | } | |
7528 | ||
6b06314c | 7529 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7530 | /* |
7531 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7532 | * the io_uring can be safely unregistered on process exit, even if we have | |
7533 | * loops in the file referencing. | |
7534 | */ | |
7535 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7536 | { | |
7537 | struct sock *sk = ctx->ring_sock->sk; | |
7538 | struct scm_fp_list *fpl; | |
7539 | struct sk_buff *skb; | |
08a45173 | 7540 | int i, nr_files; |
6b06314c | 7541 | |
6b06314c JA |
7542 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7543 | if (!fpl) | |
7544 | return -ENOMEM; | |
7545 | ||
7546 | skb = alloc_skb(0, GFP_KERNEL); | |
7547 | if (!skb) { | |
7548 | kfree(fpl); | |
7549 | return -ENOMEM; | |
7550 | } | |
7551 | ||
7552 | skb->sk = sk; | |
6b06314c | 7553 | |
08a45173 | 7554 | nr_files = 0; |
62e398be | 7555 | fpl->user = get_uid(current_user()); |
6b06314c | 7556 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7557 | struct file *file = io_file_from_index(ctx, i + offset); |
7558 | ||
7559 | if (!file) | |
08a45173 | 7560 | continue; |
65e19f54 | 7561 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7562 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7563 | nr_files++; | |
6b06314c JA |
7564 | } |
7565 | ||
08a45173 JA |
7566 | if (nr_files) { |
7567 | fpl->max = SCM_MAX_FD; | |
7568 | fpl->count = nr_files; | |
7569 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7570 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7571 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7572 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7573 | |
08a45173 JA |
7574 | for (i = 0; i < nr_files; i++) |
7575 | fput(fpl->fp[i]); | |
7576 | } else { | |
7577 | kfree_skb(skb); | |
7578 | kfree(fpl); | |
7579 | } | |
6b06314c JA |
7580 | |
7581 | return 0; | |
7582 | } | |
7583 | ||
7584 | /* | |
7585 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7586 | * causes regular reference counting to break down. We rely on the UNIX | |
7587 | * garbage collection to take care of this problem for us. | |
7588 | */ | |
7589 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7590 | { | |
7591 | unsigned left, total; | |
7592 | int ret = 0; | |
7593 | ||
7594 | total = 0; | |
7595 | left = ctx->nr_user_files; | |
7596 | while (left) { | |
7597 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7598 | |
7599 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7600 | if (ret) | |
7601 | break; | |
7602 | left -= this_files; | |
7603 | total += this_files; | |
7604 | } | |
7605 | ||
7606 | if (!ret) | |
7607 | return 0; | |
7608 | ||
7609 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7610 | struct file *file = io_file_from_index(ctx, total); |
7611 | ||
7612 | if (file) | |
7613 | fput(file); | |
6b06314c JA |
7614 | total++; |
7615 | } | |
7616 | ||
7617 | return ret; | |
7618 | } | |
7619 | #else | |
7620 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7621 | { | |
7622 | return 0; | |
7623 | } | |
7624 | #endif | |
7625 | ||
47e90392 | 7626 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7627 | { |
50238531 | 7628 | struct file *file = prsrc->file; |
05f3fb3c JA |
7629 | #if defined(CONFIG_UNIX) |
7630 | struct sock *sock = ctx->ring_sock->sk; | |
7631 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7632 | struct sk_buff *skb; | |
7633 | int i; | |
7634 | ||
7635 | __skb_queue_head_init(&list); | |
7636 | ||
7637 | /* | |
7638 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7639 | * remove this entry and rearrange the file array. | |
7640 | */ | |
7641 | skb = skb_dequeue(head); | |
7642 | while (skb) { | |
7643 | struct scm_fp_list *fp; | |
7644 | ||
7645 | fp = UNIXCB(skb).fp; | |
7646 | for (i = 0; i < fp->count; i++) { | |
7647 | int left; | |
7648 | ||
7649 | if (fp->fp[i] != file) | |
7650 | continue; | |
7651 | ||
7652 | unix_notinflight(fp->user, fp->fp[i]); | |
7653 | left = fp->count - 1 - i; | |
7654 | if (left) { | |
7655 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7656 | left * sizeof(struct file *)); | |
7657 | } | |
7658 | fp->count--; | |
7659 | if (!fp->count) { | |
7660 | kfree_skb(skb); | |
7661 | skb = NULL; | |
7662 | } else { | |
7663 | __skb_queue_tail(&list, skb); | |
7664 | } | |
7665 | fput(file); | |
7666 | file = NULL; | |
7667 | break; | |
7668 | } | |
7669 | ||
7670 | if (!file) | |
7671 | break; | |
7672 | ||
7673 | __skb_queue_tail(&list, skb); | |
7674 | ||
7675 | skb = skb_dequeue(head); | |
7676 | } | |
7677 | ||
7678 | if (skb_peek(&list)) { | |
7679 | spin_lock_irq(&head->lock); | |
7680 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7681 | __skb_queue_tail(head, skb); | |
7682 | spin_unlock_irq(&head->lock); | |
7683 | } | |
7684 | #else | |
7685 | fput(file); | |
7686 | #endif | |
7687 | } | |
7688 | ||
b895c9a6 | 7689 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7690 | { |
b895c9a6 | 7691 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7692 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7693 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7694 | |
269bbe5f BM |
7695 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7696 | list_del(&prsrc->list); | |
b60c8dce PB |
7697 | |
7698 | if (prsrc->tag) { | |
7699 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
7700 | |
7701 | io_ring_submit_lock(ctx, lock_ring); | |
79ebeaee | 7702 | spin_lock(&ctx->completion_lock); |
b60c8dce | 7703 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 7704 | ctx->cq_extra++; |
b60c8dce | 7705 | io_commit_cqring(ctx); |
79ebeaee | 7706 | spin_unlock(&ctx->completion_lock); |
b60c8dce PB |
7707 | io_cqring_ev_posted(ctx); |
7708 | io_ring_submit_unlock(ctx, lock_ring); | |
7709 | } | |
7710 | ||
40ae0ff7 | 7711 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7712 | kfree(prsrc); |
65e19f54 | 7713 | } |
05589553 | 7714 | |
28a9fe25 | 7715 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7716 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7717 | complete(&rsrc_data->done); | |
2faf852d | 7718 | } |
65e19f54 | 7719 | |
269bbe5f | 7720 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7721 | { |
7722 | struct io_ring_ctx *ctx; | |
7723 | struct llist_node *node; | |
7724 | ||
269bbe5f BM |
7725 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7726 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7727 | |
7728 | while (node) { | |
b895c9a6 | 7729 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7730 | struct llist_node *next = node->next; |
7731 | ||
b895c9a6 | 7732 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7733 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7734 | node = next; |
7735 | } | |
7736 | } | |
7737 | ||
6b06314c | 7738 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 7739 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
7740 | { |
7741 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7742 | struct file *file; |
f3baed39 | 7743 | int fd, ret; |
846a4ef2 | 7744 | unsigned i; |
6b06314c | 7745 | |
05f3fb3c | 7746 | if (ctx->file_data) |
6b06314c JA |
7747 | return -EBUSY; |
7748 | if (!nr_args) | |
7749 | return -EINVAL; | |
7750 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7751 | return -EMFILE; | |
3a1b8a4e PB |
7752 | if (nr_args > rlimit(RLIMIT_NOFILE)) |
7753 | return -EMFILE; | |
a7f0ed5a | 7754 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7755 | if (ret) |
7756 | return ret; | |
d878c816 PB |
7757 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
7758 | &ctx->file_data); | |
7759 | if (ret) | |
7760 | return ret; | |
6b06314c | 7761 | |
f3baed39 | 7762 | ret = -ENOMEM; |
aeca241b | 7763 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7764 | goto out_free; |
65e19f54 | 7765 | |
08a45173 | 7766 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 7767 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
7768 | ret = -EFAULT; |
7769 | goto out_fput; | |
7770 | } | |
08a45173 | 7771 | /* allow sparse sets */ |
792e3582 PB |
7772 | if (fd == -1) { |
7773 | ret = -EINVAL; | |
2d091d62 | 7774 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 7775 | goto out_fput; |
08a45173 | 7776 | continue; |
792e3582 | 7777 | } |
6b06314c | 7778 | |
05f3fb3c | 7779 | file = fget(fd); |
6b06314c | 7780 | ret = -EBADF; |
792e3582 | 7781 | if (unlikely(!file)) |
600cf3f8 | 7782 | goto out_fput; |
05f3fb3c | 7783 | |
6b06314c JA |
7784 | /* |
7785 | * Don't allow io_uring instances to be registered. If UNIX | |
7786 | * isn't enabled, then this causes a reference cycle and this | |
7787 | * instance can never get freed. If UNIX is enabled we'll | |
7788 | * handle it just fine, but there's still no point in allowing | |
7789 | * a ring fd as it doesn't support regular read/write anyway. | |
7790 | */ | |
05f3fb3c JA |
7791 | if (file->f_op == &io_uring_fops) { |
7792 | fput(file); | |
600cf3f8 | 7793 | goto out_fput; |
6b06314c | 7794 | } |
aeca241b | 7795 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7796 | } |
7797 | ||
6b06314c | 7798 | ret = io_sqe_files_scm(ctx); |
05589553 | 7799 | if (ret) { |
08480400 | 7800 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7801 | return ret; |
7802 | } | |
6b06314c | 7803 | |
a7f0ed5a | 7804 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7805 | return ret; |
600cf3f8 PB |
7806 | out_fput: |
7807 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7808 | file = io_file_from_index(ctx, i); | |
7809 | if (file) | |
7810 | fput(file); | |
7811 | } | |
042b0d85 | 7812 | io_free_file_tables(&ctx->file_table); |
600cf3f8 | 7813 | ctx->nr_user_files = 0; |
600cf3f8 | 7814 | out_free: |
44b31f2f | 7815 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7816 | ctx->file_data = NULL; |
6b06314c JA |
7817 | return ret; |
7818 | } | |
7819 | ||
c3a31e60 JA |
7820 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7821 | int index) | |
7822 | { | |
7823 | #if defined(CONFIG_UNIX) | |
7824 | struct sock *sock = ctx->ring_sock->sk; | |
7825 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7826 | struct sk_buff *skb; | |
7827 | ||
7828 | /* | |
7829 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7830 | * file set. If there's no room, fall back to allocating a new skb | |
7831 | * and filling it in. | |
7832 | */ | |
7833 | spin_lock_irq(&head->lock); | |
7834 | skb = skb_peek(head); | |
7835 | if (skb) { | |
7836 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7837 | ||
7838 | if (fpl->count < SCM_MAX_FD) { | |
7839 | __skb_unlink(skb, head); | |
7840 | spin_unlock_irq(&head->lock); | |
7841 | fpl->fp[fpl->count] = get_file(file); | |
7842 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7843 | fpl->count++; | |
7844 | spin_lock_irq(&head->lock); | |
7845 | __skb_queue_head(head, skb); | |
7846 | } else { | |
7847 | skb = NULL; | |
7848 | } | |
7849 | } | |
7850 | spin_unlock_irq(&head->lock); | |
7851 | ||
7852 | if (skb) { | |
7853 | fput(file); | |
7854 | return 0; | |
7855 | } | |
7856 | ||
7857 | return __io_sqe_files_scm(ctx, 1, index); | |
7858 | #else | |
7859 | return 0; | |
7860 | #endif | |
7861 | } | |
7862 | ||
b60c8dce | 7863 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 7864 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 7865 | { |
269bbe5f | 7866 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7867 | |
269bbe5f BM |
7868 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7869 | if (!prsrc) | |
a5318d3c | 7870 | return -ENOMEM; |
05f3fb3c | 7871 | |
2d091d62 | 7872 | prsrc->tag = *io_get_tag_slot(data, idx); |
50238531 | 7873 | prsrc->rsrc = rsrc; |
e7c78371 | 7874 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7875 | return 0; |
05f3fb3c JA |
7876 | } |
7877 | ||
7878 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 7879 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
7880 | unsigned nr_args) |
7881 | { | |
c3bdad02 | 7882 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 7883 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 7884 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7885 | struct io_fixed_file *file_slot; |
7886 | struct file *file; | |
98f0b3b4 PB |
7887 | int fd, i, err = 0; |
7888 | unsigned int done; | |
05589553 | 7889 | bool needs_switch = false; |
c3a31e60 | 7890 | |
98f0b3b4 PB |
7891 | if (!ctx->file_data) |
7892 | return -ENXIO; | |
7893 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
7894 | return -EINVAL; |
7895 | ||
67973b93 | 7896 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
7897 | u64 tag = 0; |
7898 | ||
7899 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
7900 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
