Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
d068b506 | 14 | * through a control-dependency in io_get_cqe (smp_store_release to |
1e84b97b SB |
15 | * store head will do). Failure to do so could lead to reading invalid |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
92 | |
93 | /* | |
94 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
95 | */ | |
96 | #define IORING_FILE_TABLE_SHIFT 9 | |
97 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
98 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
99 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
100 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
101 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 102 | |
489809e2 PB |
103 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
104 | ||
b16fed66 PB |
105 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
106 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
107 | IOSQE_BUFFER_SELECT) | |
108 | ||
2b188cc1 JA |
109 | struct io_uring { |
110 | u32 head ____cacheline_aligned_in_smp; | |
111 | u32 tail ____cacheline_aligned_in_smp; | |
112 | }; | |
113 | ||
1e84b97b | 114 | /* |
75b28aff HV |
115 | * This data is shared with the application through the mmap at offsets |
116 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
117 | * |
118 | * The offsets to the member fields are published through struct | |
119 | * io_sqring_offsets when calling io_uring_setup. | |
120 | */ | |
75b28aff | 121 | struct io_rings { |
1e84b97b SB |
122 | /* |
123 | * Head and tail offsets into the ring; the offsets need to be | |
124 | * masked to get valid indices. | |
125 | * | |
75b28aff HV |
126 | * The kernel controls head of the sq ring and the tail of the cq ring, |
127 | * and the application controls tail of the sq ring and the head of the | |
128 | * cq ring. | |
1e84b97b | 129 | */ |
75b28aff | 130 | struct io_uring sq, cq; |
1e84b97b | 131 | /* |
75b28aff | 132 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
133 | * ring_entries - 1) |
134 | */ | |
75b28aff HV |
135 | u32 sq_ring_mask, cq_ring_mask; |
136 | /* Ring sizes (constant, power of 2) */ | |
137 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
138 | /* |
139 | * Number of invalid entries dropped by the kernel due to | |
140 | * invalid index stored in array | |
141 | * | |
142 | * Written by the kernel, shouldn't be modified by the | |
143 | * application (i.e. get number of "new events" by comparing to | |
144 | * cached value). | |
145 | * | |
146 | * After a new SQ head value was read by the application this | |
147 | * counter includes all submissions that were dropped reaching | |
148 | * the new SQ head (and possibly more). | |
149 | */ | |
75b28aff | 150 | u32 sq_dropped; |
1e84b97b | 151 | /* |
0d9b5b3a | 152 | * Runtime SQ flags |
1e84b97b SB |
153 | * |
154 | * Written by the kernel, shouldn't be modified by the | |
155 | * application. | |
156 | * | |
157 | * The application needs a full memory barrier before checking | |
158 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
159 | */ | |
75b28aff | 160 | u32 sq_flags; |
0d9b5b3a SG |
161 | /* |
162 | * Runtime CQ flags | |
163 | * | |
164 | * Written by the application, shouldn't be modified by the | |
165 | * kernel. | |
166 | */ | |
167 | u32 cq_flags; | |
1e84b97b SB |
168 | /* |
169 | * Number of completion events lost because the queue was full; | |
170 | * this should be avoided by the application by making sure | |
0b4295b5 | 171 | * there are not more requests pending than there is space in |
1e84b97b SB |
172 | * the completion queue. |
173 | * | |
174 | * Written by the kernel, shouldn't be modified by the | |
175 | * application (i.e. get number of "new events" by comparing to | |
176 | * cached value). | |
177 | * | |
178 | * As completion events come in out of order this counter is not | |
179 | * ordered with any other data. | |
180 | */ | |
75b28aff | 181 | u32 cq_overflow; |
1e84b97b SB |
182 | /* |
183 | * Ring buffer of completion events. | |
184 | * | |
185 | * The kernel writes completion events fresh every time they are | |
186 | * produced, so the application is allowed to modify pending | |
187 | * entries. | |
188 | */ | |
75b28aff | 189 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
190 | }; |
191 | ||
45d189c6 PB |
192 | enum io_uring_cmd_flags { |
193 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 194 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
195 | }; |
196 | ||
edafccee JA |
197 | struct io_mapped_ubuf { |
198 | u64 ubuf; | |
4751f53d | 199 | u64 ubuf_end; |
edafccee | 200 | unsigned int nr_bvecs; |
de293938 | 201 | unsigned long acct_pages; |
41edf1a5 | 202 | struct bio_vec bvec[]; |
edafccee JA |
203 | }; |
204 | ||
50238531 BM |
205 | struct io_ring_ctx; |
206 | ||
6c2450ae PB |
207 | struct io_overflow_cqe { |
208 | struct io_uring_cqe cqe; | |
209 | struct list_head list; | |
210 | }; | |
211 | ||
a04b0ac0 PB |
212 | struct io_fixed_file { |
213 | /* file * with additional FFS_* flags */ | |
214 | unsigned long file_ptr; | |
215 | }; | |
216 | ||
269bbe5f BM |
217 | struct io_rsrc_put { |
218 | struct list_head list; | |
b60c8dce | 219 | u64 tag; |
50238531 BM |
220 | union { |
221 | void *rsrc; | |
222 | struct file *file; | |
bd54b6fe | 223 | struct io_mapped_ubuf *buf; |
50238531 | 224 | }; |
269bbe5f BM |
225 | }; |
226 | ||
aeca241b PB |
227 | struct io_file_table { |
228 | /* two level table */ | |
229 | struct io_fixed_file **files; | |
31b51510 JA |
230 | }; |
231 | ||
b895c9a6 | 232 | struct io_rsrc_node { |
05589553 XW |
233 | struct percpu_ref refs; |
234 | struct list_head node; | |
269bbe5f | 235 | struct list_head rsrc_list; |
b895c9a6 | 236 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 237 | struct llist_node llist; |
e297822b | 238 | bool done; |
05589553 XW |
239 | }; |
240 | ||
40ae0ff7 PB |
241 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
242 | ||
b895c9a6 | 243 | struct io_rsrc_data { |
05f3fb3c JA |
244 | struct io_ring_ctx *ctx; |
245 | ||
b60c8dce | 246 | u64 *tags; |
40ae0ff7 | 247 | rsrc_put_fn *do_put; |
3e942498 | 248 | atomic_t refs; |
05f3fb3c | 249 | struct completion done; |
8bad28d8 | 250 | bool quiesce; |
05f3fb3c JA |
251 | }; |
252 | ||
5a2e745d JA |
253 | struct io_buffer { |
254 | struct list_head list; | |
255 | __u64 addr; | |
d1f82808 | 256 | __u32 len; |
5a2e745d JA |
257 | __u16 bid; |
258 | }; | |
259 | ||
21b55dbc SG |
260 | struct io_restriction { |
261 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
262 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
263 | u8 sqe_flags_allowed; | |
264 | u8 sqe_flags_required; | |
7e84e1c7 | 265 | bool registered; |
21b55dbc SG |
266 | }; |
267 | ||
37d1e2e3 JA |
268 | enum { |
269 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
270 | IO_SQ_THREAD_SHOULD_PARK, | |
271 | }; | |
272 | ||
534ca6d6 JA |
273 | struct io_sq_data { |
274 | refcount_t refs; | |
9e138a48 | 275 | atomic_t park_pending; |
09a6f4ef | 276 | struct mutex lock; |
69fb2131 JA |
277 | |
278 | /* ctx's that are using this sqd */ | |
279 | struct list_head ctx_list; | |
69fb2131 | 280 | |
534ca6d6 JA |
281 | struct task_struct *thread; |
282 | struct wait_queue_head wait; | |
08369246 XW |
283 | |
284 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
285 | int sq_cpu; |
286 | pid_t task_pid; | |
5c2469e0 | 287 | pid_t task_tgid; |
37d1e2e3 JA |
288 | |
289 | unsigned long state; | |
37d1e2e3 | 290 | struct completion exited; |
534ca6d6 JA |
291 | }; |
292 | ||
258b29a9 | 293 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 294 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 295 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 296 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
297 | |
298 | struct io_comp_state { | |
6dd0be1e | 299 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 300 | unsigned int nr; |
c7dae4ba | 301 | /* inline/task_work completion list, under ->uring_lock */ |
1b4c351f | 302 | struct list_head free_list; |
258b29a9 PB |
303 | }; |
304 | ||
a1ab7b35 PB |
305 | struct io_submit_link { |
306 | struct io_kiocb *head; | |
307 | struct io_kiocb *last; | |
308 | }; | |
309 | ||
258b29a9 PB |
310 | struct io_submit_state { |
311 | struct blk_plug plug; | |
a1ab7b35 | 312 | struct io_submit_link link; |
258b29a9 PB |
313 | |
314 | /* | |
315 | * io_kiocb alloc cache | |
316 | */ | |
bf019da7 | 317 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
318 | unsigned int free_reqs; |
319 | ||
320 | bool plug_started; | |
321 | ||
322 | /* | |
323 | * Batch completion logic | |
324 | */ | |
325 | struct io_comp_state comp; | |
326 | ||
327 | /* | |
328 | * File reference cache | |
329 | */ | |
330 | struct file *file; | |
331 | unsigned int fd; | |
332 | unsigned int file_refs; | |
333 | unsigned int ios_left; | |
334 | }; | |
335 | ||
2b188cc1 JA |
336 | struct io_ring_ctx { |
337 | struct { | |
338 | struct percpu_ref refs; | |
339 | } ____cacheline_aligned_in_smp; | |
340 | ||
341 | struct { | |
342 | unsigned int flags; | |
e1d85334 | 343 | unsigned int compat: 1; |
e1d85334 RD |
344 | unsigned int drain_next: 1; |
345 | unsigned int eventfd_async: 1; | |
21b55dbc | 346 | unsigned int restricted: 1; |
2b188cc1 | 347 | |
75b28aff HV |
348 | /* |
349 | * Ring buffer of indices into array of io_uring_sqe, which is | |
350 | * mmapped by the application using the IORING_OFF_SQES offset. | |
351 | * | |
352 | * This indirection could e.g. be used to assign fixed | |
353 | * io_uring_sqe entries to operations and only submit them to | |
354 | * the queue when needed. | |
355 | * | |
356 | * The kernel modifies neither the indices array nor the entries | |
357 | * array. | |
358 | */ | |
359 | u32 *sq_array; | |
2b188cc1 JA |
360 | unsigned cached_sq_head; |
361 | unsigned sq_entries; | |
6c271ce2 | 362 | unsigned sq_thread_idle; |
498ccd9e | 363 | unsigned cached_sq_dropped; |
ad3eb2c8 | 364 | unsigned long sq_check_overflow; |
de0617e4 JA |
365 | |
366 | struct list_head defer_list; | |
5262f567 | 367 | struct list_head timeout_list; |
1d7bb1d5 | 368 | struct list_head cq_overflow_list; |
fcb323cc | 369 | |
ad3eb2c8 | 370 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
371 | } ____cacheline_aligned_in_smp; |
372 | ||
3c1a2ead JA |
373 | struct { |
374 | struct mutex uring_lock; | |
375 | wait_queue_head_t wait; | |
376 | } ____cacheline_aligned_in_smp; | |
377 | ||
378 | struct io_submit_state submit_state; | |
d0acdee2 PB |
379 | /* IRQ completion list, under ->completion_lock */ |
380 | struct list_head locked_free_list; | |
381 | unsigned int locked_free_nr; | |
3c1a2ead | 382 | |
206aefde JA |
383 | struct io_rings *rings; |
384 | ||
7c30f36a | 385 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
386 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
387 | ||
90554200 | 388 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 389 | struct list_head sqd_list; |
75b28aff | 390 | |
6b06314c | 391 | /* |
b13a8918 PB |
392 | * Fixed resources fast path, should be accessed only under uring_lock, |
393 | * and updated through io_uring_register(2) | |
6b06314c | 394 | */ |
b13a8918 PB |
395 | struct io_rsrc_node *rsrc_node; |
396 | ||
aeca241b | 397 | struct io_file_table file_table; |
6b06314c | 398 | unsigned nr_user_files; |
edafccee | 399 | unsigned nr_user_bufs; |
41edf1a5 | 400 | struct io_mapped_ubuf **user_bufs; |
edafccee | 401 | |
9e15c3a0 | 402 | struct xarray io_buffers; |
61cf9370 MWO |
403 | struct xarray personalities; |
404 | u32 pers_next; | |
071698e1 | 405 | |
206aefde JA |
406 | struct { |
407 | unsigned cached_cq_tail; | |
408 | unsigned cq_entries; | |
206aefde | 409 | atomic_t cq_timeouts; |
f010505b | 410 | unsigned cq_last_tm_flush; |
7b289c38 | 411 | unsigned cq_extra; |
ad3eb2c8 | 412 | unsigned long cq_check_overflow; |
206aefde JA |
413 | struct wait_queue_head cq_wait; |
414 | struct fasync_struct *cq_fasync; | |
415 | struct eventfd_ctx *cq_ev_fd; | |
416 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 417 | |
2b188cc1 JA |
418 | struct { |
419 | spinlock_t completion_lock; | |
e94f141b | 420 | |
def596e9 | 421 | /* |
540e32a0 | 422 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
423 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
424 | * For SQPOLL, only the single threaded io_sq_thread() will | |
425 | * manipulate the list, hence no extra locking is needed there. | |
426 | */ | |
540e32a0 | 427 | struct list_head iopoll_list; |
78076bb6 JA |
428 | struct hlist_head *cancel_hash; |
429 | unsigned cancel_hash_bits; | |
e94f141b | 430 | bool poll_multi_file; |
2b188cc1 | 431 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 432 | |
21b55dbc | 433 | struct io_restriction restrictions; |
3c1a2ead | 434 | |
b13a8918 PB |
435 | /* slow path rsrc auxilary data, used by update/register */ |
436 | struct { | |
437 | struct io_rsrc_node *rsrc_backup_node; | |
438 | struct io_mapped_ubuf *dummy_ubuf; | |
439 | struct io_rsrc_data *file_data; | |
440 | struct io_rsrc_data *buf_data; | |
441 | ||
442 | struct delayed_work rsrc_put_work; | |
443 | struct llist_head rsrc_put_llist; | |
444 | struct list_head rsrc_ref_list; | |
445 | spinlock_t rsrc_ref_lock; | |
446 | }; | |
447 | ||
3c1a2ead | 448 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
449 | struct { |
450 | #if defined(CONFIG_UNIX) | |
451 | struct socket *ring_sock; | |
452 | #endif | |
453 | /* hashed buffered write serialization */ | |
454 | struct io_wq_hash *hash_map; | |
455 | ||
456 | /* Only used for accounting purposes */ | |
457 | struct user_struct *user; | |
458 | struct mm_struct *mm_account; | |
459 | ||
460 | /* ctx exit and cancelation */ | |
461 | struct callback_head *exit_task_work; | |
462 | struct work_struct exit_work; | |
463 | struct list_head tctx_list; | |
464 | struct completion ref_comp; | |
465 | }; | |
2b188cc1 JA |
466 | }; |
467 | ||
53e043b2 SM |
468 | struct io_uring_task { |
469 | /* submission side */ | |
470 | struct xarray xa; | |
471 | struct wait_queue_head wait; | |
ee53fb2b SM |
472 | const struct io_ring_ctx *last; |
473 | struct io_wq *io_wq; | |
53e043b2 | 474 | struct percpu_counter inflight; |
b303fe2e | 475 | atomic_t inflight_tracked; |
53e043b2 | 476 | atomic_t in_idle; |
53e043b2 SM |
477 | |
478 | spinlock_t task_lock; | |
479 | struct io_wq_work_list task_list; | |
480 | unsigned long task_state; | |
481 | struct callback_head task_work; | |
482 | }; | |
483 | ||
09bb8394 JA |
484 | /* |
485 | * First field must be the file pointer in all the | |
486 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
487 | */ | |
221c5eb2 JA |
488 | struct io_poll_iocb { |
489 | struct file *file; | |
018043be | 490 | struct wait_queue_head *head; |
221c5eb2 | 491 | __poll_t events; |
8c838788 | 492 | bool done; |
221c5eb2 | 493 | bool canceled; |
392edb45 | 494 | struct wait_queue_entry wait; |
221c5eb2 JA |
495 | }; |
496 | ||
9d805892 | 497 | struct io_poll_update { |
018043be | 498 | struct file *file; |
9d805892 PB |
499 | u64 old_user_data; |
500 | u64 new_user_data; | |
501 | __poll_t events; | |
b69de288 JA |
502 | bool update_events; |
503 | bool update_user_data; | |
018043be PB |
504 | }; |
505 | ||
b5dba59e JA |
506 | struct io_close { |
507 | struct file *file; | |
b5dba59e JA |
508 | int fd; |
509 | }; | |
510 | ||
ad8a48ac JA |
511 | struct io_timeout_data { |
512 | struct io_kiocb *req; | |
513 | struct hrtimer timer; | |
514 | struct timespec64 ts; | |
515 | enum hrtimer_mode mode; | |
516 | }; | |
517 | ||
8ed8d3c3 JA |
518 | struct io_accept { |
519 | struct file *file; | |
520 | struct sockaddr __user *addr; | |
521 | int __user *addr_len; | |
522 | int flags; | |
09952e3e | 523 | unsigned long nofile; |
8ed8d3c3 JA |
524 | }; |
525 | ||
526 | struct io_sync { | |
527 | struct file *file; | |
528 | loff_t len; | |
529 | loff_t off; | |
530 | int flags; | |
d63d1b5e | 531 | int mode; |
8ed8d3c3 JA |
532 | }; |
533 | ||
fbf23849 JA |
534 | struct io_cancel { |
535 | struct file *file; | |
536 | u64 addr; | |
537 | }; | |
538 | ||
b29472ee JA |
539 | struct io_timeout { |
540 | struct file *file; | |
bfe68a22 PB |
541 | u32 off; |
542 | u32 target_seq; | |
135fcde8 | 543 | struct list_head list; |
90cd7e42 PB |
544 | /* head of the link, used by linked timeouts only */ |
545 | struct io_kiocb *head; | |
b29472ee JA |
546 | }; |
547 | ||
0bdf7a2d PB |
548 | struct io_timeout_rem { |
549 | struct file *file; | |
550 | u64 addr; | |
9c8e11b3 PB |
551 | |
552 | /* timeout update */ | |
553 | struct timespec64 ts; | |
554 | u32 flags; | |
0bdf7a2d PB |
555 | }; |
556 | ||
9adbd45d JA |
557 | struct io_rw { |
558 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
559 | struct kiocb kiocb; | |
560 | u64 addr; | |
561 | u64 len; | |
562 | }; | |
563 | ||
3fbb51c1 JA |
564 | struct io_connect { |
565 | struct file *file; | |
566 | struct sockaddr __user *addr; | |
567 | int addr_len; | |
568 | }; | |
569 | ||
e47293fd JA |
570 | struct io_sr_msg { |
571 | struct file *file; | |
fddaface | 572 | union { |
4af3417a PB |
573 | struct compat_msghdr __user *umsg_compat; |
574 | struct user_msghdr __user *umsg; | |
575 | void __user *buf; | |
fddaface | 576 | }; |
e47293fd | 577 | int msg_flags; |
bcda7baa | 578 | int bgid; |
fddaface | 579 | size_t len; |
bcda7baa | 580 | struct io_buffer *kbuf; |
e47293fd JA |
581 | }; |
582 | ||
15b71abe JA |
583 | struct io_open { |
584 | struct file *file; | |
585 | int dfd; | |
15b71abe | 586 | struct filename *filename; |
c12cedf2 | 587 | struct open_how how; |
4022e7af | 588 | unsigned long nofile; |
15b71abe JA |
589 | }; |
590 | ||
269bbe5f | 591 | struct io_rsrc_update { |
05f3fb3c JA |
592 | struct file *file; |
593 | u64 arg; | |
594 | u32 nr_args; | |
595 | u32 offset; | |
596 | }; | |
597 | ||
4840e418 JA |
598 | struct io_fadvise { |
599 | struct file *file; | |
600 | u64 offset; | |
601 | u32 len; | |
602 | u32 advice; | |
603 | }; | |
604 | ||
c1ca757b JA |
605 | struct io_madvise { |
606 | struct file *file; | |
607 | u64 addr; | |
608 | u32 len; | |
609 | u32 advice; | |
610 | }; | |
611 | ||
3e4827b0 JA |
612 | struct io_epoll { |
613 | struct file *file; | |
614 | int epfd; | |
615 | int op; | |
616 | int fd; | |
617 | struct epoll_event event; | |
e47293fd JA |
618 | }; |
619 | ||
7d67af2c PB |
620 | struct io_splice { |
621 | struct file *file_out; | |
622 | struct file *file_in; | |
623 | loff_t off_out; | |
624 | loff_t off_in; | |
625 | u64 len; | |
626 | unsigned int flags; | |
627 | }; | |
628 | ||
ddf0322d JA |
629 | struct io_provide_buf { |
630 | struct file *file; | |
631 | __u64 addr; | |
38134ada | 632 | __u32 len; |
ddf0322d JA |
633 | __u32 bgid; |
634 | __u16 nbufs; | |
635 | __u16 bid; | |
636 | }; | |
637 | ||
1d9e1288 BM |
638 | struct io_statx { |
639 | struct file *file; | |
640 | int dfd; | |
641 | unsigned int mask; | |
642 | unsigned int flags; | |
e62753e4 | 643 | const char __user *filename; |
1d9e1288 BM |
644 | struct statx __user *buffer; |
645 | }; | |
646 | ||
36f4fa68 JA |
647 | struct io_shutdown { |
648 | struct file *file; | |
649 | int how; | |
650 | }; | |
651 | ||
80a261fd JA |
652 | struct io_rename { |
653 | struct file *file; | |
654 | int old_dfd; | |
655 | int new_dfd; | |
656 | struct filename *oldpath; | |
657 | struct filename *newpath; | |
658 | int flags; | |
659 | }; | |
660 | ||
14a1143b JA |
661 | struct io_unlink { |
662 | struct file *file; | |
663 | int dfd; | |
664 | int flags; | |
665 | struct filename *filename; | |
666 | }; | |
667 | ||
3ca405eb PB |
668 | struct io_completion { |
669 | struct file *file; | |
670 | struct list_head list; | |
8c3f9cd1 | 671 | u32 cflags; |
3ca405eb PB |
672 | }; |
673 | ||
f499a021 JA |
674 | struct io_async_connect { |
675 | struct sockaddr_storage address; | |
676 | }; | |
677 | ||
03b1230c JA |
678 | struct io_async_msghdr { |
679 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
680 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
681 | struct iovec *free_iov; | |
03b1230c JA |
682 | struct sockaddr __user *uaddr; |
683 | struct msghdr msg; | |
b537916c | 684 | struct sockaddr_storage addr; |
03b1230c JA |
685 | }; |
686 | ||
f67676d1 JA |
687 | struct io_async_rw { |
688 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
689 | const struct iovec *free_iovec; |
690 | struct iov_iter iter; | |
227c0c96 | 691 | size_t bytes_done; |
bcf5a063 | 692 | struct wait_page_queue wpq; |
f67676d1 JA |
693 | }; |
694 | ||
6b47ee6e PB |
695 | enum { |
696 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
697 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
698 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
699 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
700 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 701 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 702 | |
dddca226 | 703 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 704 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
705 | REQ_F_INFLIGHT_BIT, |
706 | REQ_F_CUR_POS_BIT, | |
707 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 708 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 709 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 710 | REQ_F_POLLED_BIT, |
bcda7baa | 711 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 712 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 713 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 714 | REQ_F_REISSUE_BIT, |
8c130827 | 715 | REQ_F_DONT_REISSUE_BIT, |
7b29f92d JA |
716 | /* keep async read/write and isreg together and in order */ |
717 | REQ_F_ASYNC_READ_BIT, | |
718 | REQ_F_ASYNC_WRITE_BIT, | |
719 | REQ_F_ISREG_BIT, | |
84557871 JA |
720 | |
721 | /* not a real bit, just to check we're not overflowing the space */ | |
722 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
723 | }; |
724 | ||
725 | enum { | |
726 | /* ctx owns file */ | |
727 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
728 | /* drain existing IO first */ | |
729 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
730 | /* linked sqes */ | |
731 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
732 | /* doesn't sever on completion < 0 */ | |
733 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
734 | /* IOSQE_ASYNC */ | |
735 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
736 | /* IOSQE_BUFFER_SELECT */ |
737 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 738 | |
6b47ee6e | 739 | /* fail rest of links */ |
93d2bcd2 | 740 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 741 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
742 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
743 | /* read/write uses file position */ | |
744 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
745 | /* must not punt to workers */ | |
746 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 747 | /* has or had linked timeout */ |
6b47ee6e | 748 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
749 | /* needs cleanup */ |
750 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
751 | /* already went through poll handler */ |
752 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
753 | /* buffer already selected */ |
754 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
755 | /* linked timeout is active, i.e. prepared by link's head */ |
756 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
757 | /* completion is deferred through io_comp_state */ |
758 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
759 | /* caller should reissue async */ |
760 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
761 | /* don't attempt request reissue, see io_rw_reissue() */ |
762 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
763 | /* supports async reads */ |
764 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
765 | /* supports async writes */ | |
766 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
767 | /* regular file */ | |
768 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
769 | }; |
770 | ||
771 | struct async_poll { | |
772 | struct io_poll_iocb poll; | |
807abcb0 | 773 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
774 | }; |
775 | ||
7cbf1722 JA |
776 | struct io_task_work { |
777 | struct io_wq_work_node node; | |
778 | task_work_func_t func; | |
779 | }; | |
780 | ||
992da01a PB |
781 | enum { |
782 | IORING_RSRC_FILE = 0, | |
783 | IORING_RSRC_BUFFER = 1, | |
784 | }; | |
785 | ||
09bb8394 JA |
786 | /* |
787 | * NOTE! Each of the iocb union members has the file pointer | |
788 | * as the first entry in their struct definition. So you can | |
789 | * access the file pointer through any of the sub-structs, | |
790 | * or directly as just 'ki_filp' in this struct. | |
791 | */ | |
2b188cc1 | 792 | struct io_kiocb { |
221c5eb2 | 793 | union { |
09bb8394 | 794 | struct file *file; |
9adbd45d | 795 | struct io_rw rw; |
221c5eb2 | 796 | struct io_poll_iocb poll; |
9d805892 | 797 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
798 | struct io_accept accept; |
799 | struct io_sync sync; | |
fbf23849 | 800 | struct io_cancel cancel; |
b29472ee | 801 | struct io_timeout timeout; |
0bdf7a2d | 802 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 803 | struct io_connect connect; |
e47293fd | 804 | struct io_sr_msg sr_msg; |
15b71abe | 805 | struct io_open open; |
b5dba59e | 806 | struct io_close close; |
269bbe5f | 807 | struct io_rsrc_update rsrc_update; |
4840e418 | 808 | struct io_fadvise fadvise; |
c1ca757b | 809 | struct io_madvise madvise; |
3e4827b0 | 810 | struct io_epoll epoll; |
7d67af2c | 811 | struct io_splice splice; |
ddf0322d | 812 | struct io_provide_buf pbuf; |
1d9e1288 | 813 | struct io_statx statx; |
36f4fa68 | 814 | struct io_shutdown shutdown; |
80a261fd | 815 | struct io_rename rename; |
14a1143b | 816 | struct io_unlink unlink; |
3ca405eb PB |
817 | /* use only after cleaning per-op data, see io_clean_op() */ |
818 | struct io_completion compl; | |
221c5eb2 | 819 | }; |
2b188cc1 | 820 | |
e8c2bc1f JA |
821 | /* opcode allocated if it needs to store data for async defer */ |
822 | void *async_data; | |
d625c6ee | 823 | u8 opcode; |
65a6543d XW |
824 | /* polled IO has completed */ |
825 | u8 iopoll_completed; | |
2b188cc1 | 826 | |
4f4eeba8 | 827 | u16 buf_index; |
9cf7c104 | 828 | u32 result; |
4f4eeba8 | 829 | |
010e8e6b PB |
830 | struct io_ring_ctx *ctx; |
831 | unsigned int flags; | |
abc54d63 | 832 | atomic_t refs; |
010e8e6b PB |
833 | struct task_struct *task; |
834 | u64 user_data; | |
d7718a9d | 835 | |
f2f87370 | 836 | struct io_kiocb *link; |
269bbe5f | 837 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 838 | |
b303fe2e | 839 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 840 | struct list_head inflight_entry; |
7cbf1722 JA |
841 | union { |
842 | struct io_task_work io_task_work; | |
843 | struct callback_head task_work; | |
844 | }; | |
010e8e6b PB |
845 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
846 | struct hlist_node hash_node; | |
847 | struct async_poll *apoll; | |
848 | struct io_wq_work work; | |
eae071c9 PB |
849 | /* store used ubuf, so we can prevent reloading */ |
850 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 851 | }; |
05589553 | 852 | |
13bf43f5 PB |
853 | struct io_tctx_node { |
854 | struct list_head ctx_node; | |
855 | struct task_struct *task; | |
13bf43f5 PB |
856 | struct io_ring_ctx *ctx; |
857 | }; | |
858 | ||
27dc8338 PB |
859 | struct io_defer_entry { |
860 | struct list_head list; | |
861 | struct io_kiocb *req; | |
9cf7c104 | 862 | u32 seq; |
2b188cc1 JA |
863 | }; |
864 | ||
d3656344 | 865 | struct io_op_def { |
d3656344 JA |
866 | /* needs req->file assigned */ |
867 | unsigned needs_file : 1; | |
d3656344 JA |
868 | /* hash wq insertion if file is a regular file */ |
869 | unsigned hash_reg_file : 1; | |
870 | /* unbound wq insertion if file is a non-regular file */ | |
871 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
872 | /* opcode is not supported by this kernel */ |
873 | unsigned not_supported : 1; | |
8a72758c JA |
874 | /* set if opcode supports polled "wait" */ |
875 | unsigned pollin : 1; | |
876 | unsigned pollout : 1; | |
bcda7baa JA |
877 | /* op supports buffer selection */ |
878 | unsigned buffer_select : 1; | |
26f0505a PB |
879 | /* do prep async if is going to be punted */ |
880 | unsigned needs_async_setup : 1; | |
27926b68 JA |
881 | /* should block plug */ |
882 | unsigned plug : 1; | |
e8c2bc1f JA |
883 | /* size of async data needed, if any */ |
884 | unsigned short async_size; | |
d3656344 JA |
885 | }; |
886 | ||
0918682b | 887 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
888 | [IORING_OP_NOP] = {}, |
889 | [IORING_OP_READV] = { | |
d3656344 JA |
890 | .needs_file = 1, |
891 | .unbound_nonreg_file = 1, | |
8a72758c | 892 | .pollin = 1, |
4d954c25 | 893 | .buffer_select = 1, |
26f0505a | 894 | .needs_async_setup = 1, |
27926b68 | 895 | .plug = 1, |
e8c2bc1f | 896 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 897 | }, |
0463b6c5 | 898 | [IORING_OP_WRITEV] = { |
d3656344 JA |
899 | .needs_file = 1, |
900 | .hash_reg_file = 1, | |
901 | .unbound_nonreg_file = 1, | |
8a72758c | 902 | .pollout = 1, |
26f0505a | 903 | .needs_async_setup = 1, |
27926b68 | 904 | .plug = 1, |
e8c2bc1f | 905 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 906 | }, |
0463b6c5 | 907 | [IORING_OP_FSYNC] = { |
d3656344 JA |
908 | .needs_file = 1, |
909 | }, | |
0463b6c5 | 910 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
911 | .needs_file = 1, |
912 | .unbound_nonreg_file = 1, | |
8a72758c | 913 | .pollin = 1, |
27926b68 | 914 | .plug = 1, |
e8c2bc1f | 915 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 916 | }, |
0463b6c5 | 917 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
918 | .needs_file = 1, |
919 | .hash_reg_file = 1, | |
920 | .unbound_nonreg_file = 1, | |
8a72758c | 921 | .pollout = 1, |
27926b68 | 922 | .plug = 1, |
e8c2bc1f | 923 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 924 | }, |
0463b6c5 | 925 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
926 | .needs_file = 1, |
927 | .unbound_nonreg_file = 1, | |
928 | }, | |
0463b6c5 PB |
929 | [IORING_OP_POLL_REMOVE] = {}, |
930 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
931 | .needs_file = 1, |
932 | }, | |
0463b6c5 | 933 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
934 | .needs_file = 1, |
935 | .unbound_nonreg_file = 1, | |
8a72758c | 936 | .pollout = 1, |
26f0505a | 937 | .needs_async_setup = 1, |
e8c2bc1f | 938 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 939 | }, |
0463b6c5 | 940 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
941 | .needs_file = 1, |
942 | .unbound_nonreg_file = 1, | |
8a72758c | 943 | .pollin = 1, |
52de1fe1 | 944 | .buffer_select = 1, |
26f0505a | 945 | .needs_async_setup = 1, |
e8c2bc1f | 946 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 947 | }, |
0463b6c5 | 948 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 949 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 950 | }, |
9c8e11b3 PB |
951 | [IORING_OP_TIMEOUT_REMOVE] = { |
952 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 953 | }, |
0463b6c5 | 954 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
955 | .needs_file = 1, |
956 | .unbound_nonreg_file = 1, | |
8a72758c | 957 | .pollin = 1, |
d3656344 | 958 | }, |
0463b6c5 PB |
959 | [IORING_OP_ASYNC_CANCEL] = {}, |
960 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 961 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 962 | }, |
0463b6c5 | 963 | [IORING_OP_CONNECT] = { |
d3656344 JA |
964 | .needs_file = 1, |
965 | .unbound_nonreg_file = 1, | |
8a72758c | 966 | .pollout = 1, |
26f0505a | 967 | .needs_async_setup = 1, |
e8c2bc1f | 968 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 969 | }, |
0463b6c5 | 970 | [IORING_OP_FALLOCATE] = { |
d3656344 | 971 | .needs_file = 1, |
d3656344 | 972 | }, |
44526bed JA |
973 | [IORING_OP_OPENAT] = {}, |
974 | [IORING_OP_CLOSE] = {}, | |
975 | [IORING_OP_FILES_UPDATE] = {}, | |
976 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 977 | [IORING_OP_READ] = { |
3a6820f2 JA |
978 | .needs_file = 1, |
979 | .unbound_nonreg_file = 1, | |
8a72758c | 980 | .pollin = 1, |
bcda7baa | 981 | .buffer_select = 1, |
27926b68 | 982 | .plug = 1, |
e8c2bc1f | 983 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 984 | }, |
0463b6c5 | 985 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
986 | .needs_file = 1, |
987 | .unbound_nonreg_file = 1, | |
8a72758c | 988 | .pollout = 1, |
27926b68 | 989 | .plug = 1, |
e8c2bc1f | 990 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 991 | }, |
0463b6c5 | 992 | [IORING_OP_FADVISE] = { |
4840e418 | 993 | .needs_file = 1, |
c1ca757b | 994 | }, |
44526bed | 995 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 996 | [IORING_OP_SEND] = { |
fddaface JA |
997 | .needs_file = 1, |
998 | .unbound_nonreg_file = 1, | |
8a72758c | 999 | .pollout = 1, |
fddaface | 1000 | }, |
0463b6c5 | 1001 | [IORING_OP_RECV] = { |
fddaface JA |
1002 | .needs_file = 1, |
1003 | .unbound_nonreg_file = 1, | |
8a72758c | 1004 | .pollin = 1, |
bcda7baa | 1005 | .buffer_select = 1, |
fddaface | 1006 | }, |
0463b6c5 | 1007 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1008 | }, |
3e4827b0 JA |
1009 | [IORING_OP_EPOLL_CTL] = { |
1010 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1011 | }, |
7d67af2c PB |
1012 | [IORING_OP_SPLICE] = { |
1013 | .needs_file = 1, | |
1014 | .hash_reg_file = 1, | |
1015 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1016 | }, |
1017 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1018 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1019 | [IORING_OP_TEE] = { |
1020 | .needs_file = 1, | |
1021 | .hash_reg_file = 1, | |
1022 | .unbound_nonreg_file = 1, | |
1023 | }, | |
36f4fa68 JA |
1024 | [IORING_OP_SHUTDOWN] = { |
1025 | .needs_file = 1, | |
1026 | }, | |
44526bed JA |
1027 | [IORING_OP_RENAMEAT] = {}, |
1028 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1029 | }; |
1030 | ||
7a612350 | 1031 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1032 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1033 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1034 | struct task_struct *task, | |
3dd0c97a | 1035 | bool cancel_all); |
734551df | 1036 | static void io_uring_cancel_sqpoll(struct io_sq_data *sqd); |
b895c9a6 | 1037 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1038 | |
d4d19c19 PB |
1039 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1040 | long res, unsigned int cflags); | |
ec9c02ad | 1041 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1042 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1043 | static void io_dismantle_req(struct io_kiocb *req); |
1044 | static void io_put_task(struct task_struct *task, int nr); | |
94ae5e77 JA |
1045 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1046 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
fdecb662 | 1047 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1048 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1049 | unsigned nr_args); |
68fb8979 | 1050 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1051 | static struct file *io_file_get(struct io_submit_state *state, |
1052 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1053 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1054 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1055 | |
907d1df3 | 1056 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1057 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1058 | struct io_ring_ctx *ctx); | |
5082620f | 1059 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1060 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1061 | |
2b188cc1 JA |
1062 | static struct kmem_cache *req_cachep; |
1063 | ||
0918682b | 1064 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1065 | |
1066 | struct sock *io_uring_get_socket(struct file *file) | |
1067 | { | |
1068 | #if defined(CONFIG_UNIX) | |
1069 | if (file->f_op == &io_uring_fops) { | |
1070 | struct io_ring_ctx *ctx = file->private_data; | |
1071 | ||
1072 | return ctx->ring_sock->sk; | |
1073 | } | |
1074 | #endif | |
1075 | return NULL; | |
1076 | } | |
1077 | EXPORT_SYMBOL(io_uring_get_socket); | |
1078 | ||
f2f87370 PB |
1079 | #define io_for_each_link(pos, head) \ |
1080 | for (pos = (head); pos; pos = pos->link) | |
1081 | ||
b895c9a6 | 1082 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1083 | { |
1084 | struct io_ring_ctx *ctx = req->ctx; | |
1085 | ||
269bbe5f | 1086 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1087 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1088 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1089 | } |
1090 | } | |
1091 | ||
f70865db PB |
1092 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1093 | { | |
1094 | bool got = percpu_ref_tryget(ref); | |
1095 | ||
1096 | /* already at zero, wait for ->release() */ | |
1097 | if (!got) | |
1098 | wait_for_completion(compl); | |
1099 | percpu_ref_resurrect(ref); | |
1100 | if (got) | |
1101 | percpu_ref_put(ref); | |
1102 | } | |
1103 | ||
3dd0c97a PB |
1104 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1105 | bool cancel_all) | |
08d23634 PB |
1106 | { |
1107 | struct io_kiocb *req; | |
1108 | ||
68207680 | 1109 | if (task && head->task != task) |
08d23634 | 1110 | return false; |
3dd0c97a | 1111 | if (cancel_all) |
08d23634 PB |
1112 | return true; |
1113 | ||
1114 | io_for_each_link(req, head) { | |
b05a1bcd | 1115 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1116 | return true; |
08d23634 PB |
1117 | } |
1118 | return false; | |
1119 | } | |
1120 | ||
93d2bcd2 | 1121 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1122 | { |
93d2bcd2 | 1123 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1124 | } |
4a38aed2 | 1125 | |
2b188cc1 JA |
1126 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1127 | { | |
1128 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1129 | ||
0f158b4c | 1130 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1131 | } |
1132 | ||
8eb7e2d0 PB |
1133 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1134 | { | |
1135 | return !req->timeout.off; | |
1136 | } | |
1137 | ||
2b188cc1 JA |
1138 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1139 | { | |
1140 | struct io_ring_ctx *ctx; | |
78076bb6 | 1141 | int hash_bits; |
2b188cc1 JA |
1142 | |
1143 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1144 | if (!ctx) | |
1145 | return NULL; | |
1146 | ||
78076bb6 JA |
1147 | /* |
1148 | * Use 5 bits less than the max cq entries, that should give us around | |
1149 | * 32 entries per hash list if totally full and uniformly spread. | |
1150 | */ | |
1151 | hash_bits = ilog2(p->cq_entries); | |
1152 | hash_bits -= 5; | |
1153 | if (hash_bits <= 0) | |
1154 | hash_bits = 1; | |
1155 | ctx->cancel_hash_bits = hash_bits; | |
1156 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1157 | GFP_KERNEL); | |
1158 | if (!ctx->cancel_hash) | |
1159 | goto err; | |
1160 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1161 | ||
6224843d PB |
1162 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1163 | if (!ctx->dummy_ubuf) | |
1164 | goto err; | |
1165 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1166 | ctx->dummy_ubuf->ubuf = -1UL; | |
1167 | ||
21482896 | 1168 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1169 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1170 | goto err; | |
2b188cc1 JA |
1171 | |
1172 | ctx->flags = p->flags; | |
90554200 | 1173 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1174 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1175 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1176 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1177 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1178 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1179 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1180 | mutex_init(&ctx->uring_lock); |
1181 | init_waitqueue_head(&ctx->wait); | |
1182 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1183 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1184 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1185 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1186 | spin_lock_init(&ctx->rsrc_ref_lock); |
1187 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1188 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1189 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1190 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1191 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
d0acdee2 | 1192 | INIT_LIST_HEAD(&ctx->locked_free_list); |
2b188cc1 | 1193 | return ctx; |
206aefde | 1194 | err: |
6224843d | 1195 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1196 | kfree(ctx->cancel_hash); |
206aefde JA |
1197 | kfree(ctx); |
1198 | return NULL; | |
2b188cc1 JA |
1199 | } |
1200 | ||
8f6ed49a PB |
1201 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1202 | { | |
1203 | struct io_rings *r = ctx->rings; | |
1204 | ||
1205 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1206 | ctx->cq_extra--; | |
1207 | } | |
1208 | ||
9cf7c104 | 1209 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1210 | { |
2bc9930e JA |
1211 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1212 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1213 | |
8f6ed49a | 1214 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1215 | } |
de0617e4 | 1216 | |
9d858b21 | 1217 | return false; |
de0617e4 JA |
1218 | } |
1219 | ||
ce3d5aae PB |
1220 | static void io_req_track_inflight(struct io_kiocb *req) |
1221 | { | |
ce3d5aae | 1222 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1223 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1224 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1225 | } |
1226 | } | |
1227 | ||
1e6fa521 JA |
1228 | static void io_prep_async_work(struct io_kiocb *req) |
1229 | { | |
1230 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1231 | struct io_ring_ctx *ctx = req->ctx; |
1232 | ||
003e8dcc JA |
1233 | if (!req->work.creds) |
1234 | req->work.creds = get_current_cred(); | |
1235 | ||
e1d675df PB |
1236 | req->work.list.next = NULL; |
1237 | req->work.flags = 0; | |
feaadc4f PB |
1238 | if (req->flags & REQ_F_FORCE_ASYNC) |
1239 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1240 | ||
1e6fa521 JA |
1241 | if (req->flags & REQ_F_ISREG) { |
1242 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1243 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1244 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1245 | if (def->unbound_nonreg_file) |
1246 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1247 | } | |
e1d675df PB |
1248 | |
1249 | switch (req->opcode) { | |
1250 | case IORING_OP_SPLICE: | |
1251 | case IORING_OP_TEE: | |
e1d675df PB |
1252 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1253 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1254 | break; | |
1255 | } | |
561fb04a | 1256 | } |
cccf0ee8 | 1257 | |
cbdcb435 | 1258 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1259 | { |
cbdcb435 | 1260 | struct io_kiocb *cur; |
54a91f3b | 1261 | |
f2f87370 PB |
1262 | io_for_each_link(cur, req) |
1263 | io_prep_async_work(cur); | |
561fb04a JA |
1264 | } |
1265 | ||
ebf93667 | 1266 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1267 | { |
a197f664 | 1268 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1269 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1270 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1271 | |
3bfe6106 JA |
1272 | BUG_ON(!tctx); |
1273 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1274 | |
cbdcb435 PB |
1275 | /* init ->work of the whole link before punting */ |
1276 | io_prep_async_link(req); | |
d07f1e8a PB |
1277 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1278 | &req->work, req->flags); | |
ebf93667 | 1279 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1280 | if (link) |
1281 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1282 | } |
1283 | ||
1ee4160c | 1284 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1285 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1286 | { |
e8c2bc1f | 1287 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1288 | |
fd9c7bc5 | 1289 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1290 | atomic_set(&req->ctx->cq_timeouts, |
1291 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1292 | list_del_init(&req->timeout.list); |
d4d19c19 | 1293 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
216578e5 | 1294 | io_put_req_deferred(req, 1); |
5262f567 JA |
1295 | } |
1296 | } | |
1297 | ||
04518945 | 1298 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1299 | { |
04518945 | 1300 | do { |
27dc8338 PB |
1301 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1302 | struct io_defer_entry, list); | |
de0617e4 | 1303 | |
9cf7c104 | 1304 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1305 | break; |
27dc8338 | 1306 | list_del_init(&de->list); |
907d1df3 | 1307 | io_req_task_queue(de->req); |
27dc8338 | 1308 | kfree(de); |
04518945 PB |
1309 | } while (!list_empty(&ctx->defer_list)); |
1310 | } | |
1311 | ||
360428f8 | 1312 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1313 | { |
f010505b MDG |
1314 | u32 seq; |
1315 | ||
1316 | if (list_empty(&ctx->timeout_list)) | |
1317 | return; | |
1318 | ||
1319 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1320 | ||
1321 | do { | |
1322 | u32 events_needed, events_got; | |
360428f8 | 1323 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1324 | struct io_kiocb, timeout.list); |
de0617e4 | 1325 | |
8eb7e2d0 | 1326 | if (io_is_timeout_noseq(req)) |
360428f8 | 1327 | break; |
f010505b MDG |
1328 | |
1329 | /* | |
1330 | * Since seq can easily wrap around over time, subtract | |
1331 | * the last seq at which timeouts were flushed before comparing. | |
1332 | * Assuming not more than 2^31-1 events have happened since, | |
1333 | * these subtractions won't have wrapped, so we can check if | |
1334 | * target is in [last_seq, current_seq] by comparing the two. | |
1335 | */ | |
1336 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1337 | events_got = seq - ctx->cq_last_tm_flush; | |
1338 | if (events_got < events_needed) | |
360428f8 | 1339 | break; |
bfe68a22 | 1340 | |
135fcde8 | 1341 | list_del_init(&req->timeout.list); |
1ee4160c | 1342 | io_kill_timeout(req, 0); |
f010505b MDG |
1343 | } while (!list_empty(&ctx->timeout_list)); |
1344 | ||
1345 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1346 | } |
5262f567 | 1347 | |
360428f8 PB |
1348 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1349 | { | |
1350 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1351 | |
