]>
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 | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
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 | |
b16fed66 PB |
103 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
104 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
105 | IOSQE_BUFFER_SELECT) | |
106 | ||
2b188cc1 JA |
107 | struct io_uring { |
108 | u32 head ____cacheline_aligned_in_smp; | |
109 | u32 tail ____cacheline_aligned_in_smp; | |
110 | }; | |
111 | ||
1e84b97b | 112 | /* |
75b28aff HV |
113 | * This data is shared with the application through the mmap at offsets |
114 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
115 | * |
116 | * The offsets to the member fields are published through struct | |
117 | * io_sqring_offsets when calling io_uring_setup. | |
118 | */ | |
75b28aff | 119 | struct io_rings { |
1e84b97b SB |
120 | /* |
121 | * Head and tail offsets into the ring; the offsets need to be | |
122 | * masked to get valid indices. | |
123 | * | |
75b28aff HV |
124 | * The kernel controls head of the sq ring and the tail of the cq ring, |
125 | * and the application controls tail of the sq ring and the head of the | |
126 | * cq ring. | |
1e84b97b | 127 | */ |
75b28aff | 128 | struct io_uring sq, cq; |
1e84b97b | 129 | /* |
75b28aff | 130 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
131 | * ring_entries - 1) |
132 | */ | |
75b28aff HV |
133 | u32 sq_ring_mask, cq_ring_mask; |
134 | /* Ring sizes (constant, power of 2) */ | |
135 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
136 | /* |
137 | * Number of invalid entries dropped by the kernel due to | |
138 | * invalid index stored in array | |
139 | * | |
140 | * Written by the kernel, shouldn't be modified by the | |
141 | * application (i.e. get number of "new events" by comparing to | |
142 | * cached value). | |
143 | * | |
144 | * After a new SQ head value was read by the application this | |
145 | * counter includes all submissions that were dropped reaching | |
146 | * the new SQ head (and possibly more). | |
147 | */ | |
75b28aff | 148 | u32 sq_dropped; |
1e84b97b | 149 | /* |
0d9b5b3a | 150 | * Runtime SQ flags |
1e84b97b SB |
151 | * |
152 | * Written by the kernel, shouldn't be modified by the | |
153 | * application. | |
154 | * | |
155 | * The application needs a full memory barrier before checking | |
156 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
157 | */ | |
75b28aff | 158 | u32 sq_flags; |
0d9b5b3a SG |
159 | /* |
160 | * Runtime CQ flags | |
161 | * | |
162 | * Written by the application, shouldn't be modified by the | |
163 | * kernel. | |
164 | */ | |
165 | u32 cq_flags; | |
1e84b97b SB |
166 | /* |
167 | * Number of completion events lost because the queue was full; | |
168 | * this should be avoided by the application by making sure | |
0b4295b5 | 169 | * there are not more requests pending than there is space in |
1e84b97b SB |
170 | * the completion queue. |
171 | * | |
172 | * Written by the kernel, shouldn't be modified by the | |
173 | * application (i.e. get number of "new events" by comparing to | |
174 | * cached value). | |
175 | * | |
176 | * As completion events come in out of order this counter is not | |
177 | * ordered with any other data. | |
178 | */ | |
75b28aff | 179 | u32 cq_overflow; |
1e84b97b SB |
180 | /* |
181 | * Ring buffer of completion events. | |
182 | * | |
183 | * The kernel writes completion events fresh every time they are | |
184 | * produced, so the application is allowed to modify pending | |
185 | * entries. | |
186 | */ | |
75b28aff | 187 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
188 | }; |
189 | ||
45d189c6 PB |
190 | enum io_uring_cmd_flags { |
191 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 192 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
193 | }; |
194 | ||
edafccee JA |
195 | struct io_mapped_ubuf { |
196 | u64 ubuf; | |
4751f53d | 197 | u64 ubuf_end; |
edafccee JA |
198 | struct bio_vec *bvec; |
199 | unsigned int nr_bvecs; | |
de293938 | 200 | unsigned long acct_pages; |
edafccee JA |
201 | }; |
202 | ||
50238531 BM |
203 | struct io_ring_ctx; |
204 | ||
6c2450ae PB |
205 | struct io_overflow_cqe { |
206 | struct io_uring_cqe cqe; | |
207 | struct list_head list; | |
208 | }; | |
209 | ||
a04b0ac0 PB |
210 | struct io_fixed_file { |
211 | /* file * with additional FFS_* flags */ | |
212 | unsigned long file_ptr; | |
213 | }; | |
214 | ||
269bbe5f BM |
215 | struct io_rsrc_put { |
216 | struct list_head list; | |
50238531 BM |
217 | union { |
218 | void *rsrc; | |
219 | struct file *file; | |
220 | }; | |
269bbe5f BM |
221 | }; |
222 | ||
aeca241b PB |
223 | struct io_file_table { |
224 | /* two level table */ | |
225 | struct io_fixed_file **files; | |
31b51510 JA |
226 | }; |
227 | ||
b895c9a6 | 228 | struct io_rsrc_node { |
05589553 XW |
229 | struct percpu_ref refs; |
230 | struct list_head node; | |
269bbe5f | 231 | struct list_head rsrc_list; |
b895c9a6 | 232 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 233 | struct llist_node llist; |
e297822b | 234 | bool done; |
05589553 XW |
235 | }; |
236 | ||
40ae0ff7 PB |
237 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
238 | ||
b895c9a6 | 239 | struct io_rsrc_data { |
05f3fb3c JA |
240 | struct io_ring_ctx *ctx; |
241 | ||
40ae0ff7 | 242 | rsrc_put_fn *do_put; |
3e942498 | 243 | atomic_t refs; |
05f3fb3c | 244 | struct completion done; |
8bad28d8 | 245 | bool quiesce; |
05f3fb3c JA |
246 | }; |
247 | ||
5a2e745d JA |
248 | struct io_buffer { |
249 | struct list_head list; | |
250 | __u64 addr; | |
251 | __s32 len; | |
252 | __u16 bid; | |
253 | }; | |
254 | ||
21b55dbc SG |
255 | struct io_restriction { |
256 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
257 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
258 | u8 sqe_flags_allowed; | |
259 | u8 sqe_flags_required; | |
7e84e1c7 | 260 | bool registered; |
21b55dbc SG |
261 | }; |
262 | ||
37d1e2e3 JA |
263 | enum { |
264 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
265 | IO_SQ_THREAD_SHOULD_PARK, | |
266 | }; | |
267 | ||
534ca6d6 JA |
268 | struct io_sq_data { |
269 | refcount_t refs; | |
9e138a48 | 270 | atomic_t park_pending; |
09a6f4ef | 271 | struct mutex lock; |
69fb2131 JA |
272 | |
273 | /* ctx's that are using this sqd */ | |
274 | struct list_head ctx_list; | |
69fb2131 | 275 | |
534ca6d6 JA |
276 | struct task_struct *thread; |
277 | struct wait_queue_head wait; | |
08369246 XW |
278 | |
279 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
280 | int sq_cpu; |
281 | pid_t task_pid; | |
5c2469e0 | 282 | pid_t task_tgid; |
37d1e2e3 JA |
283 | |
284 | unsigned long state; | |
37d1e2e3 | 285 | struct completion exited; |
b7f5a0bf | 286 | struct callback_head *park_task_work; |
534ca6d6 JA |
287 | }; |
288 | ||