7901 | err = -EFAULT; |
7902 | break; | |
7903 | } | |
c3bdad02 PB |
7904 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
7905 | err = -EINVAL; | |
7906 | break; | |
7907 | } | |
4e0377a1 | 7908 | if (fd == IORING_REGISTER_FILES_SKIP) |
7909 | continue; | |
7910 | ||
67973b93 | 7911 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7912 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7913 | |
a04b0ac0 PB |
7914 | if (file_slot->file_ptr) { |
7915 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
7916 | err = io_queue_rsrc_removal(data, up->offset + done, |
7917 | ctx->rsrc_node, file); | |
a5318d3c HD |
7918 | if (err) |
7919 | break; | |
a04b0ac0 | 7920 | file_slot->file_ptr = 0; |
05589553 | 7921 | needs_switch = true; |
c3a31e60 JA |
7922 | } |
7923 | if (fd != -1) { | |
c3a31e60 JA |
7924 | file = fget(fd); |
7925 | if (!file) { | |
7926 | err = -EBADF; | |
7927 | break; | |
7928 | } | |
7929 | /* | |
7930 | * Don't allow io_uring instances to be registered. If | |
7931 | * UNIX isn't enabled, then this causes a reference | |
7932 | * cycle and this instance can never get freed. If UNIX | |
7933 | * is enabled we'll handle it just fine, but there's | |
7934 | * still no point in allowing a ring fd as it doesn't | |
7935 | * support regular read/write anyway. | |
7936 | */ | |
7937 | if (file->f_op == &io_uring_fops) { | |
7938 | fput(file); | |
7939 | err = -EBADF; | |
7940 | break; | |
7941 | } | |
2d091d62 | 7942 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 7943 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7944 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7945 | if (err) { |
a04b0ac0 | 7946 | file_slot->file_ptr = 0; |
f3bd9dae | 7947 | fput(file); |
c3a31e60 | 7948 | break; |
f3bd9dae | 7949 | } |
c3a31e60 | 7950 | } |
05f3fb3c JA |
7951 | } |
7952 | ||
a7f0ed5a PB |
7953 | if (needs_switch) |
7954 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7955 | return done ? done : err; |
7956 | } | |
05589553 | 7957 | |
685fe7fe JA |
7958 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7959 | struct task_struct *task) | |
24369c2e | 7960 | { |
e941894e | 7961 | struct io_wq_hash *hash; |
24369c2e | 7962 | struct io_wq_data data; |
24369c2e | 7963 | unsigned int concurrency; |
24369c2e | 7964 | |
362a9e65 | 7965 | mutex_lock(&ctx->uring_lock); |
e941894e JA |
7966 | hash = ctx->hash_map; |
7967 | if (!hash) { | |
7968 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
362a9e65 YY |
7969 | if (!hash) { |
7970 | mutex_unlock(&ctx->uring_lock); | |
e941894e | 7971 | return ERR_PTR(-ENOMEM); |
362a9e65 | 7972 | } |
e941894e JA |
7973 | refcount_set(&hash->refs, 1); |
7974 | init_waitqueue_head(&hash->wait); | |
7975 | ctx->hash_map = hash; | |
24369c2e | 7976 | } |
362a9e65 | 7977 | mutex_unlock(&ctx->uring_lock); |
24369c2e | 7978 | |
e941894e | 7979 | data.hash = hash; |
685fe7fe | 7980 | data.task = task; |
ebc11b6c | 7981 | data.free_work = io_wq_free_work; |
f5fa38c5 | 7982 | data.do_work = io_wq_submit_work; |
24369c2e | 7983 | |
d25e3a3d JA |
7984 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7985 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7986 | |
5aa75ed5 | 7987 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7988 | } |
7989 | ||
5aa75ed5 JA |
7990 | static int io_uring_alloc_task_context(struct task_struct *task, |
7991 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7992 | { |
7993 | struct io_uring_task *tctx; | |
d8a6df10 | 7994 | int ret; |
0f212204 | 7995 | |
09899b19 | 7996 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
7997 | if (unlikely(!tctx)) |
7998 | return -ENOMEM; | |
7999 | ||
d8a6df10 JA |
8000 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8001 | if (unlikely(ret)) { | |
8002 | kfree(tctx); | |
8003 | return ret; | |
8004 | } | |
8005 | ||
685fe7fe | 8006 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
8007 | if (IS_ERR(tctx->io_wq)) { |
8008 | ret = PTR_ERR(tctx->io_wq); | |
8009 | percpu_counter_destroy(&tctx->inflight); | |
8010 | kfree(tctx); | |
8011 | return ret; | |
8012 | } | |
8013 | ||
0f212204 JA |
8014 | xa_init(&tctx->xa); |
8015 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 8016 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 8017 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 8018 | task->io_uring = tctx; |
7cbf1722 JA |
8019 | spin_lock_init(&tctx->task_lock); |
8020 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 8021 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
8022 | return 0; |
8023 | } | |
8024 | ||
8025 | void __io_uring_free(struct task_struct *tsk) | |
8026 | { | |
8027 | struct io_uring_task *tctx = tsk->io_uring; | |
8028 | ||
8029 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 8030 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 8031 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 8032 | |
d8a6df10 | 8033 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8034 | kfree(tctx); |
8035 | tsk->io_uring = NULL; | |
8036 | } | |
8037 | ||
7e84e1c7 SG |
8038 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8039 | struct io_uring_params *p) | |
2b188cc1 JA |
8040 | { |
8041 | int ret; | |
8042 | ||
d25e3a3d JA |
8043 | /* Retain compatibility with failing for an invalid attach attempt */ |
8044 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
8045 | IORING_SETUP_ATTACH_WQ) { | |
8046 | struct fd f; | |
8047 | ||
8048 | f = fdget(p->wq_fd); | |
8049 | if (!f.file) | |
8050 | return -ENXIO; | |
0cc936f7 JA |
8051 | if (f.file->f_op != &io_uring_fops) { |
8052 | fdput(f); | |
f2a48dd0 | 8053 | return -EINVAL; |
0cc936f7 JA |
8054 | } |
8055 | fdput(f); | |
d25e3a3d | 8056 | } |
6c271ce2 | 8057 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 8058 | struct task_struct *tsk; |
534ca6d6 | 8059 | struct io_sq_data *sqd; |
26984fbf | 8060 | bool attached; |
534ca6d6 | 8061 | |
26984fbf | 8062 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8063 | if (IS_ERR(sqd)) { |
8064 | ret = PTR_ERR(sqd); | |
8065 | goto err; | |
8066 | } | |
69fb2131 | 8067 | |
7c30f36a | 8068 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8069 | ctx->sq_data = sqd; |
917257da JA |
8070 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8071 | if (!ctx->sq_thread_idle) | |
8072 | ctx->sq_thread_idle = HZ; | |
8073 | ||
78d7f6ba | 8074 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8075 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8076 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8077 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8078 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8079 | io_sq_thread_unpark(sqd); |
8080 | ||
de75a3d3 PB |
8081 | if (ret < 0) |
8082 | goto err; | |
8083 | if (attached) | |
5aa75ed5 | 8084 | return 0; |
aa06165d | 8085 | |
6c271ce2 | 8086 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8087 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8088 | |
917257da | 8089 | ret = -EINVAL; |
f2a48dd0 | 8090 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8091 | goto err_sqpoll; |
37d1e2e3 | 8092 | sqd->sq_cpu = cpu; |
6c271ce2 | 8093 | } else { |
37d1e2e3 | 8094 | sqd->sq_cpu = -1; |
6c271ce2 | 8095 | } |
37d1e2e3 JA |
8096 | |
8097 | sqd->task_pid = current->pid; | |
5c2469e0 | 8098 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8099 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8100 | if (IS_ERR(tsk)) { | |
8101 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8102 | goto err_sqpoll; |
6c271ce2 | 8103 | } |
97a73a0f | 8104 | |
46fe18b1 | 8105 | sqd->thread = tsk; |
97a73a0f | 8106 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8107 | wake_up_new_task(tsk); |
0f212204 JA |
8108 | if (ret) |
8109 | goto err; | |
6c271ce2 JA |
8110 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8111 | /* Can't have SQ_AFF without SQPOLL */ | |
8112 | ret = -EINVAL; | |
8113 | goto err; | |
8114 | } | |
8115 | ||
2b188cc1 | 8116 | return 0; |
f2a48dd0 PB |
8117 | err_sqpoll: |
8118 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8119 | err: |
37d1e2e3 | 8120 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8121 | return ret; |
8122 | } | |
8123 | ||
a087e2b5 BM |
8124 | static inline void __io_unaccount_mem(struct user_struct *user, |
8125 | unsigned long nr_pages) | |
2b188cc1 JA |
8126 | { |
8127 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8128 | } | |
8129 | ||
a087e2b5 BM |
8130 | static inline int __io_account_mem(struct user_struct *user, |
8131 | unsigned long nr_pages) | |
2b188cc1 JA |
8132 | { |
8133 | unsigned long page_limit, cur_pages, new_pages; | |
8134 | ||
8135 | /* Don't allow more pages than we can safely lock */ | |
8136 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8137 | ||
8138 | do { | |
8139 | cur_pages = atomic_long_read(&user->locked_vm); | |
8140 | new_pages = cur_pages + nr_pages; | |
8141 | if (new_pages > page_limit) | |
8142 | return -ENOMEM; | |
8143 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8144 | new_pages) != cur_pages); | |
8145 | ||
8146 | return 0; | |
8147 | } | |
8148 | ||
26bfa89e | 8149 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8150 | { |
62e398be | 8151 | if (ctx->user) |
a087e2b5 | 8152 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8153 | |
26bfa89e JA |
8154 | if (ctx->mm_account) |
8155 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8156 | } |
8157 | ||
26bfa89e | 8158 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8159 | { |
30975825 BM |
8160 | int ret; |
8161 | ||
62e398be | 8162 | if (ctx->user) { |
30975825 BM |
8163 | ret = __io_account_mem(ctx->user, nr_pages); |
8164 | if (ret) | |
8165 | return ret; | |
8166 | } | |
8167 | ||
26bfa89e JA |
8168 | if (ctx->mm_account) |
8169 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8170 | |
8171 | return 0; | |
8172 | } | |
8173 | ||
2b188cc1 JA |
8174 | static void io_mem_free(void *ptr) |
8175 | { | |
52e04ef4 MR |
8176 | struct page *page; |
8177 | ||
8178 | if (!ptr) | |
8179 | return; | |
2b188cc1 | 8180 | |
52e04ef4 | 8181 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8182 | if (put_page_testzero(page)) |
8183 | free_compound_page(page); | |
8184 | } | |
8185 | ||
8186 | static void *io_mem_alloc(size_t size) | |
8187 | { | |
8188 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8189 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8190 | |
8191 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8192 | } | |
8193 | ||
75b28aff HV |
8194 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8195 | size_t *sq_offset) | |
8196 | { | |
8197 | struct io_rings *rings; | |
8198 | size_t off, sq_array_size; | |
8199 | ||
8200 | off = struct_size(rings, cqes, cq_entries); | |
8201 | if (off == SIZE_MAX) | |
8202 | return SIZE_MAX; | |
8203 | ||
8204 | #ifdef CONFIG_SMP | |
8205 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8206 | if (off == 0) | |
8207 | return SIZE_MAX; | |
8208 | #endif | |
8209 | ||
b36200f5 DV |
8210 | if (sq_offset) |
8211 | *sq_offset = off; | |
8212 | ||
75b28aff HV |
8213 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8214 | if (sq_array_size == SIZE_MAX) | |
8215 | return SIZE_MAX; | |
8216 | ||
8217 | if (check_add_overflow(off, sq_array_size, &off)) | |
8218 | return SIZE_MAX; | |
8219 | ||
75b28aff HV |
8220 | return off; |
8221 | } | |
8222 | ||
41edf1a5 | 8223 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8224 | { |
41edf1a5 | 8225 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8226 | unsigned int i; |
8227 | ||
6224843d PB |
8228 | if (imu != ctx->dummy_ubuf) { |
8229 | for (i = 0; i < imu->nr_bvecs; i++) | |
8230 | unpin_user_page(imu->bvec[i].bv_page); | |
8231 | if (imu->acct_pages) | |
8232 | io_unaccount_mem(ctx, imu->acct_pages); | |
8233 | kvfree(imu); | |
8234 | } | |
41edf1a5 | 8235 | *slot = NULL; |
7f61a1e9 PB |
8236 | } |
8237 | ||
bd54b6fe | 8238 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8239 | { |
634d00df PB |
8240 | io_buffer_unmap(ctx, &prsrc->buf); |
8241 | prsrc->buf = NULL; | |
bd54b6fe | 8242 | } |
edafccee | 8243 | |
bd54b6fe BM |
8244 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8245 | { | |
8246 | unsigned int i; | |
edafccee | 8247 | |
7f61a1e9 PB |
8248 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8249 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8250 | kfree(ctx->user_bufs); |
bb6659cc | 8251 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8252 | ctx->user_bufs = NULL; |
bd54b6fe | 8253 | ctx->buf_data = NULL; |
edafccee | 8254 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8255 | } |
8256 | ||
0a96bbe4 | 8257 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8258 | { |