1352 | /* order cqe stores with ring update */ | |
1353 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1354 | |
04518945 PB |
1355 | if (unlikely(!list_empty(&ctx->defer_list))) |
1356 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1357 | } |
1358 | ||
90554200 JA |
1359 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1360 | { | |
1361 | struct io_rings *r = ctx->rings; | |
1362 | ||
a566c556 | 1363 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1364 | } |
1365 | ||
888aae2e PB |
1366 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1367 | { | |
1368 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1369 | } | |
1370 | ||
d068b506 | 1371 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1372 | { |
75b28aff | 1373 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1374 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1375 | |
115e12e5 SB |
1376 | /* |
1377 | * writes to the cq entry need to come after reading head; the | |
1378 | * control dependency is enough as we're using WRITE_ONCE to | |
1379 | * fill the cq entry | |
1380 | */ | |
a566c556 | 1381 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1382 | return NULL; |
1383 | ||
888aae2e | 1384 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1385 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1386 | } |
1387 | ||
f2842ab5 JA |
1388 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1389 | { | |
44c769de | 1390 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1391 | return false; |
7e55a19c SG |
1392 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1393 | return false; | |
44c769de | 1394 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1395 | } |
1396 | ||
b41e9852 | 1397 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1398 | { |
b1445e59 PB |
1399 | /* see waitqueue_active() comment */ |
1400 | smp_mb(); | |
1401 | ||
1d7bb1d5 JA |
1402 | if (waitqueue_active(&ctx->wait)) |
1403 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1404 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1405 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1406 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1407 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1408 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1409 | wake_up_interruptible(&ctx->cq_wait); |
1410 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1411 | } | |
1d7bb1d5 JA |
1412 | } |
1413 | ||
80c18e4a PB |
1414 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1415 | { | |
b1445e59 PB |
1416 | /* see waitqueue_active() comment */ |
1417 | smp_mb(); | |
1418 | ||
80c18e4a PB |
1419 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1420 | if (waitqueue_active(&ctx->wait)) | |
1421 | wake_up(&ctx->wait); | |
1422 | } | |
1423 | if (io_should_trigger_evfd(ctx)) | |
1424 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1425 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1426 | wake_up_interruptible(&ctx->cq_wait); |
1427 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1428 | } | |
80c18e4a PB |
1429 | } |
1430 | ||
c4a2ed72 | 1431 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1432 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1433 | { |
1d7bb1d5 | 1434 | unsigned long flags; |
b18032bb | 1435 | bool all_flushed, posted; |
1d7bb1d5 | 1436 | |
a566c556 | 1437 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1438 | return false; |
1d7bb1d5 | 1439 | |
b18032bb | 1440 | posted = false; |
1d7bb1d5 | 1441 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae | 1442 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1443 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1444 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1445 | |
1d7bb1d5 JA |
1446 | if (!cqe && !force) |
1447 | break; | |
6c2450ae PB |
1448 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1449 | struct io_overflow_cqe, list); | |
1450 | if (cqe) | |
1451 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1452 | else | |
8f6ed49a PB |
1453 | io_account_cq_overflow(ctx); |
1454 | ||
b18032bb | 1455 | posted = true; |
6c2450ae PB |
1456 | list_del(&ocqe->list); |
1457 | kfree(ocqe); | |
1d7bb1d5 JA |
1458 | } |
1459 | ||
09e88404 PB |
1460 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1461 | if (all_flushed) { | |
1462 | clear_bit(0, &ctx->sq_check_overflow); | |
1463 | clear_bit(0, &ctx->cq_check_overflow); | |
1464 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1465 | } | |
46930143 | 1466 | |
b18032bb JA |
1467 | if (posted) |
1468 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1469 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1470 | if (posted) |
1471 | io_cqring_ev_posted(ctx); | |
09e88404 | 1472 | return all_flushed; |
1d7bb1d5 JA |
1473 | } |
1474 | ||
6c2450ae | 1475 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1476 | { |
ca0a2651 JA |
1477 | bool ret = true; |
1478 | ||
6c503150 PB |
1479 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1480 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1481 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1482 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1483 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1484 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1485 | mutex_unlock(&ctx->uring_lock); | |
1486 | } | |
ca0a2651 JA |
1487 | |
1488 | return ret; | |
6c503150 PB |
1489 | } |
1490 | ||
abc54d63 JA |
1491 | /* |
1492 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1493 | * see commit f958d7b528b1 for details. | |
1494 | */ | |
1495 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1496 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1497 | ||
de9b4cca JA |
1498 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1499 | { | |
abc54d63 | 1500 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1501 | } |
1502 | ||
1503 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1504 | { | |
abc54d63 JA |
1505 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1506 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1507 | } |
1508 | ||
1509 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1510 | { | |
abc54d63 JA |
1511 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1512 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1513 | } |
1514 | ||
1515 | static inline void req_ref_put(struct io_kiocb *req) | |
1516 | { | |
abc54d63 | 1517 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1518 | } |
1519 | ||
1520 | static inline void req_ref_get(struct io_kiocb *req) | |
1521 | { | |
abc54d63 JA |
1522 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1523 | atomic_inc(&req->refs); | |
de9b4cca JA |
1524 | } |
1525 | ||
d4d19c19 PB |
1526 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1527 | long res, unsigned int cflags) | |
2b188cc1 | 1528 | { |
cce4b8b0 | 1529 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1530 | |
cce4b8b0 PB |
1531 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1532 | if (!ocqe) { | |
1533 | /* | |
1534 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1535 | * or cannot allocate an overflow entry, then we need to drop it | |
1536 | * on the floor. | |
1537 | */ | |
8f6ed49a | 1538 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1539 | return false; |
2b188cc1 | 1540 | } |
cce4b8b0 PB |
1541 | if (list_empty(&ctx->cq_overflow_list)) { |
1542 | set_bit(0, &ctx->sq_check_overflow); | |
1543 | set_bit(0, &ctx->cq_check_overflow); | |
1544 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; | |
1545 | } | |
d4d19c19 | 1546 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1547 | ocqe->cqe.res = res; |
1548 | ocqe->cqe.flags = cflags; | |
1549 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1550 | return true; | |
2b188cc1 JA |
1551 | } |
1552 | ||
d4d19c19 PB |
1553 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1554 | long res, unsigned int cflags) | |
2b188cc1 JA |
1555 | { |
1556 | struct io_uring_cqe *cqe; | |
1557 | ||
d4d19c19 | 1558 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1559 | |
2b188cc1 JA |
1560 | /* |
1561 | * If we can't get a cq entry, userspace overflowed the | |
1562 | * submission (by quite a lot). Increment the overflow count in | |
1563 | * the ring. | |
1564 | */ | |
d068b506 | 1565 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1566 | if (likely(cqe)) { |
d4d19c19 | 1567 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1568 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1569 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1570 | return true; |
2b188cc1 | 1571 | } |
d4d19c19 | 1572 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1573 | } |
1574 | ||
8d13326e | 1575 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1576 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1577 | long res, unsigned int cflags) | |
bcda7baa | 1578 | { |
d4d19c19 | 1579 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1580 | } |
1581 | ||
7a612350 PB |
1582 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1583 | unsigned int cflags) | |
2b188cc1 | 1584 | { |
78e19bbe | 1585 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1586 | unsigned long flags; |
1587 | ||
1588 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
d4d19c19 | 1589 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1590 | /* |
1591 | * If we're the last reference to this request, add to our locked | |
1592 | * free_list cache. | |
1593 | */ | |
de9b4cca | 1594 | if (req_ref_put_and_test(req)) { |
7a612350 | 1595 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
93d2bcd2 | 1596 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) |
7a612350 PB |
1597 | io_disarm_next(req); |
1598 | if (req->link) { | |
1599 | io_req_task_queue(req->link); | |
1600 | req->link = NULL; | |
1601 | } | |
1602 | } | |
c7dae4ba JA |
1603 | io_dismantle_req(req); |
1604 | io_put_task(req->task, 1); | |
d0acdee2 PB |
1605 | list_add(&req->compl.list, &ctx->locked_free_list); |
1606 | ctx->locked_free_nr++; | |
180f829f PB |
1607 | } else { |
1608 | if (!percpu_ref_tryget(&ctx->refs)) | |
1609 | req = NULL; | |
1610 | } | |
7a612350 | 1611 | io_commit_cqring(ctx); |
2b188cc1 | 1612 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1613 | |
180f829f PB |
1614 | if (req) { |
1615 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1616 | percpu_ref_put(&ctx->refs); |
180f829f | 1617 | } |
229a7b63 JA |
1618 | } |
1619 | ||
4e3d9ff9 JA |
1620 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1621 | { | |
75652a30 | 1622 | return req->flags & (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | |
3a0a6902 | 1623 | REQ_F_POLLED | REQ_F_INFLIGHT); |
4e3d9ff9 JA |
1624 | } |
1625 | ||
a38d68db | 1626 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1627 | unsigned int cflags) |
229a7b63 | 1628 | { |
4e3d9ff9 | 1629 | if (io_req_needs_clean(req)) |
68fb8979 | 1630 | io_clean_op(req); |
a38d68db PB |
1631 | req->result = res; |
1632 | req->compl.cflags = cflags; | |
e342c807 | 1633 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1634 | } |
1635 | ||
889fca73 PB |
1636 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1637 | long res, unsigned cflags) | |
bcda7baa | 1638 | { |
889fca73 PB |
1639 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1640 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1641 | else |
c7dae4ba | 1642 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1643 | } |
1644 | ||
a38d68db | 1645 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1646 | { |
889fca73 | 1647 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1648 | } |
1649 | ||
f41db273 PB |
1650 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1651 | { | |
93d2bcd2 | 1652 | req_set_fail(req); |
f41db273 PB |
1653 | io_put_req(req); |
1654 | io_req_complete_post(req, res, 0); | |
1655 | } | |
1656 | ||
dac7a098 PB |
1657 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1658 | struct io_comp_state *cs) | |
1659 | { | |
1660 | spin_lock_irq(&ctx->completion_lock); | |
d0acdee2 PB |
1661 | list_splice_init(&ctx->locked_free_list, &cs->free_list); |
1662 | ctx->locked_free_nr = 0; | |
dac7a098 PB |
1663 | spin_unlock_irq(&ctx->completion_lock); |
1664 | } | |
1665 | ||
dd78f492 | 1666 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1667 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1668 | { |
c7dae4ba JA |
1669 | struct io_submit_state *state = &ctx->submit_state; |
1670 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1671 | int nr; |
0ddf92e8 | 1672 | |
c7dae4ba JA |
1673 | /* |
1674 | * If we have more than a batch's worth of requests in our IRQ side | |
1675 | * locked cache, grab the lock and move them over to our submission | |
1676 | * side cache. | |
1677 | */ | |
d0acdee2 | 1678 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
dac7a098 | 1679 | io_flush_cached_locked_reqs(ctx, cs); |
0ddf92e8 | 1680 | |
dd78f492 | 1681 | nr = state->free_reqs; |
c7dae4ba | 1682 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1683 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1684 | struct io_kiocb, compl.list); | |
1685 | ||
1b4c351f | 1686 | list_del(&req->compl.list); |
dd78f492 PB |
1687 | state->reqs[nr++] = req; |
1688 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1689 | break; |
1b4c351f JA |
1690 | } |
1691 | ||
dd78f492 PB |
1692 | state->free_reqs = nr; |
1693 | return nr != 0; | |
0ddf92e8 JA |
1694 | } |
1695 | ||
e5d1bc0a | 1696 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1697 | { |
e5d1bc0a PB |
1698 | struct io_submit_state *state = &ctx->submit_state; |
1699 | ||
1700 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1701 | ||
f6b6c7d6 | 1702 | if (!state->free_reqs) { |
291b2821 | 1703 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1704 | int ret; |
1705 | ||
c7dae4ba | 1706 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1707 | goto got_req; |
1708 | ||
bf019da7 PB |
1709 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1710 | state->reqs); | |
fd6fab2c JA |
1711 | |
1712 | /* | |
1713 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1714 | * retry single alloc to be on the safe side. | |
1715 | */ | |
1716 | if (unlikely(ret <= 0)) { | |
1717 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1718 | if (!state->reqs[0]) | |
3893f39f | 1719 | return NULL; |
fd6fab2c JA |
1720 | ret = 1; |
1721 | } | |
291b2821 | 1722 | state->free_reqs = ret; |
2b188cc1 | 1723 | } |
e5d1bc0a | 1724 | got_req: |
291b2821 PB |
1725 | state->free_reqs--; |
1726 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1727 | } |
1728 | ||
e1d767f0 | 1729 | static inline void io_put_file(struct file *file) |
8da11c19 | 1730 | { |
e1d767f0 | 1731 | if (file) |
8da11c19 PB |
1732 | fput(file); |
1733 | } | |
1734 | ||
4edf20f9 | 1735 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1736 | { |
094bae49 | 1737 | unsigned int flags = req->flags; |
929a3af9 | 1738 | |
3a0a6902 PB |
1739 | if (io_req_needs_clean(req)) |
1740 | io_clean_op(req); | |
e1d767f0 PB |
1741 | if (!(flags & REQ_F_FIXED_FILE)) |
1742 | io_put_file(req->file); | |
269bbe5f BM |
1743 | if (req->fixed_rsrc_refs) |
1744 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1745 | if (req->async_data) |
1746 | kfree(req->async_data); | |
003e8dcc JA |
1747 | if (req->work.creds) { |
1748 | put_cred(req->work.creds); | |
1749 | req->work.creds = NULL; | |
1750 | } | |
e65ef56d JA |
1751 | } |
1752 | ||
b23fcf47 | 1753 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1754 | static inline void io_put_task(struct task_struct *task, int nr) |
1755 | { | |
1756 | struct io_uring_task *tctx = task->io_uring; | |
1757 | ||
1758 | percpu_counter_sub(&tctx->inflight, nr); | |
1759 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1760 | wake_up(&tctx->wait); | |
1761 | put_task_struct_many(task, nr); | |
1762 | } | |
1763 | ||
216578e5 | 1764 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1765 | { |
51a4cc11 | 1766 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1767 | |
216578e5 | 1768 | io_dismantle_req(req); |
7c660731 | 1769 | io_put_task(req->task, 1); |
c6ca97b3 | 1770 | |
3893f39f | 1771 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1772 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1773 | } |
1774 | ||
f2f87370 PB |
1775 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1776 | { | |
1777 | struct io_kiocb *nxt = req->link; | |
1778 | ||
1779 | req->link = nxt->link; | |
1780 | nxt->link = NULL; | |
1781 | } | |
1782 | ||
33cc89a9 PB |
1783 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1784 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1785 | { |
33cc89a9 | 1786 | struct io_kiocb *link = req->link; |
f2f87370 | 1787 | |
900fad45 PB |
1788 | /* |
1789 | * Can happen if a linked timeout fired and link had been like | |
1790 | * req -> link t-out -> link t-out [-> ...] | |
1791 | */ | |
c9abd7ad PB |
1792 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1793 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1794 | |
f2f87370 | 1795 | io_remove_next_linked(req); |
90cd7e42 | 1796 | link->timeout.head = NULL; |
fd9c7bc5 | 1797 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
d4d19c19 PB |
1798 | io_cqring_fill_event(link->ctx, link->user_data, |
1799 | -ECANCELED, 0); | |
33cc89a9 | 1800 | io_put_req_deferred(link, 1); |
d4729fbd | 1801 | return true; |
c9abd7ad PB |
1802 | } |
1803 | } | |
d4729fbd | 1804 | return false; |
7c86ffee PB |
1805 | } |
1806 | ||
d148ca4b | 1807 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1808 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1809 | { |
33cc89a9 | 1810 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1811 | |
f2f87370 | 1812 | req->link = NULL; |
f2f87370 PB |
1813 | while (link) { |
1814 | nxt = link->link; | |
1815 | link->link = NULL; | |
2665abfd | 1816 | |
f2f87370 | 1817 | trace_io_uring_fail_link(req, link); |
d4d19c19 | 1818 | io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0); |
1575f21a | 1819 | io_put_req_deferred(link, 2); |
f2f87370 | 1820 | link = nxt; |
9e645e11 | 1821 | } |
33cc89a9 | 1822 | } |
9e645e11 | 1823 | |
33cc89a9 PB |
1824 | static bool io_disarm_next(struct io_kiocb *req) |
1825 | __must_hold(&req->ctx->completion_lock) | |
1826 | { | |
1827 | bool posted = false; | |
1828 | ||
1829 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1830 | posted = io_kill_linked_timeout(req); | |
93d2bcd2 | 1831 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 1832 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
1833 | posted |= (req->link != NULL); |
1834 | io_fail_links(req); | |
1835 | } | |
1836 | return posted; | |
9e645e11 JA |
1837 | } |
1838 | ||
3fa5e0f3 | 1839 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1840 | { |
33cc89a9 | 1841 | struct io_kiocb *nxt; |
944e58bf | 1842 | |
9e645e11 JA |
1843 | /* |
1844 | * If LINK is set, we have dependent requests in this chain. If we | |
1845 | * didn't fail this request, queue the first one up, moving any other | |
1846 | * dependencies to the next request. In case of failure, fail the rest | |
1847 | * of the chain. | |
1848 | */ | |
93d2bcd2 | 1849 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) { |
33cc89a9 PB |
1850 | struct io_ring_ctx *ctx = req->ctx; |
1851 | unsigned long flags; | |
1852 | bool posted; | |
1853 | ||
1854 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1855 | posted = io_disarm_next(req); | |
1856 | if (posted) | |
1857 | io_commit_cqring(req->ctx); | |
1858 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1859 | if (posted) | |
1860 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1861 | } |
33cc89a9 PB |
1862 | nxt = req->link; |
1863 | req->link = NULL; | |
1864 | return nxt; | |
4d7dd462 | 1865 | } |
9e645e11 | 1866 | |
f2f87370 | 1867 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1868 | { |
cdbff982 | 1869 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1870 | return NULL; |
1871 | return __io_req_find_next(req); | |
1872 | } | |
1873 | ||
2c32395d PB |
1874 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1875 | { | |
1876 | if (!ctx) | |
1877 | return; | |
1878 | if (ctx->submit_state.comp.nr) { | |
1879 | mutex_lock(&ctx->uring_lock); | |
1880 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1881 | mutex_unlock(&ctx->uring_lock); | |
1882 | } | |
1883 | percpu_ref_put(&ctx->refs); | |
1884 | } | |
1885 | ||
7cbf1722 | 1886 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1887 | { |
65453d1e | 1888 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1889 | struct io_wq_work_list list; |
1890 | struct io_wq_work_node *node; | |
c2c4c83c | 1891 | |
7cbf1722 JA |
1892 | if (wq_list_empty(&tctx->task_list)) |
1893 | return false; | |
6200b0ae | 1894 | |
0b81e80c | 1895 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1896 | list = tctx->task_list; |
1897 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1898 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1899 | |
7cbf1722 JA |
1900 | node = list.first; |
1901 | while (node) { | |
1902 | struct io_wq_work_node *next = node->next; | |
1903 | struct io_kiocb *req; | |
0ba9c9ed | 1904 | |
7cbf1722 | 1905 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1906 | if (req->ctx != ctx) { |
1907 | ctx_flush_and_put(ctx); | |
1908 | ctx = req->ctx; | |
1909 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1910 | } |
65453d1e | 1911 | |
2c32395d PB |
1912 | req->task_work.func(&req->task_work); |
1913 | node = next; | |
7cbf1722 JA |
1914 | } |
1915 | ||
2c32395d | 1916 | ctx_flush_and_put(ctx); |
7cbf1722 | 1917 | return list.first != NULL; |
c2c4c83c JA |
1918 | } |
1919 | ||
7cbf1722 | 1920 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1921 | { |
7cbf1722 | 1922 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1923 | |
1d5f360d JA |
1924 | clear_bit(0, &tctx->task_state); |
1925 | ||
7cbf1722 JA |
1926 | while (__tctx_task_work(tctx)) |
1927 | cond_resched(); | |
7cbf1722 JA |
1928 | } |
1929 | ||
c15b79de | 1930 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1931 | { |
c15b79de | 1932 | struct task_struct *tsk = req->task; |
7cbf1722 | 1933 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1934 | enum task_work_notify_mode notify; |
7cbf1722 | 1935 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1936 | unsigned long flags; |
c15b79de PB |
1937 | int ret = 0; |
1938 | ||
1939 | if (unlikely(tsk->flags & PF_EXITING)) | |
1940 | return -ESRCH; | |
7cbf1722 JA |
1941 | |
1942 | WARN_ON_ONCE(!tctx); | |
1943 | ||
0b81e80c | 1944 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1945 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1946 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1947 | |
1948 | /* task_work already pending, we're done */ | |
1949 | if (test_bit(0, &tctx->task_state) || | |
1950 | test_and_set_bit(0, &tctx->task_state)) | |
1951 | return 0; | |
1952 | ||
c15b79de PB |
1953 | /* |
1954 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1955 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1956 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1957 | * will do the job. | |
1958 | */ | |
1959 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1960 | ||
1961 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1962 | wake_up_process(tsk); | |
7cbf1722 | 1963 | return 0; |
c15b79de | 1964 | } |
7cbf1722 JA |
1965 | |
1966 | /* | |
1967 | * Slow path - we failed, find and delete work. if the work is not | |
1968 | * in the list, it got run and we're fine. | |
1969 | */ | |
0b81e80c | 1970 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1971 | wq_list_for_each(node, prev, &tctx->task_list) { |
1972 | if (&req->io_task_work.node == node) { | |
1973 | wq_list_del(&tctx->task_list, node, prev); | |
1974 | ret = 1; | |
1975 | break; | |
1976 | } | |
1977 | } | |
0b81e80c | 1978 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1979 | clear_bit(0, &tctx->task_state); |
1980 | return ret; | |
1981 | } | |
1982 | ||
9b465711 PB |
1983 | static bool io_run_task_work_head(struct callback_head **work_head) |
1984 | { | |
1985 | struct callback_head *work, *next; | |
1986 | bool executed = false; | |
1987 | ||
1988 | do { | |
1989 | work = xchg(work_head, NULL); | |
1990 | if (!work) | |
1991 | break; | |
1992 | ||
1993 | do { | |
1994 | next = work->next; | |
1995 | work->func(work); | |
1996 | work = next; | |
1997 | cond_resched(); | |
1998 | } while (work); | |
1999 | executed = true; | |
2000 | } while (1); | |
2001 | ||
2002 | return executed; | |
2003 | } | |
2004 | ||
2005 | static void io_task_work_add_head(struct callback_head **work_head, | |
2006 | struct callback_head *task_work) | |
eab30c4d | 2007 | { |
7c25c0d1 | 2008 | struct callback_head *head; |
eab30c4d | 2009 | |
7c25c0d1 | 2010 | do { |
9b465711 PB |
2011 | head = READ_ONCE(*work_head); |
2012 | task_work->next = head; | |
2013 | } while (cmpxchg(work_head, head, task_work) != head); | |
2014 | } | |
2015 | ||
2016 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2017 | task_work_func_t cb) | |
2018 | { | |
2019 | init_task_work(&req->task_work, cb); | |
2020 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2021 | } |
2022 | ||
c40f6379 JA |
2023 | static void io_req_task_cancel(struct callback_head *cb) |
2024 | { | |
2025 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2026 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2027 | |
e83acd7d | 2028 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2029 | mutex_lock(&ctx->uring_lock); |
2593553a | 2030 | io_req_complete_failed(req, req->result); |
792bb6eb | 2031 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2032 | } |
2033 | ||
2034 | static void __io_req_task_submit(struct io_kiocb *req) | |
2035 | { | |
2036 | struct io_ring_ctx *ctx = req->ctx; | |
2037 | ||
04fc6c80 | 2038 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2039 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2040 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2041 | __io_queue_sqe(req); |
81b6d05c | 2042 | else |
2593553a | 2043 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2044 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2045 | } |
2046 | ||
2047 | static void io_req_task_submit(struct callback_head *cb) | |
2048 | { | |
2049 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2050 | ||
2051 | __io_req_task_submit(req); | |
2052 | } | |
2053 | ||
2c4b8eb6 | 2054 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2055 | { |
2c4b8eb6 PB |
2056 | req->result = ret; |
2057 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2058 | |
2c4b8eb6 | 2059 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2060 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2061 | } |
2062 | ||
2c4b8eb6 | 2063 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2064 | { |
2c4b8eb6 | 2065 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2066 | |
2067 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2068 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2069 | } |
2070 | ||
f2f87370 | 2071 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2072 | { |
9b5f7bd9 | 2073 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2074 | |
2075 | if (nxt) | |
906a8c3f | 2076 | io_req_task_queue(nxt); |
c69f8dbe JL |
2077 | } |
2078 | ||
c3524383 | 2079 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2080 | { |
c3524383 PB |
2081 | io_queue_next(req); |
2082 | __io_free_req(req); | |
2083 | } | |
8766dd51 | 2084 | |
2d6500d4 | 2085 | struct req_batch { |
5af1d13e PB |
2086 | struct task_struct *task; |
2087 | int task_refs; | |
1b4c351f | 2088 | int ctx_refs; |
2d6500d4 PB |
2089 | }; |
2090 | ||
5af1d13e PB |
2091 | static inline void io_init_req_batch(struct req_batch *rb) |
2092 | { | |
5af1d13e | 2093 | rb->task_refs = 0; |
9ae72463 | 2094 | rb->ctx_refs = 0; |
5af1d13e PB |
2095 | rb->task = NULL; |
2096 | } | |
2097 | ||
2d6500d4 PB |
2098 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2099 | struct req_batch *rb) | |
2100 | { | |
6e833d53 | 2101 | if (rb->task) |
7c660731 | 2102 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2103 | if (rb->ctx_refs) |
2104 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2105 | } |
2106 | ||
6ff119a6 PB |
2107 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2108 | struct io_submit_state *state) | |
2d6500d4 | 2109 | { |
f2f87370 | 2110 | io_queue_next(req); |
96670657 | 2111 | io_dismantle_req(req); |
2d6500d4 | 2112 | |
e3bc8e9d | 2113 | if (req->task != rb->task) { |
7c660731 PB |
2114 | if (rb->task) |
2115 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2116 | rb->task = req->task; |
2117 | rb->task_refs = 0; | |
5af1d13e | 2118 | } |
e3bc8e9d | 2119 | rb->task_refs++; |
9ae72463 | 2120 | rb->ctx_refs++; |
5af1d13e | 2121 | |
bd759045 | 2122 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2123 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2124 | else |
2125 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2126 | } |
2127 | ||
905c172f PB |
2128 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2129 | struct io_ring_ctx *ctx) | |
2130 | { | |
2131 | int i, nr = cs->nr; | |
2132 | struct io_kiocb *req; | |
2133 | struct req_batch rb; | |
2134 | ||
2135 | io_init_req_batch(&rb); | |
2136 | spin_lock_irq(&ctx->completion_lock); | |
2137 | for (i = 0; i < nr; i++) { | |
2138 | req = cs->reqs[i]; | |
d4d19c19 PB |
2139 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2140 | req->compl.cflags); | |
905c172f PB |
2141 | } |
2142 | io_commit_cqring(ctx); | |
2143 | spin_unlock_irq(&ctx->completion_lock); | |
2144 | ||
2145 | io_cqring_ev_posted(ctx); | |
2146 | for (i = 0; i < nr; i++) { | |
2147 | req = cs->reqs[i]; | |
2148 | ||
2149 | /* submission and completion refs */ | |
de9b4cca | 2150 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2151 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2152 | } |
2153 | ||
2154 | io_req_free_batch_finish(ctx, &rb); | |
2155 | cs->nr = 0; | |
7a743e22 PB |
2156 | } |
2157 | ||
ba816ad6 JA |
2158 | /* |
2159 | * Drop reference to request, return next in chain (if there is one) if this | |
2160 | * was the last reference to this request. | |
2161 | */ | |
0d85035a | 2162 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2163 | { |
9b5f7bd9 PB |
2164 | struct io_kiocb *nxt = NULL; |
2165 | ||
de9b4cca | 2166 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2167 | nxt = io_req_find_next(req); |
4d7dd462 | 2168 | __io_free_req(req); |
2a44f467 | 2169 | } |
9b5f7bd9 | 2170 | return nxt; |
2b188cc1 JA |
2171 | } |
2172 | ||
0d85035a | 2173 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2174 | { |
de9b4cca | 2175 | if (req_ref_put_and_test(req)) |
e65ef56d | 2176 | io_free_req(req); |
2b188cc1 JA |
2177 | } |
2178 | ||
216578e5 PB |
2179 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2180 | { | |
2181 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2182 | ||
2183 | io_free_req(req); | |
2184 | } | |
2185 | ||
2186 | static void io_free_req_deferred(struct io_kiocb *req) | |
2187 | { | |
7cbf1722 | 2188 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2189 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2190 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2191 | } |
2192 | ||
2193 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2194 | { | |
de9b4cca | 2195 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2196 | io_free_req_deferred(req); |
2197 | } | |
2198 | ||
6c503150 | 2199 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2200 | { |
2201 | /* See comment at the top of this file */ | |
2202 | smp_rmb(); | |
e23de15f | 2203 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2204 | } |
2205 | ||
fb5ccc98 PB |
2206 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2207 | { | |
2208 | struct io_rings *rings = ctx->rings; | |
2209 | ||
2210 | /* make sure SQ entry isn't read before tail */ | |
2211 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2212 | } | |
2213 | ||
8ff069bf | 2214 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2215 | { |
8ff069bf | 2216 | unsigned int cflags; |
e94f141b | 2217 | |
bcda7baa JA |
2218 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2219 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2220 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2221 | kfree(kbuf); |
2222 | return cflags; | |
e94f141b JA |
2223 | } |
2224 | ||
8ff069bf | 2225 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2226 | { |
4d954c25 | 2227 | struct io_buffer *kbuf; |
bcda7baa | 2228 | |
4d954c25 | 2229 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2230 | return io_put_kbuf(req, kbuf); |
2231 | } | |
2232 | ||
4c6e277c JA |
2233 | static inline bool io_run_task_work(void) |
2234 | { | |
6200b0ae JA |
2235 | /* |
2236 | * Not safe to run on exiting task, and the task_work handling will | |
2237 | * not add work to such a task. | |
2238 | */ | |
2239 | if (unlikely(current->flags & PF_EXITING)) | |
2240 | return false; | |
4c6e277c JA |
2241 | if (current->task_works) { |
2242 | __set_current_state(TASK_RUNNING); | |
2243 | task_work_run(); | |
2244 | return true; | |
2245 | } | |
2246 | ||
2247 | return false; | |
bcda7baa JA |
2248 | } |
2249 | ||
def596e9 JA |
2250 | /* |
2251 | * Find and free completed poll iocbs | |
2252 | */ | |
2253 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2254 | struct list_head *done) | |
2255 | { | |
8237e045 | 2256 | struct req_batch rb; |
def596e9 | 2257 | struct io_kiocb *req; |
bbde017a XW |
2258 | |
2259 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2260 | smp_rmb(); | |
def596e9 | 2261 | |
5af1d13e | 2262 | io_init_req_batch(&rb); |
def596e9 | 2263 | while (!list_empty(done)) { |
bcda7baa JA |
2264 | int cflags = 0; |
2265 | ||
d21ffe7e | 2266 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2267 | list_del(&req->inflight_entry); |
2268 | ||
8c130827 PB |
2269 | if (READ_ONCE(req->result) == -EAGAIN && |
2270 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2271 | req->iopoll_completed = 0; |
8c130827 PB |
2272 | req_ref_get(req); |
2273 | io_queue_async_work(req); | |
2274 | continue; | |
bbde017a | 2275 | } |
def596e9 | 2276 | |
bcda7baa | 2277 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2278 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2279 | |
d4d19c19 | 2280 | __io_cqring_fill_event(ctx, req->user_data, req->result, cflags); |
def596e9 JA |
2281 | (*nr_events)++; |
2282 | ||
de9b4cca | 2283 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2284 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2285 | } |
def596e9 | 2286 | |
09bb8394 | 2287 | io_commit_cqring(ctx); |
80c18e4a | 2288 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2289 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2290 | } |
2291 | ||
def596e9 JA |
2292 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2293 | long min) | |
2294 | { | |
2295 | struct io_kiocb *req, *tmp; | |
2296 | LIST_HEAD(done); | |
2297 | bool spin; | |
2298 | int ret; | |
2299 | ||
2300 | /* | |
2301 | * Only spin for completions if we don't have multiple devices hanging | |
2302 | * off our complete list, and we're under the requested amount. | |
2303 | */ | |
2304 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2305 | ||
2306 | ret = 0; | |
d21ffe7e | 2307 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2308 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2309 | |
2310 | /* | |
581f9810 BM |
2311 | * Move completed and retryable entries to our local lists. |
2312 | * If we find a request that requires polling, break out | |
2313 | * and complete those lists first, if we have entries there. | |
def596e9 | 2314 | */ |
65a6543d | 2315 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2316 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2317 | continue; |
2318 | } | |
2319 | if (!list_empty(&done)) | |
2320 | break; | |
2321 | ||
2322 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2323 | if (ret < 0) | |
2324 | break; | |
2325 | ||
3aadc23e PB |
2326 | /* iopoll may have completed current req */ |
2327 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2328 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2329 | |
def596e9 JA |
2330 | if (ret && spin) |
2331 | spin = false; | |
2332 | ret = 0; | |
2333 | } | |
2334 | ||
2335 | if (!list_empty(&done)) | |
2336 | io_iopoll_complete(ctx, nr_events, &done); | |
2337 | ||
2338 | return ret; | |
2339 | } | |
2340 | ||
def596e9 JA |
2341 | /* |
2342 | * We can't just wait for polled events to come to us, we have to actively | |
2343 | * find and complete them. | |
2344 | */ | |
b2edc0a7 | 2345 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2346 | { |
2347 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2348 | return; | |
2349 | ||
2350 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2351 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2352 | unsigned int nr_events = 0; |
2353 | ||
b2edc0a7 | 2354 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2355 | |
b2edc0a7 PB |
2356 | /* let it sleep and repeat later if can't complete a request */ |
2357 | if (nr_events == 0) | |
2358 | break; | |
08f5439f JA |
2359 | /* |
2360 | * Ensure we allow local-to-the-cpu processing to take place, | |
2361 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2362 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2363 | */ |
3fcee5a6 PB |
2364 | if (need_resched()) { |
2365 | mutex_unlock(&ctx->uring_lock); | |
2366 | cond_resched(); | |
2367 | mutex_lock(&ctx->uring_lock); | |
2368 | } | |
def596e9 JA |
2369 | } |
2370 | mutex_unlock(&ctx->uring_lock); | |
2371 | } | |
2372 | ||
7668b92a | 2373 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2374 | { |
7668b92a | 2375 | unsigned int nr_events = 0; |
e9979b36 | 2376 | int ret = 0; |
500f9fba | 2377 | |
c7849be9 XW |
2378 | /* |
2379 | * We disallow the app entering submit/complete with polling, but we | |
2380 | * still need to lock the ring to prevent racing with polled issue | |
2381 | * that got punted to a workqueue. | |
2382 | */ | |
2383 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2384 | /* |
2385 | * Don't enter poll loop if we already have events pending. | |
2386 | * If we do, we can potentially be spinning for commands that | |
2387 | * already triggered a CQE (eg in error). | |
2388 | */ | |
2389 | if (test_bit(0, &ctx->cq_check_overflow)) | |
2390 | __io_cqring_overflow_flush(ctx, false); | |
2391 | if (io_cqring_events(ctx)) | |
2392 | goto out; | |
def596e9 | 2393 | do { |
500f9fba JA |
2394 | /* |
2395 | * If a submit got punted to a workqueue, we can have the | |
2396 | * application entering polling for a command before it gets | |
2397 | * issued. That app will hold the uring_lock for the duration | |
2398 | * of the poll right here, so we need to take a breather every | |
2399 | * now and then to ensure that the issue has a chance to add | |
2400 | * the poll to the issued list. Otherwise we can spin here | |
2401 | * forever, while the workqueue is stuck trying to acquire the | |
2402 | * very same mutex. | |
2403 | */ | |
e9979b36 | 2404 | if (list_empty(&ctx->iopoll_list)) { |
500f9fba | 2405 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2406 | io_run_task_work(); |
500f9fba | 2407 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2408 | |
e9979b36 PB |
2409 | if (list_empty(&ctx->iopoll_list)) |
2410 | break; | |
500f9fba | 2411 | } |
f39c8a5b PB |
2412 | ret = io_do_iopoll(ctx, &nr_events, min); |
2413 | } while (!ret && nr_events < min && !need_resched()); | |
2414 | out: | |
500f9fba | 2415 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2416 | return ret; |
2417 | } | |
2418 | ||
491381ce | 2419 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2420 | { |
491381ce JA |
2421 | /* |
2422 | * Tell lockdep we inherited freeze protection from submission | |
2423 | * thread. | |
2424 | */ | |
2425 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2426 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2427 | |
1c98679d PB |
2428 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2429 | sb_end_write(sb); | |
2b188cc1 JA |
2430 | } |
2431 | } | |
2432 | ||
b63534c4 | 2433 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2434 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2435 | { |
ab454438 | 2436 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2437 | |
ab454438 PB |
2438 | if (!rw) |
2439 | return !io_req_prep_async(req); | |
2440 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2441 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2442 | return true; | |
b63534c4 | 2443 | } |
b63534c4 | 2444 | |
3e6a0d3c | 2445 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2446 | { |
355afaeb | 2447 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2448 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2449 | |
355afaeb JA |
2450 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2451 | return false; | |
3e6a0d3c JA |
2452 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2453 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2454 | return false; |
7c977a58 JA |
2455 | /* |
2456 | * If ref is dying, we might be running poll reap from the exit work. | |
2457 | * Don't attempt to reissue from that path, just let it fail with | |
2458 | * -EAGAIN. | |
2459 | */ | |
3e6a0d3c JA |
2460 | if (percpu_ref_is_dying(&ctx->refs)) |
2461 | return false; | |
2462 | return true; | |
2463 | } | |
e82ad485 | 2464 | #else |
a1ff1e3f | 2465 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2466 | { |
2467 | return false; | |
2468 | } | |
e82ad485 | 2469 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2470 | { |
b63534c4 JA |
2471 | return false; |
2472 | } | |
3e6a0d3c | 2473 | #endif |
b63534c4 | 2474 | |
a1d7c393 | 2475 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2476 | unsigned int issue_flags) |
a1d7c393 | 2477 | { |
2f8e45f1 PB |
2478 | int cflags = 0; |
2479 | ||
b65c128f PB |
2480 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2481 | kiocb_end_write(req); | |
9532b99b PB |
2482 | if (res != req->result) { |
2483 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2484 | io_rw_should_reissue(req)) { | |
2485 | req->flags |= REQ_F_REISSUE; | |
2486 | return; | |
2487 | } | |
93d2bcd2 | 2488 | req_set_fail(req); |
9532b99b | 2489 | } |
2f8e45f1 PB |
2490 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2491 | cflags = io_put_rw_kbuf(req); | |
2492 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2493 | } |
2494 | ||
2495 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2496 | { | |
9adbd45d | 2497 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2498 | |
889fca73 | 2499 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2500 | } |
2501 | ||
def596e9 JA |
2502 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2503 | { | |
9adbd45d | 2504 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2505 | |
491381ce JA |
2506 | if (kiocb->ki_flags & IOCB_WRITE) |
2507 | kiocb_end_write(req); | |
9532b99b | 2508 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2509 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2510 | io_resubmit_prep(req))) { | |
93d2bcd2 | 2511 | req_set_fail(req); |
9532b99b PB |
2512 | req->flags |= REQ_F_DONT_REISSUE; |
2513 | } | |
8c130827 | 2514 | } |
bbde017a XW |
2515 | |
2516 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2517 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2518 | smp_wmb(); |
2519 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2520 | } |
2521 | ||
2522 | /* | |
2523 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2524 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2525 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2526 | * accessing the kiocb cookie. |
2527 | */ | |
2e9dbe90 | 2528 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2529 | { |
2530 | struct io_ring_ctx *ctx = req->ctx; | |
2531 | ||
2532 | /* | |
2533 | * Track whether we have multiple files in our lists. This will impact | |
2534 | * how we do polling eventually, not spinning if we're on potentially | |
2535 | * different devices. | |
2536 | */ | |
540e32a0 | 2537 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2538 | ctx->poll_multi_file = false; |
2539 | } else if (!ctx->poll_multi_file) { | |
2540 | struct io_kiocb *list_req; | |
2541 | ||
540e32a0 | 2542 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2543 | inflight_entry); |
9adbd45d | 2544 | if (list_req->file != req->file) |
def596e9 JA |
2545 | ctx->poll_multi_file = true; |
2546 | } | |
2547 | ||
2548 | /* | |
2549 | * For fast devices, IO may have already completed. If it has, add | |
2550 | * it to the front so we find it first. | |
2551 | */ | |
65a6543d | 2552 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2553 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2554 | else |
d21ffe7e | 2555 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2556 | |
2e9dbe90 XW |
2557 | /* |
2558 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2559 | * task context or in io worker task context. If current task context is | |
2560 | * sq thread, we don't need to check whether should wake up sq thread. | |
2561 | */ | |
2562 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2563 | wq_has_sleeper(&ctx->sq_data->wait)) |
2564 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2565 | } |
2566 | ||
9f13c35b PB |
2567 | static inline void io_state_file_put(struct io_submit_state *state) |
2568 | { | |
02b23a9a PB |
2569 | if (state->file_refs) { |
2570 | fput_many(state->file, state->file_refs); | |
2571 | state->file_refs = 0; | |
2572 | } | |
9a56a232 JA |
2573 | } |
2574 | ||
2575 | /* | |
2576 | * Get as many references to a file as we have IOs left in this submission, | |
2577 | * assuming most submissions are for one file, or at least that each file | |
2578 | * has more than one submission. | |
2579 | */ | |
8da11c19 | 2580 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2581 | { |
2582 | if (!state) | |
2583 | return fget(fd); | |
2584 | ||
6e1271e6 | 2585 | if (state->file_refs) { |
9a56a232 | 2586 | if (state->fd == fd) { |
6e1271e6 | 2587 | state->file_refs--; |
9a56a232 JA |
2588 | return state->file; |
2589 | } | |
02b23a9a | 2590 | io_state_file_put(state); |
9a56a232 JA |
2591 | } |
2592 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2593 | if (unlikely(!state->file)) |
9a56a232 JA |
2594 | return NULL; |
2595 | ||
2596 | state->fd = fd; | |
6e1271e6 | 2597 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2598 | return state->file; |
2599 | } | |
2600 | ||
4503b767 JA |
2601 | static bool io_bdev_nowait(struct block_device *bdev) |
2602 | { | |