258b29a9 | 289 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 290 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 291 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 292 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
293 | |
294 | struct io_comp_state { | |
6dd0be1e | 295 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 296 | unsigned int nr; |
c7dae4ba JA |
297 | unsigned int locked_free_nr; |
298 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 299 | struct list_head free_list; |
c7dae4ba JA |
300 | /* IRQ completion list, under ->completion_lock */ |
301 | struct list_head locked_free_list; | |
258b29a9 PB |
302 | }; |
303 | ||
a1ab7b35 PB |
304 | struct io_submit_link { |
305 | struct io_kiocb *head; | |
306 | struct io_kiocb *last; | |
307 | }; | |
308 | ||
258b29a9 PB |
309 | struct io_submit_state { |
310 | struct blk_plug plug; | |
a1ab7b35 | 311 | struct io_submit_link link; |
258b29a9 PB |
312 | |
313 | /* | |
314 | * io_kiocb alloc cache | |
315 | */ | |
bf019da7 | 316 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
317 | unsigned int free_reqs; |
318 | ||
319 | bool plug_started; | |
320 | ||
321 | /* | |
322 | * Batch completion logic | |
323 | */ | |
324 | struct io_comp_state comp; | |
325 | ||
326 | /* | |
327 | * File reference cache | |
328 | */ | |
329 | struct file *file; | |
330 | unsigned int fd; | |
331 | unsigned int file_refs; | |
332 | unsigned int ios_left; | |
333 | }; | |
334 | ||
2b188cc1 JA |
335 | struct io_ring_ctx { |
336 | struct { | |
337 | struct percpu_ref refs; | |
338 | } ____cacheline_aligned_in_smp; | |
339 | ||
340 | struct { | |
341 | unsigned int flags; | |
e1d85334 | 342 | unsigned int compat: 1; |
e1d85334 RD |
343 | unsigned int drain_next: 1; |
344 | unsigned int eventfd_async: 1; | |
21b55dbc | 345 | unsigned int restricted: 1; |
2b188cc1 | 346 | |
75b28aff HV |
347 | /* |
348 | * Ring buffer of indices into array of io_uring_sqe, which is | |
349 | * mmapped by the application using the IORING_OFF_SQES offset. | |
350 | * | |
351 | * This indirection could e.g. be used to assign fixed | |
352 | * io_uring_sqe entries to operations and only submit them to | |
353 | * the queue when needed. | |
354 | * | |
355 | * The kernel modifies neither the indices array nor the entries | |
356 | * array. | |
357 | */ | |
358 | u32 *sq_array; | |
2b188cc1 JA |
359 | unsigned cached_sq_head; |
360 | unsigned sq_entries; | |
361 | unsigned sq_mask; | |
6c271ce2 | 362 | unsigned sq_thread_idle; |
498ccd9e | 363 | unsigned cached_sq_dropped; |
2c3bac6d | 364 | unsigned cached_cq_overflow; |
ad3eb2c8 | 365 | unsigned long sq_check_overflow; |
de0617e4 | 366 | |
e941894e JA |
367 | /* hashed buffered write serialization */ |
368 | struct io_wq_hash *hash_map; | |
369 | ||
de0617e4 | 370 | struct list_head defer_list; |
5262f567 | 371 | struct list_head timeout_list; |
1d7bb1d5 | 372 | struct list_head cq_overflow_list; |
fcb323cc | 373 | |
ad3eb2c8 | 374 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
375 | } ____cacheline_aligned_in_smp; |
376 | ||
3c1a2ead JA |
377 | struct { |
378 | struct mutex uring_lock; | |
379 | wait_queue_head_t wait; | |
380 | } ____cacheline_aligned_in_smp; | |
381 | ||
382 | struct io_submit_state submit_state; | |
383 | ||
206aefde JA |
384 | struct io_rings *rings; |
385 | ||
2aede0e4 JA |
386 | /* Only used for accounting purposes */ |
387 | struct mm_struct *mm_account; | |
388 | ||
7c30f36a | 389 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
390 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
391 | ||
90554200 | 392 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 393 | struct list_head sqd_list; |
75b28aff | 394 | |
6b06314c JA |
395 | /* |
396 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
397 | * readers must ensure that ->refs is alive as long as the file* is | |
398 | * used. Only updated through io_uring_register(2). | |
399 | */ | |
b895c9a6 | 400 | struct io_rsrc_data *file_data; |
aeca241b | 401 | struct io_file_table file_table; |
6b06314c JA |
402 | unsigned nr_user_files; |
403 | ||
edafccee JA |
404 | /* if used, fixed mapped user buffers */ |
405 | unsigned nr_user_bufs; | |
406 | struct io_mapped_ubuf *user_bufs; | |
407 | ||
2b188cc1 JA |
408 | struct user_struct *user; |
409 | ||
0f158b4c | 410 | struct completion ref_comp; |
206aefde JA |
411 | |
412 | #if defined(CONFIG_UNIX) | |
413 | struct socket *ring_sock; | |
414 | #endif | |
415 | ||
9e15c3a0 | 416 | struct xarray io_buffers; |
5a2e745d | 417 | |
61cf9370 MWO |
418 | struct xarray personalities; |
419 | u32 pers_next; | |
071698e1 | 420 | |
206aefde JA |
421 | struct { |
422 | unsigned cached_cq_tail; | |
423 | unsigned cq_entries; | |
424 | unsigned cq_mask; | |
425 | atomic_t cq_timeouts; | |
f010505b | 426 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 427 | unsigned long cq_check_overflow; |
206aefde JA |
428 | struct wait_queue_head cq_wait; |
429 | struct fasync_struct *cq_fasync; | |
430 | struct eventfd_ctx *cq_ev_fd; | |
431 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 432 | |
2b188cc1 JA |
433 | struct { |
434 | spinlock_t completion_lock; | |
e94f141b | 435 | |
def596e9 | 436 | /* |
540e32a0 | 437 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
438 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
439 | * For SQPOLL, only the single threaded io_sq_thread() will | |
440 | * manipulate the list, hence no extra locking is needed there. | |
441 | */ | |
540e32a0 | 442 | struct list_head iopoll_list; |
78076bb6 JA |
443 | struct hlist_head *cancel_hash; |
444 | unsigned cancel_hash_bits; | |
e94f141b | 445 | bool poll_multi_file; |
2b188cc1 | 446 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 447 | |
269bbe5f BM |
448 | struct delayed_work rsrc_put_work; |
449 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
450 | struct list_head rsrc_ref_list; |
451 | spinlock_t rsrc_ref_lock; | |
a7f0ed5a | 452 | struct io_rsrc_node *rsrc_node; |
b895c9a6 | 453 | struct io_rsrc_node *rsrc_backup_node; |
4a38aed2 | 454 | |
21b55dbc | 455 | struct io_restriction restrictions; |
3c1a2ead | 456 | |
7c25c0d1 JA |
457 | /* exit task_work */ |
458 | struct callback_head *exit_task_work; | |
459 | ||
3c1a2ead JA |
460 | /* Keep this last, we don't need it for the fast path */ |
461 | struct work_struct exit_work; | |
13bf43f5 | 462 | struct list_head tctx_list; |
2b188cc1 JA |
463 | }; |
464 | ||
53e043b2 SM |
465 | struct io_uring_task { |
466 | /* submission side */ | |
467 | struct xarray xa; | |
468 | struct wait_queue_head wait; | |
ee53fb2b SM |
469 | const struct io_ring_ctx *last; |
470 | struct io_wq *io_wq; | |
53e043b2 | 471 | struct percpu_counter inflight; |
b303fe2e | 472 | atomic_t inflight_tracked; |
53e043b2 | 473 | atomic_t in_idle; |
53e043b2 SM |
474 | |
475 | spinlock_t task_lock; | |