bd54b6fe | 8259 | int ret; |
edafccee | 8260 | |
bd54b6fe | 8261 | if (!ctx->buf_data) |
edafccee JA |
8262 | return -ENXIO; |
8263 | ||
bd54b6fe BM |
8264 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8265 | if (!ret) | |
8266 | __io_sqe_buffers_unregister(ctx); | |
8267 | return ret; | |
edafccee JA |
8268 | } |
8269 | ||
8270 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8271 | void __user *arg, unsigned index) | |
8272 | { | |
8273 | struct iovec __user *src; | |
8274 | ||
8275 | #ifdef CONFIG_COMPAT | |
8276 | if (ctx->compat) { | |
8277 | struct compat_iovec __user *ciovs; | |
8278 | struct compat_iovec ciov; | |
8279 | ||
8280 | ciovs = (struct compat_iovec __user *) arg; | |
8281 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8282 | return -EFAULT; | |
8283 | ||
d55e5f5b | 8284 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8285 | dst->iov_len = ciov.iov_len; |
8286 | return 0; | |
8287 | } | |
8288 | #endif | |
8289 | src = (struct iovec __user *) arg; | |
8290 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8291 | return -EFAULT; | |
8292 | return 0; | |
8293 | } | |
8294 | ||
de293938 JA |
8295 | /* |
8296 | * Not super efficient, but this is just a registration time. And we do cache | |
8297 | * the last compound head, so generally we'll only do a full search if we don't | |
8298 | * match that one. | |
8299 | * | |
8300 | * We check if the given compound head page has already been accounted, to | |
8301 | * avoid double accounting it. This allows us to account the full size of the | |
8302 | * page, not just the constituent pages of a huge page. | |
8303 | */ | |
8304 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8305 | int nr_pages, struct page *hpage) | |
8306 | { | |
8307 | int i, j; | |
8308 | ||
8309 | /* check current page array */ | |
8310 | for (i = 0; i < nr_pages; i++) { | |
8311 | if (!PageCompound(pages[i])) | |
8312 | continue; | |
8313 | if (compound_head(pages[i]) == hpage) | |
8314 | return true; | |
8315 | } | |
8316 | ||
8317 | /* check previously registered pages */ | |
8318 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8319 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8320 | |
8321 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8322 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8323 | continue; | |
8324 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8325 | return true; | |
8326 | } | |
8327 | } | |
8328 | ||
8329 | return false; | |
8330 | } | |
8331 | ||
8332 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8333 | int nr_pages, struct io_mapped_ubuf *imu, | |
8334 | struct page **last_hpage) | |
8335 | { | |
8336 | int i, ret; | |
8337 | ||
216e5835 | 8338 | imu->acct_pages = 0; |
de293938 JA |
8339 | for (i = 0; i < nr_pages; i++) { |
8340 | if (!PageCompound(pages[i])) { | |
8341 | imu->acct_pages++; | |
8342 | } else { | |
8343 | struct page *hpage; | |
8344 | ||
8345 | hpage = compound_head(pages[i]); | |
8346 | if (hpage == *last_hpage) | |
8347 | continue; | |
8348 | *last_hpage = hpage; | |
8349 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8350 | continue; | |
8351 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8352 | } | |
8353 | } | |
8354 | ||
8355 | if (!imu->acct_pages) | |
8356 | return 0; | |
8357 | ||
26bfa89e | 8358 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8359 | if (ret) |
8360 | imu->acct_pages = 0; | |
8361 | return ret; | |
8362 | } | |
8363 | ||
0a96bbe4 | 8364 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8365 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8366 | struct page **last_hpage) |
edafccee | 8367 | { |
41edf1a5 | 8368 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8369 | struct vm_area_struct **vmas = NULL; |
8370 | struct page **pages = NULL; | |
0a96bbe4 BM |
8371 | unsigned long off, start, end, ubuf; |
8372 | size_t size; | |
8373 | int ret, pret, nr_pages, i; | |
8374 | ||
6224843d PB |
8375 | if (!iov->iov_base) { |
8376 | *pimu = ctx->dummy_ubuf; | |
8377 | return 0; | |
8378 | } | |
8379 | ||
0a96bbe4 BM |
8380 | ubuf = (unsigned long) iov->iov_base; |
8381 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8382 | start = ubuf >> PAGE_SHIFT; | |
8383 | nr_pages = end - start; | |
8384 | ||
41edf1a5 | 8385 | *pimu = NULL; |
0a96bbe4 BM |
8386 | ret = -ENOMEM; |
8387 | ||
8388 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8389 | if (!pages) | |
8390 | goto done; | |
8391 | ||
8392 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8393 | GFP_KERNEL); | |
8394 | if (!vmas) | |
8395 | goto done; | |
edafccee | 8396 | |
41edf1a5 | 8397 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8398 | if (!imu) |
0a96bbe4 BM |
8399 | goto done; |
8400 | ||
8401 | ret = 0; | |
8402 | mmap_read_lock(current->mm); | |
8403 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8404 | pages, vmas); | |
8405 | if (pret == nr_pages) { | |
8406 | /* don't support file backed memory */ | |
8407 | for (i = 0; i < nr_pages; i++) { | |
8408 | struct vm_area_struct *vma = vmas[i]; | |
8409 | ||
40dad765 PB |
8410 | if (vma_is_shmem(vma)) |
8411 | continue; | |
0a96bbe4 BM |
8412 | if (vma->vm_file && |
8413 | !is_file_hugepages(vma->vm_file)) { | |
8414 | ret = -EOPNOTSUPP; | |
8415 | break; | |
8416 | } | |
8417 | } | |
8418 | } else { | |
8419 | ret = pret < 0 ? pret : -EFAULT; | |
8420 | } | |
8421 | mmap_read_unlock(current->mm); | |
8422 | if (ret) { | |
8423 | /* | |
8424 | * if we did partial map, or found file backed vmas, | |
8425 | * release any pages we did get | |
8426 | */ | |
8427 | if (pret > 0) | |
8428 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8429 | goto done; |
8430 | } | |
8431 | ||
8432 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8433 | if (ret) { | |
8434 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8435 | goto done; |
8436 | } | |
8437 | ||
8438 | off = ubuf & ~PAGE_MASK; | |
8439 | size = iov->iov_len; | |
8440 | for (i = 0; i < nr_pages; i++) { | |
8441 | size_t vec_len; | |
8442 | ||
8443 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8444 | imu->bvec[i].bv_page = pages[i]; | |
8445 | imu->bvec[i].bv_len = vec_len; | |
8446 | imu->bvec[i].bv_offset = off; | |
8447 | off = 0; | |
8448 | size -= vec_len; | |
8449 | } | |
8450 | /* store original address for later verification */ | |
8451 | imu->ubuf = ubuf; | |
4751f53d | 8452 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8453 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8454 | *pimu = imu; |
0a96bbe4 BM |
8455 | ret = 0; |
8456 | done: | |
41edf1a5 PB |
8457 | if (ret) |
8458 | kvfree(imu); | |
0a96bbe4 BM |
8459 | kvfree(pages); |
8460 | kvfree(vmas); | |
8461 | return ret; | |
8462 | } | |
8463 | ||
2b358604 | 8464 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8465 | { |
87094465 PB |
8466 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8467 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8468 | } |
edafccee | 8469 | |
2b358604 BM |
8470 | static int io_buffer_validate(struct iovec *iov) |
8471 | { | |
50e96989 PB |
8472 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8473 | ||
2b358604 BM |
8474 | /* |
8475 | * Don't impose further limits on the size and buffer | |
8476 | * constraints here, we'll -EINVAL later when IO is | |
8477 | * submitted if they are wrong. | |
8478 | */ | |
6224843d PB |
8479 | if (!iov->iov_base) |
8480 | return iov->iov_len ? -EFAULT : 0; | |
8481 | if (!iov->iov_len) | |
2b358604 | 8482 | return -EFAULT; |
edafccee | 8483 | |
2b358604 BM |
8484 | /* arbitrary limit, but we need something */ |
8485 | if (iov->iov_len > SZ_1G) | |
8486 | return -EFAULT; | |
edafccee | 8487 | |
50e96989 PB |
8488 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8489 | return -EOVERFLOW; | |
8490 | ||
2b358604 BM |
8491 | return 0; |
8492 | } | |
edafccee | 8493 | |
2b358604 | 8494 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 8495 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 8496 | { |
bd54b6fe BM |
8497 | struct page *last_hpage = NULL; |
8498 | struct io_rsrc_data *data; | |
2b358604 BM |
8499 | int i, ret; |
8500 | struct iovec iov; | |
edafccee | 8501 | |
87094465 PB |
8502 | if (ctx->user_bufs) |
8503 | return -EBUSY; | |
489809e2 | 8504 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 8505 | return -EINVAL; |
bd54b6fe | 8506 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8507 | if (ret) |
8508 | return ret; | |
d878c816 PB |
8509 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
8510 | if (ret) | |
8511 | return ret; | |
bd54b6fe BM |
8512 | ret = io_buffers_map_alloc(ctx, nr_args); |
8513 | if (ret) { | |
bb6659cc | 8514 | io_rsrc_data_free(data); |
bd54b6fe BM |
8515 | return ret; |
8516 | } | |
edafccee | 8517 | |
87094465 | 8518 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
8519 | ret = io_copy_iov(ctx, &iov, arg, i); |
8520 | if (ret) | |
0a96bbe4 | 8521 | break; |
2b358604 BM |
8522 | ret = io_buffer_validate(&iov); |
8523 | if (ret) | |
0a96bbe4 | 8524 | break; |
2d091d62 | 8525 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
8526 | ret = -EINVAL; |
8527 | break; | |
8528 | } | |
edafccee | 8529 | |
41edf1a5 PB |
8530 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
8531 | &last_hpage); | |
0a96bbe4 BM |
8532 | if (ret) |
8533 | break; | |
edafccee | 8534 | } |
0a96bbe4 | 8535 | |
bd54b6fe | 8536 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8537 | |
bd54b6fe BM |
8538 | ctx->buf_data = data; |
8539 | if (ret) | |
8540 | __io_sqe_buffers_unregister(ctx); | |
8541 | else | |
8542 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
8543 | return ret; |
8544 | } | |
8545 | ||
634d00df PB |
8546 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
8547 | struct io_uring_rsrc_update2 *up, | |
8548 | unsigned int nr_args) | |
8549 | { | |
8550 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
8551 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
8552 | struct page *last_hpage = NULL; |
8553 | bool needs_switch = false; | |
8554 | __u32 done; | |
8555 | int i, err; | |
8556 | ||
8557 | if (!ctx->buf_data) | |
8558 | return -ENXIO; | |
8559 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
8560 | return -EINVAL; | |
8561 | ||
8562 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
8563 | struct io_mapped_ubuf *imu; |
8564 | int offset = up->offset + done; | |
634d00df PB |
8565 | u64 tag = 0; |
8566 | ||
8567 | err = io_copy_iov(ctx, &iov, iovs, done); | |
8568 | if (err) | |
8569 | break; | |
8570 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
8571 | err = -EFAULT; | |
8572 | break; | |
8573 | } | |
0b8c0e7c PB |
8574 | err = io_buffer_validate(&iov); |
8575 | if (err) | |
8576 | break; | |
cf3770e7 CIK |
8577 | if (!iov.iov_base && tag) { |
8578 | err = -EINVAL; | |
8579 | break; | |
8580 | } | |
0b8c0e7c PB |
8581 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
8582 | if (err) | |
8583 | break; | |
634d00df | 8584 | |
0b8c0e7c | 8585 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 8586 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
8587 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
8588 | ctx->rsrc_node, ctx->user_bufs[i]); | |
8589 | if (unlikely(err)) { | |
8590 | io_buffer_unmap(ctx, &imu); | |
634d00df | 8591 | break; |
0b8c0e7c | 8592 | } |
634d00df PB |
8593 | ctx->user_bufs[i] = NULL; |
8594 | needs_switch = true; | |
8595 | } | |
8596 | ||
0b8c0e7c | 8597 | ctx->user_bufs[i] = imu; |
2d091d62 | 8598 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
8599 | } |
8600 | ||
8601 | if (needs_switch) | |
8602 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
8603 | return done ? done : err; | |
8604 | } | |
8605 | ||
9b402849 JA |
8606 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8607 | { | |
8608 | __s32 __user *fds = arg; | |
8609 | int fd; | |
8610 | ||
8611 | if (ctx->cq_ev_fd) | |
8612 | return -EBUSY; | |
8613 | ||
8614 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8615 | return -EFAULT; | |
8616 | ||
8617 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8618 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8619 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
fe7e3257 | 8620 | |
9b402849 JA |
8621 | ctx->cq_ev_fd = NULL; |
8622 | return ret; | |
8623 | } | |
8624 | ||
8625 | return 0; | |
8626 | } | |
8627 | ||
8628 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8629 | { | |
8630 | if (ctx->cq_ev_fd) { | |
8631 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8632 | ctx->cq_ev_fd = NULL; | |
8633 | return 0; | |
8634 | } | |
8635 | ||
8636 | return -ENXIO; | |
8637 | } | |
8638 | ||
5a2e745d JA |
8639 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8640 | { | |