9ba0d0c8 | 2603 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2604 | } |
2605 | ||
2b188cc1 JA |
2606 | /* |
2607 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2608 | * any file. For now, just ensure that anything potentially problematic is done | |
2609 | * inline. | |
2610 | */ | |
7b29f92d | 2611 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2612 | { |
2613 | umode_t mode = file_inode(file)->i_mode; | |
2614 | ||
4503b767 | 2615 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2616 | if (IS_ENABLED(CONFIG_BLOCK) && |
2617 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2618 | return true; |
2619 | return false; | |
2620 | } | |
2621 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2622 | return true; |
4503b767 | 2623 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2624 | if (IS_ENABLED(CONFIG_BLOCK) && |
2625 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2626 | file->f_op != &io_uring_fops) |
2627 | return true; | |
2628 | return false; | |
2629 | } | |
2b188cc1 | 2630 | |
c5b85625 JA |
2631 | /* any ->read/write should understand O_NONBLOCK */ |
2632 | if (file->f_flags & O_NONBLOCK) | |
2633 | return true; | |
2634 | ||
af197f50 JA |
2635 | if (!(file->f_mode & FMODE_NOWAIT)) |
2636 | return false; | |
2637 | ||
2638 | if (rw == READ) | |
2639 | return file->f_op->read_iter != NULL; | |
2640 | ||
2641 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2642 | } |
2643 | ||
7b29f92d JA |
2644 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2645 | { | |
2646 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2647 | return true; | |
2648 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2649 | return true; | |
2650 | ||
2651 | return __io_file_supports_async(req->file, rw); | |
2652 | } | |
2653 | ||
a88fc400 | 2654 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2655 | { |
def596e9 | 2656 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2657 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2658 | struct file *file = req->file; |
09bb8394 JA |
2659 | unsigned ioprio; |
2660 | int ret; | |
2b188cc1 | 2661 | |
7b29f92d | 2662 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2663 | req->flags |= REQ_F_ISREG; |
2664 | ||
2b188cc1 | 2665 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2666 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2667 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2668 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2669 | } |
2b188cc1 | 2670 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2671 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2672 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2673 | if (unlikely(ret)) | |
2674 | return ret; | |
2b188cc1 | 2675 | |
75c668cd PB |
2676 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2677 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2678 | req->flags |= REQ_F_NOWAIT; | |
2679 | ||
2b188cc1 JA |
2680 | ioprio = READ_ONCE(sqe->ioprio); |
2681 | if (ioprio) { | |
2682 | ret = ioprio_check_cap(ioprio); | |
2683 | if (ret) | |
09bb8394 | 2684 | return ret; |
2b188cc1 JA |
2685 | |
2686 | kiocb->ki_ioprio = ioprio; | |
2687 | } else | |
2688 | kiocb->ki_ioprio = get_current_ioprio(); | |
2689 | ||
def596e9 | 2690 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2691 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2692 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2693 | return -EOPNOTSUPP; |
2b188cc1 | 2694 | |
def596e9 JA |
2695 | kiocb->ki_flags |= IOCB_HIPRI; |
2696 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2697 | req->iopoll_completed = 0; |
def596e9 | 2698 | } else { |
09bb8394 JA |
2699 | if (kiocb->ki_flags & IOCB_HIPRI) |
2700 | return -EINVAL; | |
def596e9 JA |
2701 | kiocb->ki_complete = io_complete_rw; |
2702 | } | |
9adbd45d | 2703 | |
eae071c9 PB |
2704 | if (req->opcode == IORING_OP_READ_FIXED || |
2705 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2706 | req->imu = NULL; | |
2707 | io_req_set_rsrc_node(req); | |
2708 | } | |
2709 | ||
3529d8c2 JA |
2710 | req->rw.addr = READ_ONCE(sqe->addr); |
2711 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2712 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2713 | return 0; |
2b188cc1 JA |
2714 | } |
2715 | ||
2716 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2717 | { | |
2718 | switch (ret) { | |
2719 | case -EIOCBQUEUED: | |
2720 | break; | |
2721 | case -ERESTARTSYS: | |
2722 | case -ERESTARTNOINTR: | |
2723 | case -ERESTARTNOHAND: | |
2724 | case -ERESTART_RESTARTBLOCK: | |
2725 | /* | |
2726 | * We can't just restart the syscall, since previously | |
2727 | * submitted sqes may already be in progress. Just fail this | |
2728 | * IO with EINTR. | |
2729 | */ | |
2730 | ret = -EINTR; | |
df561f66 | 2731 | fallthrough; |
2b188cc1 JA |
2732 | default: |
2733 | kiocb->ki_complete(kiocb, ret, 0); | |
2734 | } | |
2735 | } | |
2736 | ||
a1d7c393 | 2737 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2738 | unsigned int issue_flags) |
ba816ad6 | 2739 | { |
ba04291e | 2740 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2741 | struct io_async_rw *io = req->async_data; |
97284637 | 2742 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2743 | |
227c0c96 | 2744 | /* add previously done IO, if any */ |
e8c2bc1f | 2745 | if (io && io->bytes_done > 0) { |
227c0c96 | 2746 | if (ret < 0) |
e8c2bc1f | 2747 | ret = io->bytes_done; |
227c0c96 | 2748 | else |
e8c2bc1f | 2749 | ret += io->bytes_done; |
227c0c96 JA |
2750 | } |
2751 | ||
ba04291e JA |
2752 | if (req->flags & REQ_F_CUR_POS) |
2753 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2754 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2755 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2756 | else |
2757 | io_rw_done(kiocb, ret); | |
97284637 PB |
2758 | |
2759 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2760 | req->flags &= ~REQ_F_REISSUE; | |
a7be7c23 | 2761 | if (io_resubmit_prep(req)) { |
8c130827 PB |
2762 | req_ref_get(req); |
2763 | io_queue_async_work(req); | |
2764 | } else { | |
97284637 PB |
2765 | int cflags = 0; |
2766 | ||
93d2bcd2 | 2767 | req_set_fail(req); |
97284637 PB |
2768 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2769 | cflags = io_put_rw_kbuf(req); | |
2770 | __io_req_complete(req, issue_flags, ret, cflags); | |
2771 | } | |
2772 | } | |
ba816ad6 JA |
2773 | } |
2774 | ||
eae071c9 PB |
2775 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2776 | struct io_mapped_ubuf *imu) | |
edafccee | 2777 | { |
9adbd45d | 2778 | size_t len = req->rw.len; |
75769e3f | 2779 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2780 | size_t offset; |
edafccee | 2781 | |
75769e3f | 2782 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2783 | return -EFAULT; |
2784 | /* not inside the mapped region */ | |
4751f53d | 2785 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2786 | return -EFAULT; |
2787 | ||
2788 | /* | |
2789 | * May not be a start of buffer, set size appropriately | |
2790 | * and advance us to the beginning. | |
2791 | */ | |
2792 | offset = buf_addr - imu->ubuf; | |
2793 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2794 | |
2795 | if (offset) { | |
2796 | /* | |
2797 | * Don't use iov_iter_advance() here, as it's really slow for | |
2798 | * using the latter parts of a big fixed buffer - it iterates | |
2799 | * over each segment manually. We can cheat a bit here, because | |
2800 | * we know that: | |
2801 | * | |
2802 | * 1) it's a BVEC iter, we set it up | |
2803 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2804 | * first and last bvec | |
2805 | * | |
2806 | * So just find our index, and adjust the iterator afterwards. | |
2807 | * If the offset is within the first bvec (or the whole first | |
2808 | * bvec, just use iov_iter_advance(). This makes it easier | |
2809 | * since we can just skip the first segment, which may not | |
2810 | * be PAGE_SIZE aligned. | |
2811 | */ | |
2812 | const struct bio_vec *bvec = imu->bvec; | |
2813 | ||
2814 | if (offset <= bvec->bv_len) { | |
2815 | iov_iter_advance(iter, offset); | |
2816 | } else { | |
2817 | unsigned long seg_skip; | |
2818 | ||
2819 | /* skip first vec */ | |
2820 | offset -= bvec->bv_len; | |
2821 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2822 | ||
2823 | iter->bvec = bvec + seg_skip; | |
2824 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2825 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2826 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2827 | } |
2828 | } | |
2829 | ||
847595de | 2830 | return 0; |
edafccee JA |
2831 | } |
2832 | ||
eae071c9 PB |
2833 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
2834 | { | |
2835 | struct io_ring_ctx *ctx = req->ctx; | |
2836 | struct io_mapped_ubuf *imu = req->imu; | |
2837 | u16 index, buf_index = req->buf_index; | |
2838 | ||
2839 | if (likely(!imu)) { | |
2840 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
2841 | return -EFAULT; | |
2842 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2843 | imu = READ_ONCE(ctx->user_bufs[index]); | |
2844 | req->imu = imu; | |
2845 | } | |
2846 | return __io_import_fixed(req, rw, iter, imu); | |
2847 | } | |
2848 | ||
bcda7baa JA |
2849 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2850 | { | |
2851 | if (needs_lock) | |
2852 | mutex_unlock(&ctx->uring_lock); | |
2853 | } | |
2854 | ||
2855 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2856 | { | |
2857 | /* | |
2858 | * "Normal" inline submissions always hold the uring_lock, since we | |
2859 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2860 | * The only exception is when we've detached the request and issue it | |
2861 | * from an async worker thread, grab the lock for that case. | |
2862 | */ | |
2863 | if (needs_lock) | |
2864 | mutex_lock(&ctx->uring_lock); | |
2865 | } | |
2866 | ||
2867 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2868 | int bgid, struct io_buffer *kbuf, | |
2869 | bool needs_lock) | |
2870 | { | |
2871 | struct io_buffer *head; | |
2872 | ||
2873 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2874 | return kbuf; | |
2875 | ||
2876 | io_ring_submit_lock(req->ctx, needs_lock); | |
2877 | ||
2878 | lockdep_assert_held(&req->ctx->uring_lock); | |
2879 | ||
9e15c3a0 | 2880 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2881 | if (head) { |
2882 | if (!list_empty(&head->list)) { | |
2883 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2884 | list); | |
2885 | list_del(&kbuf->list); | |
2886 | } else { | |
2887 | kbuf = head; | |
9e15c3a0 | 2888 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2889 | } |
2890 | if (*len > kbuf->len) | |
2891 | *len = kbuf->len; | |
2892 | } else { | |
2893 | kbuf = ERR_PTR(-ENOBUFS); | |
2894 | } | |
2895 | ||
2896 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2897 | ||
2898 | return kbuf; | |
2899 | } | |
2900 | ||
4d954c25 JA |
2901 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2902 | bool needs_lock) | |
2903 | { | |
2904 | struct io_buffer *kbuf; | |
4f4eeba8 | 2905 | u16 bgid; |
4d954c25 JA |
2906 | |
2907 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2908 | bgid = req->buf_index; |
4d954c25 JA |
2909 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2910 | if (IS_ERR(kbuf)) | |
2911 | return kbuf; | |
2912 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2913 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2914 | return u64_to_user_ptr(kbuf->addr); | |
2915 | } | |
2916 | ||
2917 | #ifdef CONFIG_COMPAT | |
2918 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2919 | bool needs_lock) | |
2920 | { | |
2921 | struct compat_iovec __user *uiov; | |
2922 | compat_ssize_t clen; | |
2923 | void __user *buf; | |
2924 | ssize_t len; | |
2925 | ||
2926 | uiov = u64_to_user_ptr(req->rw.addr); | |
2927 | if (!access_ok(uiov, sizeof(*uiov))) | |
2928 | return -EFAULT; | |
2929 | if (__get_user(clen, &uiov->iov_len)) | |
2930 | return -EFAULT; | |
2931 | if (clen < 0) | |
2932 | return -EINVAL; | |
2933 | ||
2934 | len = clen; | |
2935 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2936 | if (IS_ERR(buf)) | |
2937 | return PTR_ERR(buf); | |
2938 | iov[0].iov_base = buf; | |
2939 | iov[0].iov_len = (compat_size_t) len; | |
2940 | return 0; | |
2941 | } | |
2942 | #endif | |
2943 | ||
2944 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2945 | bool needs_lock) | |
2946 | { | |
2947 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2948 | void __user *buf; | |
2949 | ssize_t len; | |
2950 | ||
2951 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2952 | return -EFAULT; | |
2953 | ||
2954 | len = iov[0].iov_len; | |
2955 | if (len < 0) | |
2956 | return -EINVAL; | |
2957 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2958 | if (IS_ERR(buf)) | |
2959 | return PTR_ERR(buf); | |
2960 | iov[0].iov_base = buf; | |
2961 | iov[0].iov_len = len; | |
2962 | return 0; | |
2963 | } | |
2964 | ||
2965 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2966 | bool needs_lock) | |
2967 | { | |
dddb3e26 JA |
2968 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2969 | struct io_buffer *kbuf; | |
2970 | ||
2971 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2972 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2973 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2974 | return 0; |
dddb3e26 | 2975 | } |
dd201662 | 2976 | if (req->rw.len != 1) |
4d954c25 JA |
2977 | return -EINVAL; |
2978 | ||
2979 | #ifdef CONFIG_COMPAT | |
2980 | if (req->ctx->compat) | |
2981 | return io_compat_import(req, iov, needs_lock); | |
2982 | #endif | |
2983 | ||
2984 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2985 | } | |
2986 | ||
847595de PB |
2987 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2988 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2989 | { |
9adbd45d JA |
2990 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2991 | size_t sqe_len = req->rw.len; | |
847595de | 2992 | u8 opcode = req->opcode; |
4d954c25 | 2993 | ssize_t ret; |
edafccee | 2994 | |
7d009165 | 2995 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2996 | *iovec = NULL; |
9adbd45d | 2997 | return io_import_fixed(req, rw, iter); |
edafccee | 2998 | } |
2b188cc1 | 2999 | |
bcda7baa | 3000 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3001 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3002 | return -EINVAL; |
3003 | ||
3a6820f2 | 3004 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3005 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3006 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3007 | if (IS_ERR(buf)) |
4d954c25 | 3008 | return PTR_ERR(buf); |
3f9d6441 | 3009 | req->rw.len = sqe_len; |
bcda7baa JA |
3010 | } |
3011 | ||
3a6820f2 JA |
3012 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3013 | *iovec = NULL; | |
10fc72e4 | 3014 | return ret; |
3a6820f2 JA |
3015 | } |
3016 | ||
4d954c25 JA |
3017 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3018 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3019 | if (!ret) |
3020 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3021 | *iovec = NULL; |
3022 | return ret; | |
3023 | } | |
3024 | ||
89cd35c5 CH |
3025 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3026 | req->ctx->compat); | |
2b188cc1 JA |
3027 | } |
3028 | ||
0fef9483 JA |
3029 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3030 | { | |
5b09e37e | 3031 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3032 | } |
3033 | ||
31b51510 | 3034 | /* |
32960613 JA |
3035 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3036 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3037 | */ |
4017eb91 | 3038 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3039 | { |
4017eb91 JA |
3040 | struct kiocb *kiocb = &req->rw.kiocb; |
3041 | struct file *file = req->file; | |
32960613 JA |
3042 | ssize_t ret = 0; |
3043 | ||
3044 | /* | |
3045 | * Don't support polled IO through this interface, and we can't | |
3046 | * support non-blocking either. For the latter, this just causes | |
3047 | * the kiocb to be handled from an async context. | |
3048 | */ | |
3049 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3050 | return -EOPNOTSUPP; | |
3051 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3052 | return -EAGAIN; | |
3053 | ||
3054 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3055 | struct iovec iovec; |
32960613 JA |
3056 | ssize_t nr; |
3057 | ||
311ae9e1 PB |
3058 | if (!iov_iter_is_bvec(iter)) { |
3059 | iovec = iov_iter_iovec(iter); | |
3060 | } else { | |
4017eb91 JA |
3061 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3062 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3063 | } |
3064 | ||
32960613 JA |
3065 | if (rw == READ) { |
3066 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3067 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3068 | } else { |
3069 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3070 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3071 | } |
3072 | ||
3073 | if (nr < 0) { | |
3074 | if (!ret) | |
3075 | ret = nr; | |
3076 | break; | |
3077 | } | |
3078 | ret += nr; | |
3079 | if (nr != iovec.iov_len) | |
3080 | break; | |
4017eb91 JA |
3081 | req->rw.len -= nr; |
3082 | req->rw.addr += nr; | |
32960613 JA |
3083 | iov_iter_advance(iter, nr); |
3084 | } | |
3085 | ||
3086 | return ret; | |
3087 | } | |
3088 | ||
ff6165b2 JA |
3089 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3090 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3091 | { |
e8c2bc1f | 3092 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3093 | |
ff6165b2 | 3094 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3095 | rw->free_iovec = iovec; |
227c0c96 | 3096 | rw->bytes_done = 0; |
ff6165b2 | 3097 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3098 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3099 | return; |
b64e3444 | 3100 | if (!iovec) { |
ff6165b2 JA |
3101 | unsigned iov_off = 0; |
3102 | ||
3103 | rw->iter.iov = rw->fast_iov; | |
3104 | if (iter->iov != fast_iov) { | |
3105 | iov_off = iter->iov - fast_iov; | |
3106 | rw->iter.iov += iov_off; | |
3107 | } | |
3108 | if (rw->fast_iov != fast_iov) | |
3109 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3110 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3111 | } else { |
3112 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3113 | } |
3114 | } | |
3115 | ||
6cb78689 | 3116 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3117 | { |
e8c2bc1f JA |
3118 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3119 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3120 | return req->async_data == NULL; | |
3d9932a8 XW |
3121 | } |
3122 | ||
ff6165b2 JA |
3123 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3124 | const struct iovec *fast_iov, | |
227c0c96 | 3125 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3126 | { |
26f0505a | 3127 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3128 | return 0; |
e8c2bc1f | 3129 | if (!req->async_data) { |
6cb78689 | 3130 | if (io_alloc_async_data(req)) { |
6bf985dc | 3131 | kfree(iovec); |
5d204bcf | 3132 | return -ENOMEM; |
6bf985dc | 3133 | } |
b7bb4f7d | 3134 | |
ff6165b2 | 3135 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3136 | } |
b7bb4f7d | 3137 | return 0; |
f67676d1 JA |
3138 | } |
3139 | ||
73debe68 | 3140 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3141 | { |
e8c2bc1f | 3142 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3143 | struct iovec *iov = iorw->fast_iov; |
847595de | 3144 | int ret; |
c3e330a4 | 3145 | |
2846c481 | 3146 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3147 | if (unlikely(ret < 0)) |
3148 | return ret; | |
3149 | ||
ab0b196c PB |
3150 | iorw->bytes_done = 0; |
3151 | iorw->free_iovec = iov; | |
3152 | if (iov) | |
3153 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3154 | return 0; |
3155 | } | |
3156 | ||
73debe68 | 3157 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3158 | { |
3529d8c2 JA |
3159 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3160 | return -EBADF; | |
93642ef8 | 3161 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3162 | } |
3163 | ||
c1dd91d1 JA |
3164 | /* |
3165 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3166 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3167 | * This gets called when the page is unlocked, and we generally expect that to | |
3168 | * happen when the page IO is completed and the page is now uptodate. This will | |
3169 | * queue a task_work based retry of the operation, attempting to copy the data | |
3170 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3171 | * do a thread based blocking retry of the operation. That's the unexpected | |
3172 | * slow path. | |
3173 | */ | |
bcf5a063 JA |
3174 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3175 | int sync, void *arg) | |
3176 | { | |
3177 | struct wait_page_queue *wpq; | |
3178 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3179 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3180 | |
3181 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3182 | ||
cdc8fcb4 LT |
3183 | if (!wake_page_match(wpq, key)) |
3184 | return 0; | |
3185 | ||
c8d317aa | 3186 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3187 | list_del_init(&wait->entry); |
3188 | ||
bcf5a063 | 3189 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3190 | req_ref_get(req); |
921b9054 | 3191 | io_req_task_queue(req); |
bcf5a063 JA |
3192 | return 1; |
3193 | } | |
3194 | ||
c1dd91d1 JA |
3195 | /* |
3196 | * This controls whether a given IO request should be armed for async page | |
3197 | * based retry. If we return false here, the request is handed to the async | |
3198 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3199 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3200 | * will either succeed because the page is now uptodate and unlocked, or it | |
3201 | * will register a callback when the page is unlocked at IO completion. Through | |
3202 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3203 | * That retry will attempt the buffered read again. The retry will generally | |
3204 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3205 | * async worker threads for a blocking retry. | |
3206 | */ | |
227c0c96 | 3207 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3208 | { |
e8c2bc1f JA |
3209 | struct io_async_rw *rw = req->async_data; |
3210 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3211 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3212 | |
bcf5a063 JA |
3213 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3214 | if (req->flags & REQ_F_NOWAIT) | |
3215 | return false; | |
f67676d1 | 3216 | |
227c0c96 | 3217 | /* Only for buffered IO */ |
3b2a4439 | 3218 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3219 | return false; |
3b2a4439 | 3220 | |
bcf5a063 JA |
3221 | /* |
3222 | * just use poll if we can, and don't attempt if the fs doesn't | |
3223 | * support callback based unlocks | |
3224 | */ | |
3225 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3226 | return false; | |
f67676d1 | 3227 | |
3b2a4439 JA |
3228 | wait->wait.func = io_async_buf_func; |
3229 | wait->wait.private = req; | |
3230 | wait->wait.flags = 0; | |
3231 | INIT_LIST_HEAD(&wait->wait.entry); | |
3232 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3233 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3234 | kiocb->ki_waitq = wait; |
3b2a4439 | 3235 | return true; |
bcf5a063 JA |
3236 | } |
3237 | ||
3238 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3239 | { | |
3240 | if (req->file->f_op->read_iter) | |
3241 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3242 | else if (req->file->f_op->read) |
4017eb91 | 3243 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3244 | else |
3245 | return -EINVAL; | |
f67676d1 JA |
3246 | } |
3247 | ||
889fca73 | 3248 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3249 | { |
3250 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3251 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3252 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3253 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3254 | ssize_t io_size, ret, ret2; |
45d189c6 | 3255 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3256 | |
2846c481 | 3257 | if (rw) { |
e8c2bc1f | 3258 | iter = &rw->iter; |
2846c481 PB |
3259 | iovec = NULL; |
3260 | } else { | |
3261 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3262 | if (ret < 0) | |
3263 | return ret; | |
3264 | } | |
632546c4 | 3265 | io_size = iov_iter_count(iter); |
fa15bafb | 3266 | req->result = io_size; |
2b188cc1 | 3267 | |
fd6c2e4c JA |
3268 | /* Ensure we clear previously set non-block flag */ |
3269 | if (!force_nonblock) | |
29de5f6a | 3270 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3271 | else |
3272 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3273 | ||
24c74678 | 3274 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3275 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3276 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3277 | return ret ?: -EAGAIN; |
6713e7a6 | 3278 | } |
9e645e11 | 3279 | |
632546c4 | 3280 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3281 | if (unlikely(ret)) { |
3282 | kfree(iovec); | |
3283 | return ret; | |
3284 | } | |
2b188cc1 | 3285 | |
227c0c96 | 3286 | ret = io_iter_do_read(req, iter); |
32960613 | 3287 | |
230d50d4 | 3288 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3289 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3290 | /* IOPOLL retry should happen for io-wq threads */ |
3291 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3292 | goto done; |
75c668cd PB |
3293 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3294 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3295 | goto done; |
84216315 | 3296 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3297 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3298 | ret = 0; |
230d50d4 JA |
3299 | } else if (ret == -EIOCBQUEUED) { |
3300 | goto out_free; | |
7335e3bf | 3301 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3302 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3303 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3304 | goto done; |
227c0c96 JA |
3305 | } |
3306 | ||
227c0c96 | 3307 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3308 | if (ret2) |
3309 | return ret2; | |
3310 | ||
fe1cdd55 | 3311 | iovec = NULL; |
e8c2bc1f | 3312 | rw = req->async_data; |
227c0c96 | 3313 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3314 | iter = &rw->iter; |
227c0c96 | 3315 | |
b23df91b PB |
3316 | do { |
3317 | io_size -= ret; | |
3318 | rw->bytes_done += ret; | |
3319 | /* if we can retry, do so with the callbacks armed */ | |
3320 | if (!io_rw_should_retry(req)) { | |
3321 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3322 | return -EAGAIN; | |
3323 | } | |
3324 | ||
3325 | /* | |
3326 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3327 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3328 | * desired page gets unlocked. We can also get a partial read | |
3329 | * here, and if we do, then just retry at the new offset. | |
3330 | */ | |
3331 | ret = io_iter_do_read(req, iter); | |
3332 | if (ret == -EIOCBQUEUED) | |
3333 | return 0; | |
227c0c96 | 3334 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3335 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3336 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3337 | done: |
889fca73 | 3338 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3339 | out_free: |
3340 | /* it's faster to check here then delegate to kfree */ | |
3341 | if (iovec) | |
3342 | kfree(iovec); | |
5ea5dd45 | 3343 | return 0; |
2b188cc1 JA |
3344 | } |
3345 | ||
73debe68 | 3346 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3347 | { |
3529d8c2 JA |
3348 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3349 | return -EBADF; | |
93642ef8 | 3350 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3351 | } |
3352 | ||
889fca73 | 3353 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3354 | { |
3355 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3356 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3357 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3358 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3359 | ssize_t ret, ret2, io_size; |
45d189c6 | 3360 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3361 | |
2846c481 | 3362 | if (rw) { |
e8c2bc1f | 3363 | iter = &rw->iter; |
2846c481 PB |
3364 | iovec = NULL; |
3365 | } else { | |
3366 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3367 | if (ret < 0) | |
3368 | return ret; | |
3369 | } | |
632546c4 | 3370 | io_size = iov_iter_count(iter); |
fa15bafb | 3371 | req->result = io_size; |
2b188cc1 | 3372 | |
fd6c2e4c JA |
3373 | /* Ensure we clear previously set non-block flag */ |
3374 | if (!force_nonblock) | |
a88fc400 PB |
3375 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3376 | else | |
3377 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3378 | |
24c74678 | 3379 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3380 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3381 | goto copy_iov; |
31b51510 | 3382 | |
10d59345 JA |
3383 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3384 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3385 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3386 | goto copy_iov; |
31b51510 | 3387 | |
632546c4 | 3388 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3389 | if (unlikely(ret)) |
3390 | goto out_free; | |
4ed734b0 | 3391 | |
fa15bafb PB |
3392 | /* |
3393 | * Open-code file_start_write here to grab freeze protection, | |
3394 | * which will be released by another thread in | |
3395 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3396 | * released so that it doesn't complain about the held lock when | |
3397 | * we return to userspace. | |
3398 | */ | |
3399 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3400 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3401 | __sb_writers_release(file_inode(req->file)->i_sb, |
3402 | SB_FREEZE_WRITE); | |
3403 | } | |
3404 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3405 | |
fa15bafb | 3406 | if (req->file->f_op->write_iter) |
ff6165b2 | 3407 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3408 | else if (req->file->f_op->write) |
4017eb91 | 3409 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3410 | else |
3411 | ret2 = -EINVAL; | |
4ed734b0 | 3412 | |
6ad7f233 PB |
3413 | if (req->flags & REQ_F_REISSUE) { |
3414 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3415 | ret2 = -EAGAIN; |
6ad7f233 | 3416 | } |
230d50d4 | 3417 | |
fa15bafb PB |
3418 | /* |
3419 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3420 | * retry them without IOCB_NOWAIT. | |
3421 | */ | |
3422 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3423 | ret2 = -EAGAIN; | |
75c668cd PB |
3424 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3425 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3426 | goto done; |
fa15bafb | 3427 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3428 | /* IOPOLL retry should happen for io-wq threads */ |
3429 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3430 | goto copy_iov; | |
355afaeb | 3431 | done: |
889fca73 | 3432 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3433 | } else { |
f67676d1 | 3434 | copy_iov: |
84216315 | 3435 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3436 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3437 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3438 | return ret ?: -EAGAIN; |
2b188cc1 | 3439 | } |
31b51510 | 3440 | out_free: |
f261c168 | 3441 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3442 | if (iovec) |
6f2cc166 | 3443 | kfree(iovec); |
2b188cc1 JA |
3444 | return ret; |
3445 | } | |
3446 | ||
80a261fd JA |
3447 | static int io_renameat_prep(struct io_kiocb *req, |
3448 | const struct io_uring_sqe *sqe) | |
3449 | { | |
3450 | struct io_rename *ren = &req->rename; | |
3451 | const char __user *oldf, *newf; | |
3452 | ||
3453 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3454 | return -EBADF; | |
3455 | ||
3456 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3457 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3458 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3459 | ren->new_dfd = READ_ONCE(sqe->len); | |
3460 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3461 | ||
3462 | ren->oldpath = getname(oldf); | |
3463 | if (IS_ERR(ren->oldpath)) | |
3464 | return PTR_ERR(ren->oldpath); | |
3465 | ||
3466 | ren->newpath = getname(newf); | |
3467 | if (IS_ERR(ren->newpath)) { | |
3468 | putname(ren->oldpath); | |
3469 | return PTR_ERR(ren->newpath); | |
3470 | } | |
3471 | ||
3472 | req->flags |= REQ_F_NEED_CLEANUP; | |
3473 | return 0; | |
3474 | } | |
3475 | ||
45d189c6 | 3476 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3477 | { |
3478 | struct io_rename *ren = &req->rename; | |
3479 | int ret; | |
3480 | ||
45d189c6 | 3481 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3482 | return -EAGAIN; |
3483 | ||
3484 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3485 | ren->newpath, ren->flags); | |
3486 | ||
3487 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3488 | if (ret < 0) | |
93d2bcd2 | 3489 | req_set_fail(req); |
80a261fd JA |
3490 | io_req_complete(req, ret); |
3491 | return 0; | |
3492 | } | |
3493 | ||
14a1143b JA |
3494 | static int io_unlinkat_prep(struct io_kiocb *req, |
3495 | const struct io_uring_sqe *sqe) | |
3496 | { | |
3497 | struct io_unlink *un = &req->unlink; | |
3498 | const char __user *fname; | |
3499 | ||
3500 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3501 | return -EBADF; | |
3502 | ||
3503 | un->dfd = READ_ONCE(sqe->fd); | |
3504 | ||
3505 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3506 | if (un->flags & ~AT_REMOVEDIR) | |
3507 | return -EINVAL; | |
3508 | ||
3509 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3510 | un->filename = getname(fname); | |
3511 | if (IS_ERR(un->filename)) | |
3512 | return PTR_ERR(un->filename); | |
3513 | ||
3514 | req->flags |= REQ_F_NEED_CLEANUP; | |
3515 | return 0; | |
3516 | } | |
3517 | ||
45d189c6 | 3518 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3519 | { |
3520 | struct io_unlink *un = &req->unlink; | |
3521 | int ret; | |
3522 | ||
45d189c6 | 3523 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3524 | return -EAGAIN; |
3525 | ||
3526 | if (un->flags & AT_REMOVEDIR) | |
3527 | ret = do_rmdir(un->dfd, un->filename); | |
3528 | else | |
3529 | ret = do_unlinkat(un->dfd, un->filename); | |
3530 | ||
3531 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3532 | if (ret < 0) | |
93d2bcd2 | 3533 | req_set_fail(req); |
14a1143b JA |
3534 | io_req_complete(req, ret); |
3535 | return 0; | |
3536 | } | |
3537 | ||
36f4fa68 JA |
3538 | static int io_shutdown_prep(struct io_kiocb *req, |
3539 | const struct io_uring_sqe *sqe) | |
3540 | { | |
3541 | #if defined(CONFIG_NET) | |
3542 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3543 | return -EINVAL; | |
3544 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3545 | sqe->buf_index) | |
3546 | return -EINVAL; | |
3547 | ||
3548 | req->shutdown.how = READ_ONCE(sqe->len); | |
3549 | return 0; | |
3550 | #else | |
3551 | return -EOPNOTSUPP; | |
3552 | #endif | |
3553 | } | |
3554 | ||
45d189c6 | 3555 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3556 | { |
3557 | #if defined(CONFIG_NET) | |
3558 | struct socket *sock; | |
3559 | int ret; | |
3560 | ||
45d189c6 | 3561 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3562 | return -EAGAIN; |
3563 | ||
48aba79b | 3564 | sock = sock_from_file(req->file); |
36f4fa68 | 3565 | if (unlikely(!sock)) |
48aba79b | 3566 | return -ENOTSOCK; |
36f4fa68 JA |
3567 | |
3568 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3569 | if (ret < 0) |
93d2bcd2 | 3570 | req_set_fail(req); |
36f4fa68 JA |
3571 | io_req_complete(req, ret); |
3572 | return 0; | |
3573 | #else | |
3574 | return -EOPNOTSUPP; | |
3575 | #endif | |
3576 | } | |
3577 | ||
f2a8d5c7 PB |
3578 | static int __io_splice_prep(struct io_kiocb *req, |
3579 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3580 | { |
3581 | struct io_splice* sp = &req->splice; | |
3582 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3583 | |
3232dd02 PB |
3584 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3585 | return -EINVAL; | |
7d67af2c PB |
3586 | |
3587 | sp->file_in = NULL; | |
7d67af2c PB |
3588 | sp->len = READ_ONCE(sqe->len); |
3589 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3590 | ||
3591 | if (unlikely(sp->flags & ~valid_flags)) | |
3592 | return -EINVAL; | |
3593 | ||
8371adf5 PB |
3594 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3595 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3596 | if (!sp->file_in) | |
3597 | return -EBADF; | |
7d67af2c | 3598 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3599 | return 0; |
3600 | } | |
3601 | ||
f2a8d5c7 PB |
3602 | static int io_tee_prep(struct io_kiocb *req, |
3603 | const struct io_uring_sqe *sqe) | |
3604 | { | |
3605 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3606 | return -EINVAL; | |
3607 | return __io_splice_prep(req, sqe); | |
3608 | } | |
3609 | ||
45d189c6 | 3610 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3611 | { |
3612 | struct io_splice *sp = &req->splice; | |
3613 | struct file *in = sp->file_in; | |
3614 | struct file *out = sp->file_out; | |
3615 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3616 | long ret = 0; | |
3617 | ||
45d189c6 | 3618 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3619 | return -EAGAIN; |
3620 | if (sp->len) | |
3621 | ret = do_tee(in, out, sp->len, flags); | |
3622 | ||
e1d767f0 PB |
3623 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3624 | io_put_file(in); | |
f2a8d5c7 PB |
3625 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3626 | ||
f2a8d5c7 | 3627 | if (ret != sp->len) |
93d2bcd2 | 3628 | req_set_fail(req); |
e1e16097 | 3629 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3630 | return 0; |
3631 | } | |
3632 | ||
3633 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3634 | { | |
3635 | struct io_splice* sp = &req->splice; | |
3636 | ||
3637 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3638 | sp->off_out = READ_ONCE(sqe->off); | |
3639 | return __io_splice_prep(req, sqe); | |
3640 | } | |
3641 | ||
45d189c6 | 3642 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3643 | { |
3644 | struct io_splice *sp = &req->splice; | |
3645 | struct file *in = sp->file_in; | |
3646 | struct file *out = sp->file_out; | |
3647 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3648 | loff_t *poff_in, *poff_out; | |
c9687426 | 3649 | long ret = 0; |
7d67af2c | 3650 | |
45d189c6 | 3651 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3652 | return -EAGAIN; |
7d67af2c PB |
3653 | |
3654 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3655 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3656 | |
948a7749 | 3657 | if (sp->len) |
c9687426 | 3658 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3659 | |
e1d767f0 PB |
3660 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3661 | io_put_file(in); | |
7d67af2c PB |
3662 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3663 | ||
7d67af2c | 3664 | if (ret != sp->len) |
93d2bcd2 | 3665 | req_set_fail(req); |
e1e16097 | 3666 | io_req_complete(req, ret); |
7d67af2c PB |
3667 | return 0; |
3668 | } | |
3669 | ||
2b188cc1 JA |
3670 | /* |
3671 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3672 | */ | |
889fca73 | 3673 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3674 | { |
3675 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3676 | |
def596e9 JA |
3677 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3678 | return -EINVAL; | |
3679 | ||
889fca73 | 3680 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3681 | return 0; |
3682 | } | |
3683 | ||
1155c76a | 3684 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3685 | { |
6b06314c | 3686 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3687 | |
09bb8394 JA |
3688 | if (!req->file) |
3689 | return -EBADF; | |
c992fe29 | 3690 | |
6b06314c | 3691 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3692 | return -EINVAL; |
edafccee | 3693 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3694 | return -EINVAL; |
3695 | ||
8ed8d3c3 JA |
3696 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3697 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3698 | return -EINVAL; | |
3699 | ||
3700 | req->sync.off = READ_ONCE(sqe->off); | |
3701 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3702 | return 0; |
3703 | } | |
3704 | ||
45d189c6 | 3705 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3706 | { |
8ed8d3c3 | 3707 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3708 | int ret; |
3709 | ||
ac45abc0 | 3710 | /* fsync always requires a blocking context */ |
45d189c6 | 3711 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3712 | return -EAGAIN; |
3713 | ||
9adbd45d | 3714 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3715 | end > 0 ? end : LLONG_MAX, |
3716 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3717 | if (ret < 0) | |
93d2bcd2 | 3718 | req_set_fail(req); |
e1e16097 | 3719 | io_req_complete(req, ret); |
c992fe29 CH |
3720 | return 0; |
3721 | } | |
3722 | ||
d63d1b5e JA |
3723 | static int io_fallocate_prep(struct io_kiocb *req, |
3724 | const struct io_uring_sqe *sqe) | |
3725 | { | |
3726 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3727 | return -EINVAL; | |
3232dd02 PB |
3728 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3729 | return -EINVAL; | |
d63d1b5e JA |
3730 | |
3731 | req->sync.off = READ_ONCE(sqe->off); | |
3732 | req->sync.len = READ_ONCE(sqe->addr); | |
3733 | req->sync.mode = READ_ONCE(sqe->len); | |
3734 | return 0; | |
3735 | } | |
3736 | ||
45d189c6 | 3737 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3738 | { |
ac45abc0 PB |
3739 | int ret; |
3740 | ||
d63d1b5e | 3741 | /* fallocate always requiring blocking context */ |
45d189c6 | 3742 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3743 | return -EAGAIN; |
ac45abc0 PB |
3744 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3745 | req->sync.len); | |
ac45abc0 | 3746 | if (ret < 0) |
93d2bcd2 | 3747 | req_set_fail(req); |
e1e16097 | 3748 | io_req_complete(req, ret); |
5d17b4a4 JA |
3749 | return 0; |
3750 | } | |
3751 | ||
ec65fea5 | 3752 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3753 | { |
f8748881 | 3754 | const char __user *fname; |
15b71abe | 3755 | int ret; |
b7bb4f7d | 3756 | |
ec65fea5 | 3757 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3758 | return -EINVAL; |
ec65fea5 | 3759 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3760 | return -EBADF; |
03b1230c | 3761 | |
ec65fea5 PB |
3762 | /* open.how should be already initialised */ |
3763 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3764 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3765 | |
25e72d10 PB |
3766 | req->open.dfd = READ_ONCE(sqe->fd); |
3767 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3768 | req->open.filename = getname(fname); |
15b71abe JA |
3769 | if (IS_ERR(req->open.filename)) { |
3770 | ret = PTR_ERR(req->open.filename); | |
3771 | req->open.filename = NULL; | |
3772 | return ret; | |
3773 | } | |
4022e7af | 3774 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3775 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3776 | return 0; |
03b1230c JA |
3777 | } |
3778 | ||
ec65fea5 PB |
3779 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3780 | { | |
3781 | u64 flags, mode; | |
3782 | ||
14587a46 | 3783 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3784 | return -EINVAL; |
ec65fea5 PB |
3785 | mode = READ_ONCE(sqe->len); |
3786 | flags = READ_ONCE(sqe->open_flags); | |
3787 | req->open.how = build_open_how(flags, mode); | |
3788 | return __io_openat_prep(req, sqe); | |
3789 | } | |
3790 | ||
cebdb986 | 3791 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3792 | { |
cebdb986 | 3793 | struct open_how __user *how; |
cebdb986 | 3794 | size_t len; |
0fa03c62 JA |
3795 | int ret; |
3796 | ||
14587a46 | 3797 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3798 | return -EINVAL; |
cebdb986 JA |
3799 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3800 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3801 | if (len < OPEN_HOW_SIZE_VER0) |
3802 | return -EINVAL; | |
3529d8c2 | 3803 | |
cebdb986 JA |
3804 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3805 | len); | |
3806 | if (ret) | |
3807 | return ret; | |
3529d8c2 | 3808 | |
ec65fea5 | 3809 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3810 | } |
3811 | ||
45d189c6 | 3812 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3813 | { |
3814 | struct open_flags op; | |
15b71abe | 3815 | struct file *file; |
3a81fd02 JA |
3816 | bool nonblock_set; |
3817 | bool resolve_nonblock; | |
15b71abe JA |
3818 | int ret; |
3819 | ||
cebdb986 | 3820 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3821 | if (ret) |
3822 | goto err; | |
3a81fd02 JA |
3823 | nonblock_set = op.open_flag & O_NONBLOCK; |
3824 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3825 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3826 | /* |
3827 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3828 | * it'll always -EAGAIN | |
3829 | */ | |
3830 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3831 | return -EAGAIN; | |
3832 | op.lookup_flags |= LOOKUP_CACHED; | |
3833 | op.open_flag |= O_NONBLOCK; | |
3834 | } | |
15b71abe | 3835 | |
4022e7af | 3836 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3837 | if (ret < 0) |
3838 | goto err; | |
3839 | ||
3840 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3841 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3842 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3843 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3844 | /* |
3a81fd02 JA |
3845 | * We could hang on to this 'fd', but seems like marginal |
3846 | * gain for something that is now known to be a slower path. | |
3847 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3848 | */ |
3a81fd02 JA |
3849 | put_unused_fd(ret); |
3850 | return -EAGAIN; | |
3851 | } | |
3852 | ||
15b71abe JA |
3853 | if (IS_ERR(file)) { |
3854 | put_unused_fd(ret); | |
3855 | ret = PTR_ERR(file); | |
3856 | } else { | |
45d189c6 | 3857 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3858 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3859 | fsnotify_open(file); |
3860 | fd_install(ret, file); | |
3861 | } | |