476 | struct io_wq_work_list task_list; | |
477 | unsigned long task_state; | |
478 | struct callback_head task_work; | |
479 | }; | |
480 | ||
09bb8394 JA |
481 | /* |
482 | * First field must be the file pointer in all the | |
483 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
484 | */ | |
221c5eb2 JA |
485 | struct io_poll_iocb { |
486 | struct file *file; | |
018043be | 487 | struct wait_queue_head *head; |
221c5eb2 | 488 | __poll_t events; |
8c838788 | 489 | bool done; |
221c5eb2 | 490 | bool canceled; |
9d805892 PB |
491 | struct wait_queue_entry wait; |
492 | }; | |
493 | ||
494 | struct io_poll_update { | |
495 | struct file *file; | |
496 | u64 old_user_data; | |
497 | u64 new_user_data; | |
498 | __poll_t events; | |
b69de288 JA |
499 | bool update_events; |
500 | bool update_user_data; | |
221c5eb2 JA |
501 | }; |
502 | ||
b5dba59e JA |
503 | struct io_close { |
504 | struct file *file; | |
b5dba59e JA |
505 | int fd; |
506 | }; | |
507 | ||
ad8a48ac JA |
508 | struct io_timeout_data { |
509 | struct io_kiocb *req; | |
510 | struct hrtimer timer; | |
511 | struct timespec64 ts; | |
512 | enum hrtimer_mode mode; | |
513 | }; | |
514 | ||
8ed8d3c3 JA |
515 | struct io_accept { |
516 | struct file *file; | |
517 | struct sockaddr __user *addr; | |
518 | int __user *addr_len; | |
519 | int flags; | |
09952e3e | 520 | unsigned long nofile; |
8ed8d3c3 JA |
521 | }; |
522 | ||
523 | struct io_sync { | |
524 | struct file *file; | |
525 | loff_t len; | |
526 | loff_t off; | |
527 | int flags; | |
d63d1b5e | 528 | int mode; |
8ed8d3c3 JA |
529 | }; |
530 | ||
fbf23849 JA |
531 | struct io_cancel { |
532 | struct file *file; | |
533 | u64 addr; | |
534 | }; | |
535 | ||
b29472ee JA |
536 | struct io_timeout { |
537 | struct file *file; | |
bfe68a22 PB |
538 | u32 off; |
539 | u32 target_seq; | |
135fcde8 | 540 | struct list_head list; |
90cd7e42 PB |
541 | /* head of the link, used by linked timeouts only */ |
542 | struct io_kiocb *head; | |
b29472ee JA |
543 | }; |
544 | ||
0bdf7a2d PB |
545 | struct io_timeout_rem { |
546 | struct file *file; | |
547 | u64 addr; | |
9c8e11b3 PB |
548 | |
549 | /* timeout update */ | |
550 | struct timespec64 ts; | |
551 | u32 flags; | |
0bdf7a2d PB |
552 | }; |
553 | ||
9adbd45d JA |
554 | struct io_rw { |
555 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
556 | struct kiocb kiocb; | |
557 | u64 addr; | |
558 | u64 len; | |
559 | }; | |
560 | ||
3fbb51c1 JA |
561 | struct io_connect { |
562 | struct file *file; | |
563 | struct sockaddr __user *addr; | |
564 | int addr_len; | |
565 | }; | |
566 | ||
e47293fd JA |
567 | struct io_sr_msg { |
568 | struct file *file; | |
fddaface | 569 | union { |
4af3417a PB |
570 | struct compat_msghdr __user *umsg_compat; |
571 | struct user_msghdr __user *umsg; | |
572 | void __user *buf; | |
fddaface | 573 | }; |
e47293fd | 574 | int msg_flags; |
bcda7baa | 575 | int bgid; |
fddaface | 576 | size_t len; |
bcda7baa | 577 | struct io_buffer *kbuf; |
e47293fd JA |
578 | }; |
579 | ||
15b71abe JA |
580 | struct io_open { |
581 | struct file *file; | |
582 | int dfd; | |
15b71abe | 583 | struct filename *filename; |
c12cedf2 | 584 | struct open_how how; |
4022e7af | 585 | unsigned long nofile; |
15b71abe JA |
586 | }; |
587 | ||
269bbe5f | 588 | struct io_rsrc_update { |
05f3fb3c JA |
589 | struct file *file; |
590 | u64 arg; | |
591 | u32 nr_args; | |
592 | u32 offset; | |
593 | }; | |
594 | ||
4840e418 JA |
595 | struct io_fadvise { |
596 | struct file *file; | |
597 | u64 offset; | |
598 | u32 len; | |
599 | u32 advice; | |
600 | }; | |
601 | ||
c1ca757b JA |
602 | struct io_madvise { |
603 | struct file *file; | |
604 | u64 addr; | |
605 | u32 len; | |
606 | u32 advice; | |
607 | }; | |
608 | ||
3e4827b0 JA |
609 | struct io_epoll { |
610 | struct file *file; | |
611 | int epfd; | |
612 | int op; | |
613 | int fd; | |
614 | struct epoll_event event; | |
e47293fd JA |
615 | }; |
616 | ||
7d67af2c PB |
617 | struct io_splice { |
618 | struct file *file_out; | |
619 | struct file *file_in; | |
620 | loff_t off_out; | |
621 | loff_t off_in; | |
622 | u64 len; | |
623 | unsigned int flags; | |
624 | }; | |
625 | ||
ddf0322d JA |
626 | struct io_provide_buf { |
627 | struct file *file; | |
628 | __u64 addr; | |
38134ada | 629 | __u32 len; |
ddf0322d JA |
630 | __u32 bgid; |
631 | __u16 nbufs; | |
632 | __u16 bid; | |
633 | }; | |
634 | ||
1d9e1288 BM |
635 | struct io_statx { |
636 | struct file *file; | |
637 | int dfd; | |
638 | unsigned int mask; | |
639 | unsigned int flags; | |
e62753e4 | 640 | const char __user *filename; |
1d9e1288 BM |
641 | struct statx __user *buffer; |
642 | }; | |
643 | ||
36f4fa68 JA |
644 | struct io_shutdown { |
645 | struct file *file; | |
646 | int how; | |
647 | }; | |
648 | ||
80a261fd JA |
649 | struct io_rename { |
650 | struct file *file; | |
651 | int old_dfd; | |
652 | int new_dfd; | |
653 | struct filename *oldpath; | |
654 | struct filename *newpath; | |
655 | int flags; | |
656 | }; | |
657 | ||
14a1143b JA |
658 | struct io_unlink { |
659 | struct file *file; | |
660 | int dfd; | |
661 | int flags; | |
662 | struct filename *filename; | |
663 | }; | |
664 | ||
3ca405eb PB |
665 | struct io_completion { |
666 | struct file *file; | |
667 | struct list_head list; | |
8c3f9cd1 | 668 | u32 cflags; |
3ca405eb PB |
669 | }; |
670 | ||
f499a021 JA |
671 | struct io_async_connect { |
672 | struct sockaddr_storage address; | |
673 | }; | |
674 | ||
03b1230c JA |
675 | struct io_async_msghdr { |
676 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
677 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
678 | struct iovec *free_iov; | |
03b1230c JA |
679 | struct sockaddr __user *uaddr; |
680 | struct msghdr msg; | |
b537916c | 681 | struct sockaddr_storage addr; |
03b1230c JA |
682 | }; |
683 | ||
f67676d1 JA |
684 | struct io_async_rw { |
685 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
686 | const struct iovec *free_iovec; |
687 | struct iov_iter iter; | |
227c0c96 | 688 | size_t bytes_done; |
bcf5a063 | 689 | struct wait_page_queue wpq; |
f67676d1 JA |
690 | }; |
691 | ||
6b47ee6e PB |
692 | enum { |
693 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
694 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
695 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
696 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
697 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 698 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 699 | |
6b47ee6e PB |
700 | REQ_F_FAIL_LINK_BIT, |
701 | REQ_F_INFLIGHT_BIT, | |
702 | REQ_F_CUR_POS_BIT, | |
703 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 704 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 705 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 706 | REQ_F_POLLED_BIT, |