9e15c3a0 JA |
8641 | struct io_buffer *buf; |
8642 | unsigned long index; | |
8643 | ||
8644 | xa_for_each(&ctx->io_buffers, index, buf) | |
8645 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8646 | } |
8647 | ||
7255834e | 8648 | static void io_req_cache_free(struct list_head *list) |
1b4c351f | 8649 | { |
68e68ee6 | 8650 | struct io_kiocb *req, *nxt; |
1b4c351f | 8651 | |
bb943b82 PB |
8652 | list_for_each_entry_safe(req, nxt, list, inflight_entry) { |
8653 | list_del(&req->inflight_entry); | |
1b4c351f JA |
8654 | kmem_cache_free(req_cachep, req); |
8655 | } | |
8656 | } | |
8657 | ||
4010fec4 | 8658 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8659 | { |
cd0ca2e0 | 8660 | struct io_submit_state *state = &ctx->submit_state; |
bf019da7 | 8661 | |
9a4fdbd8 JA |
8662 | mutex_lock(&ctx->uring_lock); |
8663 | ||
cd0ca2e0 PB |
8664 | if (state->free_reqs) { |
8665 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); | |
8666 | state->free_reqs = 0; | |
8e5c66c4 | 8667 | } |
9a4fdbd8 | 8668 | |
cd0ca2e0 PB |
8669 | io_flush_cached_locked_reqs(ctx, state); |
8670 | io_req_cache_free(&state->free_list); | |
9a4fdbd8 JA |
8671 | mutex_unlock(&ctx->uring_lock); |
8672 | } | |
8673 | ||
43597aac | 8674 | static void io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 8675 | { |
43597aac | 8676 | if (data && !atomic_dec_and_test(&data->refs)) |
bd54b6fe | 8677 | wait_for_completion(&data->done); |
bd54b6fe | 8678 | } |
04fc6c80 | 8679 | |
2b188cc1 JA |
8680 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8681 | { | |
37d1e2e3 | 8682 | io_sq_thread_finish(ctx); |
2aede0e4 | 8683 | |
37d1e2e3 | 8684 | if (ctx->mm_account) { |
2aede0e4 JA |
8685 | mmdrop(ctx->mm_account); |
8686 | ctx->mm_account = NULL; | |
30975825 | 8687 | } |
def596e9 | 8688 | |
43597aac PB |
8689 | /* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */ |
8690 | io_wait_rsrc_data(ctx->buf_data); | |
8691 | io_wait_rsrc_data(ctx->file_data); | |
8692 | ||
8bad28d8 | 8693 | mutex_lock(&ctx->uring_lock); |
43597aac | 8694 | if (ctx->buf_data) |
bd54b6fe | 8695 | __io_sqe_buffers_unregister(ctx); |
43597aac | 8696 | if (ctx->file_data) |
08480400 | 8697 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
8698 | if (ctx->rings) |
8699 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8700 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8701 | io_eventfd_unregister(ctx); |
5a2e745d | 8702 | io_destroy_buffers(ctx); |
07db298a PB |
8703 | if (ctx->sq_creds) |
8704 | put_cred(ctx->sq_creds); | |
def596e9 | 8705 | |
a7f0ed5a PB |
8706 | /* there are no registered resources left, nobody uses it */ |
8707 | if (ctx->rsrc_node) | |
8708 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8709 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8710 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8711 | flush_delayed_work(&ctx->rsrc_put_work); |
8712 | ||
8713 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8714 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 8715 | |
2b188cc1 | 8716 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8717 | if (ctx->ring_sock) { |
8718 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8719 | sock_release(ctx->ring_sock); |
355e8d26 | 8720 | } |
2b188cc1 JA |
8721 | #endif |
8722 | ||
75b28aff | 8723 | io_mem_free(ctx->rings); |
2b188cc1 | 8724 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8725 | |
8726 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8727 | free_uid(ctx->user); |
4010fec4 | 8728 | io_req_caches_free(ctx); |
e941894e JA |
8729 | if (ctx->hash_map) |
8730 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8731 | kfree(ctx->cancel_hash); |
6224843d | 8732 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
8733 | kfree(ctx); |
8734 | } | |
8735 | ||
8736 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8737 | { | |
8738 | struct io_ring_ctx *ctx = file->private_data; | |
8739 | __poll_t mask = 0; | |
8740 | ||
311997b3 | 8741 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
8742 | /* |
8743 | * synchronizes with barrier from wq_has_sleeper call in | |
8744 | * io_commit_cqring | |
8745 | */ | |
2b188cc1 | 8746 | smp_rmb(); |
90554200 | 8747 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8748 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8749 | |
8750 | /* | |
8751 | * Don't flush cqring overflow list here, just do a simple check. | |
8752 | * Otherwise there could possible be ABBA deadlock: | |
8753 | * CPU0 CPU1 | |
8754 | * ---- ---- | |
8755 | * lock(&ctx->uring_lock); | |
8756 | * lock(&ep->mtx); | |
8757 | * lock(&ctx->uring_lock); | |
8758 | * lock(&ep->mtx); | |
8759 | * | |
8760 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8761 | * pushs them to do the flush. | |
8762 | */ | |
5ed7a37d | 8763 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
8764 | mask |= EPOLLIN | EPOLLRDNORM; |
8765 | ||
8766 | return mask; | |
8767 | } | |
8768 | ||
8769 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8770 | { | |
8771 | struct io_ring_ctx *ctx = file->private_data; | |
8772 | ||
8773 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8774 | } | |
8775 | ||
0bead8cd | 8776 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8777 | { |
4379bf8b | 8778 | const struct cred *creds; |
071698e1 | 8779 | |
61cf9370 | 8780 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8781 | if (creds) { |
8782 | put_cred(creds); | |
0bead8cd | 8783 | return 0; |
1e6fa521 | 8784 | } |
0bead8cd YD |
8785 | |
8786 | return -EINVAL; | |
8787 | } | |
8788 | ||
d56d938b PB |
8789 | struct io_tctx_exit { |
8790 | struct callback_head task_work; | |
8791 | struct completion completion; | |
baf186c4 | 8792 | struct io_ring_ctx *ctx; |
d56d938b PB |
8793 | }; |
8794 | ||
8795 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8796 | { | |
8797 | struct io_uring_task *tctx = current->io_uring; | |
8798 | struct io_tctx_exit *work; | |
8799 | ||
8800 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8801 | /* | |
8802 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8803 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8804 | */ | |
8805 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 8806 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
8807 | complete(&work->completion); |
8808 | } | |
8809 | ||
28090c13 PB |
8810 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8811 | { | |
8812 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8813 | ||
8814 | return req->ctx == data; | |
8815 | } | |
8816 | ||
85faa7b8 JA |
8817 | static void io_ring_exit_work(struct work_struct *work) |
8818 | { | |
d56d938b | 8819 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8820 | unsigned long timeout = jiffies + HZ * 60 * 5; |
58d3be2c | 8821 | unsigned long interval = HZ / 20; |
d56d938b PB |
8822 | struct io_tctx_exit exit; |
8823 | struct io_tctx_node *node; | |
8824 | int ret; | |
85faa7b8 | 8825 | |
56952e91 JA |
8826 | /* |
8827 | * If we're doing polled IO and end up having requests being | |
8828 | * submitted async (out-of-line), then completions can come in while | |
8829 | * we're waiting for refs to drop. We need to reap these manually, | |
8830 | * as nobody else will be looking for them. | |
8831 | */ | |
b2edc0a7 | 8832 | do { |
3dd0c97a | 8833 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
8834 | if (ctx->sq_data) { |
8835 | struct io_sq_data *sqd = ctx->sq_data; | |
8836 | struct task_struct *tsk; | |
8837 | ||
8838 | io_sq_thread_park(sqd); | |
8839 | tsk = sqd->thread; | |
8840 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
8841 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
8842 | io_cancel_ctx_cb, ctx, true); | |
8843 | io_sq_thread_unpark(sqd); | |
8844 | } | |
b5bb3a24 | 8845 | |
58d3be2c PB |
8846 | if (WARN_ON_ONCE(time_after(jiffies, timeout))) { |
8847 | /* there is little hope left, don't run it too often */ | |
8848 | interval = HZ * 60; | |
8849 | } | |
8850 | } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); | |
d56d938b | 8851 | |
7f00651a PB |
8852 | init_completion(&exit.completion); |
8853 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8854 | exit.ctx = ctx; | |
89b5066e PB |
8855 | /* |
8856 | * Some may use context even when all refs and requests have been put, | |
8857 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 8858 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
8859 | * this lock/unlock section also waits them to finish. |
8860 | */ | |
d56d938b PB |
8861 | mutex_lock(&ctx->uring_lock); |
8862 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8863 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8864 | ||
d56d938b PB |
8865 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8866 | ctx_node); | |
7f00651a PB |
8867 | /* don't spin on a single task if cancellation failed */ |
8868 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8869 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8870 | if (WARN_ON_ONCE(ret)) | |
8871 | continue; | |
8872 | wake_up_process(node->task); | |
8873 | ||
8874 | mutex_unlock(&ctx->uring_lock); | |
8875 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8876 | mutex_lock(&ctx->uring_lock); |
8877 | } | |
8878 | mutex_unlock(&ctx->uring_lock); | |
79ebeaee JA |
8879 | spin_lock(&ctx->completion_lock); |
8880 | spin_unlock(&ctx->completion_lock); | |
d56d938b | 8881 | |
85faa7b8 JA |
8882 | io_ring_ctx_free(ctx); |
8883 | } | |
8884 | ||
80c4cbdb PB |
8885 | /* Returns true if we found and killed one or more timeouts */ |
8886 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 8887 | bool cancel_all) |
80c4cbdb PB |
8888 | { |
8889 | struct io_kiocb *req, *tmp; | |
8890 | int canceled = 0; | |
8891 | ||
79ebeaee JA |
8892 | spin_lock(&ctx->completion_lock); |
8893 | spin_lock_irq(&ctx->timeout_lock); | |
80c4cbdb | 8894 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
3dd0c97a | 8895 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
8896 | io_kill_timeout(req, -ECANCELED); |
8897 | canceled++; | |
8898 | } | |
8899 | } | |
79ebeaee | 8900 | spin_unlock_irq(&ctx->timeout_lock); |
51520426 PB |
8901 | if (canceled != 0) |
8902 | io_commit_cqring(ctx); | |
79ebeaee | 8903 | spin_unlock(&ctx->completion_lock); |
80c4cbdb PB |
8904 | if (canceled != 0) |
8905 | io_cqring_ev_posted(ctx); | |
8906 | return canceled != 0; | |
8907 | } | |
8908 | ||
2b188cc1 JA |
8909 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8910 | { | |
61cf9370 MWO |
8911 | unsigned long index; |
8912 | struct creds *creds; | |
8913 | ||
2b188cc1 JA |
8914 | mutex_lock(&ctx->uring_lock); |
8915 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8916 | if (ctx->rings) |
6c2450ae | 8917 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8918 | xa_for_each(&ctx->personalities, index, creds) |
8919 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8920 | mutex_unlock(&ctx->uring_lock); |
8921 | ||
3dd0c97a PB |
8922 | io_kill_timeouts(ctx, NULL, true); |
8923 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 8924 | |
15dff286 | 8925 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8926 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8927 | |
85faa7b8 | 8928 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8929 | /* |
8930 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8931 | * if we're exiting a ton of rings at the same time. It just adds | |
8932 | * noise and overhead, there's no discernable change in runtime | |
8933 | * over using system_wq. | |
8934 | */ | |
8935 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8936 | } |
8937 | ||
8938 | static int io_uring_release(struct inode *inode, struct file *file) | |
8939 | { | |
8940 | struct io_ring_ctx *ctx = file->private_data; | |
8941 | ||
8942 | file->private_data = NULL; | |
8943 | io_ring_ctx_wait_and_kill(ctx); | |
8944 | return 0; | |
8945 | } | |
8946 | ||
f6edbabb PB |
8947 | struct io_task_cancel { |
8948 | struct task_struct *task; | |
3dd0c97a | 8949 | bool all; |
f6edbabb | 8950 | }; |
f254ac04 | 8951 | |
f6edbabb | 8952 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8953 | { |
9a472ef7 | 8954 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8955 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8956 | bool ret; |
8957 | ||
3dd0c97a | 8958 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8959 | struct io_ring_ctx *ctx = req->ctx; |
8960 | ||
8961 | /* protect against races with linked timeouts */ | |
79ebeaee | 8962 | spin_lock(&ctx->completion_lock); |
3dd0c97a | 8963 | ret = io_match_task(req, cancel->task, cancel->all); |
79ebeaee | 8964 | spin_unlock(&ctx->completion_lock); |
9a472ef7 | 8965 | } else { |