3862 | err: | |
3863 | putname(req->open.filename); | |
8fef80bf | 3864 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 3865 | if (ret < 0) |
93d2bcd2 | 3866 | req_set_fail(req); |
0bdf3398 | 3867 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3868 | return 0; |
3869 | } | |
3870 | ||
45d189c6 | 3871 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3872 | { |
e45cff58 | 3873 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3874 | } |
3875 | ||
067524e9 JA |
3876 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3877 | const struct io_uring_sqe *sqe) | |
3878 | { | |
3879 | struct io_provide_buf *p = &req->pbuf; | |
3880 | u64 tmp; | |
3881 | ||
3882 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3883 | return -EINVAL; | |
3884 | ||
3885 | tmp = READ_ONCE(sqe->fd); | |
3886 | if (!tmp || tmp > USHRT_MAX) | |
3887 | return -EINVAL; | |
3888 | ||
3889 | memset(p, 0, sizeof(*p)); | |
3890 | p->nbufs = tmp; | |
3891 | p->bgid = READ_ONCE(sqe->buf_group); | |
3892 | return 0; | |
3893 | } | |
3894 | ||
3895 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3896 | int bgid, unsigned nbufs) | |
3897 | { | |
3898 | unsigned i = 0; | |
3899 | ||
3900 | /* shouldn't happen */ | |
3901 | if (!nbufs) | |
3902 | return 0; | |
3903 | ||
3904 | /* the head kbuf is the list itself */ | |
3905 | while (!list_empty(&buf->list)) { | |
3906 | struct io_buffer *nxt; | |
3907 | ||
3908 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3909 | list_del(&nxt->list); | |
3910 | kfree(nxt); | |
3911 | if (++i == nbufs) | |
3912 | return i; | |
3913 | } | |
3914 | i++; | |
3915 | kfree(buf); | |
9e15c3a0 | 3916 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3917 | |
3918 | return i; | |
3919 | } | |
3920 | ||
889fca73 | 3921 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3922 | { |
3923 | struct io_provide_buf *p = &req->pbuf; | |
3924 | struct io_ring_ctx *ctx = req->ctx; | |
3925 | struct io_buffer *head; | |
3926 | int ret = 0; | |
45d189c6 | 3927 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3928 | |
3929 | io_ring_submit_lock(ctx, !force_nonblock); | |
3930 | ||
3931 | lockdep_assert_held(&ctx->uring_lock); | |
3932 | ||
3933 | ret = -ENOENT; | |
9e15c3a0 | 3934 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3935 | if (head) |
3936 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 3937 | if (ret < 0) |
93d2bcd2 | 3938 | req_set_fail(req); |
067524e9 | 3939 | |
9fb8cb49 PB |
3940 | /* complete before unlock, IOPOLL may need the lock */ |
3941 | __io_req_complete(req, issue_flags, ret, 0); | |
3942 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3943 | return 0; |
3944 | } | |
3945 | ||
ddf0322d JA |
3946 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3947 | const struct io_uring_sqe *sqe) | |
3948 | { | |
38134ada | 3949 | unsigned long size, tmp_check; |
ddf0322d JA |
3950 | struct io_provide_buf *p = &req->pbuf; |
3951 | u64 tmp; | |
3952 | ||
3953 | if (sqe->ioprio || sqe->rw_flags) | |
3954 | return -EINVAL; | |
3955 | ||
3956 | tmp = READ_ONCE(sqe->fd); | |
3957 | if (!tmp || tmp > USHRT_MAX) | |
3958 | return -E2BIG; | |
3959 | p->nbufs = tmp; | |
3960 | p->addr = READ_ONCE(sqe->addr); | |
3961 | p->len = READ_ONCE(sqe->len); | |
3962 | ||
38134ada PB |
3963 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
3964 | &size)) | |
3965 | return -EOVERFLOW; | |
3966 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
3967 | return -EOVERFLOW; | |
3968 | ||
d81269fe PB |
3969 | size = (unsigned long)p->len * p->nbufs; |
3970 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3971 | return -EFAULT; |
3972 | ||
3973 | p->bgid = READ_ONCE(sqe->buf_group); | |
3974 | tmp = READ_ONCE(sqe->off); | |
3975 | if (tmp > USHRT_MAX) | |
3976 | return -E2BIG; | |
3977 | p->bid = tmp; | |
3978 | return 0; | |
3979 | } | |
3980 | ||
3981 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3982 | { | |
3983 | struct io_buffer *buf; | |
3984 | u64 addr = pbuf->addr; | |
3985 | int i, bid = pbuf->bid; | |
3986 | ||
3987 | for (i = 0; i < pbuf->nbufs; i++) { | |
3988 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3989 | if (!buf) | |
3990 | break; | |
3991 | ||
3992 | buf->addr = addr; | |
d1f82808 | 3993 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
3994 | buf->bid = bid; |
3995 | addr += pbuf->len; | |
3996 | bid++; | |
3997 | if (!*head) { | |
3998 | INIT_LIST_HEAD(&buf->list); | |
3999 | *head = buf; | |
4000 | } else { | |
4001 | list_add_tail(&buf->list, &(*head)->list); | |
4002 | } | |
4003 | } | |
4004 | ||
4005 | return i ? i : -ENOMEM; | |
4006 | } | |
4007 | ||
889fca73 | 4008 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4009 | { |
4010 | struct io_provide_buf *p = &req->pbuf; | |
4011 | struct io_ring_ctx *ctx = req->ctx; | |
4012 | struct io_buffer *head, *list; | |
4013 | int ret = 0; | |
45d189c6 | 4014 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4015 | |
4016 | io_ring_submit_lock(ctx, !force_nonblock); | |
4017 | ||
4018 | lockdep_assert_held(&ctx->uring_lock); | |
4019 | ||
9e15c3a0 | 4020 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4021 | |
4022 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4023 | if (ret >= 0 && !list) { |
4024 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4025 | if (ret < 0) | |
067524e9 | 4026 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4027 | } |
ddf0322d | 4028 | if (ret < 0) |
93d2bcd2 | 4029 | req_set_fail(req); |
9fb8cb49 PB |
4030 | /* complete before unlock, IOPOLL may need the lock */ |
4031 | __io_req_complete(req, issue_flags, ret, 0); | |
4032 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4033 | return 0; |
cebdb986 JA |
4034 | } |
4035 | ||
3e4827b0 JA |
4036 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4037 | const struct io_uring_sqe *sqe) | |
4038 | { | |
4039 | #if defined(CONFIG_EPOLL) | |
4040 | if (sqe->ioprio || sqe->buf_index) | |
4041 | return -EINVAL; | |
2d74d042 | 4042 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4043 | return -EINVAL; |
3e4827b0 JA |
4044 | |
4045 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4046 | req->epoll.op = READ_ONCE(sqe->len); | |
4047 | req->epoll.fd = READ_ONCE(sqe->off); | |
4048 | ||
4049 | if (ep_op_has_event(req->epoll.op)) { | |
4050 | struct epoll_event __user *ev; | |
4051 | ||
4052 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4053 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4054 | return -EFAULT; | |
4055 | } | |
4056 | ||
4057 | return 0; | |
4058 | #else | |
4059 | return -EOPNOTSUPP; | |
4060 | #endif | |
4061 | } | |
4062 | ||
889fca73 | 4063 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4064 | { |
4065 | #if defined(CONFIG_EPOLL) | |
4066 | struct io_epoll *ie = &req->epoll; | |
4067 | int ret; | |
45d189c6 | 4068 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4069 | |
4070 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4071 | if (force_nonblock && ret == -EAGAIN) | |
4072 | return -EAGAIN; | |
4073 | ||
4074 | if (ret < 0) | |
93d2bcd2 | 4075 | req_set_fail(req); |
889fca73 | 4076 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4077 | return 0; |
4078 | #else | |
4079 | return -EOPNOTSUPP; | |
4080 | #endif | |
4081 | } | |
4082 | ||
c1ca757b JA |
4083 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4084 | { | |
4085 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4086 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4087 | return -EINVAL; | |
3232dd02 PB |
4088 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4089 | return -EINVAL; | |
c1ca757b JA |
4090 | |
4091 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4092 | req->madvise.len = READ_ONCE(sqe->len); | |
4093 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4094 | return 0; | |
4095 | #else | |
4096 | return -EOPNOTSUPP; | |
4097 | #endif | |
4098 | } | |
4099 | ||
45d189c6 | 4100 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4101 | { |
4102 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4103 | struct io_madvise *ma = &req->madvise; | |
4104 | int ret; | |
4105 | ||
45d189c6 | 4106 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4107 | return -EAGAIN; |
4108 | ||
0726b01e | 4109 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4110 | if (ret < 0) |
93d2bcd2 | 4111 | req_set_fail(req); |
e1e16097 | 4112 | io_req_complete(req, ret); |
c1ca757b JA |
4113 | return 0; |
4114 | #else | |
4115 | return -EOPNOTSUPP; | |
4116 | #endif | |
4117 | } | |
4118 | ||
4840e418 JA |
4119 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4120 | { | |
4121 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4122 | return -EINVAL; | |
3232dd02 PB |
4123 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4124 | return -EINVAL; | |
4840e418 JA |
4125 | |
4126 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4127 | req->fadvise.len = READ_ONCE(sqe->len); | |
4128 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4129 | return 0; | |
4130 | } | |
4131 | ||
45d189c6 | 4132 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4133 | { |
4134 | struct io_fadvise *fa = &req->fadvise; | |
4135 | int ret; | |
4136 | ||
45d189c6 | 4137 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4138 | switch (fa->advice) { |
4139 | case POSIX_FADV_NORMAL: | |
4140 | case POSIX_FADV_RANDOM: | |
4141 | case POSIX_FADV_SEQUENTIAL: | |
4142 | break; | |
4143 | default: | |
4144 | return -EAGAIN; | |
4145 | } | |
4146 | } | |
4840e418 JA |
4147 | |
4148 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4149 | if (ret < 0) | |
93d2bcd2 | 4150 | req_set_fail(req); |
0bdf3398 | 4151 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4152 | return 0; |
4153 | } | |
4154 | ||
eddc7ef5 JA |
4155 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4156 | { | |
2d74d042 | 4157 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4158 | return -EINVAL; |
eddc7ef5 JA |
4159 | if (sqe->ioprio || sqe->buf_index) |
4160 | return -EINVAL; | |
9c280f90 | 4161 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4162 | return -EBADF; |
eddc7ef5 | 4163 | |
1d9e1288 BM |
4164 | req->statx.dfd = READ_ONCE(sqe->fd); |
4165 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4166 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4167 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4168 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4169 | |
4170 | return 0; | |
4171 | } | |
4172 | ||
45d189c6 | 4173 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4174 | { |
1d9e1288 | 4175 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4176 | int ret; |
4177 | ||
59d70013 | 4178 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4179 | return -EAGAIN; |
4180 | ||
e62753e4 BM |
4181 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4182 | ctx->buffer); | |
eddc7ef5 | 4183 | |
eddc7ef5 | 4184 | if (ret < 0) |
93d2bcd2 | 4185 | req_set_fail(req); |
e1e16097 | 4186 | io_req_complete(req, ret); |
eddc7ef5 JA |
4187 | return 0; |
4188 | } | |
4189 | ||
b5dba59e JA |
4190 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4191 | { | |
14587a46 | 4192 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4193 | return -EINVAL; |
b5dba59e JA |
4194 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4195 | sqe->rw_flags || sqe->buf_index) | |
4196 | return -EINVAL; | |
9c280f90 | 4197 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4198 | return -EBADF; |
b5dba59e JA |
4199 | |
4200 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4201 | return 0; |
b5dba59e JA |
4202 | } |
4203 | ||
889fca73 | 4204 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4205 | { |
9eac1904 | 4206 | struct files_struct *files = current->files; |
3af73b28 | 4207 | struct io_close *close = &req->close; |
9eac1904 | 4208 | struct fdtable *fdt; |
a1fde923 PB |
4209 | struct file *file = NULL; |
4210 | int ret = -EBADF; | |
b5dba59e | 4211 | |
9eac1904 JA |
4212 | spin_lock(&files->file_lock); |
4213 | fdt = files_fdtable(files); | |
4214 | if (close->fd >= fdt->max_fds) { | |
4215 | spin_unlock(&files->file_lock); | |
4216 | goto err; | |
4217 | } | |
4218 | file = fdt->fd[close->fd]; | |
a1fde923 | 4219 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4220 | spin_unlock(&files->file_lock); |
4221 | file = NULL; | |
4222 | goto err; | |
3af73b28 | 4223 | } |
b5dba59e JA |
4224 | |
4225 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4226 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4227 | spin_unlock(&files->file_lock); |
0bf0eefd | 4228 | return -EAGAIN; |
a2100672 | 4229 | } |
b5dba59e | 4230 | |
9eac1904 JA |
4231 | ret = __close_fd_get_file(close->fd, &file); |
4232 | spin_unlock(&files->file_lock); | |
4233 | if (ret < 0) { | |
4234 | if (ret == -ENOENT) | |
4235 | ret = -EBADF; | |
4236 | goto err; | |
4237 | } | |
4238 | ||
3af73b28 | 4239 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4240 | ret = filp_close(file, current->files); |
4241 | err: | |
3af73b28 | 4242 | if (ret < 0) |
93d2bcd2 | 4243 | req_set_fail(req); |
9eac1904 JA |
4244 | if (file) |
4245 | fput(file); | |
889fca73 | 4246 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4247 | return 0; |
b5dba59e JA |
4248 | } |
4249 | ||
1155c76a | 4250 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4251 | { |
4252 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4253 | |
5d17b4a4 JA |
4254 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4255 | return -EINVAL; | |
4256 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4257 | return -EINVAL; | |
4258 | ||
8ed8d3c3 JA |
4259 | req->sync.off = READ_ONCE(sqe->off); |
4260 | req->sync.len = READ_ONCE(sqe->len); | |
4261 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4262 | return 0; |
4263 | } | |
4264 | ||
45d189c6 | 4265 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4266 | { |
8ed8d3c3 JA |
4267 | int ret; |
4268 | ||
ac45abc0 | 4269 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4270 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4271 | return -EAGAIN; |
4272 | ||
9adbd45d | 4273 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4274 | req->sync.flags); |
4275 | if (ret < 0) | |
93d2bcd2 | 4276 | req_set_fail(req); |
e1e16097 | 4277 | io_req_complete(req, ret); |
5d17b4a4 JA |
4278 | return 0; |
4279 | } | |
4280 | ||
469956e8 | 4281 | #if defined(CONFIG_NET) |
02d27d89 PB |
4282 | static int io_setup_async_msg(struct io_kiocb *req, |
4283 | struct io_async_msghdr *kmsg) | |
4284 | { | |
e8c2bc1f JA |
4285 | struct io_async_msghdr *async_msg = req->async_data; |
4286 | ||
4287 | if (async_msg) | |
02d27d89 | 4288 | return -EAGAIN; |
e8c2bc1f | 4289 | if (io_alloc_async_data(req)) { |
257e84a5 | 4290 | kfree(kmsg->free_iov); |
02d27d89 PB |
4291 | return -ENOMEM; |
4292 | } | |
e8c2bc1f | 4293 | async_msg = req->async_data; |
02d27d89 | 4294 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4295 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4296 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4297 | /* if were using fast_iov, set it to the new one */ |
4298 | if (!async_msg->free_iov) | |
4299 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4300 | ||
02d27d89 PB |
4301 | return -EAGAIN; |
4302 | } | |
4303 | ||
2ae523ed PB |
4304 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4305 | struct io_async_msghdr *iomsg) | |
4306 | { | |
2ae523ed | 4307 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4308 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4309 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4310 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4311 | } |
4312 | ||
93642ef8 PB |
4313 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4314 | { | |
4315 | int ret; | |
4316 | ||
93642ef8 PB |
4317 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4318 | if (!ret) | |
4319 | req->flags |= REQ_F_NEED_CLEANUP; | |
4320 | return ret; | |
4321 | } | |
4322 | ||
3529d8c2 | 4323 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4324 | { |
e47293fd | 4325 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4326 | |
d2b6f48b PB |
4327 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4328 | return -EINVAL; | |
4329 | ||
270a5940 | 4330 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4331 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4332 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4333 | if (sr->msg_flags & MSG_DONTWAIT) | |
4334 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4335 | |
d8768362 JA |
4336 | #ifdef CONFIG_COMPAT |
4337 | if (req->ctx->compat) | |
4338 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4339 | #endif | |
93642ef8 | 4340 | return 0; |
03b1230c JA |
4341 | } |
4342 | ||
889fca73 | 4343 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4344 | { |
6b754c8b | 4345 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4346 | struct socket *sock; |
7a7cacba | 4347 | unsigned flags; |
0031275d | 4348 | int min_ret = 0; |
0fa03c62 JA |
4349 | int ret; |
4350 | ||
dba4a925 | 4351 | sock = sock_from_file(req->file); |
7a7cacba | 4352 | if (unlikely(!sock)) |
dba4a925 | 4353 | return -ENOTSOCK; |
3529d8c2 | 4354 | |
257e84a5 PB |
4355 | kmsg = req->async_data; |
4356 | if (!kmsg) { | |
7a7cacba PB |
4357 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4358 | if (ret) | |
4359 | return ret; | |
4360 | kmsg = &iomsg; | |
0fa03c62 | 4361 | } |
0fa03c62 | 4362 | |
04411806 PB |
4363 | flags = req->sr_msg.msg_flags; |
4364 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4365 | flags |= MSG_DONTWAIT; |
0031275d SM |
4366 | if (flags & MSG_WAITALL) |
4367 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4368 | ||
7a7cacba | 4369 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4370 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4371 | return io_setup_async_msg(req, kmsg); |
4372 | if (ret == -ERESTARTSYS) | |
4373 | ret = -EINTR; | |
0fa03c62 | 4374 | |
257e84a5 PB |
4375 | /* fast path, check for non-NULL to avoid function call */ |
4376 | if (kmsg->free_iov) | |
4377 | kfree(kmsg->free_iov); | |
99bc4c38 | 4378 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4379 | if (ret < min_ret) |
93d2bcd2 | 4380 | req_set_fail(req); |
889fca73 | 4381 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4382 | return 0; |
03b1230c | 4383 | } |
aa1fa28f | 4384 | |
889fca73 | 4385 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4386 | { |
7a7cacba PB |
4387 | struct io_sr_msg *sr = &req->sr_msg; |
4388 | struct msghdr msg; | |
4389 | struct iovec iov; | |
fddaface | 4390 | struct socket *sock; |
7a7cacba | 4391 | unsigned flags; |
0031275d | 4392 | int min_ret = 0; |
fddaface JA |
4393 | int ret; |
4394 | ||
dba4a925 | 4395 | sock = sock_from_file(req->file); |
7a7cacba | 4396 | if (unlikely(!sock)) |
dba4a925 | 4397 | return -ENOTSOCK; |
fddaface | 4398 | |
7a7cacba PB |
4399 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4400 | if (unlikely(ret)) | |
14db8411 | 4401 | return ret; |
fddaface | 4402 | |
7a7cacba PB |
4403 | msg.msg_name = NULL; |
4404 | msg.msg_control = NULL; | |
4405 | msg.msg_controllen = 0; | |
4406 | msg.msg_namelen = 0; | |
fddaface | 4407 | |
04411806 PB |
4408 | flags = req->sr_msg.msg_flags; |
4409 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4410 | flags |= MSG_DONTWAIT; |
0031275d SM |
4411 | if (flags & MSG_WAITALL) |
4412 | min_ret = iov_iter_count(&msg.msg_iter); | |
4413 | ||
7a7cacba PB |
4414 | msg.msg_flags = flags; |
4415 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4416 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4417 | return -EAGAIN; |
4418 | if (ret == -ERESTARTSYS) | |
4419 | ret = -EINTR; | |
fddaface | 4420 | |
0031275d | 4421 | if (ret < min_ret) |
93d2bcd2 | 4422 | req_set_fail(req); |
889fca73 | 4423 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4424 | return 0; |
fddaface JA |
4425 | } |
4426 | ||
1400e697 PB |
4427 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4428 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4429 | { |
4430 | struct io_sr_msg *sr = &req->sr_msg; | |
4431 | struct iovec __user *uiov; | |
4432 | size_t iov_len; | |
4433 | int ret; | |
4434 | ||
1400e697 PB |
4435 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4436 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4437 | if (ret) |
4438 | return ret; | |
4439 | ||
4440 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4441 | if (iov_len > 1) | |
4442 | return -EINVAL; | |
5476dfed | 4443 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4444 | return -EFAULT; |
5476dfed | 4445 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4446 | iomsg->free_iov = NULL; |
52de1fe1 | 4447 | } else { |
257e84a5 | 4448 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4449 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4450 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4451 | false); |
52de1fe1 JA |
4452 | if (ret > 0) |
4453 | ret = 0; | |
4454 | } | |
4455 | ||
4456 | return ret; | |
4457 | } | |
4458 | ||
4459 | #ifdef CONFIG_COMPAT | |
4460 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4461 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4462 | { |
52de1fe1 JA |
4463 | struct io_sr_msg *sr = &req->sr_msg; |
4464 | struct compat_iovec __user *uiov; | |
4465 | compat_uptr_t ptr; | |
4466 | compat_size_t len; | |
4467 | int ret; | |
4468 | ||
4af3417a PB |
4469 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4470 | &ptr, &len); | |
52de1fe1 JA |
4471 | if (ret) |
4472 | return ret; | |
4473 | ||
4474 | uiov = compat_ptr(ptr); | |
4475 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4476 | compat_ssize_t clen; | |
4477 | ||
4478 | if (len > 1) | |
4479 | return -EINVAL; | |
4480 | if (!access_ok(uiov, sizeof(*uiov))) | |
4481 | return -EFAULT; | |
4482 | if (__get_user(clen, &uiov->iov_len)) | |
4483 | return -EFAULT; | |
4484 | if (clen < 0) | |
4485 | return -EINVAL; | |
2d280bc8 | 4486 | sr->len = clen; |
257e84a5 | 4487 | iomsg->free_iov = NULL; |
52de1fe1 | 4488 | } else { |
257e84a5 | 4489 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4490 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4491 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4492 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4493 | if (ret < 0) |
4494 | return ret; | |
4495 | } | |
4496 | ||
4497 | return 0; | |
4498 | } | |
4499 | #endif | |
4500 | ||
1400e697 PB |
4501 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4502 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4503 | { |
1400e697 | 4504 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4505 | |
4506 | #ifdef CONFIG_COMPAT | |
4507 | if (req->ctx->compat) | |
1400e697 | 4508 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4509 | #endif |
52de1fe1 | 4510 | |
1400e697 | 4511 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4512 | } |
4513 | ||
bcda7baa | 4514 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4515 | bool needs_lock) |
bcda7baa JA |
4516 | { |
4517 | struct io_sr_msg *sr = &req->sr_msg; | |
4518 | struct io_buffer *kbuf; | |
4519 | ||
bcda7baa JA |
4520 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4521 | if (IS_ERR(kbuf)) | |
4522 | return kbuf; | |
4523 | ||
4524 | sr->kbuf = kbuf; | |
4525 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4526 | return kbuf; |
fddaface JA |
4527 | } |
4528 | ||
7fbb1b54 PB |
4529 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4530 | { | |
4531 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4532 | } | |
4533 | ||
93642ef8 | 4534 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4535 | { |
99bc4c38 | 4536 | int ret; |
3529d8c2 | 4537 | |
93642ef8 PB |
4538 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4539 | if (!ret) | |
4540 | req->flags |= REQ_F_NEED_CLEANUP; | |
4541 | return ret; | |
4542 | } | |
4543 | ||
4544 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4545 | { | |
4546 | struct io_sr_msg *sr = &req->sr_msg; | |
4547 | ||
d2b6f48b PB |
4548 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4549 | return -EINVAL; | |
4550 | ||
270a5940 | 4551 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4552 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4553 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4554 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4555 | if (sr->msg_flags & MSG_DONTWAIT) | |
4556 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4557 | |
d8768362 JA |
4558 | #ifdef CONFIG_COMPAT |
4559 | if (req->ctx->compat) | |
4560 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4561 | #endif | |
93642ef8 | 4562 | return 0; |
aa1fa28f JA |
4563 | } |
4564 | ||
889fca73 | 4565 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4566 | { |
6b754c8b | 4567 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4568 | struct socket *sock; |
7fbb1b54 | 4569 | struct io_buffer *kbuf; |
7a7cacba | 4570 | unsigned flags; |
0031275d | 4571 | int min_ret = 0; |
52de1fe1 | 4572 | int ret, cflags = 0; |
45d189c6 | 4573 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4574 | |
dba4a925 | 4575 | sock = sock_from_file(req->file); |
7a7cacba | 4576 | if (unlikely(!sock)) |
dba4a925 | 4577 | return -ENOTSOCK; |
3529d8c2 | 4578 | |
257e84a5 PB |
4579 | kmsg = req->async_data; |
4580 | if (!kmsg) { | |
7a7cacba PB |
4581 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4582 | if (ret) | |
681fda8d | 4583 | return ret; |
7a7cacba PB |
4584 | kmsg = &iomsg; |
4585 | } | |
03b1230c | 4586 | |
bc02ef33 | 4587 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4588 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4589 | if (IS_ERR(kbuf)) |
52de1fe1 | 4590 | return PTR_ERR(kbuf); |
7a7cacba | 4591 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4592 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4593 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4594 | 1, req->sr_msg.len); |
4595 | } | |
52de1fe1 | 4596 | |
04411806 PB |
4597 | flags = req->sr_msg.msg_flags; |
4598 | if (force_nonblock) | |
7a7cacba | 4599 | flags |= MSG_DONTWAIT; |
0031275d SM |
4600 | if (flags & MSG_WAITALL) |
4601 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4602 | ||
7a7cacba PB |
4603 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4604 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4605 | if (force_nonblock && ret == -EAGAIN) |
4606 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4607 | if (ret == -ERESTARTSYS) |
4608 | ret = -EINTR; | |
03b1230c | 4609 | |
7fbb1b54 PB |
4610 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4611 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4612 | /* fast path, check for non-NULL to avoid function call */ |
4613 | if (kmsg->free_iov) | |
4614 | kfree(kmsg->free_iov); | |
99bc4c38 | 4615 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4616 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4617 | req_set_fail(req); |
889fca73 | 4618 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4619 | return 0; |
0fa03c62 | 4620 | } |
5d17b4a4 | 4621 | |
889fca73 | 4622 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4623 | { |
6b754c8b | 4624 | struct io_buffer *kbuf; |
7a7cacba PB |
4625 | struct io_sr_msg *sr = &req->sr_msg; |
4626 | struct msghdr msg; | |
4627 | void __user *buf = sr->buf; | |
fddaface | 4628 | struct socket *sock; |
7a7cacba PB |
4629 | struct iovec iov; |
4630 | unsigned flags; | |
0031275d | 4631 | int min_ret = 0; |
bcda7baa | 4632 | int ret, cflags = 0; |
45d189c6 | 4633 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4634 | |
dba4a925 | 4635 | sock = sock_from_file(req->file); |
7a7cacba | 4636 | if (unlikely(!sock)) |
dba4a925 | 4637 | return -ENOTSOCK; |
fddaface | 4638 | |
bc02ef33 | 4639 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4640 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4641 | if (IS_ERR(kbuf)) |
4642 | return PTR_ERR(kbuf); | |
7a7cacba | 4643 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4644 | } |
bcda7baa | 4645 | |
7a7cacba | 4646 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4647 | if (unlikely(ret)) |
4648 | goto out_free; | |
fddaface | 4649 | |
7a7cacba PB |
4650 | msg.msg_name = NULL; |
4651 | msg.msg_control = NULL; | |
4652 | msg.msg_controllen = 0; | |
4653 | msg.msg_namelen = 0; | |
4654 | msg.msg_iocb = NULL; | |
4655 | msg.msg_flags = 0; | |
fddaface | 4656 | |
04411806 PB |
4657 | flags = req->sr_msg.msg_flags; |
4658 | if (force_nonblock) | |
7a7cacba | 4659 | flags |= MSG_DONTWAIT; |
0031275d SM |
4660 | if (flags & MSG_WAITALL) |
4661 | min_ret = iov_iter_count(&msg.msg_iter); | |
4662 | ||
7a7cacba PB |
4663 | ret = sock_recvmsg(sock, &msg, flags); |
4664 | if (force_nonblock && ret == -EAGAIN) | |
4665 | return -EAGAIN; | |
4666 | if (ret == -ERESTARTSYS) | |
4667 | ret = -EINTR; | |
14c32eee | 4668 | out_free: |
7fbb1b54 PB |
4669 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4670 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4671 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4672 | req_set_fail(req); |
889fca73 | 4673 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4674 | return 0; |
fddaface JA |
4675 | } |
4676 | ||
3529d8c2 | 4677 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4678 | { |
8ed8d3c3 JA |
4679 | struct io_accept *accept = &req->accept; |
4680 | ||
14587a46 | 4681 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4682 | return -EINVAL; |
8042d6ce | 4683 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4684 | return -EINVAL; |
4685 | ||
d55e5f5b JA |
4686 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4687 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4688 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4689 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4690 | return 0; |
8ed8d3c3 | 4691 | } |
17f2fe35 | 4692 | |
889fca73 | 4693 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4694 | { |
4695 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4696 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4697 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4698 | int ret; |
4699 | ||
e697deed JX |
4700 | if (req->file->f_flags & O_NONBLOCK) |
4701 | req->flags |= REQ_F_NOWAIT; | |
4702 | ||
8ed8d3c3 | 4703 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4704 | accept->addr_len, accept->flags, |
4705 | accept->nofile); | |
8ed8d3c3 | 4706 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4707 | return -EAGAIN; |
ac45abc0 PB |
4708 | if (ret < 0) { |
4709 | if (ret == -ERESTARTSYS) | |
4710 | ret = -EINTR; | |
93d2bcd2 | 4711 | req_set_fail(req); |
ac45abc0 | 4712 | } |
889fca73 | 4713 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4714 | return 0; |
8ed8d3c3 JA |
4715 | } |
4716 | ||
93642ef8 PB |
4717 | static int io_connect_prep_async(struct io_kiocb *req) |
4718 | { | |
4719 | struct io_async_connect *io = req->async_data; | |
4720 | struct io_connect *conn = &req->connect; | |
4721 | ||
4722 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4723 | } | |
4724 | ||
3529d8c2 | 4725 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4726 | { |
3529d8c2 | 4727 | struct io_connect *conn = &req->connect; |
f499a021 | 4728 | |
14587a46 | 4729 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4730 | return -EINVAL; |
4731 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4732 | return -EINVAL; | |
4733 | ||
3529d8c2 JA |
4734 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4735 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4736 | return 0; |
f499a021 JA |
4737 | } |
4738 | ||
889fca73 | 4739 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4740 | { |
e8c2bc1f | 4741 | struct io_async_connect __io, *io; |
f8e85cf2 | 4742 | unsigned file_flags; |
3fbb51c1 | 4743 | int ret; |
45d189c6 | 4744 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4745 | |
e8c2bc1f JA |
4746 | if (req->async_data) { |
4747 | io = req->async_data; | |
f499a021 | 4748 | } else { |
3529d8c2 JA |
4749 | ret = move_addr_to_kernel(req->connect.addr, |
4750 | req->connect.addr_len, | |
e8c2bc1f | 4751 | &__io.address); |
f499a021 JA |
4752 | if (ret) |
4753 | goto out; | |
4754 | io = &__io; | |
4755 | } | |
4756 | ||
3fbb51c1 JA |
4757 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4758 | ||
e8c2bc1f | 4759 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4760 | req->connect.addr_len, file_flags); |
87f80d62 | 4761 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4762 | if (req->async_data) |
b7bb4f7d | 4763 | return -EAGAIN; |
e8c2bc1f | 4764 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4765 | ret = -ENOMEM; |
4766 | goto out; | |
4767 | } | |
e8c2bc1f | 4768 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4769 | return -EAGAIN; |
f499a021 | 4770 | } |
f8e85cf2 JA |
4771 | if (ret == -ERESTARTSYS) |
4772 | ret = -EINTR; | |
f499a021 | 4773 | out: |
4e88d6e7 | 4774 | if (ret < 0) |
93d2bcd2 | 4775 | req_set_fail(req); |
889fca73 | 4776 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4777 | return 0; |
469956e8 Y |
4778 | } |
4779 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4780 | #define IO_NETOP_FN(op) \ |
4781 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4782 | { \ | |
4783 | return -EOPNOTSUPP; \ | |
4784 | } | |
4785 | ||
4786 | #define IO_NETOP_PREP(op) \ | |
4787 | IO_NETOP_FN(op) \ | |
4788 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4789 | { \ | |
4790 | return -EOPNOTSUPP; \ | |
4791 | } \ | |
4792 | ||
4793 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4794 | IO_NETOP_PREP(op) \ | |
4795 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4796 | { \ | |
4797 | return -EOPNOTSUPP; \ | |
4798 | } | |
4799 | ||
4800 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4801 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4802 | IO_NETOP_PREP_ASYNC(connect); | |
4803 | IO_NETOP_PREP(accept); | |
4804 | IO_NETOP_FN(send); | |
4805 | IO_NETOP_FN(recv); | |
469956e8 | 4806 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4807 | |
d7718a9d JA |
4808 | struct io_poll_table { |
4809 | struct poll_table_struct pt; | |
4810 | struct io_kiocb *req; | |
4811 | int error; | |
4812 | }; | |
ce593a6c | 4813 | |
d7718a9d JA |
4814 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4815 | __poll_t mask, task_work_func_t func) | |
4816 | { | |
aa96bf8a | 4817 | int ret; |
d7718a9d JA |
4818 | |
4819 | /* for instances that support it check for an event match first: */ | |
4820 | if (mask && !(mask & poll->events)) | |
4821 | return 0; | |
4822 | ||
4823 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4824 | ||
4825 | list_del_init(&poll->wait.entry); | |
4826 | ||
d7718a9d | 4827 | req->result = mask; |
7cbf1722 | 4828 | req->task_work.func = func; |
6d816e08 | 4829 | |
d7718a9d | 4830 | /* |
e3aabf95 JA |
4831 | * If this fails, then the task is exiting. When a task exits, the |
4832 | * work gets canceled, so just cancel this request as well instead | |
4833 | * of executing it. We can't safely execute it anyway, as we may not | |
4834 | * have the needed state needed for it anyway. | |
d7718a9d | 4835 | */ |
355fb9e2 | 4836 | ret = io_req_task_work_add(req); |
aa96bf8a | 4837 | if (unlikely(ret)) { |
e3aabf95 | 4838 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4839 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4840 | } |
d7718a9d JA |
4841 | return 1; |
4842 | } | |
4843 | ||
74ce6ce4 JA |
4844 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4845 | __acquires(&req->ctx->completion_lock) | |
4846 | { | |
4847 | struct io_ring_ctx *ctx = req->ctx; | |
4848 | ||
4849 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4850 | struct poll_table_struct pt = { ._key = poll->events }; | |
4851 | ||
4852 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4853 | } | |
4854 | ||
4855 | spin_lock_irq(&ctx->completion_lock); | |
4856 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4857 | add_wait_queue(poll->head, &poll->wait); | |
4858 | return true; | |
4859 | } | |
4860 | ||
4861 | return false; | |
4862 | } | |
4863 | ||
d4e7cd36 | 4864 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4865 | { |
e8c2bc1f | 4866 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4867 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4868 | return req->async_data; |
d4e7cd36 JA |
4869 | return req->apoll->double_poll; |
4870 | } | |
4871 | ||
4872 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4873 | { | |
4874 | if (req->opcode == IORING_OP_POLL_ADD) | |
4875 | return &req->poll; | |
4876 | return &req->apoll->poll; | |
4877 | } | |
4878 | ||
4879 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4880 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4881 | { |
4882 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4883 | |
4884 | lockdep_assert_held(&req->ctx->completion_lock); | |
4885 | ||
4886 | if (poll && poll->head) { | |
4887 | struct wait_queue_head *head = poll->head; | |
4888 | ||
4889 | spin_lock(&head->lock); | |
4890 | list_del_init(&poll->wait.entry); | |
4891 | if (poll->wait.private) | |
de9b4cca | 4892 | req_ref_put(req); |
18bceab1 JA |
4893 | poll->head = NULL; |
4894 | spin_unlock(&head->lock); | |
4895 | } | |
4896 | } | |
4897 | ||
e27414be | 4898 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4899 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4900 | { |
4901 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4902 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4903 | int error; |
18bceab1 | 4904 | |
e27414be | 4905 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4906 | error = -ECANCELED; |
88e41cf9 | 4907 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4908 | } else { |
5082620f | 4909 | error = mangle_poll(mask); |
e27414be | 4910 | } |
b69de288 JA |
4911 | if (req->poll.events & EPOLLONESHOT) |
4912 | flags = 0; | |
d4d19c19 | 4913 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5082620f | 4914 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4915 | req->poll.done = true; |
4916 | flags = 0; | |
4917 | } | |
7b289c38 HX |
4918 | if (flags & IORING_CQE_F_MORE) |
4919 | ctx->cq_extra++; | |
18bceab1 | 4920 | |
18bceab1 | 4921 | io_commit_cqring(ctx); |
88e41cf9 | 4922 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4923 | } |
4924 | ||
dd221f46 | 4925 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4926 | { |
dd221f46 | 4927 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4928 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4929 | struct io_kiocb *nxt; |
18bceab1 JA |
4930 | |
4931 | if (io_poll_rewait(req, &req->poll)) { | |
4932 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4933 | } else { |
f40b964a | 4934 | bool done; |
18bceab1 | 4935 | |
e27414be | 4936 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4937 | if (done) { |
4938 | hash_del(&req->hash_node); | |
f40b964a | 4939 | } else { |
88e41cf9 JA |
4940 | req->result = 0; |
4941 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4942 | } | |
dd221f46 | 4943 | spin_unlock_irq(&ctx->completion_lock); |
dd221f46 | 4944 | io_cqring_ev_posted(ctx); |
18bceab1 | 4945 | |
88e41cf9 JA |
4946 | if (done) { |
4947 | nxt = io_put_req_find_next(req); | |
4948 | if (nxt) | |
4949 | __io_req_task_submit(nxt); | |
4950 | } | |
dd221f46 | 4951 | } |
18bceab1 JA |
4952 | } |
4953 | ||
4954 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4955 | int sync, void *key) | |
4956 | { | |
4957 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4958 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4959 | __poll_t mask = key_to_poll(key); |
4960 | ||
4961 | /* for instances that support it check for an event match first: */ | |
4962 | if (mask && !(mask & poll->events)) | |
4963 | return 0; | |
88e41cf9 JA |
4964 | if (!(poll->events & EPOLLONESHOT)) |
4965 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4966 | |
8706e04e JA |
4967 | list_del_init(&wait->entry); |
4968 | ||
807abcb0 | 4969 | if (poll && poll->head) { |
18bceab1 JA |
4970 | bool done; |
4971 | ||
807abcb0 JA |
4972 | spin_lock(&poll->head->lock); |
4973 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4974 | if (!done) |
807abcb0 | 4975 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4976 | /* make sure double remove sees this as being gone */ |
4977 | wait->private = NULL; | |
807abcb0 | 4978 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4979 | if (!done) { |
4980 | /* use wait func handler, so it matches the rq type */ | |
4981 | poll->wait.func(&poll->wait, mode, sync, key); | |
4982 | } | |
18bceab1 | 4983 | } |
de9b4cca | 4984 | req_ref_put(req); |
18bceab1 JA |
4985 | return 1; |
4986 | } | |
4987 | ||
4988 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4989 | wait_queue_func_t wake_func) | |
4990 | { | |
4991 | poll->head = NULL; | |
4992 | poll->done = false; | |
4993 | poll->canceled = false; | |
464dca61 JA |
4994 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
4995 | /* mask in events that we always want/need */ | |
4996 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
4997 | INIT_LIST_HEAD(&poll->wait.entry); |
4998 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4999 | } | |
5000 | ||
5001 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5002 | struct wait_queue_head *head, |
5003 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5004 | { |
5005 | struct io_kiocb *req = pt->req; | |
5006 | ||
5007 | /* | |
5008 | * If poll->head is already set, it's because the file being polled | |
5009 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5010 | * for write). Setup a separate io_poll_iocb if this happens. | |
5011 | */ | |
5012 | if (unlikely(poll->head)) { | |
58852d4d PB |
5013 | struct io_poll_iocb *poll_one = poll; |
5014 | ||
18bceab1 | 5015 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5016 | if (*poll_ptr) { |
18bceab1 JA |
5017 | pt->error = -EINVAL; |
5018 | return; | |
5019 | } | |
ea6a693d JA |
5020 | /* |
5021 | * Can't handle multishot for double wait for now, turn it | |
5022 | * into one-shot mode. | |
5023 | */ | |
7a274727 PB |
5024 | if (!(poll_one->events & EPOLLONESHOT)) |
5025 | poll_one->events |= EPOLLONESHOT; | |
1c3b3e65 | 5026 | /* double add on the same waitqueue head, ignore */ |
7a274727 | 5027 | if (poll_one->head == head) |
1c3b3e65 | 5028 | return; |
18bceab1 JA |
5029 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5030 | if (!poll) { | |
5031 | pt->error = -ENOMEM; | |
5032 | return; | |
5033 | } | |
58852d4d | 5034 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5035 | req_ref_get(req); |
18bceab1 | 5036 | poll->wait.private = req; |
807abcb0 | 5037 | *poll_ptr = poll; |
18bceab1 JA |
5038 | } |
5039 | ||
5040 | pt->error = 0; | |
5041 | poll->head = head; | |
a31eb4a2 JX |
5042 | |
5043 | if (poll->events & EPOLLEXCLUSIVE) | |
5044 | add_wait_queue_exclusive(head, &poll->wait); | |
5045 | else | |
5046 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5047 | } |
5048 | ||
5049 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5050 | struct poll_table_struct *p) | |
5051 | { | |
5052 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5053 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5054 | |
807abcb0 | 5055 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5056 | } |
5057 | ||
d7718a9d JA |
5058 | static void io_async_task_func(struct callback_head *cb) |
5059 | { | |
5060 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5061 | struct async_poll *apoll = req->apoll; | |
5062 | struct io_ring_ctx *ctx = req->ctx; | |
5063 | ||
5064 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5065 | ||
74ce6ce4 | 5066 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5067 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5068 | return; |
d7718a9d JA |
5069 | } |
5070 | ||
0ea13b44 | 5071 | hash_del(&req->hash_node); |
d4e7cd36 | 5072 | io_poll_remove_double(req); |
74ce6ce4 JA |
5073 | spin_unlock_irq(&ctx->completion_lock); |
5074 | ||
0be0b0e3 PB |
5075 | if (!READ_ONCE(apoll->poll.canceled)) |
5076 | __io_req_task_submit(req); | |
5077 | else | |
2593553a | 5078 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5079 | } |
5080 | ||
5081 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5082 | void *key) | |
5083 | { | |
5084 | struct io_kiocb *req = wait->private; | |
5085 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5086 | ||
5087 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5088 | key_to_poll(key)); | |
5089 | ||
5090 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5091 | } | |
5092 | ||
5093 | static void io_poll_req_insert(struct io_kiocb *req) | |
5094 | { | |
5095 | struct io_ring_ctx *ctx = req->ctx; | |
5096 | struct hlist_head *list; | |
5097 | ||
5098 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5099 | hlist_add_head(&req->hash_node, list); | |
5100 | } | |
5101 | ||
5102 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5103 | struct io_poll_iocb *poll, | |
5104 | struct io_poll_table *ipt, __poll_t mask, | |
5105 | wait_queue_func_t wake_func) | |
5106 | __acquires(&ctx->completion_lock) | |
5107 | { | |
5108 | struct io_ring_ctx *ctx = req->ctx; | |
5109 | bool cancel = false; | |
5110 | ||
4d52f338 | 5111 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5112 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5113 | poll->file = req->file; |
18bceab1 | 5114 | poll->wait.private = req; |
d7718a9d JA |
5115 | |
5116 | ipt->pt._key = mask; | |
5117 | ipt->req = req; | |
5118 | ipt->error = -EINVAL; | |
5119 | ||
d7718a9d JA |
5120 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5121 | ||
5122 | spin_lock_irq(&ctx->completion_lock); | |
5123 | if (likely(poll->head)) { | |
5124 | spin_lock(&poll->head->lock); | |
5125 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5126 | if (ipt->error) | |
5127 | cancel = true; | |
5128 | ipt->error = 0; | |
5129 | mask = 0; | |
5130 | } | |
88e41cf9 | 5131 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5132 | list_del_init(&poll->wait.entry); |
5133 | else if (cancel) | |
5134 | WRITE_ONCE(poll->canceled, true); | |
5135 | else if (!poll->done) /* actually waiting for an event */ | |
5136 | io_poll_req_insert(req); | |
5137 | spin_unlock(&poll->head->lock); | |
5138 | } | |
5139 | ||
5140 | return mask; | |
5141 | } | |
5142 | ||
5143 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5144 | { | |
5145 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5146 | struct io_ring_ctx *ctx = req->ctx; | |
5147 | struct async_poll *apoll; | |
5148 | struct io_poll_table ipt; | |
5149 | __poll_t mask, ret; | |
9dab14b8 | 5150 | int rw; |
d7718a9d JA |
5151 | |
5152 | if (!req->file || !file_can_poll(req->file)) | |
5153 | return false; | |
24c74678 | 5154 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5155 | return false; |