bcda7baa | 707 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 708 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 709 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 710 | REQ_F_REISSUE_BIT, |
8c130827 | 711 | REQ_F_DONT_REISSUE_BIT, |
7b29f92d JA |
712 | /* keep async read/write and isreg together and in order */ |
713 | REQ_F_ASYNC_READ_BIT, | |
714 | REQ_F_ASYNC_WRITE_BIT, | |
715 | REQ_F_ISREG_BIT, | |
84557871 JA |
716 | |
717 | /* not a real bit, just to check we're not overflowing the space */ | |
718 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
719 | }; |
720 | ||
721 | enum { | |
722 | /* ctx owns file */ | |
723 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
724 | /* drain existing IO first */ | |
725 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
726 | /* linked sqes */ | |
727 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
728 | /* doesn't sever on completion < 0 */ | |
729 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
730 | /* IOSQE_ASYNC */ | |
731 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
732 | /* IOSQE_BUFFER_SELECT */ |
733 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 734 | |
6b47ee6e PB |
735 | /* fail rest of links */ |
736 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 737 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
738 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
739 | /* read/write uses file position */ | |
740 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
741 | /* must not punt to workers */ | |
742 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 743 | /* has or had linked timeout */ |
6b47ee6e | 744 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
745 | /* needs cleanup */ |
746 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
747 | /* already went through poll handler */ |
748 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
749 | /* buffer already selected */ |
750 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
751 | /* linked timeout is active, i.e. prepared by link's head */ |
752 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
753 | /* completion is deferred through io_comp_state */ |
754 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
755 | /* caller should reissue async */ |
756 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
757 | /* don't attempt request reissue, see io_rw_reissue() */ |
758 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
759 | /* supports async reads */ |
760 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
761 | /* supports async writes */ | |
762 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
763 | /* regular file */ | |
764 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
765 | }; |
766 | ||
767 | struct async_poll { | |
768 | struct io_poll_iocb poll; | |
807abcb0 | 769 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
770 | }; |
771 | ||
7cbf1722 JA |
772 | struct io_task_work { |
773 | struct io_wq_work_node node; | |
774 | task_work_func_t func; | |
775 | }; | |
776 | ||
09bb8394 JA |
777 | /* |
778 | * NOTE! Each of the iocb union members has the file pointer | |
779 | * as the first entry in their struct definition. So you can | |
780 | * access the file pointer through any of the sub-structs, | |
781 | * or directly as just 'ki_filp' in this struct. | |
782 | */ | |
2b188cc1 | 783 | struct io_kiocb { |
221c5eb2 | 784 | union { |
09bb8394 | 785 | struct file *file; |
9adbd45d | 786 | struct io_rw rw; |
221c5eb2 | 787 | struct io_poll_iocb poll; |
9d805892 | 788 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
789 | struct io_accept accept; |
790 | struct io_sync sync; | |
fbf23849 | 791 | struct io_cancel cancel; |
b29472ee | 792 | struct io_timeout timeout; |
0bdf7a2d | 793 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 794 | struct io_connect connect; |
e47293fd | 795 | struct io_sr_msg sr_msg; |
15b71abe | 796 | struct io_open open; |
b5dba59e | 797 | struct io_close close; |
269bbe5f | 798 | struct io_rsrc_update rsrc_update; |
4840e418 | 799 | struct io_fadvise fadvise; |
c1ca757b | 800 | struct io_madvise madvise; |
3e4827b0 | 801 | struct io_epoll epoll; |
7d67af2c | 802 | struct io_splice splice; |
ddf0322d | 803 | struct io_provide_buf pbuf; |
1d9e1288 | 804 | struct io_statx statx; |
36f4fa68 | 805 | struct io_shutdown shutdown; |
80a261fd | 806 | struct io_rename rename; |
14a1143b | 807 | struct io_unlink unlink; |
3ca405eb PB |
808 | /* use only after cleaning per-op data, see io_clean_op() */ |
809 | struct io_completion compl; | |
221c5eb2 | 810 | }; |
2b188cc1 | 811 | |
e8c2bc1f JA |
812 | /* opcode allocated if it needs to store data for async defer */ |
813 | void *async_data; | |
d625c6ee | 814 | u8 opcode; |
65a6543d XW |
815 | /* polled IO has completed */ |
816 | u8 iopoll_completed; | |
2b188cc1 | 817 | |
4f4eeba8 | 818 | u16 buf_index; |
9cf7c104 | 819 | u32 result; |
4f4eeba8 | 820 | |
010e8e6b PB |
821 | struct io_ring_ctx *ctx; |
822 | unsigned int flags; | |
abc54d63 | 823 | atomic_t refs; |
010e8e6b PB |
824 | struct task_struct *task; |
825 | u64 user_data; | |
d7718a9d | 826 | |
f2f87370 | 827 | struct io_kiocb *link; |
269bbe5f | 828 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 829 | |
b303fe2e | 830 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 831 | struct list_head inflight_entry; |
7cbf1722 JA |
832 | union { |
833 | struct io_task_work io_task_work; | |
834 | struct callback_head task_work; | |
835 | }; | |
010e8e6b PB |
836 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
837 | struct hlist_node hash_node; | |
838 | struct async_poll *apoll; | |
839 | struct io_wq_work work; | |
2b188cc1 | 840 | }; |
05589553 | 841 | |
13bf43f5 PB |
842 | struct io_tctx_node { |
843 | struct list_head ctx_node; | |
844 | struct task_struct *task; | |
13bf43f5 PB |
845 | struct io_ring_ctx *ctx; |
846 | }; | |
847 | ||
27dc8338 PB |
848 | struct io_defer_entry { |
849 | struct list_head list; | |
850 | struct io_kiocb *req; | |
9cf7c104 | 851 | u32 seq; |
2b188cc1 JA |
852 | }; |
853 | ||
d3656344 | 854 | struct io_op_def { |
d3656344 JA |
855 | /* needs req->file assigned */ |
856 | unsigned needs_file : 1; | |
d3656344 JA |
857 | /* hash wq insertion if file is a regular file */ |
858 | unsigned hash_reg_file : 1; | |
859 | /* unbound wq insertion if file is a non-regular file */ | |
860 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
861 | /* opcode is not supported by this kernel */ |
862 | unsigned not_supported : 1; | |
8a72758c JA |
863 | /* set if opcode supports polled "wait" */ |
864 | unsigned pollin : 1; | |
865 | unsigned pollout : 1; | |
bcda7baa JA |
866 | /* op supports buffer selection */ |
867 | unsigned buffer_select : 1; | |
26f0505a PB |
868 | /* do prep async if is going to be punted */ |
869 | unsigned needs_async_setup : 1; | |