3dd0c97a | 8966 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8967 | } |
8968 | return ret; | |
b711d4ea JA |
8969 | } |
8970 | ||
e1915f76 | 8971 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 8972 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 8973 | { |
e1915f76 | 8974 | struct io_defer_entry *de; |
b7ddce3c PB |
8975 | LIST_HEAD(list); |
8976 | ||
79ebeaee | 8977 | spin_lock(&ctx->completion_lock); |
b7ddce3c | 8978 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { |
3dd0c97a | 8979 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
8980 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8981 | break; | |
8982 | } | |
8983 | } | |
79ebeaee | 8984 | spin_unlock(&ctx->completion_lock); |
e1915f76 PB |
8985 | if (list_empty(&list)) |
8986 | return false; | |
b7ddce3c PB |
8987 | |
8988 | while (!list_empty(&list)) { | |
8989 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8990 | list_del_init(&de->list); | |
f41db273 | 8991 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8992 | kfree(de); |
8993 | } | |
e1915f76 | 8994 | return true; |
b7ddce3c PB |
8995 | } |
8996 | ||
1b00764f PB |
8997 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
8998 | { | |
8999 | struct io_tctx_node *node; | |
9000 | enum io_wq_cancel cret; | |
9001 | bool ret = false; | |
9002 | ||
9003 | mutex_lock(&ctx->uring_lock); | |
9004 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
9005 | struct io_uring_task *tctx = node->task->io_uring; | |
9006 | ||
9007 | /* | |
9008 | * io_wq will stay alive while we hold uring_lock, because it's | |
9009 | * killed after ctx nodes, which requires to take the lock. | |
9010 | */ | |
9011 | if (!tctx || !tctx->io_wq) | |
9012 | continue; | |
9013 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
9014 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9015 | } | |
9016 | mutex_unlock(&ctx->uring_lock); | |
9017 | ||
9018 | return ret; | |
9019 | } | |
9020 | ||
9936c7c2 PB |
9021 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
9022 | struct task_struct *task, | |
3dd0c97a | 9023 | bool cancel_all) |
9936c7c2 | 9024 | { |
3dd0c97a | 9025 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 9026 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
9027 | |
9028 | while (1) { | |
9029 | enum io_wq_cancel cret; | |
9030 | bool ret = false; | |
9031 | ||
1b00764f PB |
9032 | if (!task) { |
9033 | ret |= io_uring_try_cancel_iowq(ctx); | |
9034 | } else if (tctx && tctx->io_wq) { | |
9035 | /* | |
9036 | * Cancels requests of all rings, not only @ctx, but | |
9037 | * it's fine as the task is in exit/exec. | |
9038 | */ | |
5aa75ed5 | 9039 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
9040 | &cancel, true); |
9041 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
9042 | } | |
9043 | ||
9044 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 9045 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 9046 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
9047 | while (!list_empty_careful(&ctx->iopoll_list)) { |
9048 | io_iopoll_try_reap_events(ctx); | |
9049 | ret = true; | |
9050 | } | |
9051 | } | |
9052 | ||
3dd0c97a PB |
9053 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
9054 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
9055 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
9056 | if (task) |
9057 | ret |= io_run_task_work(); | |
9936c7c2 PB |
9058 | if (!ret) |
9059 | break; | |
9060 | cond_resched(); | |
9061 | } | |
9062 | } | |
9063 | ||
eef51daa | 9064 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 9065 | { |
236434c3 | 9066 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 9067 | struct io_tctx_node *node; |
a528b04e | 9068 | int ret; |
236434c3 MWO |
9069 | |
9070 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9071 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9072 | if (unlikely(ret)) |
9073 | return ret; | |
236434c3 | 9074 | tctx = current->io_uring; |
0f212204 | 9075 | } |
cf27f3b1 PB |
9076 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9077 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9078 | if (!node) | |
9079 | return -ENOMEM; | |
9080 | node->ctx = ctx; | |
9081 | node->task = current; | |
13bf43f5 | 9082 | |
cf27f3b1 PB |
9083 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9084 | node, GFP_KERNEL)); | |
9085 | if (ret) { | |
9086 | kfree(node); | |
9087 | return ret; | |
0f212204 | 9088 | } |
cf27f3b1 PB |
9089 | |
9090 | mutex_lock(&ctx->uring_lock); | |
9091 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9092 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9093 | } |
cf27f3b1 | 9094 | tctx->last = ctx; |
0f212204 JA |
9095 | return 0; |
9096 | } | |
9097 | ||
cf27f3b1 PB |
9098 | /* |
9099 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9100 | */ | |
eef51daa | 9101 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9102 | { |
9103 | struct io_uring_task *tctx = current->io_uring; | |
9104 | ||
9105 | if (likely(tctx && tctx->last == ctx)) | |
9106 | return 0; | |
eef51daa | 9107 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9108 | } |
9109 | ||
0f212204 JA |
9110 | /* |
9111 | * Remove this io_uring_file -> task mapping. | |
9112 | */ | |
eef51daa | 9113 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9114 | { |
9115 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9116 | struct io_tctx_node *node; |
2941267b | 9117 | |
eebd2e37 PB |
9118 | if (!tctx) |
9119 | return; | |
13bf43f5 PB |
9120 | node = xa_erase(&tctx->xa, index); |
9121 | if (!node) | |
2941267b | 9122 | return; |
0f212204 | 9123 | |
13bf43f5 PB |
9124 | WARN_ON_ONCE(current != node->task); |
9125 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9126 | ||
9127 | mutex_lock(&node->ctx->uring_lock); | |
9128 | list_del(&node->ctx_node); | |
9129 | mutex_unlock(&node->ctx->uring_lock); | |
9130 | ||
baf186c4 | 9131 | if (tctx->last == node->ctx) |
0f212204 | 9132 | tctx->last = NULL; |
13bf43f5 | 9133 | kfree(node); |
0f212204 JA |
9134 | } |
9135 | ||
8452d4a6 | 9136 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9137 | { |
ba5ef6dc | 9138 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9139 | struct io_tctx_node *node; |
de7f1d9e PB |
9140 | unsigned long index; |
9141 | ||
13bf43f5 | 9142 | xa_for_each(&tctx->xa, index, node) |
eef51daa | 9143 | io_uring_del_tctx_node(index); |
b16ef427 ME |
9144 | if (wq) { |
9145 | /* | |
9146 | * Must be after io_uring_del_task_file() (removes nodes under | |
9147 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9148 | */ | |
9149 | tctx->io_wq = NULL; | |
ba5ef6dc | 9150 | io_wq_put_and_exit(wq); |
b16ef427 | 9151 | } |
de7f1d9e PB |
9152 | } |
9153 | ||
3f48cf18 | 9154 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9155 | { |
3f48cf18 PB |
9156 | if (tracked) |
9157 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9158 | return percpu_counter_sum(&tctx->inflight); |
9159 | } | |
9160 | ||
09899b19 PB |
9161 | static void io_uring_drop_tctx_refs(struct task_struct *task) |
9162 | { | |
9163 | struct io_uring_task *tctx = task->io_uring; | |
9164 | unsigned int refs = tctx->cached_refs; | |
9165 | ||
e9dbe221 PB |
9166 | if (refs) { |
9167 | tctx->cached_refs = 0; | |
9168 | percpu_counter_sub(&tctx->inflight, refs); | |
9169 | put_task_struct_many(task, refs); | |
9170 | } | |
09899b19 PB |
9171 | } |
9172 | ||
78cc687b PB |
9173 | /* |
9174 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
9175 | * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation. | |
9176 | */ | |
9177 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9178 | { |
521d6a73 | 9179 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9180 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9181 | s64 inflight; |
9182 | DEFINE_WAIT(wait); | |
fdaf083c | 9183 | |
78cc687b PB |
9184 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9185 | ||
6d042ffb PO |
9186 | if (!current->io_uring) |
9187 | return; | |
17a91051 PB |
9188 | if (tctx->io_wq) |
9189 | io_wq_exit_start(tctx->io_wq); | |
9190 | ||
0e9ddb39 PB |
9191 | atomic_inc(&tctx->in_idle); |
9192 | do { | |
e9dbe221 | 9193 | io_uring_drop_tctx_refs(current); |
0e9ddb39 | 9194 | /* read completions before cancelations */ |
78cc687b | 9195 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9196 | if (!inflight) |
9197 | break; | |
fdaf083c | 9198 | |
78cc687b PB |
9199 | if (!sqd) { |
9200 | struct io_tctx_node *node; | |
9201 | unsigned long index; | |
0f212204 | 9202 | |
78cc687b PB |
9203 | xa_for_each(&tctx->xa, index, node) { |
9204 | /* sqpoll task will cancel all its requests */ | |
9205 | if (node->ctx->sq_data) | |
9206 | continue; | |
9207 | io_uring_try_cancel_requests(node->ctx, current, | |
9208 | cancel_all); | |
9209 | } | |
9210 | } else { | |
9211 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9212 | io_uring_try_cancel_requests(ctx, current, | |
9213 | cancel_all); | |
9214 | } | |
17a91051 | 9215 | |
0f212204 | 9216 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
e9dbe221 | 9217 | io_uring_drop_tctx_refs(current); |
0f212204 | 9218 | /* |
a1bb3cd5 PB |
9219 | * If we've seen completions, retry without waiting. This |
9220 | * avoids a race where a completion comes in before we did | |
9221 | * prepare_to_wait(). | |
0f212204 | 9222 | */ |
3dd0c97a | 9223 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9224 | schedule(); |
f57555ed | 9225 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9226 | } while (1); |
fdaf083c | 9227 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9228 | |
8452d4a6 | 9229 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9230 | if (cancel_all) { |
3f48cf18 PB |
9231 | /* for exec all current's requests should be gone, kill tctx */ |
9232 | __io_uring_free(current); | |
9233 | } | |
44e728b8 PB |
9234 | } |
9235 | ||
f552a27a | 9236 | void __io_uring_cancel(bool cancel_all) |
78cc687b | 9237 | { |
f552a27a | 9238 | io_uring_cancel_generic(cancel_all, NULL); |
78cc687b PB |
9239 | } |
9240 | ||
6c5c240e RP |
9241 | static void *io_uring_validate_mmap_request(struct file *file, |
9242 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9243 | { |
2b188cc1 | 9244 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9245 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9246 | struct page *page; |
9247 | void *ptr; | |
9248 | ||
9249 | switch (offset) { | |
9250 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9251 | case IORING_OFF_CQ_RING: |
9252 | ptr = ctx->rings; | |
2b188cc1 JA |
9253 | break; |
9254 | case IORING_OFF_SQES: | |
9255 | ptr = ctx->sq_sqes; | |
9256 | break; | |
2b188cc1 | 9257 | default: |
6c5c240e | 9258 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9259 | } |
9260 | ||
9261 | page = virt_to_head_page(ptr); | |
a50b854e | 9262 | if (sz > page_size(page)) |
6c5c240e RP |
9263 | return ERR_PTR(-EINVAL); |
9264 | ||
9265 | return ptr; | |
9266 | } | |
9267 | ||
9268 | #ifdef CONFIG_MMU | |
9269 | ||
9270 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9271 | { | |
9272 | size_t sz = vma->vm_end - vma->vm_start; | |
9273 | unsigned long pfn; | |
9274 | void *ptr; | |
9275 | ||
9276 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9277 | if (IS_ERR(ptr)) | |
9278 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9279 | |
9280 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9281 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9282 | } | |
9283 | ||
6c5c240e RP |
9284 | #else /* !CONFIG_MMU */ |
9285 | ||
9286 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9287 | { | |
9288 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9289 | } | |
9290 | ||
9291 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9292 | { | |
9293 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9294 | } | |
9295 | ||
9296 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9297 | unsigned long addr, unsigned long len, | |
9298 | unsigned long pgoff, unsigned long flags) | |
9299 | { | |
9300 | void *ptr; | |
9301 | ||
9302 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9303 | if (IS_ERR(ptr)) | |
9304 | return PTR_ERR(ptr); | |
9305 | ||
9306 | return (unsigned long) ptr; | |
9307 | } | |
9308 | ||
9309 | #endif /* !CONFIG_MMU */ | |
9310 | ||
d9d05217 | 9311 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9312 | { |
9313 | DEFINE_WAIT(wait); | |
9314 | ||
9315 | do { | |
9316 | if (!io_sqring_full(ctx)) | |
9317 | break; | |
90554200 JA |
9318 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9319 | ||
9320 | if (!io_sqring_full(ctx)) | |
9321 | break; | |
90554200 JA |
9322 | schedule(); |
9323 | } while (!signal_pending(current)); | |
9324 | ||