9dab14b8 JA |
5156 | if (def->pollin) |
5157 | rw = READ; | |
5158 | else if (def->pollout) | |
5159 | rw = WRITE; | |
5160 | else | |
5161 | return false; | |
5162 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5163 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5164 | return false; |
5165 | ||
5166 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5167 | if (unlikely(!apoll)) | |
5168 | return false; | |
807abcb0 | 5169 | apoll->double_poll = NULL; |
d7718a9d JA |
5170 | |
5171 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5172 | req->apoll = apoll; |
d7718a9d | 5173 | |
88e41cf9 | 5174 | mask = EPOLLONESHOT; |
d7718a9d | 5175 | if (def->pollin) |
8755d97a | 5176 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5177 | if (def->pollout) |
5178 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5179 | |
5180 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5181 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5182 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5183 | mask &= ~POLLIN; | |
5184 | ||
d7718a9d JA |
5185 | mask |= POLLERR | POLLPRI; |
5186 | ||
5187 | ipt.pt._qproc = io_async_queue_proc; | |
5188 | ||
5189 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5190 | io_async_wake); | |
a36da65c | 5191 | if (ret || ipt.error) { |
d4e7cd36 | 5192 | io_poll_remove_double(req); |
d7718a9d | 5193 | spin_unlock_irq(&ctx->completion_lock); |
d7718a9d JA |
5194 | return false; |
5195 | } | |
5196 | spin_unlock_irq(&ctx->completion_lock); | |
5197 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5198 | apoll->poll.events); | |
5199 | return true; | |
5200 | } | |
5201 | ||
5202 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5203 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5204 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5205 | { |
b41e9852 | 5206 | bool do_complete = false; |
221c5eb2 | 5207 | |
5082620f JA |
5208 | if (!poll->head) |
5209 | return false; | |
221c5eb2 | 5210 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5211 | if (do_cancel) |
5212 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5213 | if (!list_empty(&poll->wait.entry)) { |
5214 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5215 | do_complete = true; |
221c5eb2 JA |
5216 | } |
5217 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5218 | hash_del(&req->hash_node); |
d7718a9d JA |
5219 | return do_complete; |
5220 | } | |
5221 | ||
b2c3f7e1 | 5222 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5223 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5224 | { |
5225 | bool do_complete; | |
5226 | ||
d4e7cd36 | 5227 | io_poll_remove_double(req); |
e31001a3 | 5228 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5229 | |
e31001a3 | 5230 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
d7718a9d | 5231 | /* non-poll requests have submit ref still */ |
e31001a3 | 5232 | req_ref_put(req); |
b1f573bd | 5233 | } |
b2c3f7e1 JA |
5234 | return do_complete; |
5235 | } | |
5236 | ||
5237 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5238 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5239 | { |
5240 | bool do_complete; | |
b1f573bd | 5241 | |
b2c3f7e1 | 5242 | do_complete = io_poll_remove_waitqs(req); |
b41e9852 | 5243 | if (do_complete) { |
d4d19c19 | 5244 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5245 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5246 | req_set_fail(req); |
216578e5 | 5247 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5248 | } |
5249 | ||
5250 | return do_complete; | |
221c5eb2 JA |
5251 | } |
5252 | ||
76e1b642 JA |
5253 | /* |
5254 | * Returns true if we found and killed one or more poll requests | |
5255 | */ | |
6b81928d | 5256 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5257 | bool cancel_all) |
221c5eb2 | 5258 | { |
78076bb6 | 5259 | struct hlist_node *tmp; |
221c5eb2 | 5260 | struct io_kiocb *req; |
8e2e1faf | 5261 | int posted = 0, i; |
221c5eb2 JA |
5262 | |
5263 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5264 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5265 | struct hlist_head *list; | |
5266 | ||
5267 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5268 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5269 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5270 | posted += io_poll_remove_one(req); |
5271 | } | |
221c5eb2 JA |
5272 | } |
5273 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5274 | |
8e2e1faf JA |
5275 | if (posted) |
5276 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5277 | |
5278 | return posted != 0; | |
221c5eb2 JA |
5279 | } |
5280 | ||
9ba5fac8 PB |
5281 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5282 | bool poll_only) | |
e07785b0 | 5283 | __must_hold(&ctx->completion_lock) |
47f46768 | 5284 | { |
78076bb6 | 5285 | struct hlist_head *list; |
47f46768 JA |
5286 | struct io_kiocb *req; |
5287 | ||
78076bb6 JA |
5288 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5289 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5290 | if (sqe_addr != req->user_data) |
5291 | continue; | |
9ba5fac8 PB |
5292 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5293 | continue; | |
b2cb805f | 5294 | return req; |
47f46768 | 5295 | } |
b2cb805f JA |
5296 | return NULL; |
5297 | } | |
5298 | ||
9ba5fac8 PB |
5299 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5300 | bool poll_only) | |
e07785b0 | 5301 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5302 | { |
5303 | struct io_kiocb *req; | |
5304 | ||
9ba5fac8 | 5305 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5306 | if (!req) |
5307 | return -ENOENT; | |
5308 | if (io_poll_remove_one(req)) | |
5309 | return 0; | |
5310 | ||
5311 | return -EALREADY; | |
47f46768 JA |
5312 | } |
5313 | ||
9096af3e PB |
5314 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5315 | unsigned int flags) | |
5316 | { | |
5317 | u32 events; | |
47f46768 | 5318 | |
9096af3e PB |
5319 | events = READ_ONCE(sqe->poll32_events); |
5320 | #ifdef __BIG_ENDIAN | |
5321 | events = swahw32(events); | |
5322 | #endif | |
5323 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5324 | events |= EPOLLONESHOT; | |
5325 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5326 | } |
5327 | ||
c5de0036 | 5328 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5329 | const struct io_uring_sqe *sqe) |
0969e783 | 5330 | { |
c5de0036 PB |
5331 | struct io_poll_update *upd = &req->poll_update; |
5332 | u32 flags; | |
5333 | ||
0969e783 JA |
5334 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5335 | return -EINVAL; | |
c5de0036 PB |
5336 | if (sqe->ioprio || sqe->buf_index) |
5337 | return -EINVAL; | |
5338 | flags = READ_ONCE(sqe->len); | |
5339 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5340 | IORING_POLL_ADD_MULTI)) | |
5341 | return -EINVAL; | |
5342 | /* meaningless without update */ | |
5343 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5344 | return -EINVAL; |
5345 | ||
c5de0036 PB |
5346 | upd->old_user_data = READ_ONCE(sqe->addr); |
5347 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5348 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5349 | |
c5de0036 PB |
5350 | upd->new_user_data = READ_ONCE(sqe->off); |
5351 | if (!upd->update_user_data && upd->new_user_data) | |
5352 | return -EINVAL; | |
5353 | if (upd->update_events) | |
5354 | upd->events = io_poll_parse_events(sqe, flags); | |
5355 | else if (sqe->poll32_events) | |
5356 | return -EINVAL; | |
221c5eb2 | 5357 | |
221c5eb2 JA |
5358 | return 0; |
5359 | } | |
5360 | ||
221c5eb2 JA |
5361 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5362 | void *key) | |
5363 | { | |
c2f2eb7d JA |
5364 | struct io_kiocb *req = wait->private; |
5365 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5366 | |
d7718a9d | 5367 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5368 | } |
5369 | ||
221c5eb2 JA |
5370 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5371 | struct poll_table_struct *p) | |
5372 | { | |
5373 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5374 | ||
e8c2bc1f | 5375 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5376 | } |
5377 | ||
3529d8c2 | 5378 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5379 | { |
5380 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5381 | u32 flags; |
221c5eb2 JA |
5382 | |
5383 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5384 | return -EINVAL; | |
c5de0036 | 5385 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5386 | return -EINVAL; |
5387 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5388 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5389 | return -EINVAL; |
5390 | ||
c5de0036 | 5391 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5392 | return 0; |
5393 | } | |
5394 | ||
61e98203 | 5395 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5396 | { |
5397 | struct io_poll_iocb *poll = &req->poll; | |
5398 | struct io_ring_ctx *ctx = req->ctx; | |
5399 | struct io_poll_table ipt; | |
0969e783 | 5400 | __poll_t mask; |
0969e783 | 5401 | |
d7718a9d | 5402 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5403 | |
d7718a9d JA |
5404 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5405 | io_poll_wake); | |
221c5eb2 | 5406 | |
8c838788 | 5407 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5408 | ipt.error = 0; |
e27414be | 5409 | io_poll_complete(req, mask); |
221c5eb2 | 5410 | } |
221c5eb2 JA |
5411 | spin_unlock_irq(&ctx->completion_lock); |
5412 | ||
8c838788 JA |
5413 | if (mask) { |
5414 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5415 | if (poll->events & EPOLLONESHOT) |
5416 | io_put_req(req); | |
221c5eb2 | 5417 | } |
8c838788 | 5418 | return ipt.error; |
221c5eb2 JA |
5419 | } |
5420 | ||
c5de0036 | 5421 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5422 | { |
5423 | struct io_ring_ctx *ctx = req->ctx; | |
5424 | struct io_kiocb *preq; | |
cb3b200e | 5425 | bool completing; |
b69de288 JA |
5426 | int ret; |
5427 | ||
5428 | spin_lock_irq(&ctx->completion_lock); | |
9ba5fac8 | 5429 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5430 | if (!preq) { |
5431 | ret = -ENOENT; | |
5432 | goto err; | |
b69de288 | 5433 | } |
cb3b200e | 5434 | |
c5de0036 PB |
5435 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5436 | completing = true; | |
5437 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5438 | goto err; | |
5439 | } | |
5440 | ||
cb3b200e JA |
5441 | /* |
5442 | * Don't allow racy completion with singleshot, as we cannot safely | |
5443 | * update those. For multishot, if we're racing with completion, just | |
5444 | * let completion re-add it. | |
5445 | */ | |
5446 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5447 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5448 | ret = -EALREADY; | |
5449 | goto err; | |
b69de288 JA |
5450 | } |
5451 | /* we now have a detached poll request. reissue. */ | |
5452 | ret = 0; | |
5453 | err: | |
b69de288 | 5454 | if (ret < 0) { |
cb3b200e | 5455 | spin_unlock_irq(&ctx->completion_lock); |
93d2bcd2 | 5456 | req_set_fail(req); |
b69de288 JA |
5457 | io_req_complete(req, ret); |
5458 | return 0; | |
5459 | } | |
5460 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5461 | if (req->poll_update.update_events) { |
b69de288 | 5462 | preq->poll.events &= ~0xffff; |
9d805892 | 5463 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5464 | preq->poll.events |= IO_POLL_UNMASK; |
5465 | } | |
9d805892 PB |
5466 | if (req->poll_update.update_user_data) |
5467 | preq->user_data = req->poll_update.new_user_data; | |
cb3b200e JA |
5468 | spin_unlock_irq(&ctx->completion_lock); |
5469 | ||
b69de288 JA |
5470 | /* complete update request, we're done with it */ |
5471 | io_req_complete(req, ret); | |
5472 | ||
cb3b200e | 5473 | if (!completing) { |
c5de0036 | 5474 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5475 | if (ret < 0) { |
93d2bcd2 | 5476 | req_set_fail(preq); |
cb3b200e JA |
5477 | io_req_complete(preq, ret); |
5478 | } | |
b69de288 JA |
5479 | } |
5480 | return 0; | |
5481 | } | |
5482 | ||
5262f567 JA |
5483 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5484 | { | |
ad8a48ac JA |
5485 | struct io_timeout_data *data = container_of(timer, |
5486 | struct io_timeout_data, timer); | |
5487 | struct io_kiocb *req = data->req; | |
5488 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5489 | unsigned long flags; |
5490 | ||
5262f567 | 5491 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5492 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5493 | atomic_set(&req->ctx->cq_timeouts, |
5494 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5495 | ||
d4d19c19 | 5496 | io_cqring_fill_event(ctx, req->user_data, -ETIME, 0); |
5262f567 JA |
5497 | io_commit_cqring(ctx); |
5498 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5499 | ||
5500 | io_cqring_ev_posted(ctx); | |
93d2bcd2 | 5501 | req_set_fail(req); |
5262f567 JA |
5502 | io_put_req(req); |
5503 | return HRTIMER_NORESTART; | |
5504 | } | |
5505 | ||
fbd15848 PB |
5506 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5507 | __u64 user_data) | |
e07785b0 | 5508 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5509 | { |
fbd15848 | 5510 | struct io_timeout_data *io; |
47f46768 | 5511 | struct io_kiocb *req; |
fd9c7bc5 | 5512 | bool found = false; |
f254ac04 | 5513 | |
135fcde8 | 5514 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5515 | found = user_data == req->user_data; |
5516 | if (found) | |
47f46768 | 5517 | break; |
47f46768 | 5518 | } |
fd9c7bc5 PB |
5519 | if (!found) |
5520 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5521 | |
5522 | io = req->async_data; | |
fd9c7bc5 | 5523 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5524 | return ERR_PTR(-EALREADY); |
a71976f3 | 5525 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5526 | return req; |
5527 | } | |
47f46768 | 5528 | |
fbd15848 | 5529 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5530 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5531 | { |
5532 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5533 | ||
5534 | if (IS_ERR(req)) | |
5535 | return PTR_ERR(req); | |
f254ac04 | 5536 | |
93d2bcd2 | 5537 | req_set_fail(req); |
d4d19c19 | 5538 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
216578e5 | 5539 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5540 | return 0; |
5541 | } | |
5542 | ||
9c8e11b3 PB |
5543 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5544 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5545 | __must_hold(&ctx->completion_lock) |
47f46768 | 5546 | { |
9c8e11b3 PB |
5547 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5548 | struct io_timeout_data *data; | |
47f46768 | 5549 | |
9c8e11b3 PB |
5550 | if (IS_ERR(req)) |
5551 | return PTR_ERR(req); | |
47f46768 | 5552 | |
9c8e11b3 PB |
5553 | req->timeout.off = 0; /* noseq */ |
5554 | data = req->async_data; | |
5555 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5556 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5557 | data->timer.function = io_timeout_fn; | |
5558 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5559 | return 0; | |
47f46768 JA |
5560 | } |
5561 | ||
3529d8c2 JA |
5562 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5563 | const struct io_uring_sqe *sqe) | |
b29472ee | 5564 | { |
9c8e11b3 PB |
5565 | struct io_timeout_rem *tr = &req->timeout_rem; |
5566 | ||
b29472ee JA |
5567 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5568 | return -EINVAL; | |
61710e43 DA |
5569 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5570 | return -EINVAL; | |
9c8e11b3 | 5571 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5572 | return -EINVAL; |
5573 | ||
9c8e11b3 PB |
5574 | tr->addr = READ_ONCE(sqe->addr); |
5575 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5576 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5577 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5578 | return -EINVAL; | |
5579 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5580 | return -EFAULT; | |
5581 | } else if (tr->flags) { | |
5582 | /* timeout removal doesn't support flags */ | |
b29472ee | 5583 | return -EINVAL; |
9c8e11b3 | 5584 | } |
b29472ee | 5585 | |
b29472ee JA |
5586 | return 0; |
5587 | } | |
5588 | ||
8662daec PB |
5589 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5590 | { | |
5591 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5592 | : HRTIMER_MODE_REL; | |
5593 | } | |
5594 | ||
11365043 JA |
5595 | /* |
5596 | * Remove or update an existing timeout command | |
5597 | */ | |
61e98203 | 5598 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5599 | { |
9c8e11b3 | 5600 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5601 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5602 | int ret; |
11365043 | 5603 | |
11365043 | 5604 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5605 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5606 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5607 | else |
5608 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5609 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5610 | |
d4d19c19 | 5611 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
11365043 JA |
5612 | io_commit_cqring(ctx); |
5613 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5614 | io_cqring_ev_posted(ctx); |
4e88d6e7 | 5615 | if (ret < 0) |
93d2bcd2 | 5616 | req_set_fail(req); |
ec9c02ad | 5617 | io_put_req(req); |
11365043 | 5618 | return 0; |
5262f567 JA |
5619 | } |
5620 | ||
3529d8c2 | 5621 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5622 | bool is_timeout_link) |
5262f567 | 5623 | { |
ad8a48ac | 5624 | struct io_timeout_data *data; |
a41525ab | 5625 | unsigned flags; |
56080b02 | 5626 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5627 | |
ad8a48ac | 5628 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5629 | return -EINVAL; |
ad8a48ac | 5630 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5631 | return -EINVAL; |
56080b02 | 5632 | if (off && is_timeout_link) |
2d28390a | 5633 | return -EINVAL; |
a41525ab JA |
5634 | flags = READ_ONCE(sqe->timeout_flags); |
5635 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5636 | return -EINVAL; |
bdf20073 | 5637 | |
bfe68a22 | 5638 | req->timeout.off = off; |
26a61679 | 5639 | |
e8c2bc1f | 5640 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5641 | return -ENOMEM; |
5642 | ||
e8c2bc1f | 5643 | data = req->async_data; |
ad8a48ac | 5644 | data->req = req; |
ad8a48ac JA |
5645 | |
5646 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5647 | return -EFAULT; |
5648 | ||
8662daec | 5649 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5650 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5651 | if (is_timeout_link) |
5652 | io_req_track_inflight(req); | |
ad8a48ac JA |
5653 | return 0; |
5654 | } | |
5655 | ||
61e98203 | 5656 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5657 | { |
ad8a48ac | 5658 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5659 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5660 | struct list_head *entry; |
bfe68a22 | 5661 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5662 | |
733f5c95 | 5663 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5664 | |
5262f567 JA |
5665 | /* |
5666 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5667 | * timeout event to be satisfied. If it isn't set, then this is |
5668 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5669 | */ |
8eb7e2d0 | 5670 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5671 | entry = ctx->timeout_list.prev; |
5672 | goto add; | |
5673 | } | |
5262f567 | 5674 | |
bfe68a22 PB |
5675 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5676 | req->timeout.target_seq = tail + off; | |
5262f567 | 5677 | |
f010505b MDG |
5678 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5679 | * This is safe because ->completion_lock is held, and submissions | |
5680 | * and completions are never mixed in the same ->completion_lock section. | |
5681 | */ | |
5682 | ctx->cq_last_tm_flush = tail; | |
5683 | ||
5262f567 JA |
5684 | /* |
5685 | * Insertion sort, ensuring the first entry in the list is always | |
5686 | * the one we need first. | |
5687 | */ | |
5262f567 | 5688 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5689 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5690 | timeout.list); | |
5262f567 | 5691 | |
8eb7e2d0 | 5692 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5693 | continue; |
bfe68a22 PB |
5694 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5695 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5696 | break; |
5697 | } | |
93bd25bb | 5698 | add: |
135fcde8 | 5699 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5700 | data->timer.function = io_timeout_fn; |
5701 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5702 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5703 | return 0; |
5704 | } | |
5262f567 | 5705 | |
f458dd84 PB |
5706 | struct io_cancel_data { |
5707 | struct io_ring_ctx *ctx; | |
5708 | u64 user_data; | |
5709 | }; | |
5710 | ||
62755e35 JA |
5711 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5712 | { | |
5713 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5714 | struct io_cancel_data *cd = data; |
62755e35 | 5715 | |
f458dd84 | 5716 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5717 | } |
5718 | ||
f458dd84 PB |
5719 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5720 | struct io_ring_ctx *ctx) | |
62755e35 | 5721 | { |
f458dd84 | 5722 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5723 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5724 | int ret = 0; |
5725 | ||
f458dd84 | 5726 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5727 | return -ENOENT; |
5728 | ||
f458dd84 | 5729 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5730 | switch (cancel_ret) { |
5731 | case IO_WQ_CANCEL_OK: | |
5732 | ret = 0; | |
5733 | break; | |
5734 | case IO_WQ_CANCEL_RUNNING: | |
5735 | ret = -EALREADY; | |
5736 | break; | |
5737 | case IO_WQ_CANCEL_NOTFOUND: | |
5738 | ret = -ENOENT; | |
5739 | break; | |
5740 | } | |
5741 | ||
e977d6d3 JA |
5742 | return ret; |
5743 | } | |
5744 | ||
47f46768 JA |
5745 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5746 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5747 | int success_ret) |
47f46768 JA |
5748 | { |
5749 | unsigned long flags; | |
5750 | int ret; | |
5751 | ||
f458dd84 | 5752 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5753 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5754 | if (ret != -ENOENT) |
5755 | goto done; | |
47f46768 JA |
5756 | ret = io_timeout_cancel(ctx, sqe_addr); |
5757 | if (ret != -ENOENT) | |
5758 | goto done; | |
9ba5fac8 | 5759 | ret = io_poll_cancel(ctx, sqe_addr, false); |
47f46768 | 5760 | done: |
b0dd8a41 JA |
5761 | if (!ret) |
5762 | ret = success_ret; | |
d4d19c19 | 5763 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
47f46768 JA |
5764 | io_commit_cqring(ctx); |
5765 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5766 | io_cqring_ev_posted(ctx); | |
5767 | ||
4e88d6e7 | 5768 | if (ret < 0) |
93d2bcd2 | 5769 | req_set_fail(req); |
47f46768 JA |
5770 | } |
5771 | ||
3529d8c2 JA |
5772 | static int io_async_cancel_prep(struct io_kiocb *req, |
5773 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5774 | { |
fbf23849 | 5775 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5776 | return -EINVAL; |
61710e43 DA |
5777 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5778 | return -EINVAL; | |
5779 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5780 | return -EINVAL; |
5781 | ||
fbf23849 JA |
5782 | req->cancel.addr = READ_ONCE(sqe->addr); |
5783 | return 0; | |
5784 | } | |
5785 | ||
61e98203 | 5786 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5787 | { |
5788 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5789 | u64 sqe_addr = req->cancel.addr; |
5790 | struct io_tctx_node *node; | |
5791 | int ret; | |
5792 | ||
5793 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5794 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5795 | spin_lock_irq(&ctx->completion_lock); | |
5796 | if (ret != -ENOENT) | |
5797 | goto done; | |
5798 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5799 | if (ret != -ENOENT) | |
5800 | goto done; | |
9ba5fac8 | 5801 | ret = io_poll_cancel(ctx, sqe_addr, false); |
58f99373 PB |
5802 | if (ret != -ENOENT) |
5803 | goto done; | |
5804 | spin_unlock_irq(&ctx->completion_lock); | |
5805 | ||
5806 | /* slow path, try all io-wq's */ | |
5807 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5808 | ret = -ENOENT; | |
5809 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5810 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5811 | |
58f99373 PB |
5812 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5813 | if (ret != -ENOENT) | |
5814 | break; | |
5815 | } | |
5816 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5817 | ||
5818 | spin_lock_irq(&ctx->completion_lock); | |
5819 | done: | |
d4d19c19 | 5820 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
58f99373 PB |
5821 | io_commit_cqring(ctx); |
5822 | spin_unlock_irq(&ctx->completion_lock); | |
5823 | io_cqring_ev_posted(ctx); | |
5824 | ||
5825 | if (ret < 0) | |
93d2bcd2 | 5826 | req_set_fail(req); |
58f99373 | 5827 | io_put_req(req); |
5262f567 JA |
5828 | return 0; |
5829 | } | |
5830 | ||
269bbe5f | 5831 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5832 | const struct io_uring_sqe *sqe) |
5833 | { | |
61710e43 DA |
5834 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5835 | return -EINVAL; | |
5836 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5837 | return -EINVAL; |
5838 | ||
269bbe5f BM |
5839 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5840 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5841 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5842 | return -EINVAL; |
269bbe5f | 5843 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5844 | return 0; |
5845 | } | |
5846 | ||
889fca73 | 5847 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5848 | { |
5849 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 5850 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 5851 | int ret; |
fbf23849 | 5852 | |
45d189c6 | 5853 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5854 | return -EAGAIN; |
05f3fb3c | 5855 | |
269bbe5f BM |
5856 | up.offset = req->rsrc_update.offset; |
5857 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
5858 | up.nr = 0; |
5859 | up.tags = 0; | |
615cee49 | 5860 | up.resv = 0; |
05f3fb3c JA |
5861 | |
5862 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 5863 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 5864 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5865 | mutex_unlock(&ctx->uring_lock); |
5866 | ||
5867 | if (ret < 0) | |
93d2bcd2 | 5868 | req_set_fail(req); |
889fca73 | 5869 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5870 | return 0; |
5871 | } | |
5872 | ||
bfe76559 | 5873 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5874 | { |
d625c6ee | 5875 | switch (req->opcode) { |
e781573e | 5876 | case IORING_OP_NOP: |
bfe76559 | 5877 | return 0; |
f67676d1 JA |
5878 | case IORING_OP_READV: |
5879 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5880 | case IORING_OP_READ: |
bfe76559 | 5881 | return io_read_prep(req, sqe); |
f67676d1 JA |
5882 | case IORING_OP_WRITEV: |
5883 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5884 | case IORING_OP_WRITE: |
bfe76559 | 5885 | return io_write_prep(req, sqe); |
0969e783 | 5886 | case IORING_OP_POLL_ADD: |
bfe76559 | 5887 | return io_poll_add_prep(req, sqe); |
0969e783 | 5888 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5889 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5890 | case IORING_OP_FSYNC: |
1155c76a | 5891 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5892 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5893 | return io_sfr_prep(req, sqe); |
03b1230c | 5894 | case IORING_OP_SENDMSG: |
fddaface | 5895 | case IORING_OP_SEND: |
bfe76559 | 5896 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5897 | case IORING_OP_RECVMSG: |
fddaface | 5898 | case IORING_OP_RECV: |
bfe76559 | 5899 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5900 | case IORING_OP_CONNECT: |
bfe76559 | 5901 | return io_connect_prep(req, sqe); |
2d28390a | 5902 | case IORING_OP_TIMEOUT: |
bfe76559 | 5903 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5904 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5905 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5906 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5907 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5908 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5909 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5910 | case IORING_OP_ACCEPT: |
bfe76559 | 5911 | return io_accept_prep(req, sqe); |
d63d1b5e | 5912 | case IORING_OP_FALLOCATE: |
bfe76559 | 5913 | return io_fallocate_prep(req, sqe); |
15b71abe | 5914 | case IORING_OP_OPENAT: |
bfe76559 | 5915 | return io_openat_prep(req, sqe); |
b5dba59e | 5916 | case IORING_OP_CLOSE: |
bfe76559 | 5917 | return io_close_prep(req, sqe); |
05f3fb3c | 5918 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5919 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5920 | case IORING_OP_STATX: |
bfe76559 | 5921 | return io_statx_prep(req, sqe); |
4840e418 | 5922 | case IORING_OP_FADVISE: |
bfe76559 | 5923 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5924 | case IORING_OP_MADVISE: |
bfe76559 | 5925 | return io_madvise_prep(req, sqe); |
cebdb986 | 5926 | case IORING_OP_OPENAT2: |
bfe76559 | 5927 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5928 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5929 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5930 | case IORING_OP_SPLICE: |
bfe76559 | 5931 | return io_splice_prep(req, sqe); |
ddf0322d | 5932 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5933 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5934 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5935 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5936 | case IORING_OP_TEE: |
bfe76559 | 5937 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5938 | case IORING_OP_SHUTDOWN: |
5939 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5940 | case IORING_OP_RENAMEAT: |
5941 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5942 | case IORING_OP_UNLINKAT: |
5943 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5944 | } |
5945 | ||
bfe76559 PB |
5946 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5947 | req->opcode); | |
bd54b6fe | 5948 | return -EINVAL; |
bfe76559 PB |
5949 | } |
5950 | ||
93642ef8 | 5951 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5952 | { |
b7e298d2 PB |
5953 | if (!io_op_defs[req->opcode].needs_async_setup) |
5954 | return 0; | |
5955 | if (WARN_ON_ONCE(req->async_data)) | |
5956 | return -EFAULT; | |
5957 | if (io_alloc_async_data(req)) | |
5958 | return -EAGAIN; | |
5959 | ||
93642ef8 PB |
5960 | switch (req->opcode) { |
5961 | case IORING_OP_READV: | |
93642ef8 PB |
5962 | return io_rw_prep_async(req, READ); |
5963 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5964 | return io_rw_prep_async(req, WRITE); |
5965 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5966 | return io_sendmsg_prep_async(req); |
5967 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5968 | return io_recvmsg_prep_async(req); |
5969 | case IORING_OP_CONNECT: | |
5970 | return io_connect_prep_async(req); | |
5971 | } | |
b7e298d2 PB |
5972 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5973 | req->opcode); | |
5974 | return -EFAULT; | |
f67676d1 JA |
5975 | } |
5976 | ||
9cf7c104 PB |
5977 | static u32 io_get_sequence(struct io_kiocb *req) |
5978 | { | |
5979 | struct io_kiocb *pos; | |
5980 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5981 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5982 | |
f2f87370 PB |
5983 | io_for_each_link(pos, req) |
5984 | nr_reqs++; | |
9cf7c104 PB |
5985 | |
5986 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5987 | return total_submitted - nr_reqs; | |
5988 | } | |
5989 | ||
be7053b7 | 5990 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5991 | { |
a197f664 | 5992 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5993 | struct io_defer_entry *de; |
f67676d1 | 5994 | int ret; |
9cf7c104 | 5995 | u32 seq; |
de0617e4 | 5996 | |
9d858b21 | 5997 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5998 | if (likely(list_empty_careful(&ctx->defer_list) && |
5999 | !(req->flags & REQ_F_IO_DRAIN))) | |
6000 | return 0; | |
6001 | ||
6002 | seq = io_get_sequence(req); | |
6003 | /* Still a chance to pass the sequence check */ | |
6004 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6005 | return 0; |
6006 | ||
b7e298d2 | 6007 | ret = io_req_prep_async(req); |
be7053b7 PB |
6008 | if (ret) |
6009 | return ret; | |
cbdcb435 | 6010 | io_prep_async_link(req); |
27dc8338 PB |
6011 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6012 | if (!de) | |
6013 | return -ENOMEM; | |
2d28390a | 6014 | |
de0617e4 | 6015 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6016 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6017 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6018 | kfree(de); |
ae34817b PB |
6019 | io_queue_async_work(req); |
6020 | return -EIOCBQUEUED; | |
de0617e4 JA |
6021 | } |
6022 | ||
915967f6 | 6023 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6024 | de->req = req; |
9cf7c104 | 6025 | de->seq = seq; |
27dc8338 | 6026 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6027 | spin_unlock_irq(&ctx->completion_lock); |
6028 | return -EIOCBQUEUED; | |
6029 | } | |
6030 | ||
68fb8979 | 6031 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6032 | { |
0e1b6fe3 PB |
6033 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6034 | switch (req->opcode) { | |
6035 | case IORING_OP_READV: | |
6036 | case IORING_OP_READ_FIXED: | |
6037 | case IORING_OP_READ: | |
bcda7baa | 6038 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6039 | break; |
6040 | case IORING_OP_RECVMSG: | |
6041 | case IORING_OP_RECV: | |
bcda7baa | 6042 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6043 | break; |
6044 | } | |
6045 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6046 | } |
6047 | ||
0e1b6fe3 PB |
6048 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6049 | switch (req->opcode) { | |
6050 | case IORING_OP_READV: | |
6051 | case IORING_OP_READ_FIXED: | |
6052 | case IORING_OP_READ: | |
6053 | case IORING_OP_WRITEV: | |
6054 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6055 | case IORING_OP_WRITE: { |
6056 | struct io_async_rw *io = req->async_data; | |
6057 | if (io->free_iovec) | |
6058 | kfree(io->free_iovec); | |
0e1b6fe3 | 6059 | break; |
e8c2bc1f | 6060 | } |
0e1b6fe3 | 6061 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6062 | case IORING_OP_SENDMSG: { |
6063 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6064 | |
6065 | kfree(io->free_iov); | |
0e1b6fe3 | 6066 | break; |
e8c2bc1f | 6067 | } |
0e1b6fe3 PB |
6068 | case IORING_OP_SPLICE: |
6069 | case IORING_OP_TEE: | |
e1d767f0 PB |
6070 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6071 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6072 | break; |
f3cd4850 JA |
6073 | case IORING_OP_OPENAT: |
6074 | case IORING_OP_OPENAT2: | |
6075 | if (req->open.filename) | |
6076 | putname(req->open.filename); | |
6077 | break; | |
80a261fd JA |
6078 | case IORING_OP_RENAMEAT: |
6079 | putname(req->rename.oldpath); | |
6080 | putname(req->rename.newpath); | |
6081 | break; | |
14a1143b JA |
6082 | case IORING_OP_UNLINKAT: |
6083 | putname(req->unlink.filename); | |
6084 | break; | |
0e1b6fe3 PB |
6085 | } |
6086 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6087 | } |
75652a30 JA |
6088 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6089 | kfree(req->apoll->double_poll); | |
6090 | kfree(req->apoll); | |
6091 | req->apoll = NULL; | |
6092 | } | |
3a0a6902 PB |
6093 | if (req->flags & REQ_F_INFLIGHT) { |
6094 | struct io_uring_task *tctx = req->task->io_uring; | |
6095 | ||
6096 | atomic_dec(&tctx->inflight_tracked); | |
6097 | req->flags &= ~REQ_F_INFLIGHT; | |
6098 | } | |
99bc4c38 PB |
6099 | } |
6100 | ||
889fca73 | 6101 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6102 | { |
a197f664 | 6103 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6104 | const struct cred *creds = NULL; |
d625c6ee | 6105 | int ret; |
2b188cc1 | 6106 | |
003e8dcc JA |
6107 | if (req->work.creds && req->work.creds != current_cred()) |
6108 | creds = override_creds(req->work.creds); | |
5730b27e | 6109 | |
d625c6ee | 6110 | switch (req->opcode) { |
2b188cc1 | 6111 | case IORING_OP_NOP: |
889fca73 | 6112 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6113 | break; |
6114 | case IORING_OP_READV: | |
edafccee | 6115 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6116 | case IORING_OP_READ: |
889fca73 | 6117 | ret = io_read(req, issue_flags); |
edafccee | 6118 | break; |
3529d8c2 | 6119 | case IORING_OP_WRITEV: |
edafccee | 6120 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6121 | case IORING_OP_WRITE: |
889fca73 | 6122 | ret = io_write(req, issue_flags); |
2b188cc1 | 6123 | break; |
c992fe29 | 6124 | case IORING_OP_FSYNC: |
45d189c6 | 6125 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6126 | break; |
221c5eb2 | 6127 | case IORING_OP_POLL_ADD: |
61e98203 | 6128 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6129 | break; |
6130 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6131 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6132 | break; |
5d17b4a4 | 6133 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6134 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6135 | break; |
0fa03c62 | 6136 | case IORING_OP_SENDMSG: |
889fca73 | 6137 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6138 | break; |
fddaface | 6139 | case IORING_OP_SEND: |
889fca73 | 6140 | ret = io_send(req, issue_flags); |
0fa03c62 | 6141 | break; |
aa1fa28f | 6142 | case IORING_OP_RECVMSG: |
889fca73 | 6143 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6144 | break; |
fddaface | 6145 | case IORING_OP_RECV: |
889fca73 | 6146 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6147 | break; |
5262f567 | 6148 | case IORING_OP_TIMEOUT: |
61e98203 | 6149 | ret = io_timeout(req, issue_flags); |
5262f567 | 6150 | break; |
11365043 | 6151 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6152 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6153 | break; |
17f2fe35 | 6154 | case IORING_OP_ACCEPT: |
889fca73 | 6155 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6156 | break; |
f8e85cf2 | 6157 | case IORING_OP_CONNECT: |
889fca73 | 6158 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6159 | break; |
62755e35 | 6160 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6161 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6162 | break; |
d63d1b5e | 6163 | case IORING_OP_FALLOCATE: |
45d189c6 | 6164 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6165 | break; |
15b71abe | 6166 | case IORING_OP_OPENAT: |
45d189c6 | 6167 | ret = io_openat(req, issue_flags); |
15b71abe | 6168 | break; |
b5dba59e | 6169 | case IORING_OP_CLOSE: |
889fca73 | 6170 | ret = io_close(req, issue_flags); |
b5dba59e | 6171 | break; |
05f3fb3c | 6172 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6173 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6174 | break; |
eddc7ef5 | 6175 | case IORING_OP_STATX: |
45d189c6 | 6176 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6177 | break; |
4840e418 | 6178 | case IORING_OP_FADVISE: |
45d189c6 | 6179 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6180 | break; |
c1ca757b | 6181 | case IORING_OP_MADVISE: |
45d189c6 | 6182 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6183 | break; |
cebdb986 | 6184 | case IORING_OP_OPENAT2: |
45d189c6 | 6185 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6186 | break; |
3e4827b0 | 6187 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6188 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6189 | break; |
7d67af2c | 6190 | case IORING_OP_SPLICE: |
45d189c6 | 6191 | ret = io_splice(req, issue_flags); |
7d67af2c | 6192 | break; |
ddf0322d | 6193 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6194 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6195 | break; |
067524e9 | 6196 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6197 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6198 | break; |
f2a8d5c7 | 6199 | case IORING_OP_TEE: |
45d189c6 | 6200 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6201 | break; |
36f4fa68 | 6202 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6203 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6204 | break; |
80a261fd | 6205 | case IORING_OP_RENAMEAT: |
45d189c6 | 6206 | ret = io_renameat(req, issue_flags); |
80a261fd | 6207 | break; |
14a1143b | 6208 | case IORING_OP_UNLINKAT: |
45d189c6 | 6209 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6210 | break; |
2b188cc1 JA |
6211 | default: |
6212 | ret = -EINVAL; | |
6213 | break; | |
6214 | } | |
6215 | ||
5730b27e JA |
6216 | if (creds) |
6217 | revert_creds(creds); | |
6218 | ||
def596e9 JA |
6219 | if (ret) |
6220 | return ret; | |
6221 | ||
b532576e JA |
6222 | /* If the op doesn't have a file, we're not polling for it */ |
6223 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6224 | const bool in_async = io_wq_current_is_worker(); |
6225 | ||
11ba820b JA |
6226 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6227 | if (in_async) | |
6228 | mutex_lock(&ctx->uring_lock); | |
6229 | ||
2e9dbe90 | 6230 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6231 | |
6232 | if (in_async) | |
6233 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6234 | } |
6235 | ||
6236 | return 0; | |
2b188cc1 JA |
6237 | } |
6238 | ||
5280f7e5 | 6239 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6240 | { |
6241 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6242 | struct io_kiocb *timeout; |
561fb04a | 6243 | int ret = 0; |
2b188cc1 | 6244 | |
6df1db6b PB |
6245 | timeout = io_prep_linked_timeout(req); |
6246 | if (timeout) | |
6247 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6248 | |
4014d943 | 6249 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6250 | ret = -ECANCELED; |
31b51510 | 6251 | |
561fb04a | 6252 | if (!ret) { |
561fb04a | 6253 | do { |
889fca73 | 6254 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6255 | /* |
6256 | * We can get EAGAIN for polled IO even though we're | |
6257 | * forcing a sync submission from here, since we can't | |
6258 | * wait for request slots on the block side. | |
6259 | */ | |
6260 | if (ret != -EAGAIN) | |
6261 | break; | |
6262 | cond_resched(); | |
6263 | } while (1); | |
6264 | } | |
31b51510 | 6265 | |
a3df7698 | 6266 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6267 | if (ret) { |
a3df7698 | 6268 | /* io-wq is going to take one down */ |
de9b4cca | 6269 | req_ref_get(req); |
a3df7698 | 6270 | io_req_task_queue_fail(req, ret); |
edafccee | 6271 | } |
2b188cc1 JA |
6272 | } |
6273 | ||
7b29f92d JA |
6274 | #define FFS_ASYNC_READ 0x1UL |
6275 | #define FFS_ASYNC_WRITE 0x2UL | |
6276 | #ifdef CONFIG_64BIT | |
6277 | #define FFS_ISREG 0x4UL | |
6278 | #else | |
6279 | #define FFS_ISREG 0x0UL | |
6280 | #endif | |
6281 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6282 | ||
aeca241b | 6283 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6284 | unsigned i) |
65e19f54 | 6285 | { |
aeca241b | 6286 | struct io_fixed_file *table_l2; |
65e19f54 | 6287 | |
aeca241b PB |
6288 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6289 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6290 | } |
6291 | ||
65e19f54 JA |
6292 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6293 | int index) | |
6294 | { | |
aeca241b | 6295 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6296 | |
a04b0ac0 | 6297 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6298 | } |
6299 | ||
a04b0ac0 | 6300 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6301 | { |
6302 | unsigned long file_ptr = (unsigned long) file; | |
6303 | ||
6304 | if (__io_file_supports_async(file, READ)) | |
6305 | file_ptr |= FFS_ASYNC_READ; | |
6306 | if (__io_file_supports_async(file, WRITE)) | |
6307 | file_ptr |= FFS_ASYNC_WRITE; | |
6308 | if (S_ISREG(file_inode(file)->i_mode)) | |
6309 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6310 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6311 | } |
6312 | ||
8371adf5 PB |
6313 | static struct file *io_file_get(struct io_submit_state *state, |
6314 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6315 | { |
a197f664 | 6316 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6317 | struct file *file; |
09bb8394 | 6318 | |
8da11c19 | 6319 | if (fixed) { |
7b29f92d JA |
6320 | unsigned long file_ptr; |
6321 | ||
479f517b | 6322 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6323 | return NULL; |
b7620121 | 6324 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6325 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6326 | file = (struct file *) (file_ptr & FFS_MASK); |
6327 | file_ptr &= ~FFS_MASK; | |
6328 | /* mask in overlapping REQ_F and FFS bits */ | |
6329 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6330 | io_req_set_rsrc_node(req); |
09bb8394 | 6331 | } else { |
c826bd7a | 6332 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6333 | file = __io_file_get(state, fd); |
d44f554e JA |
6334 | |
6335 | /* we don't allow fixed io_uring files */ | |
6336 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6337 | io_req_track_inflight(req); | |
09bb8394 JA |
6338 | } |
6339 | ||
8371adf5 | 6340 | return file; |
09bb8394 JA |
6341 | } |
6342 | ||
2665abfd | 6343 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6344 | { |
ad8a48ac JA |
6345 | struct io_timeout_data *data = container_of(timer, |
6346 | struct io_timeout_data, timer); | |
90cd7e42 | 6347 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6348 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6349 | unsigned long flags; |
2665abfd JA |
6350 | |
6351 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6352 | prev = req->timeout.head; |
6353 | req->timeout.head = NULL; | |
2665abfd JA |
6354 | |
6355 | /* | |
6356 | * We don't expect the list to be empty, that will only happen if we | |
6357 | * race with the completion of the linked work. | |
6358 | */ | |
447c19f3 | 6359 | if (prev) { |
f2f87370 | 6360 | io_remove_next_linked(prev); |