27926b68 JA |
870 | /* should block plug */ |
871 | unsigned plug : 1; | |
e8c2bc1f JA |
872 | /* size of async data needed, if any */ |
873 | unsigned short async_size; | |
d3656344 JA |
874 | }; |
875 | ||
0918682b | 876 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
877 | [IORING_OP_NOP] = {}, |
878 | [IORING_OP_READV] = { | |
d3656344 JA |
879 | .needs_file = 1, |
880 | .unbound_nonreg_file = 1, | |
8a72758c | 881 | .pollin = 1, |
4d954c25 | 882 | .buffer_select = 1, |
26f0505a | 883 | .needs_async_setup = 1, |
27926b68 | 884 | .plug = 1, |
e8c2bc1f | 885 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 886 | }, |
0463b6c5 | 887 | [IORING_OP_WRITEV] = { |
d3656344 JA |
888 | .needs_file = 1, |
889 | .hash_reg_file = 1, | |
890 | .unbound_nonreg_file = 1, | |
8a72758c | 891 | .pollout = 1, |
26f0505a | 892 | .needs_async_setup = 1, |
27926b68 | 893 | .plug = 1, |
e8c2bc1f | 894 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 895 | }, |
0463b6c5 | 896 | [IORING_OP_FSYNC] = { |
d3656344 JA |
897 | .needs_file = 1, |
898 | }, | |
0463b6c5 | 899 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
900 | .needs_file = 1, |
901 | .unbound_nonreg_file = 1, | |
8a72758c | 902 | .pollin = 1, |
27926b68 | 903 | .plug = 1, |
e8c2bc1f | 904 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 905 | }, |
0463b6c5 | 906 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
907 | .needs_file = 1, |
908 | .hash_reg_file = 1, | |
909 | .unbound_nonreg_file = 1, | |
8a72758c | 910 | .pollout = 1, |
27926b68 | 911 | .plug = 1, |
e8c2bc1f | 912 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 913 | }, |
0463b6c5 | 914 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
915 | .needs_file = 1, |
916 | .unbound_nonreg_file = 1, | |
917 | }, | |
0463b6c5 PB |
918 | [IORING_OP_POLL_REMOVE] = {}, |
919 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
920 | .needs_file = 1, |
921 | }, | |
0463b6c5 | 922 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
923 | .needs_file = 1, |
924 | .unbound_nonreg_file = 1, | |
8a72758c | 925 | .pollout = 1, |
26f0505a | 926 | .needs_async_setup = 1, |
e8c2bc1f | 927 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 928 | }, |
0463b6c5 | 929 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
930 | .needs_file = 1, |
931 | .unbound_nonreg_file = 1, | |
8a72758c | 932 | .pollin = 1, |
52de1fe1 | 933 | .buffer_select = 1, |
26f0505a | 934 | .needs_async_setup = 1, |
e8c2bc1f | 935 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 936 | }, |
0463b6c5 | 937 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 938 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 939 | }, |
9c8e11b3 PB |
940 | [IORING_OP_TIMEOUT_REMOVE] = { |
941 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
944 | .needs_file = 1, |
945 | .unbound_nonreg_file = 1, | |
8a72758c | 946 | .pollin = 1, |
d3656344 | 947 | }, |
0463b6c5 PB |
948 | [IORING_OP_ASYNC_CANCEL] = {}, |
949 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 950 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 951 | }, |
0463b6c5 | 952 | [IORING_OP_CONNECT] = { |
d3656344 JA |
953 | .needs_file = 1, |
954 | .unbound_nonreg_file = 1, | |
8a72758c | 955 | .pollout = 1, |
26f0505a | 956 | .needs_async_setup = 1, |
e8c2bc1f | 957 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 958 | }, |
0463b6c5 | 959 | [IORING_OP_FALLOCATE] = { |
d3656344 | 960 | .needs_file = 1, |
d3656344 | 961 | }, |
44526bed JA |
962 | [IORING_OP_OPENAT] = {}, |
963 | [IORING_OP_CLOSE] = {}, | |
964 | [IORING_OP_FILES_UPDATE] = {}, | |
965 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 966 | [IORING_OP_READ] = { |
3a6820f2 JA |
967 | .needs_file = 1, |
968 | .unbound_nonreg_file = 1, | |
8a72758c | 969 | .pollin = 1, |
bcda7baa | 970 | .buffer_select = 1, |
27926b68 | 971 | .plug = 1, |
e8c2bc1f | 972 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 973 | }, |
0463b6c5 | 974 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
975 | .needs_file = 1, |
976 | .unbound_nonreg_file = 1, | |
8a72758c | 977 | .pollout = 1, |
27926b68 | 978 | .plug = 1, |
e8c2bc1f | 979 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 980 | }, |
0463b6c5 | 981 | [IORING_OP_FADVISE] = { |
4840e418 | 982 | .needs_file = 1, |
c1ca757b | 983 | }, |
44526bed | 984 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 985 | [IORING_OP_SEND] = { |
fddaface JA |
986 | .needs_file = 1, |
987 | .unbound_nonreg_file = 1, | |
8a72758c | 988 | .pollout = 1, |
fddaface | 989 | }, |
0463b6c5 | 990 | [IORING_OP_RECV] = { |
fddaface JA |
991 | .needs_file = 1, |
992 | .unbound_nonreg_file = 1, | |
8a72758c | 993 | .pollin = 1, |
bcda7baa | 994 | .buffer_select = 1, |
fddaface | 995 | }, |
0463b6c5 | 996 | [IORING_OP_OPENAT2] = { |
cebdb986 | 997 | }, |
3e4827b0 JA |
998 | [IORING_OP_EPOLL_CTL] = { |
999 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1000 | }, |
7d67af2c PB |
1001 | [IORING_OP_SPLICE] = { |
1002 | .needs_file = 1, | |
1003 | .hash_reg_file = 1, | |
1004 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1005 | }, |
1006 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1007 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1008 | [IORING_OP_TEE] = { |
1009 | .needs_file = 1, | |
1010 | .hash_reg_file = 1, | |
1011 | .unbound_nonreg_file = 1, | |
1012 | }, | |
36f4fa68 JA |
1013 | [IORING_OP_SHUTDOWN] = { |
1014 | .needs_file = 1, | |
1015 | }, | |
44526bed JA |
1016 | [IORING_OP_RENAMEAT] = {}, |
1017 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1018 | }; |
1019 | ||
7a612350 | 1020 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1021 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1022 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1023 | struct task_struct *task, | |
1024 | struct files_struct *files); | |
37d1e2e3 | 1025 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
b895c9a6 | 1026 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1027 | |
ff642164 | 1028 | static bool io_cqring_fill_event(struct io_kiocb *req, long res, unsigned cflags); |
ec9c02ad | 1029 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1030 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1031 | static void io_dismantle_req(struct io_kiocb *req); |
1032 | static void io_put_task(struct task_struct *task, int nr); | |
94ae5e77 JA |
1033 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1034 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c | 1035 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1036 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1037 | unsigned nr_args); |
68fb8979 | 1038 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1039 | static struct file *io_file_get(struct io_submit_state *state, |