9325 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9326 | return 0; |
90554200 JA |
9327 | } |
9328 | ||
c73ebb68 HX |
9329 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9330 | struct __kernel_timespec __user **ts, | |
9331 | const sigset_t __user **sig) | |
9332 | { | |
9333 | struct io_uring_getevents_arg arg; | |
9334 | ||
9335 | /* | |
9336 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9337 | * is just a pointer to the sigset_t. | |
9338 | */ | |
9339 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9340 | *sig = (const sigset_t __user *) argp; | |
9341 | *ts = NULL; | |
9342 | return 0; | |
9343 | } | |
9344 | ||
9345 | /* | |
9346 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9347 | * timespec and sigset_t pointers if good. | |
9348 | */ | |
9349 | if (*argsz != sizeof(arg)) | |
9350 | return -EINVAL; | |
9351 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9352 | return -EFAULT; | |
9353 | *sig = u64_to_user_ptr(arg.sigmask); | |
9354 | *argsz = arg.sigmask_sz; | |
9355 | *ts = u64_to_user_ptr(arg.ts); | |
9356 | return 0; | |
9357 | } | |
9358 | ||
2b188cc1 | 9359 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9360 | u32, min_complete, u32, flags, const void __user *, argp, |
9361 | size_t, argsz) | |
2b188cc1 JA |
9362 | { |
9363 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9364 | int submitted = 0; |
9365 | struct fd f; | |
33f993da | 9366 | long ret; |
2b188cc1 | 9367 | |
4c6e277c | 9368 | io_run_task_work(); |
b41e9852 | 9369 | |
33f993da PB |
9370 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9371 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9372 | return -EINVAL; |
9373 | ||
9374 | f = fdget(fd); | |
33f993da | 9375 | if (unlikely(!f.file)) |
2b188cc1 JA |
9376 | return -EBADF; |
9377 | ||
9378 | ret = -EOPNOTSUPP; | |
33f993da | 9379 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9380 | goto out_fput; |
9381 | ||
9382 | ret = -ENXIO; | |
9383 | ctx = f.file->private_data; | |
33f993da | 9384 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9385 | goto out_fput; |
9386 | ||
7e84e1c7 | 9387 | ret = -EBADFD; |
33f993da | 9388 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9389 | goto out; |
9390 | ||
6c271ce2 JA |
9391 | /* |
9392 | * For SQ polling, the thread will do all submissions and completions. | |
9393 | * Just return the requested submit count, and wake the thread if | |
9394 | * we were asked to. | |
9395 | */ | |
b2a9eada | 9396 | ret = 0; |
6c271ce2 | 9397 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
90f67366 | 9398 | io_cqring_overflow_flush(ctx); |
89448c47 | 9399 | |
21f96522 JA |
9400 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9401 | ret = -EOWNERDEAD; | |
04147488 | 9402 | goto out; |
21f96522 | 9403 | } |
6c271ce2 | 9404 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9405 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9406 | if (flags & IORING_ENTER_SQ_WAIT) { |
9407 | ret = io_sqpoll_wait_sq(ctx); | |
9408 | if (ret) | |
9409 | goto out; | |
9410 | } | |
6c271ce2 | 9411 | submitted = to_submit; |
b2a9eada | 9412 | } else if (to_submit) { |
eef51daa | 9413 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
9414 | if (unlikely(ret)) |
9415 | goto out; | |
2b188cc1 | 9416 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9417 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9418 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9419 | |
9420 | if (submitted != to_submit) | |
9421 | goto out; | |
2b188cc1 JA |
9422 | } |
9423 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9424 | const sigset_t __user *sig; |
9425 | struct __kernel_timespec __user *ts; | |
9426 | ||
9427 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9428 | if (unlikely(ret)) | |
9429 | goto out; | |
9430 | ||
2b188cc1 JA |
9431 | min_complete = min(min_complete, ctx->cq_entries); |
9432 | ||
32b2244a XW |
9433 | /* |
9434 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9435 | * space applications don't need to do io completion events | |
9436 | * polling again, they can rely on io_sq_thread to do polling | |
9437 | * work, which can reduce cpu usage and uring_lock contention. | |
9438 | */ | |
9439 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9440 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9441 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9442 | } else { |
c73ebb68 | 9443 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9444 | } |
2b188cc1 JA |
9445 | } |
9446 | ||
7c504e65 | 9447 | out: |
6805b32e | 9448 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9449 | out_fput: |
9450 | fdput(f); | |
9451 | return submitted ? submitted : ret; | |
9452 | } | |
9453 | ||
bebdb65e | 9454 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9455 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9456 | const struct cred *cred) | |
87ce955b | 9457 | { |
87ce955b JA |
9458 | struct user_namespace *uns = seq_user_ns(m); |
9459 | struct group_info *gi; | |
9460 | kernel_cap_t cap; | |
9461 | unsigned __capi; | |
9462 | int g; | |
9463 | ||
9464 | seq_printf(m, "%5d\n", id); | |
9465 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9466 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9467 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9468 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9469 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9470 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9471 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9472 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9473 | seq_puts(m, "\n\tGroups:\t"); | |
9474 | gi = cred->group_info; | |
9475 | for (g = 0; g < gi->ngroups; g++) { | |
9476 | seq_put_decimal_ull(m, g ? " " : "", | |
9477 | from_kgid_munged(uns, gi->gid[g])); | |
9478 | } | |
9479 | seq_puts(m, "\n\tCapEff:\t"); | |
9480 | cap = cred->cap_effective; | |
9481 | CAP_FOR_EACH_U32(__capi) | |
9482 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9483 | seq_putc(m, '\n'); | |
9484 | return 0; | |
9485 | } | |
9486 | ||
9487 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9488 | { | |
dbbe9c64 | 9489 | struct io_sq_data *sq = NULL; |
fad8e0de | 9490 | bool has_lock; |
87ce955b JA |
9491 | int i; |
9492 | ||
fad8e0de JA |
9493 | /* |
9494 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9495 | * since fdinfo case grabs it in the opposite direction of normal use | |
9496 | * cases. If we fail to get the lock, we just don't iterate any | |
9497 | * structures that could be going away outside the io_uring mutex. | |
9498 | */ | |
9499 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9500 | ||
5f3f26f9 | 9501 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9502 | sq = ctx->sq_data; |
5f3f26f9 JA |
9503 | if (!sq->thread) |
9504 | sq = NULL; | |
9505 | } | |
dbbe9c64 JQ |
9506 | |
9507 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9508 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9509 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9510 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9511 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9512 | |
87ce955b JA |
9513 | if (f) |
9514 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9515 | else | |
9516 | seq_printf(m, "%5u: <none>\n", i); | |
9517 | } | |
9518 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9519 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9520 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9521 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9522 | |
4751f53d | 9523 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9524 | } |
61cf9370 MWO |
9525 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9526 | unsigned long index; | |
9527 | const struct cred *cred; | |
9528 | ||
87ce955b | 9529 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9530 | xa_for_each(&ctx->personalities, index, cred) |
9531 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9532 | } |
d7718a9d | 9533 | seq_printf(m, "PollList:\n"); |
79ebeaee | 9534 | spin_lock(&ctx->completion_lock); |
d7718a9d JA |
9535 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
9536 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9537 | struct io_kiocb *req; | |
9538 | ||
9539 | hlist_for_each_entry(req, list, hash_node) | |
9540 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9541 | req->task->task_works != NULL); | |
9542 | } | |
79ebeaee | 9543 | spin_unlock(&ctx->completion_lock); |
fad8e0de JA |
9544 | if (has_lock) |
9545 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9546 | } |
9547 | ||
9548 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9549 | { | |
9550 | struct io_ring_ctx *ctx = f->private_data; | |
9551 | ||
9552 | if (percpu_ref_tryget(&ctx->refs)) { | |
9553 | __io_uring_show_fdinfo(ctx, m); | |
9554 | percpu_ref_put(&ctx->refs); | |
9555 | } | |
9556 | } | |
bebdb65e | 9557 | #endif |
87ce955b | 9558 | |
2b188cc1 JA |
9559 | static const struct file_operations io_uring_fops = { |
9560 | .release = io_uring_release, | |
9561 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9562 | #ifndef CONFIG_MMU |
9563 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9564 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9565 | #endif | |
2b188cc1 JA |
9566 | .poll = io_uring_poll, |
9567 | .fasync = io_uring_fasync, | |
bebdb65e | 9568 | #ifdef CONFIG_PROC_FS |
87ce955b | 9569 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9570 | #endif |
2b188cc1 JA |
9571 | }; |
9572 | ||
9573 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9574 | struct io_uring_params *p) | |
9575 | { | |
75b28aff HV |
9576 | struct io_rings *rings; |
9577 | size_t size, sq_array_offset; | |
2b188cc1 | 9578 | |
bd740481 JA |
9579 | /* make sure these are sane, as we already accounted them */ |
9580 | ctx->sq_entries = p->sq_entries; | |
9581 | ctx->cq_entries = p->cq_entries; | |
9582 | ||
75b28aff HV |
9583 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9584 | if (size == SIZE_MAX) | |
9585 | return -EOVERFLOW; | |
9586 | ||
9587 | rings = io_mem_alloc(size); | |
9588 | if (!rings) | |
2b188cc1 JA |
9589 | return -ENOMEM; |
9590 | ||
75b28aff HV |
9591 | ctx->rings = rings; |
9592 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9593 | rings->sq_ring_mask = p->sq_entries - 1; | |
9594 | rings->cq_ring_mask = p->cq_entries - 1; | |
9595 | rings->sq_ring_entries = p->sq_entries; | |
9596 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
9597 | |
9598 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9599 | if (size == SIZE_MAX) { |
9600 | io_mem_free(ctx->rings); | |
9601 | ctx->rings = NULL; | |
2b188cc1 | 9602 | return -EOVERFLOW; |
eb065d30 | 9603 | } |
2b188cc1 JA |
9604 | |
9605 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9606 | if (!ctx->sq_sqes) { |
9607 | io_mem_free(ctx->rings); | |
9608 | ctx->rings = NULL; | |
2b188cc1 | 9609 | return -ENOMEM; |
eb065d30 | 9610 | } |
2b188cc1 | 9611 | |
2b188cc1 JA |
9612 | return 0; |
9613 | } | |
9614 | ||
9faadcc8 PB |
9615 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9616 | { | |
9617 | int ret, fd; | |
9618 | ||
9619 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9620 | if (fd < 0) | |
9621 | return fd; | |
9622 | ||
eef51daa | 9623 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
9624 | if (ret) { |
9625 | put_unused_fd(fd); | |
9626 | return ret; | |
9627 | } | |
9628 | fd_install(fd, file); | |
9629 | return fd; | |
9630 | } | |
9631 | ||
2b188cc1 JA |
9632 | /* |
9633 | * Allocate an anonymous fd, this is what constitutes the application | |
9634 | * visible backing of an io_uring instance. The application mmaps this | |
9635 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9636 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9637 | */ | |
9faadcc8 | 9638 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9639 | { |
9640 | struct file *file; | |
9faadcc8 | 9641 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9642 | int ret; |
9643 | ||
2b188cc1 JA |
9644 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9645 | &ctx->ring_sock); | |
9646 | if (ret) | |
9faadcc8 | 9647 | return ERR_PTR(ret); |
2b188cc1 JA |
9648 | #endif |
9649 | ||
2b188cc1 JA |
9650 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9651 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9652 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9653 | if (IS_ERR(file)) { |
9654 | sock_release(ctx->ring_sock); | |
9655 | ctx->ring_sock = NULL; | |
9656 | } else { | |
9657 | ctx->ring_sock->file = file; | |
0f212204 | 9658 | } |
2b188cc1 | 9659 | #endif |
9faadcc8 | 9660 | return file; |
2b188cc1 JA |
9661 | } |
9662 | ||
7f13657d XW |