447c19f3 PB |
6361 | if (!req_ref_inc_not_zero(prev)) |
6362 | prev = NULL; | |
6363 | } | |
2665abfd JA |
6364 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6365 | ||
6366 | if (prev) { | |
014db007 | 6367 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6368 | io_put_req_deferred(prev, 1); |
a298232e | 6369 | io_put_req_deferred(req, 1); |
47f46768 | 6370 | } else { |
9ae1f8dd | 6371 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6372 | } |
2665abfd JA |
6373 | return HRTIMER_NORESTART; |
6374 | } | |
6375 | ||
de968c18 | 6376 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6377 | { |
de968c18 PB |
6378 | struct io_ring_ctx *ctx = req->ctx; |
6379 | ||
6380 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6381 | /* |
f2f87370 PB |
6382 | * If the back reference is NULL, then our linked request finished |
6383 | * before we got a chance to setup the timer | |
76a46e06 | 6384 | */ |
90cd7e42 | 6385 | if (req->timeout.head) { |
e8c2bc1f | 6386 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6387 | |
ad8a48ac JA |
6388 | data->timer.function = io_link_timeout_fn; |
6389 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6390 | data->mode); | |
2665abfd | 6391 | } |
76a46e06 | 6392 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6393 | /* drop submission reference */ |
76a46e06 JA |
6394 | io_put_req(req); |
6395 | } | |
2665abfd | 6396 | |
ad8a48ac | 6397 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6398 | { |
f2f87370 | 6399 | struct io_kiocb *nxt = req->link; |
2665abfd | 6400 | |
f2f87370 PB |
6401 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6402 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6403 | return NULL; |
2665abfd | 6404 | |
90cd7e42 | 6405 | nxt->timeout.head = req; |
900fad45 | 6406 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6407 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6408 | return nxt; |
2665abfd JA |
6409 | } |
6410 | ||
c5eef2b9 | 6411 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6412 | { |
d3d7298d | 6413 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6414 | int ret; |
2b188cc1 | 6415 | |
c5eef2b9 | 6416 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6417 | |
491381ce JA |
6418 | /* |
6419 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6420 | * doesn't support non-blocking read/write attempts | |
6421 | */ | |
1840038e | 6422 | if (likely(!ret)) { |
0d63c148 | 6423 | /* drop submission reference */ |
e342c807 | 6424 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6425 | struct io_ring_ctx *ctx = req->ctx; |
6426 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6427 | |
6dd0be1e | 6428 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6429 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6430 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6431 | } else { |
d3d7298d | 6432 | io_put_req(req); |
0d63c148 | 6433 | } |
1840038e PB |
6434 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6435 | if (!io_arm_poll_handler(req)) { | |
6436 | /* | |
6437 | * Queued up for async execution, worker will release | |
6438 | * submit reference when the iocb is actually submitted. | |
6439 | */ | |
6440 | io_queue_async_work(req); | |
6441 | } | |
0d63c148 | 6442 | } else { |
f41db273 | 6443 | io_req_complete_failed(req, ret); |
9e645e11 | 6444 | } |
d3d7298d PB |
6445 | if (linked_timeout) |
6446 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6447 | } |
6448 | ||
be7053b7 | 6449 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6450 | { |
6451 | int ret; | |
6452 | ||
be7053b7 | 6453 | ret = io_req_defer(req); |
4fe2c963 JL |
6454 | if (ret) { |
6455 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6456 | fail_req: |
f41db273 | 6457 | io_req_complete_failed(req, ret); |
4fe2c963 | 6458 | } |
2550878f | 6459 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
b7e298d2 | 6460 | ret = io_req_prep_async(req); |
be7053b7 PB |
6461 | if (unlikely(ret)) |
6462 | goto fail_req; | |
ce35a47a JA |
6463 | io_queue_async_work(req); |
6464 | } else { | |
c5eef2b9 | 6465 | __io_queue_sqe(req); |
ce35a47a | 6466 | } |
4fe2c963 JL |
6467 | } |
6468 | ||
b16fed66 PB |
6469 | /* |
6470 | * Check SQE restrictions (opcode and flags). | |
6471 | * | |
6472 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6473 | */ | |
6474 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6475 | struct io_kiocb *req, | |
6476 | unsigned int sqe_flags) | |
4fe2c963 | 6477 | { |
b16fed66 PB |
6478 | if (!ctx->restricted) |
6479 | return true; | |
6480 | ||
6481 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6482 | return false; | |
6483 | ||
6484 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6485 | ctx->restrictions.sqe_flags_required) | |
6486 | return false; | |
6487 | ||
6488 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6489 | ctx->restrictions.sqe_flags_required)) | |
6490 | return false; | |
6491 | ||
6492 | return true; | |
4fe2c963 JL |
6493 | } |
6494 | ||
b16fed66 PB |
6495 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6496 | const struct io_uring_sqe *sqe) | |
6497 | { | |
6498 | struct io_submit_state *state; | |
6499 | unsigned int sqe_flags; | |
003e8dcc | 6500 | int personality, ret = 0; |
b16fed66 PB |
6501 | |
6502 | req->opcode = READ_ONCE(sqe->opcode); | |
6503 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6504 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6505 | req->user_data = READ_ONCE(sqe->user_data); | |
6506 | req->async_data = NULL; | |
6507 | req->file = NULL; | |
6508 | req->ctx = ctx; | |
6509 | req->link = NULL; | |
6510 | req->fixed_rsrc_refs = NULL; | |
6511 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6512 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6513 | req->task = current; |
6514 | req->result = 0; | |
93e68e03 | 6515 | req->work.creds = NULL; |
b16fed66 PB |
6516 | |
6517 | /* enforce forwards compatibility on users */ | |
dddca226 | 6518 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 6519 | return -EINVAL; |
b16fed66 PB |
6520 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
6521 | return -EINVAL; | |
b16fed66 PB |
6522 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6523 | return -EACCES; | |
6524 | ||
6525 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6526 | !io_op_defs[req->opcode].buffer_select) | |
6527 | return -EOPNOTSUPP; | |
863e0560 | 6528 | |
003e8dcc JA |
6529 | personality = READ_ONCE(sqe->personality); |
6530 | if (personality) { | |
61cf9370 | 6531 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6532 | if (!req->work.creds) |
6533 | return -EINVAL; | |
6534 | get_cred(req->work.creds); | |
003e8dcc | 6535 | } |
b16fed66 PB |
6536 | state = &ctx->submit_state; |
6537 | ||
6538 | /* | |
6539 | * Plug now if we have more than 1 IO left after this, and the target | |
6540 | * is potentially a read/write to block based storage. | |
6541 | */ | |
6542 | if (!state->plug_started && state->ios_left > 1 && | |
6543 | io_op_defs[req->opcode].plug) { | |
6544 | blk_start_plug(&state->plug); | |
6545 | state->plug_started = true; | |
6546 | } | |
6547 | ||
6548 | if (io_op_defs[req->opcode].needs_file) { | |
6549 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6550 | ||
6551 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6552 | if (unlikely(!req->file)) | |
6553 | ret = -EBADF; | |
6554 | } | |
6555 | ||
6556 | state->ios_left--; | |
6557 | return ret; | |
6558 | } | |
6559 | ||
a6b8cadc | 6560 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6561 | const struct io_uring_sqe *sqe) |
9e645e11 | 6562 | { |
a1ab7b35 | 6563 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6564 | int ret; |
9e645e11 | 6565 | |
a6b8cadc PB |
6566 | ret = io_init_req(ctx, req, sqe); |
6567 | if (unlikely(ret)) { | |
6568 | fail_req: | |
de59bc10 PB |
6569 | if (link->head) { |
6570 | /* fail even hard links since we don't submit */ | |
93d2bcd2 | 6571 | req_set_fail(link->head); |
f41db273 | 6572 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6573 | link->head = NULL; |
6574 | } | |
f41db273 | 6575 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6576 | return ret; |
6577 | } | |
be7053b7 PB |
6578 | ret = io_req_prep(req, sqe); |
6579 | if (unlikely(ret)) | |
6580 | goto fail_req; | |
a6b8cadc | 6581 | |
be7053b7 | 6582 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6583 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6584 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6585 | ||
9e645e11 JA |
6586 | /* |
6587 | * If we already have a head request, queue this one for async | |
6588 | * submittal once the head completes. If we don't have a head but | |
6589 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6590 | * submitted sync once the chain is complete. If none of those | |
6591 | * conditions are true (normal request), then just queue it. | |
6592 | */ | |
863e0560 PB |
6593 | if (link->head) { |
6594 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6595 | |
8cdf2193 PB |
6596 | /* |
6597 | * Taking sequential execution of a link, draining both sides | |
6598 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6599 | * requests in the link. So, it drains the head and the | |
6600 | * next after the link request. The last one is done via | |
6601 | * drain_next flag to persist the effect across calls. | |
6602 | */ | |
ef4ff581 | 6603 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6604 | head->flags |= REQ_F_IO_DRAIN; |
6605 | ctx->drain_next = 1; | |
6606 | } | |
b7e298d2 | 6607 | ret = io_req_prep_async(req); |
cf109604 | 6608 | if (unlikely(ret)) |
a6b8cadc | 6609 | goto fail_req; |
9d76377f | 6610 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6611 | link->last->link = req; |
863e0560 | 6612 | link->last = req; |
32fe525b PB |
6613 | |
6614 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6615 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6616 | io_queue_sqe(head); |
863e0560 | 6617 | link->head = NULL; |
32fe525b | 6618 | } |
9e645e11 | 6619 | } else { |
711be031 PB |
6620 | if (unlikely(ctx->drain_next)) { |
6621 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6622 | ctx->drain_next = 0; |
711be031 | 6623 | } |
ef4ff581 | 6624 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6625 | link->head = req; |
6626 | link->last = req; | |
711be031 | 6627 | } else { |
be7053b7 | 6628 | io_queue_sqe(req); |
711be031 | 6629 | } |
9e645e11 | 6630 | } |
2e6e1fde | 6631 | |
1d4240cc | 6632 | return 0; |
9e645e11 JA |
6633 | } |
6634 | ||
9a56a232 JA |
6635 | /* |
6636 | * Batched submission is done, ensure local IO is flushed out. | |
6637 | */ | |
ba88ff11 PB |
6638 | static void io_submit_state_end(struct io_submit_state *state, |
6639 | struct io_ring_ctx *ctx) | |
9a56a232 | 6640 | { |
a1ab7b35 | 6641 | if (state->link.head) |
de59bc10 | 6642 | io_queue_sqe(state->link.head); |
6dd0be1e | 6643 | if (state->comp.nr) |
ba88ff11 | 6644 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6645 | if (state->plug_started) |
6646 | blk_finish_plug(&state->plug); | |
9f13c35b | 6647 | io_state_file_put(state); |
9a56a232 JA |
6648 | } |
6649 | ||
6650 | /* | |
6651 | * Start submission side cache. | |
6652 | */ | |
6653 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6654 | unsigned int max_ios) |
9a56a232 | 6655 | { |
27926b68 | 6656 | state->plug_started = false; |
9a56a232 | 6657 | state->ios_left = max_ios; |
a1ab7b35 PB |
6658 | /* set only head, no need to init link_last in advance */ |
6659 | state->link.head = NULL; | |
9a56a232 JA |
6660 | } |
6661 | ||
2b188cc1 JA |
6662 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6663 | { | |
75b28aff | 6664 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6665 | |
caf582c6 PB |
6666 | /* |
6667 | * Ensure any loads from the SQEs are done at this point, | |
6668 | * since once we write the new head, the application could | |
6669 | * write new data to them. | |
6670 | */ | |
6671 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6672 | } |
6673 | ||
2b188cc1 | 6674 | /* |
3529d8c2 | 6675 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6676 | * that is mapped by userspace. This means that care needs to be taken to |
6677 | * ensure that reads are stable, as we cannot rely on userspace always | |
6678 | * being a good citizen. If members of the sqe are validated and then later | |
6679 | * used, it's important that those reads are done through READ_ONCE() to | |
6680 | * prevent a re-load down the line. | |
6681 | */ | |
709b302f | 6682 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6683 | { |
75b28aff | 6684 | u32 *sq_array = ctx->sq_array; |
ea5ab3b5 | 6685 | unsigned head, mask = ctx->sq_entries - 1; |
2b188cc1 JA |
6686 | |
6687 | /* | |
6688 | * The cached sq head (or cq tail) serves two purposes: | |
6689 | * | |
6690 | * 1) allows us to batch the cost of updating the user visible | |
6691 | * head updates. | |
6692 | * 2) allows the kernel side to track the head on its own, even | |
6693 | * though the application is the one updating it. | |
6694 | */ | |
ea5ab3b5 | 6695 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & mask]); |
709b302f PB |
6696 | if (likely(head < ctx->sq_entries)) |
6697 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6698 | |
6699 | /* drop invalid entries */ | |
498ccd9e | 6700 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6701 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6702 | return NULL; |
6703 | } | |
6704 | ||
0f212204 | 6705 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6706 | { |
46c4e16a | 6707 | int submitted = 0; |
6c271ce2 | 6708 | |
ee7d46d9 PB |
6709 | /* make sure SQ entry isn't read before tail */ |
6710 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6711 | |
2b85edfc PB |
6712 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6713 | return -EAGAIN; | |
6c271ce2 | 6714 | |
d8a6df10 | 6715 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6716 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6717 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6718 | |
46c4e16a | 6719 | while (submitted < nr) { |
3529d8c2 | 6720 | const struct io_uring_sqe *sqe; |
196be95c | 6721 | struct io_kiocb *req; |
fb5ccc98 | 6722 | |
258b29a9 | 6723 | req = io_alloc_req(ctx); |
196be95c PB |
6724 | if (unlikely(!req)) { |
6725 | if (!submitted) | |
6726 | submitted = -EAGAIN; | |
fb5ccc98 | 6727 | break; |
196be95c | 6728 | } |
4fccfcbb PB |
6729 | sqe = io_get_sqe(ctx); |
6730 | if (unlikely(!sqe)) { | |
6731 | kmem_cache_free(req_cachep, req); | |
6732 | break; | |
6733 | } | |
d3656344 JA |
6734 | /* will complete beyond this point, count as submitted */ |
6735 | submitted++; | |
a1ab7b35 | 6736 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6737 | break; |
6c271ce2 JA |
6738 | } |
6739 | ||
9466f437 PB |
6740 | if (unlikely(submitted != nr)) { |
6741 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6742 | struct io_uring_task *tctx = current->io_uring; |
6743 | int unused = nr - ref_used; | |
9466f437 | 6744 | |
d8a6df10 JA |
6745 | percpu_ref_put_many(&ctx->refs, unused); |
6746 | percpu_counter_sub(&tctx->inflight, unused); | |
6747 | put_task_struct_many(current, unused); | |
9466f437 | 6748 | } |
6c271ce2 | 6749 | |
a1ab7b35 | 6750 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6751 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6752 | io_commit_sqring(ctx); | |
6753 | ||
6c271ce2 JA |
6754 | return submitted; |
6755 | } | |
6756 | ||
e4b6d902 PB |
6757 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
6758 | { | |
6759 | return READ_ONCE(sqd->state); | |
6760 | } | |
6761 | ||
23b3628e XW |
6762 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6763 | { | |
6764 | /* Tell userspace we may need a wakeup call */ | |
6765 | spin_lock_irq(&ctx->completion_lock); | |
6766 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6767 | spin_unlock_irq(&ctx->completion_lock); | |
6768 | } | |
6769 | ||
6770 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6771 | { | |
6772 | spin_lock_irq(&ctx->completion_lock); | |
6773 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6774 | spin_unlock_irq(&ctx->completion_lock); | |
6775 | } | |
6776 | ||
08369246 | 6777 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6778 | { |
c8d1ba58 | 6779 | unsigned int to_submit; |
bdcd3eab | 6780 | int ret = 0; |
6c271ce2 | 6781 | |
c8d1ba58 | 6782 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6783 | /* if we're handling multiple rings, cap submit size for fairness */ |
6784 | if (cap_entries && to_submit > 8) | |
6785 | to_submit = 8; | |
6786 | ||
906a3c6f | 6787 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6788 | unsigned nr_events = 0; |
a4c0b3de | 6789 | |
c8d1ba58 | 6790 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6791 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6792 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6793 | |
3b763ba1 PB |
6794 | /* |
6795 | * Don't submit if refs are dying, good for io_uring_register(), | |
6796 | * but also it is relied upon by io_ring_exit_work() | |
6797 | */ | |
0298ef96 PB |
6798 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6799 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6800 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 6801 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 6802 | |
acfb381d PB |
6803 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6804 | wake_up(&ctx->sqo_sq_wait); | |
6805 | } | |
6c271ce2 | 6806 | |
08369246 XW |
6807 | return ret; |
6808 | } | |
6c271ce2 | 6809 | |
08369246 XW |
6810 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6811 | { | |
6812 | struct io_ring_ctx *ctx; | |
6813 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6814 | |
c9dca27d PB |
6815 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6816 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6817 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6818 | } |
6c271ce2 | 6819 | |
e4b6d902 PB |
6820 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
6821 | { | |
6822 | bool did_sig = false; | |
6823 | struct ksignal ksig; | |
6824 | ||
6825 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
6826 | signal_pending(current)) { | |
6827 | mutex_unlock(&sqd->lock); | |
6828 | if (signal_pending(current)) | |
6829 | did_sig = get_signal(&ksig); | |
6830 | cond_resched(); | |
6831 | mutex_lock(&sqd->lock); | |
6832 | } | |
6833 | io_run_task_work(); | |
e4b6d902 PB |
6834 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
6835 | } | |
6836 | ||
c8d1ba58 JA |
6837 | static int io_sq_thread(void *data) |
6838 | { | |
69fb2131 JA |
6839 | struct io_sq_data *sqd = data; |
6840 | struct io_ring_ctx *ctx; | |
a0d9205f | 6841 | unsigned long timeout = 0; |
37d1e2e3 | 6842 | char buf[TASK_COMM_LEN]; |
08369246 | 6843 | DEFINE_WAIT(wait); |
6c271ce2 | 6844 | |
696ee88a | 6845 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6846 | set_task_comm(current, buf); |
37d1e2e3 JA |
6847 | |
6848 | if (sqd->sq_cpu != -1) | |
6849 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6850 | else | |
6851 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6852 | current->flags |= PF_NO_SETAFFINITY; | |
6853 | ||
09a6f4ef | 6854 | mutex_lock(&sqd->lock); |
e4b6d902 | 6855 | while (1) { |
08369246 XW |
6856 | int ret; |
6857 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6858 | |
e4b6d902 PB |
6859 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
6860 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 6861 | break; |
08369246 | 6862 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6863 | continue; |
08369246 | 6864 | } |
e4b6d902 | 6865 | |
08369246 | 6866 | sqt_spin = false; |
e95eee2d | 6867 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6868 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6869 | const struct cred *creds = NULL; |
6870 | ||
6871 | if (ctx->sq_creds != current_cred()) | |
6872 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6873 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6874 | if (creds) |
6875 | revert_creds(creds); | |
08369246 XW |
6876 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6877 | sqt_spin = true; | |
69fb2131 | 6878 | } |
6c271ce2 | 6879 | |
08369246 | 6880 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6881 | io_run_task_work(); |
6882 | cond_resched(); | |
08369246 XW |
6883 | if (sqt_spin) |
6884 | timeout = jiffies + sqd->sq_thread_idle; | |
6885 | continue; | |
6886 | } | |
6887 | ||
08369246 | 6888 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
e4b6d902 | 6889 | if (!io_sqd_events_pending(sqd)) { |
724cb4f9 HX |
6890 | needs_sched = true; |
6891 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
aaa9f0f4 PB |
6892 | io_ring_set_wakeup_flag(ctx); |
6893 | ||
724cb4f9 HX |
6894 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
6895 | !list_empty_careful(&ctx->iopoll_list)) { | |
6896 | needs_sched = false; | |
6897 | break; | |
6898 | } | |
6899 | if (io_sqring_entries(ctx)) { | |
6900 | needs_sched = false; | |
6901 | break; | |
6902 | } | |
6903 | } | |
6904 | ||
6905 | if (needs_sched) { | |
6906 | mutex_unlock(&sqd->lock); | |
6907 | schedule(); | |
6908 | mutex_lock(&sqd->lock); | |
6909 | } | |
69fb2131 JA |
6910 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6911 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6912 | } |
08369246 XW |
6913 | |
6914 | finish_wait(&sqd->wait, &wait); | |
6915 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6916 | } |
28cea78a | 6917 | |
734551df | 6918 | io_uring_cancel_sqpoll(sqd); |
37d1e2e3 | 6919 | sqd->thread = NULL; |
05962f95 | 6920 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6921 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 6922 | io_run_task_work(); |
734551df PB |
6923 | mutex_unlock(&sqd->lock); |
6924 | ||
37d1e2e3 JA |
6925 | complete(&sqd->exited); |
6926 | do_exit(0); | |
6c271ce2 JA |
6927 | } |
6928 | ||
bda52162 JA |
6929 | struct io_wait_queue { |
6930 | struct wait_queue_entry wq; | |
6931 | struct io_ring_ctx *ctx; | |
6932 | unsigned to_wait; | |
6933 | unsigned nr_timeouts; | |
6934 | }; | |
6935 | ||
6c503150 | 6936 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6937 | { |
6938 | struct io_ring_ctx *ctx = iowq->ctx; | |
6939 | ||
6940 | /* | |
d195a66e | 6941 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6942 | * started waiting. For timeouts, we always want to return to userspace, |
6943 | * regardless of event count. | |
6944 | */ | |
6c503150 | 6945 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6946 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6947 | } | |
6948 | ||
6949 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6950 | int wake_flags, void *key) | |
6951 | { | |
6952 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6953 | wq); | |
6954 | ||
6c503150 PB |
6955 | /* |
6956 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6957 | * the task, and the next invocation will do it. | |
6958 | */ | |
6959 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6960 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6961 | return -1; | |
bda52162 JA |
6962 | } |
6963 | ||
af9c1a44 JA |
6964 | static int io_run_task_work_sig(void) |
6965 | { | |
6966 | if (io_run_task_work()) | |
6967 | return 1; | |
6968 | if (!signal_pending(current)) | |
6969 | return 0; | |
0b8cfa97 | 6970 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6971 | return -ERESTARTSYS; |
af9c1a44 JA |
6972 | return -EINTR; |
6973 | } | |
6974 | ||
eeb60b9a PB |
6975 | /* when returns >0, the caller should retry */ |
6976 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6977 | struct io_wait_queue *iowq, | |
6978 | signed long *timeout) | |
6979 | { | |
6980 | int ret; | |
6981 | ||
6982 | /* make sure we run task_work before checking for signals */ | |
6983 | ret = io_run_task_work_sig(); | |
6984 | if (ret || io_should_wake(iowq)) | |
6985 | return ret; | |
6986 | /* let the caller flush overflows, retry */ | |
6987 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6988 | return 1; | |
6989 | ||
6990 | *timeout = schedule_timeout(*timeout); | |
6991 | return !*timeout ? -ETIME : 1; | |
6992 | } | |
6993 | ||
2b188cc1 JA |
6994 | /* |
6995 | * Wait until events become available, if we don't already have some. The | |
6996 | * application must reap them itself, as they reside on the shared cq ring. | |
6997 | */ | |
6998 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6999 | const sigset_t __user *sig, size_t sigsz, |
7000 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7001 | { |
bda52162 JA |
7002 | struct io_wait_queue iowq = { |
7003 | .wq = { | |
7004 | .private = current, | |
7005 | .func = io_wake_function, | |
7006 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7007 | }, | |
7008 | .ctx = ctx, | |
7009 | .to_wait = min_events, | |
7010 | }; | |
75b28aff | 7011 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7012 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7013 | int ret; | |
2b188cc1 | 7014 | |
b41e9852 | 7015 | do { |
6c2450ae | 7016 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 7017 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7018 | return 0; |
4c6e277c | 7019 | if (!io_run_task_work()) |
b41e9852 | 7020 | break; |
b41e9852 | 7021 | } while (1); |
2b188cc1 JA |
7022 | |
7023 | if (sig) { | |
9e75ad5d AB |
7024 | #ifdef CONFIG_COMPAT |
7025 | if (in_compat_syscall()) | |
7026 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7027 | sigsz); |
9e75ad5d AB |
7028 | else |
7029 | #endif | |
b772434b | 7030 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7031 | |
2b188cc1 JA |
7032 | if (ret) |
7033 | return ret; | |
7034 | } | |
7035 | ||
c73ebb68 | 7036 | if (uts) { |
c1d5a224 PB |
7037 | struct timespec64 ts; |
7038 | ||
c73ebb68 HX |
7039 | if (get_timespec64(&ts, uts)) |
7040 | return -EFAULT; | |
7041 | timeout = timespec64_to_jiffies(&ts); | |
7042 | } | |
7043 | ||
bda52162 | 7044 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7045 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7046 | do { |
ca0a2651 | 7047 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7048 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7049 | ret = -EBUSY; |
7050 | break; | |
7051 | } | |
bda52162 JA |
7052 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7053 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7054 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7055 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 7056 | cond_resched(); |
eeb60b9a | 7057 | } while (ret > 0); |
bda52162 | 7058 | |
b7db41c9 | 7059 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7060 | |
75b28aff | 7061 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7062 | } |
7063 | ||
aeca241b | 7064 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) |
05f3fb3c | 7065 | { |
846a4ef2 | 7066 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); |
05f3fb3c | 7067 | |
846a4ef2 | 7068 | for (i = 0; i < nr_tables; i++) |
aeca241b PB |
7069 | kfree(table->files[i]); |
7070 | kfree(table->files); | |
7071 | table->files = NULL; | |
05f3fb3c JA |
7072 | } |
7073 | ||
2a63b2d9 | 7074 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7075 | { |
2a63b2d9 | 7076 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7077 | } |
7078 | ||
2a63b2d9 | 7079 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7080 | { |
2a63b2d9 BM |
7081 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7082 | } | |
65e19f54 | 7083 | |
28a9fe25 | 7084 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7085 | { |
28a9fe25 PB |
7086 | percpu_ref_exit(&ref_node->refs); |
7087 | kfree(ref_node); | |
1642b445 PB |
7088 | } |
7089 | ||
a7f0ed5a PB |
7090 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7091 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7092 | { |
a7f0ed5a PB |
7093 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7094 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7095 | |
a7f0ed5a PB |
7096 | if (data_to_kill) { |
7097 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7098 | |
a7f0ed5a PB |
7099 | rsrc_node->rsrc_data = data_to_kill; |
7100 | io_rsrc_ref_lock(ctx); | |
7101 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7102 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7103 | |
3e942498 | 7104 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7105 | percpu_ref_kill(&rsrc_node->refs); |
7106 | ctx->rsrc_node = NULL; | |
7107 | } | |
6b06314c | 7108 | |
a7f0ed5a PB |
7109 | if (!ctx->rsrc_node) { |
7110 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7111 | ctx->rsrc_backup_node = NULL; | |
7112 | } | |
8bad28d8 HX |
7113 | } |
7114 | ||
a7f0ed5a | 7115 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7116 | { |
7117 | if (ctx->rsrc_backup_node) | |
7118 | return 0; | |
b895c9a6 | 7119 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7120 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7121 | } |
7122 | ||
40ae0ff7 | 7123 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7124 | { |
7125 | int ret; | |
05589553 | 7126 | |
215c3902 | 7127 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7128 | if (data->quiesce) |
7129 | return -ENXIO; | |
05589553 | 7130 | |
8bad28d8 | 7131 | data->quiesce = true; |
1ffc5422 | 7132 | do { |
a7f0ed5a | 7133 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7134 | if (ret) |
f2303b1f | 7135 | break; |
a7f0ed5a | 7136 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7137 | |
3e942498 PB |
7138 | /* kill initial ref, already quiesced if zero */ |
7139 | if (atomic_dec_and_test(&data->refs)) | |
7140 | break; | |
8bad28d8 | 7141 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7142 | ret = wait_for_completion_interruptible(&data->done); |
7143 | if (!ret) | |
7144 | break; | |
8bad28d8 | 7145 | |
3e942498 PB |
7146 | atomic_inc(&data->refs); |
7147 | /* wait for all works potentially completing data->done */ | |
7148 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7149 | reinit_completion(&data->done); |
8dd03afe | 7150 | |
8bad28d8 | 7151 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7152 | ret = io_run_task_work_sig(); |
8bad28d8 | 7153 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7154 | } while (ret >= 0); |
8bad28d8 | 7155 | data->quiesce = false; |
05f3fb3c | 7156 | |
8bad28d8 | 7157 | return ret; |
d7954b2b BM |
7158 | } |
7159 | ||
44b31f2f | 7160 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7161 | { |
b60c8dce | 7162 | kvfree(data->tags); |
44b31f2f PB |
7163 | kfree(data); |
7164 | } | |
7165 | ||
40ae0ff7 | 7166 | static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx, |
b60c8dce PB |
7167 | rsrc_put_fn *do_put, |
7168 | unsigned nr) | |
1ad555c6 | 7169 | { |
b895c9a6 | 7170 | struct io_rsrc_data *data; |
1ad555c6 BM |
7171 | |
7172 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7173 | if (!data) | |
7174 | return NULL; | |
7175 | ||
b60c8dce PB |
7176 | data->tags = kvcalloc(nr, sizeof(*data->tags), GFP_KERNEL); |
7177 | if (!data->tags) { | |
1ad555c6 BM |
7178 | kfree(data); |
7179 | return NULL; | |
7180 | } | |
b60c8dce | 7181 | |
3e942498 | 7182 | atomic_set(&data->refs, 1); |
1ad555c6 | 7183 | data->ctx = ctx; |
40ae0ff7 | 7184 | data->do_put = do_put; |
1ad555c6 BM |
7185 | init_completion(&data->done); |
7186 | return data; | |
7187 | } | |
7188 | ||
fff4db76 | 7189 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7190 | { |
fff4db76 PB |
7191 | #if defined(CONFIG_UNIX) |
7192 | if (ctx->ring_sock) { | |
7193 | struct sock *sock = ctx->ring_sock->sk; | |
7194 | struct sk_buff *skb; | |
7195 | ||
7196 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7197 | kfree_skb(skb); | |
7198 | } | |
7199 | #else | |
7200 | int i; | |
7201 | ||
7202 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7203 | struct file *file; | |
7204 | ||
7205 | file = io_file_from_index(ctx, i); | |
7206 | if (file) | |
7207 | fput(file); | |
7208 | } | |
7209 | #endif | |
7210 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); | |
44b31f2f | 7211 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7212 | ctx->file_data = NULL; |
7213 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7214 | } |
7215 | ||
d7954b2b BM |
7216 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7217 | { | |
d7954b2b BM |
7218 | int ret; |
7219 | ||
08480400 | 7220 | if (!ctx->file_data) |
d7954b2b | 7221 | return -ENXIO; |
08480400 PB |
7222 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7223 | if (!ret) | |
7224 | __io_sqe_files_unregister(ctx); | |
7225 | return ret; | |
6b06314c JA |
7226 | } |
7227 | ||
37d1e2e3 | 7228 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7229 | __releases(&sqd->lock) |
37d1e2e3 | 7230 | { |
521d6a73 PB |
7231 | WARN_ON_ONCE(sqd->thread == current); |
7232 | ||
9e138a48 PB |
7233 | /* |
7234 | * Do the dance but not conditional clear_bit() because it'd race with | |
7235 | * other threads incrementing park_pending and setting the bit. | |
7236 | */ | |
37d1e2e3 | 7237 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7238 | if (atomic_dec_return(&sqd->park_pending)) |
7239 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7240 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7241 | } |
7242 | ||
86e0d676 | 7243 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7244 | __acquires(&sqd->lock) |
37d1e2e3 | 7245 | { |
521d6a73 PB |
7246 | WARN_ON_ONCE(sqd->thread == current); |
7247 | ||
9e138a48 | 7248 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7249 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7250 | mutex_lock(&sqd->lock); |
05962f95 | 7251 | if (sqd->thread) |
86e0d676 | 7252 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7253 | } |
7254 | ||
7255 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7256 | { | |
521d6a73 | 7257 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7258 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7259 | |
05962f95 | 7260 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7261 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7262 | if (sqd->thread) |
7263 | wake_up_process(sqd->thread); | |
09a6f4ef | 7264 | mutex_unlock(&sqd->lock); |
05962f95 | 7265 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7266 | } |
7267 | ||
534ca6d6 | 7268 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7269 | { |
534ca6d6 | 7270 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7271 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7272 | ||
37d1e2e3 JA |
7273 | io_sq_thread_stop(sqd); |
7274 | kfree(sqd); | |
7275 | } | |
7276 | } | |
7277 | ||
7278 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7279 | { | |
7280 | struct io_sq_data *sqd = ctx->sq_data; | |
7281 | ||
7282 | if (sqd) { | |
05962f95 | 7283 | io_sq_thread_park(sqd); |
521d6a73 | 7284 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7285 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7286 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7287 | |
7288 | io_put_sq_data(sqd); | |
7289 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7290 | } |
7291 | } | |
7292 | ||
aa06165d JA |
7293 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7294 | { | |
7295 | struct io_ring_ctx *ctx_attach; | |
7296 | struct io_sq_data *sqd; | |
7297 | struct fd f; | |
7298 | ||
7299 | f = fdget(p->wq_fd); | |
7300 | if (!f.file) | |
7301 | return ERR_PTR(-ENXIO); | |
7302 | if (f.file->f_op != &io_uring_fops) { | |
7303 | fdput(f); | |
7304 | return ERR_PTR(-EINVAL); | |
7305 | } | |
7306 | ||
7307 | ctx_attach = f.file->private_data; | |
7308 | sqd = ctx_attach->sq_data; | |
7309 | if (!sqd) { | |
7310 | fdput(f); | |
7311 | return ERR_PTR(-EINVAL); | |
7312 | } | |
5c2469e0 JA |
7313 | if (sqd->task_tgid != current->tgid) { |
7314 | fdput(f); | |
7315 | return ERR_PTR(-EPERM); | |
7316 | } | |
aa06165d JA |
7317 | |
7318 | refcount_inc(&sqd->refs); | |
7319 | fdput(f); | |
7320 | return sqd; | |
7321 | } | |
7322 | ||
26984fbf PB |
7323 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7324 | bool *attached) | |
534ca6d6 JA |
7325 | { |
7326 | struct io_sq_data *sqd; | |
7327 | ||
26984fbf | 7328 | *attached = false; |
5c2469e0 JA |
7329 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7330 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7331 | if (!IS_ERR(sqd)) { |
7332 | *attached = true; | |
5c2469e0 | 7333 | return sqd; |
26984fbf | 7334 | } |
5c2469e0 JA |
7335 | /* fall through for EPERM case, setup new sqd/task */ |
7336 | if (PTR_ERR(sqd) != -EPERM) | |
7337 | return sqd; | |
7338 | } | |
aa06165d | 7339 | |
534ca6d6 JA |
7340 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7341 | if (!sqd) | |
7342 | return ERR_PTR(-ENOMEM); | |
7343 | ||
9e138a48 | 7344 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7345 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7346 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7347 | mutex_init(&sqd->lock); |
534ca6d6 | 7348 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7349 | init_completion(&sqd->exited); |
534ca6d6 JA |
7350 | return sqd; |
7351 | } | |
7352 | ||
6b06314c | 7353 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7354 | /* |
7355 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7356 | * the io_uring can be safely unregistered on process exit, even if we have | |
7357 | * loops in the file referencing. | |
7358 | */ | |
7359 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7360 | { | |
7361 | struct sock *sk = ctx->ring_sock->sk; | |
7362 | struct scm_fp_list *fpl; | |
7363 | struct sk_buff *skb; | |
08a45173 | 7364 | int i, nr_files; |
6b06314c | 7365 | |
6b06314c JA |
7366 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7367 | if (!fpl) | |
7368 | return -ENOMEM; | |
7369 | ||
7370 | skb = alloc_skb(0, GFP_KERNEL); | |
7371 | if (!skb) { | |
7372 | kfree(fpl); | |
7373 | return -ENOMEM; | |
7374 | } | |
7375 | ||
7376 | skb->sk = sk; | |
6b06314c | 7377 | |
08a45173 | 7378 | nr_files = 0; |
62e398be | 7379 | fpl->user = get_uid(current_user()); |
6b06314c | 7380 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7381 | struct file *file = io_file_from_index(ctx, i + offset); |
7382 | ||
7383 | if (!file) | |
08a45173 | 7384 | continue; |
65e19f54 | 7385 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7386 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7387 | nr_files++; | |
6b06314c JA |
7388 | } |
7389 | ||
08a45173 JA |
7390 | if (nr_files) { |
7391 | fpl->max = SCM_MAX_FD; | |
7392 | fpl->count = nr_files; | |
7393 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7394 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7395 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7396 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7397 | |
08a45173 JA |
7398 | for (i = 0; i < nr_files; i++) |
7399 | fput(fpl->fp[i]); | |
7400 | } else { | |
7401 | kfree_skb(skb); | |
7402 | kfree(fpl); | |
7403 | } | |
6b06314c JA |
7404 | |
7405 | return 0; | |
7406 | } | |
7407 | ||
7408 | /* | |
7409 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7410 | * causes regular reference counting to break down. We rely on the UNIX | |
7411 | * garbage collection to take care of this problem for us. | |
7412 | */ | |
7413 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7414 | { | |
7415 | unsigned left, total; | |
7416 | int ret = 0; | |
7417 | ||
7418 | total = 0; | |
7419 | left = ctx->nr_user_files; | |
7420 | while (left) { | |
7421 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7422 | |
7423 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7424 | if (ret) | |
7425 | break; | |
7426 | left -= this_files; | |
7427 | total += this_files; | |
7428 | } | |
7429 | ||
7430 | if (!ret) | |
7431 | return 0; | |
7432 | ||
7433 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7434 | struct file *file = io_file_from_index(ctx, total); |
7435 | ||
7436 | if (file) | |
7437 | fput(file); | |
6b06314c JA |
7438 | total++; |
7439 | } | |
7440 | ||
7441 | return ret; | |
7442 | } | |
7443 | #else | |
7444 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7445 | { | |
7446 | return 0; | |
7447 | } | |
7448 | #endif | |
7449 | ||
aeca241b | 7450 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
65e19f54 | 7451 | { |
846a4ef2 PB |
7452 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); |
7453 | ||
aeca241b PB |
7454 | table->files = kcalloc(nr_tables, sizeof(*table->files), GFP_KERNEL); |
7455 | if (!table->files) | |
846a4ef2 | 7456 | return false; |
65e19f54 JA |
7457 | |
7458 | for (i = 0; i < nr_tables; i++) { | |
846a4ef2 | 7459 | unsigned int this_files = min(nr_files, IORING_MAX_FILES_TABLE); |
65e19f54 | 7460 | |
aeca241b | 7461 | table->files[i] = kcalloc(this_files, sizeof(*table->files[i]), |
65e19f54 | 7462 | GFP_KERNEL); |
aeca241b | 7463 | if (!table->files[i]) |
65e19f54 JA |
7464 | break; |
7465 | nr_files -= this_files; | |
7466 | } | |
7467 | ||
7468 | if (i == nr_tables) | |
846a4ef2 | 7469 | return true; |
65e19f54 | 7470 | |
aeca241b | 7471 | io_free_file_tables(table, nr_tables * IORING_MAX_FILES_TABLE); |
846a4ef2 | 7472 | return false; |
65e19f54 JA |
7473 | } |
7474 | ||
47e90392 | 7475 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7476 | { |
50238531 | 7477 | struct file *file = prsrc->file; |
05f3fb3c JA |
7478 | #if defined(CONFIG_UNIX) |
7479 | struct sock *sock = ctx->ring_sock->sk; | |
7480 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7481 | struct sk_buff *skb; | |
7482 | int i; | |
7483 | ||
7484 | __skb_queue_head_init(&list); | |
7485 | ||
7486 | /* | |
7487 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7488 | * remove this entry and rearrange the file array. | |
7489 | */ | |
7490 | skb = skb_dequeue(head); | |
7491 | while (skb) { | |
7492 | struct scm_fp_list *fp; | |
7493 | ||
7494 | fp = UNIXCB(skb).fp; | |
7495 | for (i = 0; i < fp->count; i++) { | |
7496 | int left; | |
7497 | ||
7498 | if (fp->fp[i] != file) | |
7499 | continue; | |
7500 | ||
7501 | unix_notinflight(fp->user, fp->fp[i]); | |
7502 | left = fp->count - 1 - i; | |
7503 | if (left) { | |
7504 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7505 | left * sizeof(struct file *)); | |
7506 | } | |
7507 | fp->count--; | |
7508 | if (!fp->count) { | |
7509 | kfree_skb(skb); | |
7510 | skb = NULL; | |
7511 | } else { | |
7512 | __skb_queue_tail(&list, skb); | |
7513 | } | |
7514 | fput(file); | |
7515 | file = NULL; | |
7516 | break; | |
7517 | } | |
7518 | ||
7519 | if (!file) | |
7520 | break; | |
7521 | ||
7522 | __skb_queue_tail(&list, skb); | |
7523 | ||
7524 | skb = skb_dequeue(head); | |
7525 | } | |
7526 | ||
7527 | if (skb_peek(&list)) { | |
7528 | spin_lock_irq(&head->lock); | |
7529 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7530 | __skb_queue_tail(head, skb); | |
7531 | spin_unlock_irq(&head->lock); | |
7532 | } | |
7533 | #else | |
7534 | fput(file); | |
7535 | #endif | |
7536 | } | |
7537 | ||
b895c9a6 | 7538 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7539 | { |
b895c9a6 | 7540 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7541 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7542 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7543 | |
269bbe5f BM |
7544 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7545 | list_del(&prsrc->list); | |
b60c8dce PB |
7546 | |
7547 | if (prsrc->tag) { | |
7548 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
7549 | unsigned long flags; | |
7550 | ||
7551 | io_ring_submit_lock(ctx, lock_ring); | |
7552 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
7553 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); | |
2840f710 | 7554 | ctx->cq_extra++; |
b60c8dce PB |
7555 | io_commit_cqring(ctx); |
7556 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
7557 | io_cqring_ev_posted(ctx); | |
7558 | io_ring_submit_unlock(ctx, lock_ring); | |
7559 | } | |
7560 | ||
40ae0ff7 | 7561 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7562 | kfree(prsrc); |
65e19f54 | 7563 | } |
05589553 | 7564 | |
28a9fe25 | 7565 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7566 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7567 | complete(&rsrc_data->done); | |
2faf852d | 7568 | } |
65e19f54 | 7569 | |
269bbe5f | 7570 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7571 | { |
7572 | struct io_ring_ctx *ctx; | |
7573 | struct llist_node *node; | |
7574 | ||
269bbe5f BM |
7575 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7576 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7577 | |
7578 | while (node) { | |
b895c9a6 | 7579 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7580 | struct llist_node *next = node->next; |
7581 | ||
b895c9a6 | 7582 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7583 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7584 | node = next; |
7585 | } | |
7586 | } | |
7587 | ||
00835dce | 7588 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7589 | { |
b895c9a6 | 7590 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7591 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7592 | bool first_add = false; |
e297822b | 7593 | |
2a63b2d9 | 7594 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7595 | node->done = true; |
e297822b | 7596 | |
d67d2263 | 7597 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7598 | node = list_first_entry(&ctx->rsrc_ref_list, |
7599 | struct io_rsrc_node, node); | |
e297822b | 7600 | /* recycle ref nodes in order */ |
b895c9a6 | 7601 | if (!node->done) |
e297822b | 7602 | break; |
b895c9a6 PB |
7603 | list_del(&node->node); |
7604 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7605 | } |
2a63b2d9 | 7606 | io_rsrc_ref_unlock(ctx); |
05589553 | 7607 | |
3e942498 PB |
7608 | if (first_add) |
7609 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7610 | } |
65e19f54 | 7611 | |
b895c9a6 | 7612 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7613 | { |
b895c9a6 | 7614 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7615 | |
05589553 XW |
7616 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7617 | if (!ref_node) | |
3e2224c5 | 7618 | return NULL; |
05f3fb3c | 7619 | |
00835dce | 7620 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7621 | 0, GFP_KERNEL)) { |
7622 | kfree(ref_node); | |
3e2224c5 | 7623 | return NULL; |
05589553 XW |
7624 | } |
7625 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7626 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7627 | ref_node->done = false; |
05589553 | 7628 | return ref_node; |
05589553 XW |
7629 | } |
7630 | ||
6b06314c | 7631 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 7632 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
7633 | { |