1040 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1041 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1042 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1043 | |
907d1df3 | 1044 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1045 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1046 | struct io_ring_ctx *ctx); | |
5082620f | 1047 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1048 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1049 | |
2b188cc1 JA |
1050 | static struct kmem_cache *req_cachep; |
1051 | ||
0918682b | 1052 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1053 | |
1054 | struct sock *io_uring_get_socket(struct file *file) | |
1055 | { | |
1056 | #if defined(CONFIG_UNIX) | |
1057 | if (file->f_op == &io_uring_fops) { | |
1058 | struct io_ring_ctx *ctx = file->private_data; | |
1059 | ||
1060 | return ctx->ring_sock->sk; | |
1061 | } | |
1062 | #endif | |
1063 | return NULL; | |
1064 | } | |
1065 | EXPORT_SYMBOL(io_uring_get_socket); | |
1066 | ||
f2f87370 PB |
1067 | #define io_for_each_link(pos, head) \ |
1068 | for (pos = (head); pos; pos = pos->link) | |
1069 | ||
b895c9a6 | 1070 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1071 | { |
1072 | struct io_ring_ctx *ctx = req->ctx; | |
1073 | ||
269bbe5f | 1074 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1075 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1076 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1077 | } |
1078 | } | |
1079 | ||
f70865db PB |
1080 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1081 | { | |
1082 | bool got = percpu_ref_tryget(ref); | |
1083 | ||
1084 | /* already at zero, wait for ->release() */ | |
1085 | if (!got) | |
1086 | wait_for_completion(compl); | |
1087 | percpu_ref_resurrect(ref); | |
1088 | if (got) | |
1089 | percpu_ref_put(ref); | |
1090 | } | |
1091 | ||
08d23634 PB |
1092 | static bool io_match_task(struct io_kiocb *head, |
1093 | struct task_struct *task, | |
1094 | struct files_struct *files) | |
1095 | { | |
1096 | struct io_kiocb *req; | |
1097 | ||
68207680 | 1098 | if (task && head->task != task) |
08d23634 PB |
1099 | return false; |
1100 | if (!files) | |
1101 | return true; | |
1102 | ||
1103 | io_for_each_link(req, head) { | |
b05a1bcd | 1104 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1105 | return true; |
08d23634 PB |
1106 | } |
1107 | return false; | |
1108 | } | |
1109 | ||
c40f6379 JA |
1110 | static inline void req_set_fail_links(struct io_kiocb *req) |
1111 | { | |
e4335ed3 | 1112 | if (req->flags & REQ_F_LINK) |
c40f6379 JA |
1113 | req->flags |= REQ_F_FAIL_LINK; |
1114 | } | |
4a38aed2 | 1115 | |
2b188cc1 JA |
1116 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1117 | { | |
1118 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1119 | ||
0f158b4c | 1120 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1121 | } |
1122 | ||
8eb7e2d0 PB |
1123 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1124 | { | |
1125 | return !req->timeout.off; | |
1126 | } | |
1127 | ||
2b188cc1 JA |
1128 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1129 | { | |
1130 | struct io_ring_ctx *ctx; | |
78076bb6 | 1131 | int hash_bits; |
2b188cc1 JA |
1132 | |
1133 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1134 | if (!ctx) | |
1135 | return NULL; | |
1136 | ||
78076bb6 JA |
1137 | /* |
1138 | * Use 5 bits less than the max cq entries, that should give us around | |
1139 | * 32 entries per hash list if totally full and uniformly spread. | |
1140 | */ | |
1141 | hash_bits = ilog2(p->cq_entries); | |
1142 | hash_bits -= 5; | |
1143 | if (hash_bits <= 0) | |
1144 | hash_bits = 1; | |
1145 | ctx->cancel_hash_bits = hash_bits; | |
1146 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1147 | GFP_KERNEL); | |
1148 | if (!ctx->cancel_hash) | |
1149 | goto err; | |
1150 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1151 | ||
21482896 | 1152 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1153 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1154 | goto err; | |
2b188cc1 JA |
1155 | |
1156 | ctx->flags = p->flags; | |
90554200 | 1157 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1158 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1159 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1160 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1161 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1162 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1163 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1164 | mutex_init(&ctx->uring_lock); |
1165 | init_waitqueue_head(&ctx->wait); | |
1166 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1167 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1168 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1169 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1170 | spin_lock_init(&ctx->rsrc_ref_lock); |
1171 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1172 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1173 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1174 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1175 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1176 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1177 | return ctx; |
206aefde | 1178 | err: |
78076bb6 | 1179 | kfree(ctx->cancel_hash); |
206aefde JA |
1180 | kfree(ctx); |
1181 | return NULL; | |
2b188cc1 JA |
1182 | } |
1183 | ||
9cf7c104 | 1184 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1185 | { |
2bc9930e JA |
1186 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1187 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1188 | |
9cf7c104 | 1189 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1190 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1191 | } |
de0617e4 | 1192 | |
9d858b21 | 1193 | return false; |
de0617e4 JA |
1194 | } |
1195 | ||
ce3d5aae PB |
1196 | static void io_req_track_inflight(struct io_kiocb *req) |
1197 | { | |
ce3d5aae | 1198 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1199 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1200 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1201 | } |
1202 | } | |
1203 | ||
1e6fa521 JA |
1204 | static void io_prep_async_work(struct io_kiocb *req) |
1205 | { | |
1206 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1207 | struct io_ring_ctx *ctx = req->ctx; |
1208 | ||
003e8dcc JA |
1209 | if (!req->work.creds) |
1210 | req->work.creds = get_current_cred(); | |
1211 | ||
e1d675df PB |
1212 | req->work.list.next = NULL; |
1213 | req->work.flags = 0; | |
feaadc4f PB |
1214 | if (req->flags & REQ_F_FORCE_ASYNC) |
1215 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1216 | ||
1e6fa521 JA |
1217 | if (req->flags & REQ_F_ISREG) { |
1218 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1219 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1220 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1221 | if (def->unbound_nonreg_file) |