9663 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9664 | struct io_uring_params __user *params) | |
2b188cc1 | 9665 | { |
2b188cc1 | 9666 | struct io_ring_ctx *ctx; |
9faadcc8 | 9667 | struct file *file; |
2b188cc1 JA |
9668 | int ret; |
9669 | ||
8110c1a6 | 9670 | if (!entries) |
2b188cc1 | 9671 | return -EINVAL; |
8110c1a6 JA |
9672 | if (entries > IORING_MAX_ENTRIES) { |
9673 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9674 | return -EINVAL; | |
9675 | entries = IORING_MAX_ENTRIES; | |
9676 | } | |
2b188cc1 JA |
9677 | |
9678 | /* | |
9679 | * Use twice as many entries for the CQ ring. It's possible for the | |
9680 | * application to drive a higher depth than the size of the SQ ring, | |
9681 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9682 | * some flexibility in overcommitting a bit. If the application has |
9683 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9684 | * of CQ ring entries manually. | |
2b188cc1 JA |
9685 | */ |
9686 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9687 | if (p->flags & IORING_SETUP_CQSIZE) { |
9688 | /* | |
9689 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9690 | * to a power-of-two, if it isn't already. We do NOT impose | |
9691 | * any cq vs sq ring sizing. | |
9692 | */ | |
eb2667b3 | 9693 | if (!p->cq_entries) |
33a107f0 | 9694 | return -EINVAL; |
8110c1a6 JA |
9695 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9696 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9697 | return -EINVAL; | |
9698 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9699 | } | |
eb2667b3 JQ |
9700 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9701 | if (p->cq_entries < p->sq_entries) | |
9702 | return -EINVAL; | |
33a107f0 JA |
9703 | } else { |
9704 | p->cq_entries = 2 * p->sq_entries; | |
9705 | } | |
2b188cc1 | 9706 | |
2b188cc1 | 9707 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9708 | if (!ctx) |
2b188cc1 | 9709 | return -ENOMEM; |
2b188cc1 | 9710 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9711 | if (!capable(CAP_IPC_LOCK)) |
9712 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9713 | |
9714 | /* | |
9715 | * This is just grabbed for accounting purposes. When a process exits, | |
9716 | * the mm is exited and dropped before the files, hence we need to hang | |
9717 | * on to this mm purely for the purposes of being able to unaccount | |
9718 | * memory (locked/pinned vm). It's not used for anything else. | |
9719 | */ | |
6b7898eb | 9720 | mmgrab(current->mm); |
2aede0e4 | 9721 | ctx->mm_account = current->mm; |
6b7898eb | 9722 | |
2b188cc1 JA |
9723 | ret = io_allocate_scq_urings(ctx, p); |
9724 | if (ret) | |
9725 | goto err; | |
9726 | ||
7e84e1c7 | 9727 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9728 | if (ret) |
9729 | goto err; | |
eae071c9 | 9730 | /* always set a rsrc node */ |
47b228ce PB |
9731 | ret = io_rsrc_node_switch_start(ctx); |
9732 | if (ret) | |
9733 | goto err; | |
eae071c9 | 9734 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 9735 | |
2b188cc1 | 9736 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9737 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9738 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9739 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9740 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9741 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9742 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9743 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9744 | |
9745 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9746 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9747 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9748 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9749 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9750 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9751 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9752 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9753 | |
7f13657d XW |
9754 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9755 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9756 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9757 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
9758 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
9759 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
9760 | |
9761 | if (copy_to_user(params, p, sizeof(*p))) { | |
9762 | ret = -EFAULT; | |
9763 | goto err; | |
9764 | } | |
d1719f70 | 9765 | |
9faadcc8 PB |
9766 | file = io_uring_get_file(ctx); |
9767 | if (IS_ERR(file)) { | |
9768 | ret = PTR_ERR(file); | |
9769 | goto err; | |
9770 | } | |
9771 | ||
044c1ab3 JA |
9772 | /* |
9773 | * Install ring fd as the very last thing, so we don't risk someone | |
9774 | * having closed it before we finish setup | |
9775 | */ | |
9faadcc8 PB |
9776 | ret = io_uring_install_fd(ctx, file); |
9777 | if (ret < 0) { | |
9778 | /* fput will clean it up */ | |
9779 | fput(file); | |
9780 | return ret; | |
9781 | } | |
044c1ab3 | 9782 | |
c826bd7a | 9783 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9784 | return ret; |
9785 | err: | |
9786 | io_ring_ctx_wait_and_kill(ctx); | |
9787 | return ret; | |
9788 | } | |
9789 | ||
9790 | /* | |
9791 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9792 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9793 | * params structure passed in. | |
9794 | */ | |
9795 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9796 | { | |
9797 | struct io_uring_params p; | |
2b188cc1 JA |
9798 | int i; |
9799 | ||
9800 | if (copy_from_user(&p, params, sizeof(p))) | |
9801 | return -EFAULT; | |
9802 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9803 | if (p.resv[i]) | |
9804 | return -EINVAL; | |
9805 | } | |
9806 | ||
6c271ce2 | 9807 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9808 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9809 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9810 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9811 | return -EINVAL; |
9812 | ||
7f13657d | 9813 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9814 | } |
9815 | ||
9816 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9817 | struct io_uring_params __user *, params) | |
9818 | { | |
9819 | return io_uring_setup(entries, params); | |
9820 | } | |
9821 | ||
66f4af93 JA |
9822 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9823 | { | |
9824 | struct io_uring_probe *p; | |
9825 | size_t size; | |
9826 | int i, ret; | |
9827 | ||
9828 | size = struct_size(p, ops, nr_args); | |
9829 | if (size == SIZE_MAX) | |
9830 | return -EOVERFLOW; | |
9831 | p = kzalloc(size, GFP_KERNEL); | |
9832 | if (!p) | |
9833 | return -ENOMEM; | |
9834 | ||
9835 | ret = -EFAULT; | |
9836 | if (copy_from_user(p, arg, size)) | |
9837 | goto out; | |
9838 | ret = -EINVAL; | |
9839 | if (memchr_inv(p, 0, size)) | |
9840 | goto out; | |
9841 | ||
9842 | p->last_op = IORING_OP_LAST - 1; | |
9843 | if (nr_args > IORING_OP_LAST) | |
9844 | nr_args = IORING_OP_LAST; | |
9845 | ||
9846 | for (i = 0; i < nr_args; i++) { | |
9847 | p->ops[i].op = i; | |
9848 | if (!io_op_defs[i].not_supported) | |
9849 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9850 | } | |
9851 | p->ops_len = i; | |
9852 | ||
9853 | ret = 0; | |
9854 | if (copy_to_user(arg, p, size)) | |
9855 | ret = -EFAULT; | |
9856 | out: | |
9857 | kfree(p); | |
9858 | return ret; | |
9859 | } | |
9860 | ||
071698e1 JA |
9861 | static int io_register_personality(struct io_ring_ctx *ctx) |
9862 | { | |
4379bf8b | 9863 | const struct cred *creds; |
61cf9370 | 9864 | u32 id; |
1e6fa521 | 9865 | int ret; |
071698e1 | 9866 | |
4379bf8b | 9867 | creds = get_current_cred(); |
1e6fa521 | 9868 | |
61cf9370 MWO |
9869 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9870 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
a30f895a JA |
9871 | if (ret < 0) { |
9872 | put_cred(creds); | |
9873 | return ret; | |
9874 | } | |
9875 | return id; | |
071698e1 JA |
9876 | } |
9877 | ||
21b55dbc SG |
9878 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9879 | unsigned int nr_args) | |
9880 | { | |
9881 | struct io_uring_restriction *res; | |
9882 | size_t size; | |
9883 | int i, ret; | |
9884 | ||
7e84e1c7 SG |
9885 | /* Restrictions allowed only if rings started disabled */ |
9886 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9887 | return -EBADFD; | |
9888 | ||
21b55dbc | 9889 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9890 | if (ctx->restrictions.registered) |
21b55dbc SG |
9891 | return -EBUSY; |
9892 | ||
9893 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9894 | return -EINVAL; | |
9895 | ||
9896 | size = array_size(nr_args, sizeof(*res)); | |
9897 | if (size == SIZE_MAX) | |
9898 | return -EOVERFLOW; | |
9899 | ||
9900 | res = memdup_user(arg, size); | |
9901 | if (IS_ERR(res)) | |
9902 | return PTR_ERR(res); | |
9903 | ||
9904 | ret = 0; | |
9905 | ||
9906 | for (i = 0; i < nr_args; i++) { | |
9907 | switch (res[i].opcode) { | |
9908 | case IORING_RESTRICTION_REGISTER_OP: | |
9909 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9910 | ret = -EINVAL; | |
9911 | goto out; | |
9912 | } | |
9913 | ||
9914 | __set_bit(res[i].register_op, | |
9915 | ctx->restrictions.register_op); | |
9916 | break; | |
9917 | case IORING_RESTRICTION_SQE_OP: | |
9918 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9919 | ret = -EINVAL; | |
9920 | goto out; | |
9921 | } | |
9922 | ||
9923 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9924 | break; | |
9925 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9926 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9927 | break; | |
9928 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9929 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9930 | break; | |
9931 | default: | |
9932 | ret = -EINVAL; | |
9933 | goto out; | |
9934 | } | |
9935 | } | |
9936 | ||
9937 | out: | |
9938 | /* Reset all restrictions if an error happened */ | |
9939 | if (ret != 0) | |
9940 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9941 | else | |
7e84e1c7 | 9942 | ctx->restrictions.registered = true; |
21b55dbc SG |
9943 | |
9944 | kfree(res); | |
9945 | return ret; | |
9946 | } | |
9947 | ||
7e84e1c7 SG |
9948 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9949 | { | |
9950 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9951 | return -EBADFD; | |
9952 | ||
9953 | if (ctx->restrictions.registered) | |
9954 | ctx->restricted = 1; | |
9955 | ||
0298ef96 PB |
9956 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9957 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9958 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9959 | return 0; |
9960 | } | |
9961 | ||
fdecb662 | 9962 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 9963 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
9964 | unsigned nr_args) |
9965 | { | |
9966 | __u32 tmp; | |
9967 | int err; | |
9968 | ||
c3bdad02 PB |
9969 | if (up->resv) |
9970 | return -EINVAL; | |
98f0b3b4 PB |
9971 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
9972 | return -EOVERFLOW; | |
9973 | err = io_rsrc_node_switch_start(ctx); | |
9974 | if (err) | |
9975 | return err; | |
9976 | ||
fdecb662 PB |
9977 | switch (type) { |
9978 | case IORING_RSRC_FILE: | |
98f0b3b4 | 9979 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
9980 | case IORING_RSRC_BUFFER: |
9981 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
9982 | } |
9983 | return -EINVAL; | |
9984 | } | |
9985 | ||
c3bdad02 PB |
9986 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
9987 | unsigned nr_args) | |
98f0b3b4 | 9988 | { |
c3bdad02 | 9989 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
9990 | |
9991 | if (!nr_args) | |
9992 | return -EINVAL; | |
c3bdad02 PB |
9993 | memset(&up, 0, sizeof(up)); |
9994 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
9995 | return -EFAULT; | |
9996 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
9997 | } | |
9998 | ||
9999 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 10000 | unsigned size, unsigned type) |
c3bdad02 PB |
10001 | { |
10002 | struct io_uring_rsrc_update2 up; | |
10003 | ||
10004 | if (size != sizeof(up)) | |
10005 | return -EINVAL; | |
98f0b3b4 PB |
10006 | if (copy_from_user(&up, arg, sizeof(up))) |
10007 | return -EFAULT; | |
992da01a | 10008 | if (!up.nr || up.resv) |
98f0b3b4 | 10009 | return -EINVAL; |
992da01a | 10010 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
10011 | } |
10012 | ||
792e3582 | 10013 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 10014 | unsigned int size, unsigned int type) |
792e3582 PB |
10015 | { |
10016 | struct io_uring_rsrc_register rr; | |