7634 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7635 | struct file *file; |
f3baed39 | 7636 | int fd, ret; |
846a4ef2 | 7637 | unsigned i; |
b895c9a6 | 7638 | struct io_rsrc_data *file_data; |
6b06314c | 7639 | |
05f3fb3c | 7640 | if (ctx->file_data) |
6b06314c JA |
7641 | return -EBUSY; |
7642 | if (!nr_args) | |
7643 | return -EINVAL; | |
7644 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7645 | return -EMFILE; | |
a7f0ed5a | 7646 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7647 | if (ret) |
7648 | return ret; | |
6b06314c | 7649 | |
b60c8dce | 7650 | file_data = io_rsrc_data_alloc(ctx, io_rsrc_file_put, nr_args); |
5398ae69 | 7651 | if (!file_data) |
05f3fb3c | 7652 | return -ENOMEM; |
13770a71 | 7653 | ctx->file_data = file_data; |
f3baed39 | 7654 | ret = -ENOMEM; |
aeca241b | 7655 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7656 | goto out_free; |
65e19f54 | 7657 | |
08a45173 | 7658 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
792e3582 PB |
7659 | u64 tag = 0; |
7660 | ||
7661 | if ((tags && copy_from_user(&tag, &tags[i], sizeof(tag))) || | |
7662 | copy_from_user(&fd, &fds[i], sizeof(fd))) { | |
600cf3f8 PB |
7663 | ret = -EFAULT; |
7664 | goto out_fput; | |
7665 | } | |
08a45173 | 7666 | /* allow sparse sets */ |
792e3582 PB |
7667 | if (fd == -1) { |
7668 | ret = -EINVAL; | |
7669 | if (unlikely(tag)) | |
7670 | goto out_fput; | |
08a45173 | 7671 | continue; |
792e3582 | 7672 | } |
6b06314c | 7673 | |
05f3fb3c | 7674 | file = fget(fd); |
6b06314c | 7675 | ret = -EBADF; |
792e3582 | 7676 | if (unlikely(!file)) |
600cf3f8 | 7677 | goto out_fput; |
05f3fb3c | 7678 | |
6b06314c JA |
7679 | /* |
7680 | * Don't allow io_uring instances to be registered. If UNIX | |
7681 | * isn't enabled, then this causes a reference cycle and this | |
7682 | * instance can never get freed. If UNIX is enabled we'll | |
7683 | * handle it just fine, but there's still no point in allowing | |
7684 | * a ring fd as it doesn't support regular read/write anyway. | |
7685 | */ | |
05f3fb3c JA |
7686 | if (file->f_op == &io_uring_fops) { |
7687 | fput(file); | |
600cf3f8 | 7688 | goto out_fput; |
6b06314c | 7689 | } |
792e3582 | 7690 | ctx->file_data->tags[i] = tag; |
aeca241b | 7691 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7692 | } |
7693 | ||
6b06314c | 7694 | ret = io_sqe_files_scm(ctx); |
05589553 | 7695 | if (ret) { |
08480400 | 7696 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7697 | return ret; |
7698 | } | |
6b06314c | 7699 | |
a7f0ed5a | 7700 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7701 | return ret; |
600cf3f8 PB |
7702 | out_fput: |
7703 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7704 | file = io_file_from_index(ctx, i); | |
7705 | if (file) | |
7706 | fput(file); | |
7707 | } | |
aeca241b | 7708 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7709 | ctx->nr_user_files = 0; |
600cf3f8 | 7710 | out_free: |
44b31f2f | 7711 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7712 | ctx->file_data = NULL; |
6b06314c JA |
7713 | return ret; |
7714 | } | |
7715 | ||
c3a31e60 JA |
7716 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7717 | int index) | |
7718 | { | |
7719 | #if defined(CONFIG_UNIX) | |
7720 | struct sock *sock = ctx->ring_sock->sk; | |
7721 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7722 | struct sk_buff *skb; | |
7723 | ||
7724 | /* | |
7725 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7726 | * file set. If there's no room, fall back to allocating a new skb | |
7727 | * and filling it in. | |
7728 | */ | |
7729 | spin_lock_irq(&head->lock); | |
7730 | skb = skb_peek(head); | |
7731 | if (skb) { | |
7732 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7733 | ||
7734 | if (fpl->count < SCM_MAX_FD) { | |
7735 | __skb_unlink(skb, head); | |
7736 | spin_unlock_irq(&head->lock); | |
7737 | fpl->fp[fpl->count] = get_file(file); | |
7738 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7739 | fpl->count++; | |
7740 | spin_lock_irq(&head->lock); | |
7741 | __skb_queue_head(head, skb); | |
7742 | } else { | |
7743 | skb = NULL; | |
7744 | } | |
7745 | } | |
7746 | spin_unlock_irq(&head->lock); | |
7747 | ||
7748 | if (skb) { | |
7749 | fput(file); | |
7750 | return 0; | |
7751 | } | |
7752 | ||
7753 | return __io_sqe_files_scm(ctx, 1, index); | |
7754 | #else | |
7755 | return 0; | |
7756 | #endif | |
7757 | } | |
7758 | ||
b60c8dce | 7759 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 7760 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 7761 | { |
269bbe5f | 7762 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7763 | |
269bbe5f BM |
7764 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7765 | if (!prsrc) | |
a5318d3c | 7766 | return -ENOMEM; |
05f3fb3c | 7767 | |
b60c8dce | 7768 | prsrc->tag = data->tags[idx]; |
50238531 | 7769 | prsrc->rsrc = rsrc; |
e7c78371 | 7770 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7771 | return 0; |
05f3fb3c JA |
7772 | } |
7773 | ||
7774 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 7775 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
7776 | unsigned nr_args) |
7777 | { | |
c3bdad02 | 7778 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 7779 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 7780 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7781 | struct io_fixed_file *file_slot; |
7782 | struct file *file; | |
98f0b3b4 PB |
7783 | int fd, i, err = 0; |
7784 | unsigned int done; | |
05589553 | 7785 | bool needs_switch = false; |
c3a31e60 | 7786 | |
98f0b3b4 PB |
7787 | if (!ctx->file_data) |
7788 | return -ENXIO; | |
7789 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
7790 | return -EINVAL; |
7791 | ||
67973b93 | 7792 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
7793 | u64 tag = 0; |
7794 | ||
7795 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
7796 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
7797 | err = -EFAULT; |
7798 | break; | |
7799 | } | |
c3bdad02 PB |
7800 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
7801 | err = -EINVAL; | |
7802 | break; | |
7803 | } | |
4e0377a1 | 7804 | if (fd == IORING_REGISTER_FILES_SKIP) |
7805 | continue; | |
7806 | ||
67973b93 | 7807 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7808 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7809 | |
a04b0ac0 PB |
7810 | if (file_slot->file_ptr) { |
7811 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
7812 | err = io_queue_rsrc_removal(data, up->offset + done, |
7813 | ctx->rsrc_node, file); | |
a5318d3c HD |
7814 | if (err) |
7815 | break; | |
a04b0ac0 | 7816 | file_slot->file_ptr = 0; |
05589553 | 7817 | needs_switch = true; |
c3a31e60 JA |
7818 | } |
7819 | if (fd != -1) { | |
c3a31e60 JA |
7820 | file = fget(fd); |
7821 | if (!file) { | |
7822 | err = -EBADF; | |
7823 | break; | |
7824 | } | |
7825 | /* | |
7826 | * Don't allow io_uring instances to be registered. If | |
7827 | * UNIX isn't enabled, then this causes a reference | |
7828 | * cycle and this instance can never get freed. If UNIX | |
7829 | * is enabled we'll handle it just fine, but there's | |
7830 | * still no point in allowing a ring fd as it doesn't | |
7831 | * support regular read/write anyway. | |
7832 | */ | |
7833 | if (file->f_op == &io_uring_fops) { | |
7834 | fput(file); | |
7835 | err = -EBADF; | |
7836 | break; | |
7837 | } | |
c3bdad02 | 7838 | data->tags[up->offset + done] = tag; |
9a321c98 | 7839 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7840 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7841 | if (err) { |
a04b0ac0 | 7842 | file_slot->file_ptr = 0; |
f3bd9dae | 7843 | fput(file); |
c3a31e60 | 7844 | break; |
f3bd9dae | 7845 | } |
c3a31e60 | 7846 | } |
05f3fb3c JA |
7847 | } |
7848 | ||
a7f0ed5a PB |
7849 | if (needs_switch) |
7850 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7851 | return done ? done : err; |
7852 | } | |
05589553 | 7853 | |
5280f7e5 | 7854 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7855 | { |
7856 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7857 | ||
5280f7e5 PB |
7858 | req = io_put_req_find_next(req); |
7859 | return req ? &req->work : NULL; | |
7d723065 JA |
7860 | } |
7861 | ||
685fe7fe JA |
7862 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7863 | struct task_struct *task) | |
24369c2e | 7864 | { |
e941894e | 7865 | struct io_wq_hash *hash; |
24369c2e | 7866 | struct io_wq_data data; |
24369c2e | 7867 | unsigned int concurrency; |
24369c2e | 7868 | |
e941894e JA |
7869 | hash = ctx->hash_map; |
7870 | if (!hash) { | |
7871 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7872 | if (!hash) | |
7873 | return ERR_PTR(-ENOMEM); | |
7874 | refcount_set(&hash->refs, 1); | |
7875 | init_waitqueue_head(&hash->wait); | |
7876 | ctx->hash_map = hash; | |
24369c2e PB |
7877 | } |
7878 | ||
e941894e | 7879 | data.hash = hash; |
685fe7fe | 7880 | data.task = task; |
e9fd9396 | 7881 | data.free_work = io_free_work; |
f5fa38c5 | 7882 | data.do_work = io_wq_submit_work; |
24369c2e | 7883 | |
d25e3a3d JA |
7884 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7885 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7886 | |
5aa75ed5 | 7887 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7888 | } |
7889 | ||
5aa75ed5 JA |
7890 | static int io_uring_alloc_task_context(struct task_struct *task, |
7891 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7892 | { |
7893 | struct io_uring_task *tctx; | |
d8a6df10 | 7894 | int ret; |
0f212204 JA |
7895 | |
7896 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7897 | if (unlikely(!tctx)) | |
7898 | return -ENOMEM; | |
7899 | ||
d8a6df10 JA |
7900 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7901 | if (unlikely(ret)) { | |
7902 | kfree(tctx); | |
7903 | return ret; | |
7904 | } | |
7905 | ||
685fe7fe | 7906 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7907 | if (IS_ERR(tctx->io_wq)) { |
7908 | ret = PTR_ERR(tctx->io_wq); | |
7909 | percpu_counter_destroy(&tctx->inflight); | |
7910 | kfree(tctx); | |
7911 | return ret; | |
7912 | } | |
7913 | ||
0f212204 JA |
7914 | xa_init(&tctx->xa); |
7915 | init_waitqueue_head(&tctx->wait); | |
7916 | tctx->last = NULL; | |
fdaf083c | 7917 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7918 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7919 | task->io_uring = tctx; |
7cbf1722 JA |
7920 | spin_lock_init(&tctx->task_lock); |
7921 | INIT_WQ_LIST(&tctx->task_list); | |
7922 | tctx->task_state = 0; | |
7923 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7924 | return 0; |
7925 | } | |
7926 | ||
7927 | void __io_uring_free(struct task_struct *tsk) | |
7928 | { | |
7929 | struct io_uring_task *tctx = tsk->io_uring; | |
7930 | ||
7931 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7932 | WARN_ON_ONCE(tctx->io_wq); |
7933 | ||
d8a6df10 | 7934 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7935 | kfree(tctx); |
7936 | tsk->io_uring = NULL; | |
7937 | } | |
7938 | ||
7e84e1c7 SG |
7939 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7940 | struct io_uring_params *p) | |
2b188cc1 JA |
7941 | { |
7942 | int ret; | |
7943 | ||
d25e3a3d JA |
7944 | /* Retain compatibility with failing for an invalid attach attempt */ |
7945 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7946 | IORING_SETUP_ATTACH_WQ) { | |
7947 | struct fd f; | |
7948 | ||
7949 | f = fdget(p->wq_fd); | |
7950 | if (!f.file) | |
7951 | return -ENXIO; | |
d25e3a3d | 7952 | fdput(f); |
f2a48dd0 PB |
7953 | if (f.file->f_op != &io_uring_fops) |
7954 | return -EINVAL; | |
d25e3a3d | 7955 | } |
6c271ce2 | 7956 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7957 | struct task_struct *tsk; |
534ca6d6 | 7958 | struct io_sq_data *sqd; |
26984fbf | 7959 | bool attached; |
534ca6d6 | 7960 | |
26984fbf | 7961 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7962 | if (IS_ERR(sqd)) { |
7963 | ret = PTR_ERR(sqd); | |
7964 | goto err; | |
7965 | } | |
69fb2131 | 7966 | |
7c30f36a | 7967 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7968 | ctx->sq_data = sqd; |
917257da JA |
7969 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7970 | if (!ctx->sq_thread_idle) | |
7971 | ctx->sq_thread_idle = HZ; | |
7972 | ||
78d7f6ba | 7973 | io_sq_thread_park(sqd); |
de75a3d3 PB |
7974 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
7975 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 7976 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 7977 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
7978 | io_sq_thread_unpark(sqd); |
7979 | ||
de75a3d3 PB |
7980 | if (ret < 0) |
7981 | goto err; | |
7982 | if (attached) | |
5aa75ed5 | 7983 | return 0; |
aa06165d | 7984 | |
6c271ce2 | 7985 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7986 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7987 | |
917257da | 7988 | ret = -EINVAL; |
f2a48dd0 | 7989 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 7990 | goto err_sqpoll; |
37d1e2e3 | 7991 | sqd->sq_cpu = cpu; |
6c271ce2 | 7992 | } else { |
37d1e2e3 | 7993 | sqd->sq_cpu = -1; |
6c271ce2 | 7994 | } |
37d1e2e3 JA |
7995 | |
7996 | sqd->task_pid = current->pid; | |
5c2469e0 | 7997 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7998 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7999 | if (IS_ERR(tsk)) { | |
8000 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8001 | goto err_sqpoll; |
6c271ce2 | 8002 | } |
97a73a0f | 8003 | |
46fe18b1 | 8004 | sqd->thread = tsk; |
97a73a0f | 8005 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8006 | wake_up_new_task(tsk); |
0f212204 JA |
8007 | if (ret) |
8008 | goto err; | |
6c271ce2 JA |
8009 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8010 | /* Can't have SQ_AFF without SQPOLL */ | |
8011 | ret = -EINVAL; | |
8012 | goto err; | |
8013 | } | |
8014 | ||
2b188cc1 | 8015 | return 0; |
f2a48dd0 PB |
8016 | err_sqpoll: |
8017 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8018 | err: |
37d1e2e3 | 8019 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8020 | return ret; |
8021 | } | |
8022 | ||
a087e2b5 BM |
8023 | static inline void __io_unaccount_mem(struct user_struct *user, |
8024 | unsigned long nr_pages) | |
2b188cc1 JA |
8025 | { |
8026 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8027 | } | |
8028 | ||
a087e2b5 BM |
8029 | static inline int __io_account_mem(struct user_struct *user, |
8030 | unsigned long nr_pages) | |
2b188cc1 JA |
8031 | { |
8032 | unsigned long page_limit, cur_pages, new_pages; | |
8033 | ||
8034 | /* Don't allow more pages than we can safely lock */ | |
8035 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8036 | ||
8037 | do { | |
8038 | cur_pages = atomic_long_read(&user->locked_vm); | |
8039 | new_pages = cur_pages + nr_pages; | |
8040 | if (new_pages > page_limit) | |
8041 | return -ENOMEM; | |
8042 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8043 | new_pages) != cur_pages); | |
8044 | ||
8045 | return 0; | |
8046 | } | |
8047 | ||
26bfa89e | 8048 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8049 | { |
62e398be | 8050 | if (ctx->user) |
a087e2b5 | 8051 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8052 | |
26bfa89e JA |
8053 | if (ctx->mm_account) |
8054 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8055 | } |
8056 | ||
26bfa89e | 8057 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8058 | { |
30975825 BM |
8059 | int ret; |
8060 | ||
62e398be | 8061 | if (ctx->user) { |
30975825 BM |
8062 | ret = __io_account_mem(ctx->user, nr_pages); |
8063 | if (ret) | |
8064 | return ret; | |
8065 | } | |
8066 | ||
26bfa89e JA |
8067 | if (ctx->mm_account) |
8068 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8069 | |
8070 | return 0; | |
8071 | } | |
8072 | ||
2b188cc1 JA |
8073 | static void io_mem_free(void *ptr) |
8074 | { | |
52e04ef4 MR |
8075 | struct page *page; |
8076 | ||
8077 | if (!ptr) | |
8078 | return; | |
2b188cc1 | 8079 | |
52e04ef4 | 8080 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8081 | if (put_page_testzero(page)) |
8082 | free_compound_page(page); | |
8083 | } | |
8084 | ||
8085 | static void *io_mem_alloc(size_t size) | |
8086 | { | |
8087 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8088 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8089 | |
8090 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8091 | } | |
8092 | ||
75b28aff HV |
8093 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8094 | size_t *sq_offset) | |
8095 | { | |
8096 | struct io_rings *rings; | |
8097 | size_t off, sq_array_size; | |
8098 | ||
8099 | off = struct_size(rings, cqes, cq_entries); | |
8100 | if (off == SIZE_MAX) | |
8101 | return SIZE_MAX; | |
8102 | ||
8103 | #ifdef CONFIG_SMP | |
8104 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8105 | if (off == 0) | |
8106 | return SIZE_MAX; | |
8107 | #endif | |
8108 | ||
b36200f5 DV |
8109 | if (sq_offset) |
8110 | *sq_offset = off; | |
8111 | ||
75b28aff HV |
8112 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8113 | if (sq_array_size == SIZE_MAX) | |
8114 | return SIZE_MAX; | |
8115 | ||
8116 | if (check_add_overflow(off, sq_array_size, &off)) | |
8117 | return SIZE_MAX; | |
8118 | ||
75b28aff HV |
8119 | return off; |
8120 | } | |
8121 | ||
41edf1a5 | 8122 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8123 | { |
41edf1a5 | 8124 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8125 | unsigned int i; |
8126 | ||
6224843d PB |
8127 | if (imu != ctx->dummy_ubuf) { |
8128 | for (i = 0; i < imu->nr_bvecs; i++) | |
8129 | unpin_user_page(imu->bvec[i].bv_page); | |
8130 | if (imu->acct_pages) | |
8131 | io_unaccount_mem(ctx, imu->acct_pages); | |
8132 | kvfree(imu); | |
8133 | } | |
41edf1a5 | 8134 | *slot = NULL; |
7f61a1e9 PB |
8135 | } |
8136 | ||
bd54b6fe | 8137 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8138 | { |
634d00df PB |
8139 | io_buffer_unmap(ctx, &prsrc->buf); |
8140 | prsrc->buf = NULL; | |
bd54b6fe | 8141 | } |
edafccee | 8142 | |
bd54b6fe BM |
8143 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8144 | { | |
8145 | unsigned int i; | |
edafccee | 8146 | |
7f61a1e9 PB |
8147 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8148 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8149 | kfree(ctx->user_bufs); |
bb6659cc | 8150 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8151 | ctx->user_bufs = NULL; |
bd54b6fe | 8152 | ctx->buf_data = NULL; |
edafccee | 8153 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8154 | } |
8155 | ||
0a96bbe4 | 8156 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8157 | { |
bd54b6fe | 8158 | int ret; |
edafccee | 8159 | |
bd54b6fe | 8160 | if (!ctx->buf_data) |
edafccee JA |
8161 | return -ENXIO; |
8162 | ||
bd54b6fe BM |
8163 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8164 | if (!ret) | |
8165 | __io_sqe_buffers_unregister(ctx); | |
8166 | return ret; | |
edafccee JA |
8167 | } |
8168 | ||
8169 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8170 | void __user *arg, unsigned index) | |
8171 | { | |
8172 | struct iovec __user *src; | |
8173 | ||
8174 | #ifdef CONFIG_COMPAT | |
8175 | if (ctx->compat) { | |
8176 | struct compat_iovec __user *ciovs; | |
8177 | struct compat_iovec ciov; | |
8178 | ||
8179 | ciovs = (struct compat_iovec __user *) arg; | |
8180 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8181 | return -EFAULT; | |
8182 | ||
d55e5f5b | 8183 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8184 | dst->iov_len = ciov.iov_len; |
8185 | return 0; | |
8186 | } | |
8187 | #endif | |
8188 | src = (struct iovec __user *) arg; | |
8189 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8190 | return -EFAULT; | |
8191 | return 0; | |
8192 | } | |
8193 | ||
de293938 JA |
8194 | /* |
8195 | * Not super efficient, but this is just a registration time. And we do cache | |
8196 | * the last compound head, so generally we'll only do a full search if we don't | |
8197 | * match that one. | |
8198 | * | |
8199 | * We check if the given compound head page has already been accounted, to | |
8200 | * avoid double accounting it. This allows us to account the full size of the | |
8201 | * page, not just the constituent pages of a huge page. | |
8202 | */ | |
8203 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8204 | int nr_pages, struct page *hpage) | |
8205 | { | |
8206 | int i, j; | |
8207 | ||
8208 | /* check current page array */ | |
8209 | for (i = 0; i < nr_pages; i++) { | |
8210 | if (!PageCompound(pages[i])) | |
8211 | continue; | |
8212 | if (compound_head(pages[i]) == hpage) | |
8213 | return true; | |
8214 | } | |
8215 | ||
8216 | /* check previously registered pages */ | |
8217 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8218 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8219 | |
8220 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8221 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8222 | continue; | |
8223 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8224 | return true; | |
8225 | } | |
8226 | } | |
8227 | ||
8228 | return false; | |
8229 | } | |
8230 | ||
8231 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8232 | int nr_pages, struct io_mapped_ubuf *imu, | |
8233 | struct page **last_hpage) | |
8234 | { | |
8235 | int i, ret; | |
8236 | ||
216e5835 | 8237 | imu->acct_pages = 0; |
de293938 JA |
8238 | for (i = 0; i < nr_pages; i++) { |
8239 | if (!PageCompound(pages[i])) { | |
8240 | imu->acct_pages++; | |
8241 | } else { | |
8242 | struct page *hpage; | |
8243 | ||
8244 | hpage = compound_head(pages[i]); | |
8245 | if (hpage == *last_hpage) | |
8246 | continue; | |
8247 | *last_hpage = hpage; | |
8248 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8249 | continue; | |
8250 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8251 | } | |
8252 | } | |
8253 | ||
8254 | if (!imu->acct_pages) | |
8255 | return 0; | |
8256 | ||
26bfa89e | 8257 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8258 | if (ret) |
8259 | imu->acct_pages = 0; | |
8260 | return ret; | |
8261 | } | |
8262 | ||
0a96bbe4 | 8263 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8264 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8265 | struct page **last_hpage) |
edafccee | 8266 | { |
41edf1a5 | 8267 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8268 | struct vm_area_struct **vmas = NULL; |
8269 | struct page **pages = NULL; | |
0a96bbe4 BM |
8270 | unsigned long off, start, end, ubuf; |
8271 | size_t size; | |
8272 | int ret, pret, nr_pages, i; | |
8273 | ||
6224843d PB |
8274 | if (!iov->iov_base) { |
8275 | *pimu = ctx->dummy_ubuf; | |
8276 | return 0; | |
8277 | } | |
8278 | ||
0a96bbe4 BM |
8279 | ubuf = (unsigned long) iov->iov_base; |
8280 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8281 | start = ubuf >> PAGE_SHIFT; | |
8282 | nr_pages = end - start; | |
8283 | ||
41edf1a5 | 8284 | *pimu = NULL; |
0a96bbe4 BM |
8285 | ret = -ENOMEM; |
8286 | ||
8287 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8288 | if (!pages) | |
8289 | goto done; | |
8290 | ||
8291 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8292 | GFP_KERNEL); | |
8293 | if (!vmas) | |
8294 | goto done; | |
edafccee | 8295 | |
41edf1a5 | 8296 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8297 | if (!imu) |
0a96bbe4 BM |
8298 | goto done; |
8299 | ||
8300 | ret = 0; | |
8301 | mmap_read_lock(current->mm); | |
8302 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8303 | pages, vmas); | |
8304 | if (pret == nr_pages) { | |
8305 | /* don't support file backed memory */ | |
8306 | for (i = 0; i < nr_pages; i++) { | |
8307 | struct vm_area_struct *vma = vmas[i]; | |
8308 | ||
8309 | if (vma->vm_file && | |
8310 | !is_file_hugepages(vma->vm_file)) { | |
8311 | ret = -EOPNOTSUPP; | |
8312 | break; | |
8313 | } | |
8314 | } | |
8315 | } else { | |
8316 | ret = pret < 0 ? pret : -EFAULT; | |
8317 | } | |
8318 | mmap_read_unlock(current->mm); | |
8319 | if (ret) { | |
8320 | /* | |
8321 | * if we did partial map, or found file backed vmas, | |
8322 | * release any pages we did get | |
8323 | */ | |
8324 | if (pret > 0) | |
8325 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8326 | goto done; |
8327 | } | |
8328 | ||
8329 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8330 | if (ret) { | |
8331 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8332 | goto done; |
8333 | } | |
8334 | ||
8335 | off = ubuf & ~PAGE_MASK; | |
8336 | size = iov->iov_len; | |
8337 | for (i = 0; i < nr_pages; i++) { | |
8338 | size_t vec_len; | |
8339 | ||
8340 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8341 | imu->bvec[i].bv_page = pages[i]; | |
8342 | imu->bvec[i].bv_len = vec_len; | |
8343 | imu->bvec[i].bv_offset = off; | |
8344 | off = 0; | |
8345 | size -= vec_len; | |
8346 | } | |
8347 | /* store original address for later verification */ | |
8348 | imu->ubuf = ubuf; | |
4751f53d | 8349 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8350 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8351 | *pimu = imu; |
0a96bbe4 BM |
8352 | ret = 0; |
8353 | done: | |
41edf1a5 PB |
8354 | if (ret) |
8355 | kvfree(imu); | |
0a96bbe4 BM |
8356 | kvfree(pages); |
8357 | kvfree(vmas); | |
8358 | return ret; | |
8359 | } | |
8360 | ||
2b358604 | 8361 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8362 | { |
87094465 PB |
8363 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8364 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8365 | } |
edafccee | 8366 | |
2b358604 BM |
8367 | static int io_buffer_validate(struct iovec *iov) |
8368 | { | |
50e96989 PB |
8369 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8370 | ||
2b358604 BM |
8371 | /* |
8372 | * Don't impose further limits on the size and buffer | |
8373 | * constraints here, we'll -EINVAL later when IO is | |
8374 | * submitted if they are wrong. | |
8375 | */ | |
6224843d PB |
8376 | if (!iov->iov_base) |
8377 | return iov->iov_len ? -EFAULT : 0; | |
8378 | if (!iov->iov_len) | |
2b358604 | 8379 | return -EFAULT; |
edafccee | 8380 | |
2b358604 BM |
8381 | /* arbitrary limit, but we need something */ |
8382 | if (iov->iov_len > SZ_1G) | |
8383 | return -EFAULT; | |
edafccee | 8384 | |
50e96989 PB |
8385 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8386 | return -EOVERFLOW; | |
8387 | ||
2b358604 BM |
8388 | return 0; |
8389 | } | |
edafccee | 8390 | |
2b358604 | 8391 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 8392 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 8393 | { |
bd54b6fe BM |
8394 | struct page *last_hpage = NULL; |
8395 | struct io_rsrc_data *data; | |
2b358604 BM |
8396 | int i, ret; |
8397 | struct iovec iov; | |
edafccee | 8398 | |
87094465 PB |
8399 | if (ctx->user_bufs) |
8400 | return -EBUSY; | |
489809e2 | 8401 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 8402 | return -EINVAL; |
bd54b6fe | 8403 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8404 | if (ret) |
8405 | return ret; | |
bd54b6fe BM |
8406 | data = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, nr_args); |
8407 | if (!data) | |
8408 | return -ENOMEM; | |
8409 | ret = io_buffers_map_alloc(ctx, nr_args); | |
8410 | if (ret) { | |
bb6659cc | 8411 | io_rsrc_data_free(data); |
bd54b6fe BM |
8412 | return ret; |
8413 | } | |
edafccee | 8414 | |
87094465 | 8415 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
634d00df | 8416 | u64 tag = 0; |
edafccee | 8417 | |
634d00df PB |
8418 | if (tags && copy_from_user(&tag, &tags[i], sizeof(tag))) { |
8419 | ret = -EFAULT; | |
8420 | break; | |
8421 | } | |
edafccee JA |
8422 | ret = io_copy_iov(ctx, &iov, arg, i); |
8423 | if (ret) | |
0a96bbe4 | 8424 | break; |
2b358604 BM |
8425 | ret = io_buffer_validate(&iov); |
8426 | if (ret) | |
0a96bbe4 | 8427 | break; |
cf3770e7 CIK |
8428 | if (!iov.iov_base && tag) { |
8429 | ret = -EINVAL; | |
8430 | break; | |
8431 | } | |
edafccee | 8432 | |
41edf1a5 PB |
8433 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
8434 | &last_hpage); | |
0a96bbe4 BM |
8435 | if (ret) |
8436 | break; | |
634d00df | 8437 | data->tags[i] = tag; |
edafccee | 8438 | } |
0a96bbe4 | 8439 | |
bd54b6fe | 8440 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8441 | |
bd54b6fe BM |
8442 | ctx->buf_data = data; |
8443 | if (ret) | |
8444 | __io_sqe_buffers_unregister(ctx); | |
8445 | else | |
8446 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
8447 | return ret; |
8448 | } | |
8449 | ||
634d00df PB |
8450 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
8451 | struct io_uring_rsrc_update2 *up, | |
8452 | unsigned int nr_args) | |
8453 | { | |
8454 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
8455 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
8456 | struct page *last_hpage = NULL; |
8457 | bool needs_switch = false; | |
8458 | __u32 done; | |
8459 | int i, err; | |
8460 | ||
8461 | if (!ctx->buf_data) | |
8462 | return -ENXIO; | |
8463 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
8464 | return -EINVAL; | |
8465 | ||
8466 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
8467 | struct io_mapped_ubuf *imu; |
8468 | int offset = up->offset + done; | |
634d00df PB |
8469 | u64 tag = 0; |
8470 | ||
8471 | err = io_copy_iov(ctx, &iov, iovs, done); | |
8472 | if (err) | |
8473 | break; | |
8474 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
8475 | err = -EFAULT; | |
8476 | break; | |
8477 | } | |
0b8c0e7c PB |
8478 | err = io_buffer_validate(&iov); |
8479 | if (err) | |
8480 | break; | |
cf3770e7 CIK |
8481 | if (!iov.iov_base && tag) { |
8482 | err = -EINVAL; | |
8483 | break; | |
8484 | } | |
0b8c0e7c PB |
8485 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
8486 | if (err) | |
8487 | break; | |
634d00df | 8488 | |
0b8c0e7c | 8489 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 8490 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
8491 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
8492 | ctx->rsrc_node, ctx->user_bufs[i]); | |
8493 | if (unlikely(err)) { | |
8494 | io_buffer_unmap(ctx, &imu); | |
634d00df | 8495 | break; |
0b8c0e7c | 8496 | } |
634d00df PB |
8497 | ctx->user_bufs[i] = NULL; |
8498 | needs_switch = true; | |
8499 | } | |
8500 | ||
0b8c0e7c PB |
8501 | ctx->user_bufs[i] = imu; |
8502 | ctx->buf_data->tags[offset] = tag; | |
634d00df PB |
8503 | } |
8504 | ||
8505 | if (needs_switch) | |
8506 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
8507 | return done ? done : err; | |
8508 | } | |
8509 | ||
9b402849 JA |
8510 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8511 | { | |
8512 | __s32 __user *fds = arg; | |
8513 | int fd; | |
8514 | ||
8515 | if (ctx->cq_ev_fd) | |
8516 | return -EBUSY; | |
8517 | ||
8518 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8519 | return -EFAULT; | |
8520 | ||
8521 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8522 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8523 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8524 | ctx->cq_ev_fd = NULL; | |
8525 | return ret; | |
8526 | } | |
8527 | ||
8528 | return 0; | |
8529 | } | |
8530 | ||
8531 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8532 | { | |
8533 | if (ctx->cq_ev_fd) { | |
8534 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8535 | ctx->cq_ev_fd = NULL; | |
8536 | return 0; | |
8537 | } | |
8538 | ||
8539 | return -ENXIO; | |
8540 | } | |
8541 | ||
5a2e745d JA |
8542 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8543 | { | |
9e15c3a0 JA |
8544 | struct io_buffer *buf; |
8545 | unsigned long index; | |
8546 | ||
8547 | xa_for_each(&ctx->io_buffers, index, buf) | |
8548 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8549 | } |
8550 | ||
68e68ee6 | 8551 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8552 | { |
68e68ee6 | 8553 | struct io_kiocb *req, *nxt; |
1b4c351f | 8554 | |
68e68ee6 JA |
8555 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8556 | if (tsk && req->task != tsk) | |
8557 | continue; | |
1b4c351f JA |
8558 | list_del(&req->compl.list); |
8559 | kmem_cache_free(req_cachep, req); | |
8560 | } | |
8561 | } | |
8562 | ||
4010fec4 | 8563 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8564 | { |
bf019da7 | 8565 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8566 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8567 | |
9a4fdbd8 JA |
8568 | mutex_lock(&ctx->uring_lock); |
8569 | ||
8e5c66c4 | 8570 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8571 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8572 | submit_state->reqs); | |
8e5c66c4 PB |
8573 | submit_state->free_reqs = 0; |
8574 | } | |
9a4fdbd8 | 8575 | |
dac7a098 | 8576 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8577 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8578 | mutex_unlock(&ctx->uring_lock); |
8579 | } | |
8580 | ||
bd54b6fe | 8581 | static bool io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 8582 | { |
bd54b6fe BM |
8583 | if (!data) |
8584 | return false; | |
8585 | if (!atomic_dec_and_test(&data->refs)) | |
8586 | wait_for_completion(&data->done); | |
8587 | return true; | |
8588 | } | |
04fc6c80 | 8589 | |
2b188cc1 JA |
8590 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8591 | { | |
37d1e2e3 | 8592 | io_sq_thread_finish(ctx); |
2aede0e4 | 8593 | |
37d1e2e3 | 8594 | if (ctx->mm_account) { |
2aede0e4 JA |
8595 | mmdrop(ctx->mm_account); |
8596 | ctx->mm_account = NULL; | |
30975825 | 8597 | } |
def596e9 | 8598 | |
8bad28d8 | 8599 | mutex_lock(&ctx->uring_lock); |
bd54b6fe BM |
8600 | if (io_wait_rsrc_data(ctx->buf_data)) |
8601 | __io_sqe_buffers_unregister(ctx); | |
8602 | if (io_wait_rsrc_data(ctx->file_data)) | |
08480400 | 8603 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
8604 | if (ctx->rings) |
8605 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8606 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8607 | io_eventfd_unregister(ctx); |
5a2e745d | 8608 | io_destroy_buffers(ctx); |
07db298a PB |
8609 | if (ctx->sq_creds) |
8610 | put_cred(ctx->sq_creds); | |
def596e9 | 8611 | |
a7f0ed5a PB |
8612 | /* there are no registered resources left, nobody uses it */ |
8613 | if (ctx->rsrc_node) | |
8614 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8615 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8616 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8617 | flush_delayed_work(&ctx->rsrc_put_work); |
8618 | ||
8619 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8620 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 8621 | |
2b188cc1 | 8622 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8623 | if (ctx->ring_sock) { |
8624 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8625 | sock_release(ctx->ring_sock); |
355e8d26 | 8626 | } |
2b188cc1 JA |
8627 | #endif |
8628 | ||
75b28aff | 8629 | io_mem_free(ctx->rings); |
2b188cc1 | 8630 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8631 | |
8632 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8633 | free_uid(ctx->user); |
4010fec4 | 8634 | io_req_caches_free(ctx); |
e941894e JA |
8635 | if (ctx->hash_map) |
8636 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8637 | kfree(ctx->cancel_hash); |
6224843d | 8638 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
8639 | kfree(ctx); |
8640 | } | |
8641 | ||
8642 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8643 | { | |
8644 | struct io_ring_ctx *ctx = file->private_data; | |
8645 | __poll_t mask = 0; | |
8646 | ||
8647 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8648 | /* |
8649 | * synchronizes with barrier from wq_has_sleeper call in | |
8650 | * io_commit_cqring | |
8651 | */ | |
2b188cc1 | 8652 | smp_rmb(); |
90554200 | 8653 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8654 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8655 | |
8656 | /* | |
8657 | * Don't flush cqring overflow list here, just do a simple check. | |
8658 | * Otherwise there could possible be ABBA deadlock: | |
8659 | * CPU0 CPU1 | |
8660 | * ---- ---- | |
8661 | * lock(&ctx->uring_lock); | |
8662 | * lock(&ep->mtx); | |
8663 | * lock(&ctx->uring_lock); | |
8664 | * lock(&ep->mtx); | |
8665 | * | |
8666 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8667 | * pushs them to do the flush. | |
8668 | */ | |
8669 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8670 | mask |= EPOLLIN | EPOLLRDNORM; |
8671 | ||
8672 | return mask; | |
8673 | } | |
8674 | ||
8675 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8676 | { | |
8677 | struct io_ring_ctx *ctx = file->private_data; | |
8678 | ||
8679 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8680 | } | |
8681 | ||
0bead8cd | 8682 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8683 | { |
4379bf8b | 8684 | const struct cred *creds; |
071698e1 | 8685 | |
61cf9370 | 8686 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8687 | if (creds) { |
8688 | put_cred(creds); | |
0bead8cd | 8689 | return 0; |
1e6fa521 | 8690 | } |
0bead8cd YD |
8691 | |
8692 | return -EINVAL; | |
8693 | } | |
8694 | ||
9b465711 | 8695 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8696 | { |
9b465711 | 8697 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8698 | } |
8699 | ||
d56d938b PB |
8700 | struct io_tctx_exit { |
8701 | struct callback_head task_work; | |
8702 | struct completion completion; | |
baf186c4 | 8703 | struct io_ring_ctx *ctx; |
d56d938b PB |
8704 | }; |
8705 | ||
8706 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8707 | { | |
8708 | struct io_uring_task *tctx = current->io_uring; | |
8709 | struct io_tctx_exit *work; | |
8710 | ||
8711 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8712 | /* | |
8713 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8714 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8715 | */ | |
8716 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8717 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8718 | complete(&work->completion); |
8719 | } | |
8720 | ||
28090c13 PB |
8721 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8722 | { | |
8723 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8724 | ||
8725 | return req->ctx == data; | |
8726 | } | |
8727 | ||
85faa7b8 JA |
8728 | static void io_ring_exit_work(struct work_struct *work) |
8729 | { | |
d56d938b | 8730 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8731 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8732 | struct io_tctx_exit exit; |
8733 | struct io_tctx_node *node; | |
8734 | int ret; | |
85faa7b8 | 8735 | |
56952e91 JA |
8736 | /* |
8737 | * If we're doing polled IO and end up having requests being | |
8738 | * submitted async (out-of-line), then completions can come in while | |
8739 | * we're waiting for refs to drop. We need to reap these manually, | |
8740 | * as nobody else will be looking for them. | |
8741 | */ | |
b2edc0a7 | 8742 | do { |
3dd0c97a | 8743 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
8744 | if (ctx->sq_data) { |
8745 | struct io_sq_data *sqd = ctx->sq_data; | |
8746 | struct task_struct *tsk; | |
8747 | ||
8748 | io_sq_thread_park(sqd); | |
8749 | tsk = sqd->thread; | |
8750 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
8751 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
8752 | io_cancel_ctx_cb, ctx, true); | |
8753 | io_sq_thread_unpark(sqd); | |
8754 | } | |
b5bb3a24 PB |
8755 | |
8756 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8757 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8758 | |
7f00651a PB |
8759 | init_completion(&exit.completion); |
8760 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8761 | exit.ctx = ctx; | |
89b5066e PB |
8762 | /* |
8763 | * Some may use context even when all refs and requests have been put, | |
8764 | * and they are free to do so while still holding uring_lock or | |
8765 | * completion_lock, see __io_req_task_submit(). Apart from other work, | |
8766 | * this lock/unlock section also waits them to finish. | |
8767 | */ | |
d56d938b PB |
8768 | mutex_lock(&ctx->uring_lock); |
8769 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8770 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8771 | ||
d56d938b PB |
8772 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8773 | ctx_node); | |
7f00651a PB |
8774 | /* don't spin on a single task if cancellation failed */ |
8775 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8776 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8777 | if (WARN_ON_ONCE(ret)) | |
8778 | continue; | |
8779 | wake_up_process(node->task); | |
8780 | ||
8781 | mutex_unlock(&ctx->uring_lock); | |
8782 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8783 | mutex_lock(&ctx->uring_lock); |
8784 | } | |
8785 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8786 | spin_lock_irq(&ctx->completion_lock); |
8787 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8788 | |
85faa7b8 JA |
8789 | io_ring_ctx_free(ctx); |
8790 | } | |
8791 | ||
80c4cbdb PB |
8792 | /* Returns true if we found and killed one or more timeouts */ |
8793 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 8794 | bool cancel_all) |
80c4cbdb PB |
8795 | { |
8796 | struct io_kiocb *req, *tmp; | |
8797 | int canceled = 0; | |
8798 | ||
8799 | spin_lock_irq(&ctx->completion_lock); | |
8800 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
3dd0c97a | 8801 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
8802 | io_kill_timeout(req, -ECANCELED); |
8803 | canceled++; | |
8804 | } | |
8805 | } | |
51520426 PB |
8806 | if (canceled != 0) |
8807 | io_commit_cqring(ctx); | |
80c4cbdb | 8808 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8809 | if (canceled != 0) |
8810 | io_cqring_ev_posted(ctx); | |
8811 | return canceled != 0; | |
8812 | } | |
8813 | ||
2b188cc1 JA |
8814 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8815 | { | |
61cf9370 MWO |
8816 | unsigned long index; |
8817 | struct creds *creds; | |
8818 | ||
2b188cc1 JA |
8819 | mutex_lock(&ctx->uring_lock); |
8820 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8821 | if (ctx->rings) |
6c2450ae | 8822 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8823 | xa_for_each(&ctx->personalities, index, creds) |
8824 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8825 | mutex_unlock(&ctx->uring_lock); |
8826 | ||
3dd0c97a PB |
8827 | io_kill_timeouts(ctx, NULL, true); |
8828 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 8829 | |
15dff286 | 8830 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8831 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8832 | |
85faa7b8 | 8833 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8834 | /* |
8835 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8836 | * if we're exiting a ton of rings at the same time. It just adds | |
8837 | * noise and overhead, there's no discernable change in runtime | |
8838 | * over using system_wq. | |
8839 | */ | |
8840 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8841 | } |
8842 | ||
8843 | static int io_uring_release(struct inode *inode, struct file *file) | |
8844 | { | |
8845 | struct io_ring_ctx *ctx = file->private_data; | |
8846 | ||
8847 | file->private_data = NULL; | |
8848 | io_ring_ctx_wait_and_kill(ctx); | |
8849 | return 0; | |
8850 | } | |
8851 | ||
f6edbabb PB |
8852 | struct io_task_cancel { |
8853 | struct task_struct *task; | |
3dd0c97a | 8854 | bool all; |
f6edbabb | 8855 | }; |
f254ac04 | 8856 | |
f6edbabb | 8857 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8858 | { |
9a472ef7 | 8859 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8860 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8861 | bool ret; |
8862 | ||
3dd0c97a | 8863 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8864 | unsigned long flags; |
8865 | struct io_ring_ctx *ctx = req->ctx; | |
8866 | ||
8867 | /* protect against races with linked timeouts */ | |
8868 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
3dd0c97a | 8869 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8870 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8871 | } else { | |
3dd0c97a | 8872 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8873 | } |
8874 | return ret; | |
b711d4ea JA |
8875 | } |
8876 | ||
e1915f76 | 8877 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 8878 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 8879 | { |
e1915f76 | 8880 | struct io_defer_entry *de; |
b7ddce3c PB |
8881 | LIST_HEAD(list); |
8882 | ||
8883 | spin_lock_irq(&ctx->completion_lock); | |
8884 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
3dd0c97a | 8885 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
8886 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8887 | break; | |
8888 | } | |
8889 | } | |
8890 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8891 | if (list_empty(&list)) |
8892 | return false; | |
b7ddce3c PB |
8893 | |
8894 | while (!list_empty(&list)) { | |
8895 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8896 | list_del_init(&de->list); | |
f41db273 | 8897 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8898 | kfree(de); |
8899 | } | |
e1915f76 | 8900 | return true; |
b7ddce3c PB |
8901 | } |
8902 | ||
1b00764f PB |
8903 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
8904 | { | |
8905 | struct io_tctx_node *node; | |
8906 | enum io_wq_cancel cret; | |
8907 | bool ret = false; | |
8908 | ||
8909 | mutex_lock(&ctx->uring_lock); | |
8910 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8911 | struct io_uring_task *tctx = node->task->io_uring; | |
8912 | ||
8913 | /* | |
8914 | * io_wq will stay alive while we hold uring_lock, because it's | |
8915 | * killed after ctx nodes, which requires to take the lock. | |
8916 | */ | |
8917 | if (!tctx || !tctx->io_wq) | |
8918 | continue; | |
8919 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8920 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8921 | } | |
8922 | mutex_unlock(&ctx->uring_lock); | |
8923 | ||
8924 | return ret; | |
8925 | } | |
8926 | ||
9936c7c2 PB |
8927 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8928 | struct task_struct *task, | |
3dd0c97a | 8929 | bool cancel_all) |
9936c7c2 | 8930 | { |
3dd0c97a | 8931 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 8932 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8933 | |
8934 | while (1) { | |
8935 | enum io_wq_cancel cret; | |
8936 | bool ret = false; | |
8937 | ||
1b00764f PB |
8938 | if (!task) { |
8939 | ret |= io_uring_try_cancel_iowq(ctx); | |
8940 | } else if (tctx && tctx->io_wq) { | |
8941 | /* | |
8942 | * Cancels requests of all rings, not only @ctx, but | |
8943 | * it's fine as the task is in exit/exec. | |
8944 | */ | |
5aa75ed5 | 8945 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8946 | &cancel, true); |
8947 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8948 | } | |
8949 | ||