1222 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1223 | } | |
e1d675df PB |
1224 | |
1225 | switch (req->opcode) { | |
1226 | case IORING_OP_SPLICE: | |
1227 | case IORING_OP_TEE: | |
e1d675df PB |
1228 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1229 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1230 | break; | |
1231 | } | |
561fb04a | 1232 | } |
cccf0ee8 | 1233 | |
cbdcb435 | 1234 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1235 | { |
cbdcb435 | 1236 | struct io_kiocb *cur; |
54a91f3b | 1237 | |
f2f87370 PB |
1238 | io_for_each_link(cur, req) |
1239 | io_prep_async_work(cur); | |
561fb04a JA |
1240 | } |
1241 | ||
ebf93667 | 1242 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1243 | { |
a197f664 | 1244 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1245 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1246 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1247 | |
3bfe6106 JA |
1248 | BUG_ON(!tctx); |
1249 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1250 | |
cbdcb435 PB |
1251 | /* init ->work of the whole link before punting */ |
1252 | io_prep_async_link(req); | |
d07f1e8a PB |
1253 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1254 | &req->work, req->flags); | |
ebf93667 | 1255 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1256 | if (link) |
1257 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1258 | } |
1259 | ||
1ee4160c | 1260 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1261 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1262 | { |
e8c2bc1f | 1263 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1264 | |
fd9c7bc5 | 1265 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1266 | atomic_set(&req->ctx->cq_timeouts, |
1267 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1268 | list_del_init(&req->timeout.list); |
ff642164 | 1269 | io_cqring_fill_event(req, status, 0); |
216578e5 | 1270 | io_put_req_deferred(req, 1); |
5262f567 JA |
1271 | } |
1272 | } | |
1273 | ||
04518945 | 1274 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1275 | { |
04518945 | 1276 | do { |
27dc8338 PB |
1277 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1278 | struct io_defer_entry, list); | |
de0617e4 | 1279 | |
9cf7c104 | 1280 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1281 | break; |
27dc8338 | 1282 | list_del_init(&de->list); |
907d1df3 | 1283 | io_req_task_queue(de->req); |
27dc8338 | 1284 | kfree(de); |
04518945 PB |
1285 | } while (!list_empty(&ctx->defer_list)); |
1286 | } | |
1287 | ||
360428f8 | 1288 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1289 | { |
f010505b MDG |
1290 | u32 seq; |
1291 | ||
1292 | if (list_empty(&ctx->timeout_list)) | |
1293 | return; | |
1294 | ||
1295 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1296 | ||
1297 | do { | |
1298 | u32 events_needed, events_got; | |
360428f8 | 1299 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1300 | struct io_kiocb, timeout.list); |
de0617e4 | 1301 | |
8eb7e2d0 | 1302 | if (io_is_timeout_noseq(req)) |
360428f8 | 1303 | break; |
f010505b MDG |
1304 | |
1305 | /* | |
1306 | * Since seq can easily wrap around over time, subtract | |
1307 | * the last seq at which timeouts were flushed before comparing. | |
1308 | * Assuming not more than 2^31-1 events have happened since, | |
1309 | * these subtractions won't have wrapped, so we can check if | |
1310 | * target is in [last_seq, current_seq] by comparing the two. | |
1311 | */ | |
1312 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1313 | events_got = seq - ctx->cq_last_tm_flush; | |
1314 | if (events_got < events_needed) | |
360428f8 | 1315 | break; |
bfe68a22 | 1316 | |
135fcde8 | 1317 | list_del_init(&req->timeout.list); |
1ee4160c | 1318 | io_kill_timeout(req, 0); |
f010505b MDG |
1319 | } while (!list_empty(&ctx->timeout_list)); |
1320 | ||
1321 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1322 | } |
5262f567 | 1323 | |
360428f8 PB |
1324 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1325 | { | |
1326 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1327 | |
1328 | /* order cqe stores with ring update */ | |
1329 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1330 | |
04518945 PB |
1331 | if (unlikely(!list_empty(&ctx->defer_list))) |
1332 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1333 | } |
1334 | ||
90554200 JA |
1335 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1336 | { | |
1337 | struct io_rings *r = ctx->rings; | |
1338 | ||
1339 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1340 | } | |
1341 | ||
888aae2e PB |
1342 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1343 | { | |
1344 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1345 | } | |
1346 | ||
8d13326e | 1347 | static inline struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1348 | { |
75b28aff | 1349 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1350 | unsigned tail; |
1351 | ||
115e12e5 SB |
1352 | /* |
1353 | * writes to the cq entry need to come after reading head; the | |
1354 | * control dependency is enough as we're using WRITE_ONCE to | |
1355 | * fill the cq entry | |
1356 | */ | |
888aae2e | 1357 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1358 | return NULL; |
1359 | ||
888aae2e | 1360 | tail = ctx->cached_cq_tail++; |
75b28aff | 1361 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1362 | } |
1363 | ||
f2842ab5 JA |
1364 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1365 | { | |
44c769de | 1366 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1367 | return false; |
7e55a19c SG |
1368 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1369 | return false; | |
44c769de | 1370 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1371 | } |
1372 | ||
b41e9852 | 1373 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1374 | { |
b1445e59 PB |
1375 | /* see waitqueue_active() comment */ |
1376 | smp_mb(); | |
1377 | ||
1d7bb1d5 JA |
1378 | if (waitqueue_active(&ctx->wait)) |
1379 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1380 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1381 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1382 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1383 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1384 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1385 | wake_up_interruptible(&ctx->cq_wait); |
1386 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1387 | } | |
1d7bb1d5 JA |
1388 | } |
1389 | ||
80c18e4a PB |
1390 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1391 | { | |
b1445e59 PB |
1392 | /* see waitqueue_active() comment */ |
1393 | smp_mb(); | |
1394 | ||
80c18e4a PB |
1395 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1396 | if (waitqueue_active(&ctx->wait)) | |