10017 | ||
10018 | /* keep it extendible */ | |
10019 | if (size != sizeof(rr)) | |
10020 | return -EINVAL; | |
10021 | ||
10022 | memset(&rr, 0, sizeof(rr)); | |
10023 | if (copy_from_user(&rr, arg, size)) | |
10024 | return -EFAULT; | |
992da01a | 10025 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
10026 | return -EINVAL; |
10027 | ||
992da01a | 10028 | switch (type) { |
792e3582 PB |
10029 | case IORING_RSRC_FILE: |
10030 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
10031 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
10032 | case IORING_RSRC_BUFFER: |
10033 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
10034 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
10035 | } |
10036 | return -EINVAL; | |
10037 | } | |
10038 | ||
fe76421d JA |
10039 | static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg, |
10040 | unsigned len) | |
10041 | { | |
10042 | struct io_uring_task *tctx = current->io_uring; | |
10043 | cpumask_var_t new_mask; | |
10044 | int ret; | |
10045 | ||
10046 | if (!tctx || !tctx->io_wq) | |
10047 | return -EINVAL; | |
10048 | ||
10049 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
10050 | return -ENOMEM; | |
10051 | ||
10052 | cpumask_clear(new_mask); | |
10053 | if (len > cpumask_size()) | |
10054 | len = cpumask_size(); | |
10055 | ||
10056 | if (copy_from_user(new_mask, arg, len)) { | |
10057 | free_cpumask_var(new_mask); | |
10058 | return -EFAULT; | |
10059 | } | |
10060 | ||
10061 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
10062 | free_cpumask_var(new_mask); | |
10063 | return ret; | |
10064 | } | |
10065 | ||
10066 | static int io_unregister_iowq_aff(struct io_ring_ctx *ctx) | |
10067 | { | |
10068 | struct io_uring_task *tctx = current->io_uring; | |
10069 | ||
10070 | if (!tctx || !tctx->io_wq) | |
10071 | return -EINVAL; | |
10072 | ||
10073 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
10074 | } | |
10075 | ||
071698e1 JA |
10076 | static bool io_register_op_must_quiesce(int op) |
10077 | { | |
10078 | switch (op) { | |
bd54b6fe BM |
10079 | case IORING_REGISTER_BUFFERS: |
10080 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 10081 | case IORING_REGISTER_FILES: |
071698e1 JA |
10082 | case IORING_UNREGISTER_FILES: |
10083 | case IORING_REGISTER_FILES_UPDATE: | |
10084 | case IORING_REGISTER_PROBE: | |
10085 | case IORING_REGISTER_PERSONALITY: | |
10086 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
10087 | case IORING_REGISTER_FILES2: |
10088 | case IORING_REGISTER_FILES_UPDATE2: | |
10089 | case IORING_REGISTER_BUFFERS2: | |
10090 | case IORING_REGISTER_BUFFERS_UPDATE: | |
fe76421d JA |
10091 | case IORING_REGISTER_IOWQ_AFF: |
10092 | case IORING_UNREGISTER_IOWQ_AFF: | |
071698e1 JA |
10093 | return false; |
10094 | default: | |
10095 | return true; | |
10096 | } | |
10097 | } | |
10098 | ||
e73c5c7c PB |
10099 | static int io_ctx_quiesce(struct io_ring_ctx *ctx) |
10100 | { | |
10101 | long ret; | |
10102 | ||
10103 | percpu_ref_kill(&ctx->refs); | |
10104 | ||
10105 | /* | |
10106 | * Drop uring mutex before waiting for references to exit. If another | |
10107 | * thread is currently inside io_uring_enter() it might need to grab the | |
10108 | * uring_lock to make progress. If we hold it here across the drain | |
10109 | * wait, then we can deadlock. It's safe to drop the mutex here, since | |
10110 | * no new references will come in after we've killed the percpu ref. | |
10111 | */ | |
10112 | mutex_unlock(&ctx->uring_lock); | |
10113 | do { | |
10114 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10115 | if (!ret) | |
10116 | break; | |
10117 | ret = io_run_task_work_sig(); | |
10118 | } while (ret >= 0); | |
10119 | mutex_lock(&ctx->uring_lock); | |
10120 | ||
10121 | if (ret) | |
10122 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); | |
10123 | return ret; | |
10124 | } | |
10125 | ||
edafccee JA |
10126 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10127 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10128 | __releases(ctx->uring_lock) |
10129 | __acquires(ctx->uring_lock) | |
edafccee JA |
10130 | { |
10131 | int ret; | |
10132 | ||
35fa71a0 JA |
10133 | /* |
10134 | * We're inside the ring mutex, if the ref is already dying, then | |
10135 | * someone else killed the ctx or is already going through | |
10136 | * io_uring_register(). | |
10137 | */ | |
10138 | if (percpu_ref_is_dying(&ctx->refs)) | |
10139 | return -ENXIO; | |
10140 | ||
75c4021a PB |
10141 | if (ctx->restricted) { |
10142 | if (opcode >= IORING_REGISTER_LAST) | |
10143 | return -EINVAL; | |
10144 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10145 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10146 | return -EACCES; | |
10147 | } | |
10148 | ||
071698e1 | 10149 | if (io_register_op_must_quiesce(opcode)) { |
e73c5c7c PB |
10150 | ret = io_ctx_quiesce(ctx); |
10151 | if (ret) | |
f70865db | 10152 | return ret; |
05f3fb3c | 10153 | } |
edafccee JA |
10154 | |
10155 | switch (opcode) { | |
10156 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10157 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10158 | break; |
10159 | case IORING_UNREGISTER_BUFFERS: | |
10160 | ret = -EINVAL; | |
10161 | if (arg || nr_args) | |
10162 | break; | |
0a96bbe4 | 10163 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10164 | break; |
6b06314c | 10165 | case IORING_REGISTER_FILES: |
792e3582 | 10166 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10167 | break; |
10168 | case IORING_UNREGISTER_FILES: | |
10169 | ret = -EINVAL; | |
10170 | if (arg || nr_args) | |
10171 | break; | |
10172 | ret = io_sqe_files_unregister(ctx); | |
10173 | break; | |
c3a31e60 | 10174 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10175 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10176 | break; |
9b402849 | 10177 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10178 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10179 | ret = -EINVAL; |
10180 | if (nr_args != 1) | |
10181 | break; | |
10182 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10183 | if (ret) |
10184 | break; | |
10185 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10186 | ctx->eventfd_async = 1; | |
10187 | else | |
10188 | ctx->eventfd_async = 0; | |
9b402849 JA |
10189 | break; |
10190 | case IORING_UNREGISTER_EVENTFD: | |
10191 | ret = -EINVAL; | |
10192 | if (arg || nr_args) | |
10193 | break; | |
10194 | ret = io_eventfd_unregister(ctx); | |
10195 | break; | |
66f4af93 JA |
10196 | case IORING_REGISTER_PROBE: |
10197 | ret = -EINVAL; | |
10198 | if (!arg || nr_args > 256) | |
10199 | break; | |
10200 | ret = io_probe(ctx, arg, nr_args); | |
10201 | break; | |
071698e1 JA |
10202 | case IORING_REGISTER_PERSONALITY: |
10203 | ret = -EINVAL; | |
10204 | if (arg || nr_args) | |
10205 | break; | |
10206 | ret = io_register_personality(ctx); | |
10207 | break; | |
10208 | case IORING_UNREGISTER_PERSONALITY: | |
10209 | ret = -EINVAL; | |
10210 | if (arg) | |
10211 | break; | |
10212 | ret = io_unregister_personality(ctx, nr_args); | |
10213 | break; | |
7e84e1c7 SG |
10214 | case IORING_REGISTER_ENABLE_RINGS: |
10215 | ret = -EINVAL; | |
10216 | if (arg || nr_args) | |
10217 | break; | |
10218 | ret = io_register_enable_rings(ctx); | |
10219 | break; | |
21b55dbc SG |
10220 | case IORING_REGISTER_RESTRICTIONS: |
10221 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10222 | break; | |
992da01a PB |
10223 | case IORING_REGISTER_FILES2: |
10224 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10225 | break; | |
10226 | case IORING_REGISTER_FILES_UPDATE2: | |
10227 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10228 | IORING_RSRC_FILE); | |
10229 | break; | |
10230 | case IORING_REGISTER_BUFFERS2: | |
10231 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10232 | break; |
992da01a PB |
10233 | case IORING_REGISTER_BUFFERS_UPDATE: |
10234 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10235 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10236 | break; |
fe76421d JA |
10237 | case IORING_REGISTER_IOWQ_AFF: |
10238 | ret = -EINVAL; | |
10239 | if (!arg || !nr_args) | |
10240 | break; | |
10241 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
10242 | break; | |
10243 | case IORING_UNREGISTER_IOWQ_AFF: | |
10244 | ret = -EINVAL; | |
10245 | if (arg || nr_args) | |
10246 | break; | |
10247 | ret = io_unregister_iowq_aff(ctx); | |
10248 | break; | |
edafccee JA |
10249 | default: |
10250 | ret = -EINVAL; | |
10251 | break; | |
10252 | } | |
10253 | ||
071698e1 | 10254 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10255 | /* bring the ctx back to life */ |
05f3fb3c | 10256 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10257 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10258 | } |
edafccee JA |
10259 | return ret; |
10260 | } | |
10261 | ||
10262 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10263 | void __user *, arg, unsigned int, nr_args) | |
10264 | { | |
10265 | struct io_ring_ctx *ctx; | |
10266 | long ret = -EBADF; | |
10267 | struct fd f; | |
10268 | ||
10269 | f = fdget(fd); | |
10270 | if (!f.file) | |
10271 | return -EBADF; | |
10272 | ||
10273 | ret = -EOPNOTSUPP; | |
10274 | if (f.file->f_op != &io_uring_fops) | |
10275 | goto out_fput; | |
10276 | ||
10277 | ctx = f.file->private_data; | |
10278 | ||
b6c23dd5 PB |
10279 | io_run_task_work(); |
10280 | ||
edafccee JA |
10281 | mutex_lock(&ctx->uring_lock); |
10282 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10283 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10284 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10285 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10286 | out_fput: |
10287 | fdput(f); | |
10288 | return ret; | |
10289 | } | |
10290 | ||
2b188cc1 JA |
10291 | static int __init io_uring_init(void) |
10292 | { | |
d7f62e82 SM |
10293 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10294 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10295 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10296 | } while (0) | |
10297 | ||
10298 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10299 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10300 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10301 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10302 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10303 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10304 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10305 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10306 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10307 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10308 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10309 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10310 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10311 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10312 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10313 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10314 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10315 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10316 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10317 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10318 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10319 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10320 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10321 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10322 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10323 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10324 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10325 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10326 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 10327 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 10328 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 10329 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10330 | |
b0d658ec PB |
10331 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10332 | sizeof(struct io_uring_rsrc_update)); | |
10333 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10334 | sizeof(struct io_uring_rsrc_update2)); | |
10335 | /* should fit into one byte */ | |
10336 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10337 | ||
d3656344 | 10338 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10339 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
16340eab | 10340 | |
91f245d5 JA |
10341 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10342 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10343 | return 0; |
10344 | }; | |
10345 | __initcall(io_uring_init); |