8950 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 8951 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 8952 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
8953 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8954 | io_iopoll_try_reap_events(ctx); | |
8955 | ret = true; | |
8956 | } | |
8957 | } | |
8958 | ||
3dd0c97a PB |
8959 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
8960 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
8961 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
9936c7c2 | 8962 | ret |= io_run_task_work(); |
ba50a036 | 8963 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8964 | if (!ret) |
8965 | break; | |
8966 | cond_resched(); | |
8967 | } | |
8968 | } | |
8969 | ||
cf27f3b1 | 8970 | static int __io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8971 | { |
236434c3 | 8972 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8973 | struct io_tctx_node *node; |
a528b04e | 8974 | int ret; |
236434c3 MWO |
8975 | |
8976 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8977 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8978 | if (unlikely(ret)) |
8979 | return ret; | |
236434c3 | 8980 | tctx = current->io_uring; |
0f212204 | 8981 | } |
cf27f3b1 PB |
8982 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
8983 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
8984 | if (!node) | |
8985 | return -ENOMEM; | |
8986 | node->ctx = ctx; | |
8987 | node->task = current; | |
13bf43f5 | 8988 | |
cf27f3b1 PB |
8989 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
8990 | node, GFP_KERNEL)); | |
8991 | if (ret) { | |
8992 | kfree(node); | |
8993 | return ret; | |
0f212204 | 8994 | } |
cf27f3b1 PB |
8995 | |
8996 | mutex_lock(&ctx->uring_lock); | |
8997 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8998 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8999 | } |
cf27f3b1 | 9000 | tctx->last = ctx; |
0f212204 JA |
9001 | return 0; |
9002 | } | |
9003 | ||
cf27f3b1 PB |
9004 | /* |
9005 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9006 | */ | |
9007 | static inline int io_uring_add_task_file(struct io_ring_ctx *ctx) | |
9008 | { | |
9009 | struct io_uring_task *tctx = current->io_uring; | |
9010 | ||
9011 | if (likely(tctx && tctx->last == ctx)) | |
9012 | return 0; | |
9013 | return __io_uring_add_task_file(ctx); | |
9014 | } | |
9015 | ||
0f212204 JA |
9016 | /* |
9017 | * Remove this io_uring_file -> task mapping. | |
9018 | */ | |
2941267b | 9019 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
9020 | { |
9021 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9022 | struct io_tctx_node *node; |
2941267b | 9023 | |
eebd2e37 PB |
9024 | if (!tctx) |
9025 | return; | |
13bf43f5 PB |
9026 | node = xa_erase(&tctx->xa, index); |
9027 | if (!node) | |
2941267b | 9028 | return; |
0f212204 | 9029 | |
13bf43f5 PB |
9030 | WARN_ON_ONCE(current != node->task); |
9031 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9032 | ||
9033 | mutex_lock(&node->ctx->uring_lock); | |
9034 | list_del(&node->ctx_node); | |
9035 | mutex_unlock(&node->ctx->uring_lock); | |
9036 | ||
baf186c4 | 9037 | if (tctx->last == node->ctx) |
0f212204 | 9038 | tctx->last = NULL; |
13bf43f5 | 9039 | kfree(node); |
0f212204 JA |
9040 | } |
9041 | ||
8452d4a6 | 9042 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9043 | { |
ba5ef6dc | 9044 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9045 | struct io_tctx_node *node; |
de7f1d9e PB |
9046 | unsigned long index; |
9047 | ||
13bf43f5 | 9048 | xa_for_each(&tctx->xa, index, node) |
2941267b | 9049 | io_uring_del_task_file(index); |
b16ef427 ME |
9050 | if (wq) { |
9051 | /* | |
9052 | * Must be after io_uring_del_task_file() (removes nodes under | |
9053 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9054 | */ | |
9055 | tctx->io_wq = NULL; | |
ba5ef6dc | 9056 | io_wq_put_and_exit(wq); |
b16ef427 | 9057 | } |
de7f1d9e PB |
9058 | } |
9059 | ||
3f48cf18 | 9060 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9061 | { |
3f48cf18 PB |
9062 | if (tracked) |
9063 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9064 | return percpu_counter_sum(&tctx->inflight); |
9065 | } | |
9066 | ||
3dd0c97a | 9067 | static void io_uring_try_cancel(bool cancel_all) |
0f212204 JA |
9068 | { |
9069 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9070 | struct io_tctx_node *node; |
ce765372 | 9071 | unsigned long index; |
0f212204 | 9072 | |
521d6a73 PB |
9073 | xa_for_each(&tctx->xa, index, node) { |
9074 | struct io_ring_ctx *ctx = node->ctx; | |
9075 | ||
9f59a9d8 PB |
9076 | /* sqpoll task will cancel all its requests */ |
9077 | if (!ctx->sq_data) | |
3dd0c97a | 9078 | io_uring_try_cancel_requests(ctx, current, cancel_all); |
521d6a73 | 9079 | } |
fdaf083c JA |
9080 | } |
9081 | ||
521d6a73 | 9082 | /* should only be called by SQPOLL task */ |
734551df | 9083 | static void io_uring_cancel_sqpoll(struct io_sq_data *sqd) |
0e9ddb39 | 9084 | { |
521d6a73 | 9085 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9086 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9087 | s64 inflight; |
9088 | DEFINE_WAIT(wait); | |
fdaf083c | 9089 | |
6d042ffb PO |
9090 | if (!current->io_uring) |
9091 | return; | |
17a91051 PB |
9092 | if (tctx->io_wq) |
9093 | io_wq_exit_start(tctx->io_wq); | |
9094 | ||
734551df | 9095 | WARN_ON_ONCE(!sqd || sqd->thread != current); |
521d6a73 | 9096 | |
0e9ddb39 PB |
9097 | atomic_inc(&tctx->in_idle); |
9098 | do { | |
9099 | /* read completions before cancelations */ | |
3f48cf18 | 9100 | inflight = tctx_inflight(tctx, false); |
0e9ddb39 PB |
9101 | if (!inflight) |
9102 | break; | |
734551df | 9103 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
3dd0c97a | 9104 | io_uring_try_cancel_requests(ctx, current, true); |
fdaf083c | 9105 | |
0e9ddb39 PB |
9106 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9107 | /* | |
9108 | * If we've seen completions, retry without waiting. This | |
9109 | * avoids a race where a completion comes in before we did | |
9110 | * prepare_to_wait(). | |
9111 | */ | |
3f48cf18 | 9112 | if (inflight == tctx_inflight(tctx, false)) |
0e9ddb39 PB |
9113 | schedule(); |
9114 | finish_wait(&tctx->wait, &wait); | |
9115 | } while (1); | |
9116 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9117 | } |
9118 | ||
0f212204 JA |
9119 | /* |
9120 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9121 | * requests. | |
9122 | */ | |
3f48cf18 | 9123 | void __io_uring_cancel(struct files_struct *files) |
0f212204 JA |
9124 | { |
9125 | struct io_uring_task *tctx = current->io_uring; | |
9126 | DEFINE_WAIT(wait); | |
d8a6df10 | 9127 | s64 inflight; |
3dd0c97a | 9128 | bool cancel_all = !files; |
0f212204 | 9129 | |
17a91051 PB |
9130 | if (tctx->io_wq) |
9131 | io_wq_exit_start(tctx->io_wq); | |
9132 | ||
0f212204 | 9133 | /* make sure overflow events are dropped */ |
fdaf083c | 9134 | atomic_inc(&tctx->in_idle); |
d8a6df10 | 9135 | do { |
0f212204 | 9136 | /* read completions before cancelations */ |
3dd0c97a | 9137 | inflight = tctx_inflight(tctx, !cancel_all); |
d8a6df10 JA |
9138 | if (!inflight) |
9139 | break; | |
3dd0c97a | 9140 | io_uring_try_cancel(cancel_all); |
0f212204 JA |
9141 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9142 | ||
9143 | /* | |
a1bb3cd5 PB |
9144 | * If we've seen completions, retry without waiting. This |
9145 | * avoids a race where a completion comes in before we did | |
9146 | * prepare_to_wait(). | |
0f212204 | 9147 | */ |
3dd0c97a | 9148 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9149 | schedule(); |
f57555ed | 9150 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9151 | } while (1); |
fdaf083c | 9152 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9153 | |
8452d4a6 | 9154 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9155 | if (cancel_all) { |
3f48cf18 PB |
9156 | /* for exec all current's requests should be gone, kill tctx */ |
9157 | __io_uring_free(current); | |
9158 | } | |
44e728b8 PB |
9159 | } |
9160 | ||
6c5c240e RP |
9161 | static void *io_uring_validate_mmap_request(struct file *file, |
9162 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9163 | { |
2b188cc1 | 9164 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9165 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9166 | struct page *page; |
9167 | void *ptr; | |
9168 | ||
9169 | switch (offset) { | |
9170 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9171 | case IORING_OFF_CQ_RING: |
9172 | ptr = ctx->rings; | |
2b188cc1 JA |
9173 | break; |
9174 | case IORING_OFF_SQES: | |
9175 | ptr = ctx->sq_sqes; | |
9176 | break; | |
2b188cc1 | 9177 | default: |
6c5c240e | 9178 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9179 | } |
9180 | ||
9181 | page = virt_to_head_page(ptr); | |
a50b854e | 9182 | if (sz > page_size(page)) |
6c5c240e RP |
9183 | return ERR_PTR(-EINVAL); |
9184 | ||
9185 | return ptr; | |
9186 | } | |
9187 | ||
9188 | #ifdef CONFIG_MMU | |
9189 | ||
9190 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9191 | { | |
9192 | size_t sz = vma->vm_end - vma->vm_start; | |
9193 | unsigned long pfn; | |
9194 | void *ptr; | |
9195 | ||
9196 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9197 | if (IS_ERR(ptr)) | |
9198 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9199 | |
9200 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9201 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9202 | } | |
9203 | ||
6c5c240e RP |
9204 | #else /* !CONFIG_MMU */ |
9205 | ||
9206 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9207 | { | |
9208 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9209 | } | |
9210 | ||
9211 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9212 | { | |
9213 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9214 | } | |
9215 | ||
9216 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9217 | unsigned long addr, unsigned long len, | |
9218 | unsigned long pgoff, unsigned long flags) | |
9219 | { | |
9220 | void *ptr; | |
9221 | ||
9222 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9223 | if (IS_ERR(ptr)) | |
9224 | return PTR_ERR(ptr); | |
9225 | ||
9226 | return (unsigned long) ptr; | |
9227 | } | |
9228 | ||
9229 | #endif /* !CONFIG_MMU */ | |
9230 | ||
d9d05217 | 9231 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9232 | { |
9233 | DEFINE_WAIT(wait); | |
9234 | ||
9235 | do { | |
9236 | if (!io_sqring_full(ctx)) | |
9237 | break; | |
90554200 JA |
9238 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9239 | ||
9240 | if (!io_sqring_full(ctx)) | |
9241 | break; | |
90554200 JA |
9242 | schedule(); |
9243 | } while (!signal_pending(current)); | |
9244 | ||
9245 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9246 | return 0; |
90554200 JA |
9247 | } |
9248 | ||
c73ebb68 HX |
9249 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9250 | struct __kernel_timespec __user **ts, | |
9251 | const sigset_t __user **sig) | |
9252 | { | |
9253 | struct io_uring_getevents_arg arg; | |
9254 | ||
9255 | /* | |
9256 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9257 | * is just a pointer to the sigset_t. | |
9258 | */ | |
9259 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9260 | *sig = (const sigset_t __user *) argp; | |
9261 | *ts = NULL; | |
9262 | return 0; | |
9263 | } | |
9264 | ||
9265 | /* | |
9266 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9267 | * timespec and sigset_t pointers if good. | |
9268 | */ | |
9269 | if (*argsz != sizeof(arg)) | |
9270 | return -EINVAL; | |
9271 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9272 | return -EFAULT; | |
9273 | *sig = u64_to_user_ptr(arg.sigmask); | |
9274 | *argsz = arg.sigmask_sz; | |
9275 | *ts = u64_to_user_ptr(arg.ts); | |
9276 | return 0; | |
9277 | } | |
9278 | ||
2b188cc1 | 9279 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9280 | u32, min_complete, u32, flags, const void __user *, argp, |
9281 | size_t, argsz) | |
2b188cc1 JA |
9282 | { |
9283 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9284 | int submitted = 0; |
9285 | struct fd f; | |
33f993da | 9286 | long ret; |
2b188cc1 | 9287 | |
4c6e277c | 9288 | io_run_task_work(); |
b41e9852 | 9289 | |
33f993da PB |
9290 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9291 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9292 | return -EINVAL; |
9293 | ||
9294 | f = fdget(fd); | |
33f993da | 9295 | if (unlikely(!f.file)) |
2b188cc1 JA |
9296 | return -EBADF; |
9297 | ||
9298 | ret = -EOPNOTSUPP; | |
33f993da | 9299 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9300 | goto out_fput; |
9301 | ||
9302 | ret = -ENXIO; | |
9303 | ctx = f.file->private_data; | |
33f993da | 9304 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9305 | goto out_fput; |
9306 | ||
7e84e1c7 | 9307 | ret = -EBADFD; |
33f993da | 9308 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9309 | goto out; |
9310 | ||
6c271ce2 JA |
9311 | /* |
9312 | * For SQ polling, the thread will do all submissions and completions. | |
9313 | * Just return the requested submit count, and wake the thread if | |
9314 | * we were asked to. | |
9315 | */ | |
b2a9eada | 9316 | ret = 0; |
6c271ce2 | 9317 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9318 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9319 | |
d9d05217 | 9320 | ret = -EOWNERDEAD; |
04147488 SM |
9321 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9322 | goto out; | |
9323 | } | |
6c271ce2 | 9324 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9325 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9326 | if (flags & IORING_ENTER_SQ_WAIT) { |
9327 | ret = io_sqpoll_wait_sq(ctx); | |
9328 | if (ret) | |
9329 | goto out; | |
9330 | } | |
6c271ce2 | 9331 | submitted = to_submit; |
b2a9eada | 9332 | } else if (to_submit) { |
baf186c4 | 9333 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9334 | if (unlikely(ret)) |
9335 | goto out; | |
2b188cc1 | 9336 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9337 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9338 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9339 | |
9340 | if (submitted != to_submit) | |
9341 | goto out; | |
2b188cc1 JA |
9342 | } |
9343 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9344 | const sigset_t __user *sig; |
9345 | struct __kernel_timespec __user *ts; | |
9346 | ||
9347 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9348 | if (unlikely(ret)) | |
9349 | goto out; | |
9350 | ||
2b188cc1 JA |
9351 | min_complete = min(min_complete, ctx->cq_entries); |
9352 | ||
32b2244a XW |
9353 | /* |
9354 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9355 | * space applications don't need to do io completion events | |
9356 | * polling again, they can rely on io_sq_thread to do polling | |
9357 | * work, which can reduce cpu usage and uring_lock contention. | |
9358 | */ | |
9359 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9360 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9361 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9362 | } else { |
c73ebb68 | 9363 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9364 | } |
2b188cc1 JA |
9365 | } |
9366 | ||
7c504e65 | 9367 | out: |
6805b32e | 9368 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9369 | out_fput: |
9370 | fdput(f); | |
9371 | return submitted ? submitted : ret; | |
9372 | } | |
9373 | ||
bebdb65e | 9374 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9375 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9376 | const struct cred *cred) | |
87ce955b | 9377 | { |
87ce955b JA |
9378 | struct user_namespace *uns = seq_user_ns(m); |
9379 | struct group_info *gi; | |
9380 | kernel_cap_t cap; | |
9381 | unsigned __capi; | |
9382 | int g; | |
9383 | ||
9384 | seq_printf(m, "%5d\n", id); | |
9385 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9386 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9387 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9388 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9389 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9390 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9391 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9392 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9393 | seq_puts(m, "\n\tGroups:\t"); | |
9394 | gi = cred->group_info; | |
9395 | for (g = 0; g < gi->ngroups; g++) { | |
9396 | seq_put_decimal_ull(m, g ? " " : "", | |
9397 | from_kgid_munged(uns, gi->gid[g])); | |
9398 | } | |
9399 | seq_puts(m, "\n\tCapEff:\t"); | |
9400 | cap = cred->cap_effective; | |
9401 | CAP_FOR_EACH_U32(__capi) | |
9402 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9403 | seq_putc(m, '\n'); | |
9404 | return 0; | |
9405 | } | |
9406 | ||
9407 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9408 | { | |
dbbe9c64 | 9409 | struct io_sq_data *sq = NULL; |
fad8e0de | 9410 | bool has_lock; |
87ce955b JA |
9411 | int i; |
9412 | ||
fad8e0de JA |
9413 | /* |
9414 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9415 | * since fdinfo case grabs it in the opposite direction of normal use | |
9416 | * cases. If we fail to get the lock, we just don't iterate any | |
9417 | * structures that could be going away outside the io_uring mutex. | |
9418 | */ | |
9419 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9420 | ||
5f3f26f9 | 9421 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9422 | sq = ctx->sq_data; |
5f3f26f9 JA |
9423 | if (!sq->thread) |
9424 | sq = NULL; | |
9425 | } | |
dbbe9c64 JQ |
9426 | |
9427 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9428 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9429 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9430 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9431 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9432 | |
87ce955b JA |
9433 | if (f) |
9434 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9435 | else | |
9436 | seq_printf(m, "%5u: <none>\n", i); | |
9437 | } | |
9438 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9439 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9440 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9441 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9442 | |
4751f53d | 9443 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9444 | } |
61cf9370 MWO |
9445 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9446 | unsigned long index; | |
9447 | const struct cred *cred; | |
9448 | ||
87ce955b | 9449 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9450 | xa_for_each(&ctx->personalities, index, cred) |
9451 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9452 | } |
d7718a9d JA |
9453 | seq_printf(m, "PollList:\n"); |
9454 | spin_lock_irq(&ctx->completion_lock); | |
9455 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9456 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9457 | struct io_kiocb *req; | |
9458 | ||
9459 | hlist_for_each_entry(req, list, hash_node) | |
9460 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9461 | req->task->task_works != NULL); | |
9462 | } | |
9463 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9464 | if (has_lock) |
9465 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9466 | } |
9467 | ||
9468 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9469 | { | |
9470 | struct io_ring_ctx *ctx = f->private_data; | |
9471 | ||
9472 | if (percpu_ref_tryget(&ctx->refs)) { | |
9473 | __io_uring_show_fdinfo(ctx, m); | |
9474 | percpu_ref_put(&ctx->refs); | |
9475 | } | |
9476 | } | |
bebdb65e | 9477 | #endif |
87ce955b | 9478 | |
2b188cc1 JA |
9479 | static const struct file_operations io_uring_fops = { |
9480 | .release = io_uring_release, | |
9481 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9482 | #ifndef CONFIG_MMU |
9483 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9484 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9485 | #endif | |
2b188cc1 JA |
9486 | .poll = io_uring_poll, |
9487 | .fasync = io_uring_fasync, | |
bebdb65e | 9488 | #ifdef CONFIG_PROC_FS |
87ce955b | 9489 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9490 | #endif |
2b188cc1 JA |
9491 | }; |
9492 | ||
9493 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9494 | struct io_uring_params *p) | |
9495 | { | |
75b28aff HV |
9496 | struct io_rings *rings; |
9497 | size_t size, sq_array_offset; | |
2b188cc1 | 9498 | |
bd740481 JA |
9499 | /* make sure these are sane, as we already accounted them */ |
9500 | ctx->sq_entries = p->sq_entries; | |
9501 | ctx->cq_entries = p->cq_entries; | |
9502 | ||
75b28aff HV |
9503 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9504 | if (size == SIZE_MAX) | |
9505 | return -EOVERFLOW; | |
9506 | ||
9507 | rings = io_mem_alloc(size); | |
9508 | if (!rings) | |
2b188cc1 JA |
9509 | return -ENOMEM; |
9510 | ||
75b28aff HV |
9511 | ctx->rings = rings; |
9512 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9513 | rings->sq_ring_mask = p->sq_entries - 1; | |
9514 | rings->cq_ring_mask = p->cq_entries - 1; | |
9515 | rings->sq_ring_entries = p->sq_entries; | |
9516 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
9517 | |
9518 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9519 | if (size == SIZE_MAX) { |
9520 | io_mem_free(ctx->rings); | |
9521 | ctx->rings = NULL; | |
2b188cc1 | 9522 | return -EOVERFLOW; |
eb065d30 | 9523 | } |
2b188cc1 JA |
9524 | |
9525 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9526 | if (!ctx->sq_sqes) { |
9527 | io_mem_free(ctx->rings); | |
9528 | ctx->rings = NULL; | |
2b188cc1 | 9529 | return -ENOMEM; |
eb065d30 | 9530 | } |
2b188cc1 | 9531 | |
2b188cc1 JA |
9532 | return 0; |
9533 | } | |
9534 | ||
9faadcc8 PB |
9535 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9536 | { | |
9537 | int ret, fd; | |
9538 | ||
9539 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9540 | if (fd < 0) | |
9541 | return fd; | |
9542 | ||
baf186c4 | 9543 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9544 | if (ret) { |
9545 | put_unused_fd(fd); | |
9546 | return ret; | |
9547 | } | |
9548 | fd_install(fd, file); | |
9549 | return fd; | |
9550 | } | |
9551 | ||
2b188cc1 JA |
9552 | /* |
9553 | * Allocate an anonymous fd, this is what constitutes the application | |
9554 | * visible backing of an io_uring instance. The application mmaps this | |
9555 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9556 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9557 | */ | |
9faadcc8 | 9558 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9559 | { |
9560 | struct file *file; | |
9faadcc8 | 9561 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9562 | int ret; |
9563 | ||
2b188cc1 JA |
9564 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9565 | &ctx->ring_sock); | |
9566 | if (ret) | |
9faadcc8 | 9567 | return ERR_PTR(ret); |
2b188cc1 JA |
9568 | #endif |
9569 | ||
2b188cc1 JA |
9570 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9571 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9572 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9573 | if (IS_ERR(file)) { |
9574 | sock_release(ctx->ring_sock); | |
9575 | ctx->ring_sock = NULL; | |
9576 | } else { | |
9577 | ctx->ring_sock->file = file; | |
0f212204 | 9578 | } |
2b188cc1 | 9579 | #endif |
9faadcc8 | 9580 | return file; |
2b188cc1 JA |
9581 | } |
9582 | ||
7f13657d XW |
9583 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9584 | struct io_uring_params __user *params) | |
2b188cc1 | 9585 | { |
2b188cc1 | 9586 | struct io_ring_ctx *ctx; |
9faadcc8 | 9587 | struct file *file; |
2b188cc1 JA |
9588 | int ret; |
9589 | ||
8110c1a6 | 9590 | if (!entries) |
2b188cc1 | 9591 | return -EINVAL; |
8110c1a6 JA |
9592 | if (entries > IORING_MAX_ENTRIES) { |
9593 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9594 | return -EINVAL; | |
9595 | entries = IORING_MAX_ENTRIES; | |
9596 | } | |
2b188cc1 JA |
9597 | |
9598 | /* | |
9599 | * Use twice as many entries for the CQ ring. It's possible for the | |
9600 | * application to drive a higher depth than the size of the SQ ring, | |
9601 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9602 | * some flexibility in overcommitting a bit. If the application has |
9603 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9604 | * of CQ ring entries manually. | |
2b188cc1 JA |
9605 | */ |
9606 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9607 | if (p->flags & IORING_SETUP_CQSIZE) { |
9608 | /* | |
9609 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9610 | * to a power-of-two, if it isn't already. We do NOT impose | |
9611 | * any cq vs sq ring sizing. | |
9612 | */ | |
eb2667b3 | 9613 | if (!p->cq_entries) |
33a107f0 | 9614 | return -EINVAL; |
8110c1a6 JA |
9615 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9616 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9617 | return -EINVAL; | |
9618 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9619 | } | |
eb2667b3 JQ |
9620 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9621 | if (p->cq_entries < p->sq_entries) | |
9622 | return -EINVAL; | |
33a107f0 JA |
9623 | } else { |
9624 | p->cq_entries = 2 * p->sq_entries; | |
9625 | } | |
2b188cc1 | 9626 | |
2b188cc1 | 9627 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9628 | if (!ctx) |
2b188cc1 | 9629 | return -ENOMEM; |
2b188cc1 | 9630 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9631 | if (!capable(CAP_IPC_LOCK)) |
9632 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9633 | |
9634 | /* | |
9635 | * This is just grabbed for accounting purposes. When a process exits, | |
9636 | * the mm is exited and dropped before the files, hence we need to hang | |
9637 | * on to this mm purely for the purposes of being able to unaccount | |
9638 | * memory (locked/pinned vm). It's not used for anything else. | |
9639 | */ | |
6b7898eb | 9640 | mmgrab(current->mm); |
2aede0e4 | 9641 | ctx->mm_account = current->mm; |
6b7898eb | 9642 | |
2b188cc1 JA |
9643 | ret = io_allocate_scq_urings(ctx, p); |
9644 | if (ret) | |
9645 | goto err; | |
9646 | ||
7e84e1c7 | 9647 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9648 | if (ret) |
9649 | goto err; | |
eae071c9 | 9650 | /* always set a rsrc node */ |
47b228ce PB |
9651 | ret = io_rsrc_node_switch_start(ctx); |
9652 | if (ret) | |
9653 | goto err; | |
eae071c9 | 9654 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 9655 | |
2b188cc1 | 9656 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9657 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9658 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9659 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9660 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9661 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9662 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9663 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9664 | |
9665 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9666 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9667 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9668 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9669 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9670 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9671 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9672 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9673 | |
7f13657d XW |
9674 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9675 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9676 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9677 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
9678 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
9679 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
9680 | |
9681 | if (copy_to_user(params, p, sizeof(*p))) { | |
9682 | ret = -EFAULT; | |
9683 | goto err; | |
9684 | } | |
d1719f70 | 9685 | |
9faadcc8 PB |
9686 | file = io_uring_get_file(ctx); |
9687 | if (IS_ERR(file)) { | |
9688 | ret = PTR_ERR(file); | |
9689 | goto err; | |
9690 | } | |
9691 | ||
044c1ab3 JA |
9692 | /* |
9693 | * Install ring fd as the very last thing, so we don't risk someone | |
9694 | * having closed it before we finish setup | |
9695 | */ | |
9faadcc8 PB |
9696 | ret = io_uring_install_fd(ctx, file); |
9697 | if (ret < 0) { | |
9698 | /* fput will clean it up */ | |
9699 | fput(file); | |
9700 | return ret; | |
9701 | } | |
044c1ab3 | 9702 | |
c826bd7a | 9703 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9704 | return ret; |
9705 | err: | |
9706 | io_ring_ctx_wait_and_kill(ctx); | |
9707 | return ret; | |
9708 | } | |
9709 | ||
9710 | /* | |
9711 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9712 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9713 | * params structure passed in. | |
9714 | */ | |
9715 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9716 | { | |
9717 | struct io_uring_params p; | |
2b188cc1 JA |
9718 | int i; |
9719 | ||
9720 | if (copy_from_user(&p, params, sizeof(p))) | |
9721 | return -EFAULT; | |
9722 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9723 | if (p.resv[i]) | |
9724 | return -EINVAL; | |
9725 | } | |
9726 | ||
6c271ce2 | 9727 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9728 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9729 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9730 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9731 | return -EINVAL; |
9732 | ||
7f13657d | 9733 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9734 | } |
9735 | ||
9736 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9737 | struct io_uring_params __user *, params) | |
9738 | { | |
9739 | return io_uring_setup(entries, params); | |
9740 | } | |
9741 | ||
66f4af93 JA |
9742 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9743 | { | |
9744 | struct io_uring_probe *p; | |
9745 | size_t size; | |
9746 | int i, ret; | |
9747 | ||
9748 | size = struct_size(p, ops, nr_args); | |
9749 | if (size == SIZE_MAX) | |
9750 | return -EOVERFLOW; | |
9751 | p = kzalloc(size, GFP_KERNEL); | |
9752 | if (!p) | |
9753 | return -ENOMEM; | |
9754 | ||
9755 | ret = -EFAULT; | |
9756 | if (copy_from_user(p, arg, size)) | |
9757 | goto out; | |
9758 | ret = -EINVAL; | |
9759 | if (memchr_inv(p, 0, size)) | |
9760 | goto out; | |
9761 | ||
9762 | p->last_op = IORING_OP_LAST - 1; | |
9763 | if (nr_args > IORING_OP_LAST) | |
9764 | nr_args = IORING_OP_LAST; | |
9765 | ||
9766 | for (i = 0; i < nr_args; i++) { | |
9767 | p->ops[i].op = i; | |
9768 | if (!io_op_defs[i].not_supported) | |
9769 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9770 | } | |
9771 | p->ops_len = i; | |
9772 | ||
9773 | ret = 0; | |
9774 | if (copy_to_user(arg, p, size)) | |
9775 | ret = -EFAULT; | |
9776 | out: | |
9777 | kfree(p); | |
9778 | return ret; | |
9779 | } | |
9780 | ||
071698e1 JA |
9781 | static int io_register_personality(struct io_ring_ctx *ctx) |
9782 | { | |
4379bf8b | 9783 | const struct cred *creds; |
61cf9370 | 9784 | u32 id; |
1e6fa521 | 9785 | int ret; |
071698e1 | 9786 | |
4379bf8b | 9787 | creds = get_current_cred(); |
1e6fa521 | 9788 | |
61cf9370 MWO |
9789 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9790 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9791 | if (!ret) | |
9792 | return id; | |
9793 | put_cred(creds); | |
1e6fa521 | 9794 | return ret; |
071698e1 JA |
9795 | } |
9796 | ||
21b55dbc SG |
9797 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9798 | unsigned int nr_args) | |
9799 | { | |
9800 | struct io_uring_restriction *res; | |
9801 | size_t size; | |
9802 | int i, ret; | |
9803 | ||
7e84e1c7 SG |
9804 | /* Restrictions allowed only if rings started disabled */ |
9805 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9806 | return -EBADFD; | |
9807 | ||
21b55dbc | 9808 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9809 | if (ctx->restrictions.registered) |
21b55dbc SG |
9810 | return -EBUSY; |
9811 | ||
9812 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9813 | return -EINVAL; | |
9814 | ||
9815 | size = array_size(nr_args, sizeof(*res)); | |
9816 | if (size == SIZE_MAX) | |
9817 | return -EOVERFLOW; | |
9818 | ||
9819 | res = memdup_user(arg, size); | |
9820 | if (IS_ERR(res)) | |
9821 | return PTR_ERR(res); | |
9822 | ||
9823 | ret = 0; | |
9824 | ||
9825 | for (i = 0; i < nr_args; i++) { | |
9826 | switch (res[i].opcode) { | |
9827 | case IORING_RESTRICTION_REGISTER_OP: | |
9828 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9829 | ret = -EINVAL; | |
9830 | goto out; | |
9831 | } | |
9832 | ||
9833 | __set_bit(res[i].register_op, | |
9834 | ctx->restrictions.register_op); | |
9835 | break; | |
9836 | case IORING_RESTRICTION_SQE_OP: | |
9837 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9838 | ret = -EINVAL; | |
9839 | goto out; | |
9840 | } | |
9841 | ||
9842 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9843 | break; | |
9844 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9845 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9846 | break; | |
9847 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9848 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9849 | break; | |
9850 | default: | |
9851 | ret = -EINVAL; | |
9852 | goto out; | |
9853 | } | |
9854 | } | |
9855 | ||
9856 | out: | |
9857 | /* Reset all restrictions if an error happened */ | |
9858 | if (ret != 0) | |
9859 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9860 | else | |
7e84e1c7 | 9861 | ctx->restrictions.registered = true; |
21b55dbc SG |
9862 | |
9863 | kfree(res); | |
9864 | return ret; | |
9865 | } | |
9866 | ||
7e84e1c7 SG |
9867 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9868 | { | |
9869 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9870 | return -EBADFD; | |
9871 | ||
9872 | if (ctx->restrictions.registered) | |
9873 | ctx->restricted = 1; | |
9874 | ||
0298ef96 PB |
9875 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9876 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9877 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9878 | return 0; |
9879 | } | |
9880 | ||
fdecb662 | 9881 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 9882 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
9883 | unsigned nr_args) |
9884 | { | |
9885 | __u32 tmp; | |
9886 | int err; | |
9887 | ||
c3bdad02 PB |
9888 | if (up->resv) |
9889 | return -EINVAL; | |
98f0b3b4 PB |
9890 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
9891 | return -EOVERFLOW; | |
9892 | err = io_rsrc_node_switch_start(ctx); | |
9893 | if (err) | |
9894 | return err; | |
9895 | ||
fdecb662 PB |
9896 | switch (type) { |
9897 | case IORING_RSRC_FILE: | |
98f0b3b4 | 9898 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
9899 | case IORING_RSRC_BUFFER: |
9900 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
9901 | } |
9902 | return -EINVAL; | |
9903 | } | |
9904 | ||
c3bdad02 PB |
9905 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
9906 | unsigned nr_args) | |
98f0b3b4 | 9907 | { |
c3bdad02 | 9908 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
9909 | |
9910 | if (!nr_args) | |
9911 | return -EINVAL; | |
c3bdad02 PB |
9912 | memset(&up, 0, sizeof(up)); |
9913 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
9914 | return -EFAULT; | |
9915 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
9916 | } | |
9917 | ||
9918 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 9919 | unsigned size, unsigned type) |
c3bdad02 PB |
9920 | { |
9921 | struct io_uring_rsrc_update2 up; | |
9922 | ||
9923 | if (size != sizeof(up)) | |
9924 | return -EINVAL; | |
98f0b3b4 PB |
9925 | if (copy_from_user(&up, arg, sizeof(up))) |
9926 | return -EFAULT; | |
992da01a | 9927 | if (!up.nr || up.resv) |
98f0b3b4 | 9928 | return -EINVAL; |
992da01a | 9929 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
9930 | } |
9931 | ||
792e3582 | 9932 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 9933 | unsigned int size, unsigned int type) |
792e3582 PB |
9934 | { |
9935 | struct io_uring_rsrc_register rr; | |
9936 | ||
9937 | /* keep it extendible */ | |
9938 | if (size != sizeof(rr)) | |
9939 | return -EINVAL; | |
9940 | ||
9941 | memset(&rr, 0, sizeof(rr)); | |
9942 | if (copy_from_user(&rr, arg, size)) | |
9943 | return -EFAULT; | |
992da01a | 9944 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
9945 | return -EINVAL; |
9946 | ||
992da01a | 9947 | switch (type) { |
792e3582 PB |
9948 | case IORING_RSRC_FILE: |
9949 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
9950 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
9951 | case IORING_RSRC_BUFFER: |
9952 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
9953 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
9954 | } |
9955 | return -EINVAL; | |
9956 | } | |
9957 | ||
071698e1 JA |
9958 | static bool io_register_op_must_quiesce(int op) |
9959 | { | |
9960 | switch (op) { | |
bd54b6fe BM |
9961 | case IORING_REGISTER_BUFFERS: |
9962 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 9963 | case IORING_REGISTER_FILES: |
071698e1 JA |
9964 | case IORING_UNREGISTER_FILES: |
9965 | case IORING_REGISTER_FILES_UPDATE: | |
9966 | case IORING_REGISTER_PROBE: | |
9967 | case IORING_REGISTER_PERSONALITY: | |
9968 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
9969 | case IORING_REGISTER_FILES2: |
9970 | case IORING_REGISTER_FILES_UPDATE2: | |
9971 | case IORING_REGISTER_BUFFERS2: | |
9972 | case IORING_REGISTER_BUFFERS_UPDATE: | |
071698e1 JA |
9973 | return false; |
9974 | default: | |
9975 | return true; | |
9976 | } | |
9977 | } | |
9978 | ||
edafccee JA |
9979 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9980 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9981 | __releases(ctx->uring_lock) |
9982 | __acquires(ctx->uring_lock) | |
edafccee JA |
9983 | { |
9984 | int ret; | |
9985 | ||
35fa71a0 JA |
9986 | /* |
9987 | * We're inside the ring mutex, if the ref is already dying, then | |
9988 | * someone else killed the ctx or is already going through | |
9989 | * io_uring_register(). | |
9990 | */ | |
9991 | if (percpu_ref_is_dying(&ctx->refs)) | |
9992 | return -ENXIO; | |
9993 | ||
75c4021a PB |
9994 | if (ctx->restricted) { |
9995 | if (opcode >= IORING_REGISTER_LAST) | |
9996 | return -EINVAL; | |
9997 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
9998 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
9999 | return -EACCES; | |
10000 | } | |
10001 | ||
071698e1 | 10002 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10003 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10004 | |
05f3fb3c JA |
10005 | /* |
10006 | * Drop uring mutex before waiting for references to exit. If | |
10007 | * another thread is currently inside io_uring_enter() it might | |
10008 | * need to grab the uring_lock to make progress. If we hold it | |
10009 | * here across the drain wait, then we can deadlock. It's safe | |
10010 | * to drop the mutex here, since no new references will come in | |
10011 | * after we've killed the percpu ref. | |
10012 | */ | |
10013 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10014 | do { |
10015 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10016 | if (!ret) | |
10017 | break; | |
ed6930c9 JA |
10018 | ret = io_run_task_work_sig(); |
10019 | if (ret < 0) | |
10020 | break; | |
af9c1a44 | 10021 | } while (1); |
05f3fb3c | 10022 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10023 | |
c150368b | 10024 | if (ret) { |
f70865db PB |
10025 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
10026 | return ret; | |
c150368b | 10027 | } |
05f3fb3c | 10028 | } |
edafccee JA |
10029 | |
10030 | switch (opcode) { | |
10031 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10032 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10033 | break; |
10034 | case IORING_UNREGISTER_BUFFERS: | |
10035 | ret = -EINVAL; | |
10036 | if (arg || nr_args) | |
10037 | break; | |
0a96bbe4 | 10038 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10039 | break; |
6b06314c | 10040 | case IORING_REGISTER_FILES: |
792e3582 | 10041 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10042 | break; |
10043 | case IORING_UNREGISTER_FILES: | |
10044 | ret = -EINVAL; | |
10045 | if (arg || nr_args) | |
10046 | break; | |
10047 | ret = io_sqe_files_unregister(ctx); | |
10048 | break; | |
c3a31e60 | 10049 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10050 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10051 | break; |
9b402849 | 10052 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10053 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10054 | ret = -EINVAL; |
10055 | if (nr_args != 1) | |
10056 | break; | |
10057 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10058 | if (ret) |
10059 | break; | |
10060 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10061 | ctx->eventfd_async = 1; | |
10062 | else | |
10063 | ctx->eventfd_async = 0; | |
9b402849 JA |
10064 | break; |
10065 | case IORING_UNREGISTER_EVENTFD: | |
10066 | ret = -EINVAL; | |
10067 | if (arg || nr_args) | |
10068 | break; | |
10069 | ret = io_eventfd_unregister(ctx); | |
10070 | break; | |
66f4af93 JA |
10071 | case IORING_REGISTER_PROBE: |
10072 | ret = -EINVAL; | |
10073 | if (!arg || nr_args > 256) | |
10074 | break; | |
10075 | ret = io_probe(ctx, arg, nr_args); | |
10076 | break; | |
071698e1 JA |
10077 | case IORING_REGISTER_PERSONALITY: |
10078 | ret = -EINVAL; | |
10079 | if (arg || nr_args) | |
10080 | break; | |
10081 | ret = io_register_personality(ctx); | |
10082 | break; | |
10083 | case IORING_UNREGISTER_PERSONALITY: | |
10084 | ret = -EINVAL; | |
10085 | if (arg) | |
10086 | break; | |
10087 | ret = io_unregister_personality(ctx, nr_args); | |
10088 | break; | |
7e84e1c7 SG |
10089 | case IORING_REGISTER_ENABLE_RINGS: |
10090 | ret = -EINVAL; | |
10091 | if (arg || nr_args) | |
10092 | break; | |
10093 | ret = io_register_enable_rings(ctx); | |
10094 | break; | |
21b55dbc SG |
10095 | case IORING_REGISTER_RESTRICTIONS: |
10096 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10097 | break; | |
992da01a PB |
10098 | case IORING_REGISTER_FILES2: |
10099 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10100 | break; | |
10101 | case IORING_REGISTER_FILES_UPDATE2: | |
10102 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10103 | IORING_RSRC_FILE); | |
10104 | break; | |
10105 | case IORING_REGISTER_BUFFERS2: | |
10106 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10107 | break; |
992da01a PB |
10108 | case IORING_REGISTER_BUFFERS_UPDATE: |
10109 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10110 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10111 | break; |
edafccee JA |
10112 | default: |
10113 | ret = -EINVAL; | |
10114 | break; | |
10115 | } | |
10116 | ||
071698e1 | 10117 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10118 | /* bring the ctx back to life */ |
05f3fb3c | 10119 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10120 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10121 | } |
edafccee JA |
10122 | return ret; |
10123 | } | |
10124 | ||
10125 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10126 | void __user *, arg, unsigned int, nr_args) | |
10127 | { | |
10128 | struct io_ring_ctx *ctx; | |
10129 | long ret = -EBADF; | |
10130 | struct fd f; | |
10131 | ||
10132 | f = fdget(fd); | |
10133 | if (!f.file) | |
10134 | return -EBADF; | |
10135 | ||
10136 | ret = -EOPNOTSUPP; | |
10137 | if (f.file->f_op != &io_uring_fops) | |
10138 | goto out_fput; | |
10139 | ||
10140 | ctx = f.file->private_data; | |
10141 | ||
b6c23dd5 PB |
10142 | io_run_task_work(); |
10143 | ||
edafccee JA |
10144 | mutex_lock(&ctx->uring_lock); |
10145 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10146 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10147 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10148 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10149 | out_fput: |
10150 | fdput(f); | |
10151 | return ret; | |
10152 | } | |
10153 | ||
2b188cc1 JA |
10154 | static int __init io_uring_init(void) |
10155 | { | |
d7f62e82 SM |
10156 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10157 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10158 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10159 | } while (0) | |
10160 | ||
10161 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10162 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10163 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10164 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10165 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10166 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10167 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10168 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10169 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10170 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10171 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10172 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10173 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10174 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10175 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10176 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10177 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10178 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10179 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10180 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10181 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10182 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10183 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10184 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10185 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10186 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10187 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10188 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10189 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10190 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10191 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10192 | |
b0d658ec PB |
10193 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10194 | sizeof(struct io_uring_rsrc_update)); | |
10195 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10196 | sizeof(struct io_uring_rsrc_update2)); | |
10197 | /* should fit into one byte */ | |
10198 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10199 | ||
d3656344 | 10200 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10201 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10202 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10203 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10204 | return 0; |
10205 | }; | |
10206 | __initcall(io_uring_init); |