1397 | wake_up(&ctx->wait); | |
1398 | } | |
1399 | if (io_should_trigger_evfd(ctx)) | |
1400 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1401 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1402 | wake_up_interruptible(&ctx->cq_wait); |
1403 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1404 | } | |
80c18e4a PB |
1405 | } |
1406 | ||
c4a2ed72 | 1407 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1408 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 JA |
1409 | { |
1410 | struct io_rings *rings = ctx->rings; | |
1d7bb1d5 | 1411 | unsigned long flags; |
b18032bb | 1412 | bool all_flushed, posted; |
1d7bb1d5 | 1413 | |
e23de15f PB |
1414 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1415 | return false; | |
1d7bb1d5 | 1416 | |
b18032bb | 1417 | posted = false; |
1d7bb1d5 | 1418 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae PB |
1419 | while (!list_empty(&ctx->cq_overflow_list)) { |
1420 | struct io_uring_cqe *cqe = io_get_cqring(ctx); | |
1421 | struct io_overflow_cqe *ocqe; | |
e6c8aa9a | 1422 | |
1d7bb1d5 JA |
1423 | if (!cqe && !force) |
1424 | break; | |
6c2450ae PB |
1425 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1426 | struct io_overflow_cqe, list); | |
1427 | if (cqe) | |
1428 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1429 | else | |
1d7bb1d5 | 1430 | WRITE_ONCE(ctx->rings->cq_overflow, |
6c2450ae | 1431 | ++ctx->cached_cq_overflow); |
b18032bb | 1432 | posted = true; |
6c2450ae PB |
1433 | list_del(&ocqe->list); |
1434 | kfree(ocqe); | |
1d7bb1d5 JA |
1435 | } |
1436 | ||
09e88404 PB |
1437 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1438 | if (all_flushed) { | |
1439 | clear_bit(0, &ctx->sq_check_overflow); | |
1440 | clear_bit(0, &ctx->cq_check_overflow); | |
1441 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1442 | } | |
46930143 | 1443 | |
b18032bb JA |
1444 | if (posted) |
1445 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1446 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1447 | if (posted) |
1448 | io_cqring_ev_posted(ctx); | |
09e88404 | 1449 | return all_flushed; |
1d7bb1d5 JA |
1450 | } |
1451 | ||
6c2450ae | 1452 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1453 | { |
ca0a2651 JA |
1454 | bool ret = true; |
1455 | ||
6c503150 PB |
1456 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1457 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1458 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1459 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1460 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1461 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1462 | mutex_unlock(&ctx->uring_lock); | |
1463 | } | |
ca0a2651 JA |
1464 | |
1465 | return ret; | |
6c503150 PB |
1466 | } |
1467 | ||
abc54d63 JA |
1468 | /* |
1469 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1470 | * see commit f958d7b528b1 for details. | |
1471 | */ | |
1472 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1473 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1474 | ||
de9b4cca JA |
1475 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1476 | { | |
abc54d63 | 1477 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1478 | } |
1479 | ||
1480 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1481 | { | |
abc54d63 JA |
1482 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1483 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1484 | } |
1485 | ||
1486 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1487 | { | |
abc54d63 JA |
1488 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1489 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1490 | } |
1491 | ||
1492 | static inline void req_ref_put(struct io_kiocb *req) | |
1493 | { | |
abc54d63 | 1494 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1495 | } |
1496 | ||
1497 | static inline void req_ref_get(struct io_kiocb *req) | |
1498 | { | |
abc54d63 JA |
1499 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1500 | atomic_inc(&req->refs); | |
de9b4cca JA |
1501 | } |
1502 | ||
8d13326e PB |
1503 | static bool io_cqring_event_overflow(struct io_kiocb *req, long res, |
1504 | unsigned int cflags) | |
2b188cc1 | 1505 | { |
78e19bbe | 1506 | struct io_ring_ctx *ctx = req->ctx; |
cce4b8b0 | 1507 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1508 | |
cce4b8b0 PB |
1509 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1510 | if (!ocqe) { | |
1511 | /* | |
1512 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1513 | * or cannot allocate an overflow entry, then we need to drop it | |
1514 | * on the floor. | |
1515 | */ | |
1516 | WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow); | |
1517 | return false; | |
2b188cc1 | 1518 | } |
cce4b8b0 PB |
1519 | if (list_empty(&ctx->cq_overflow_list)) { |
1520 | set_bit(0, &ctx->sq_check_overflow); | |
1521 | set_bit(0, &ctx->cq_check_overflow); | |
1522 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; | |
1523 | } | |
1524 | ocqe->cqe.user_data = req->user_data; | |
1525 | ocqe->cqe.res = res; | |
1526 | ocqe->cqe.flags = cflags; | |
1527 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1528 | return true; | |
2b188cc1 JA |
1529 | } |
1530 | ||
8d13326e PB |
1531 | static inline bool __io_cqring_fill_event(struct io_kiocb *req, long res, |
1532 | unsigned int cflags) | |
1533 | { | |
1534 | struct io_ring_ctx *ctx = req->ctx; | |
1535 | struct io_uring_cqe *cqe; | |
1536 | ||
1537 | trace_io_uring_complete(ctx, req->user_data, res, cflags); | |
1538 | ||
1539 | /* | |
1540 | * If we can't get a cq entry, userspace overflowed the | |
1541 | * submission (by quite a lot). Increment the overflow count in | |
1542 | * the ring. | |
1543 | */ | |
1544 | cqe = io_get_cqring(ctx); | |
1545 | if (likely(cqe)) { | |
1546 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1547 | WRITE_ONCE(cqe->res, res); | |
1548 | WRITE_ONCE(cqe->flags, cflags); | |
1549 | return true; | |
1550 | } | |
1551 | return io_cqring_event_overflow(req, res, cflags); | |
1552 | } | |
1553 | ||
1554 | /* not as hot to bloat with inlining */ | |
1555 | static noinline bool io_cqring_fill_event(struct io_kiocb *req, long res, | |
1556 | unsigned int cflags) | |
1557 | { | |
1558 | return __io_cqring_fill_event(req, res, cflags); | |
1559 | } | |
1560 | ||
7a612350 PB |
1561 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1562 | unsigned int cflags) | |
2b188cc1 | 1563 | { |
78e19bbe | 1564 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1565 | unsigned long flags; |
1566 | ||
1567 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
8d13326e | 1568 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1569 | /* |
1570 | * If we're the last reference to this request, add to our locked | |
1571 | * free_list cache. | |
1572 | */ | |
de9b4cca | 1573 | if (req_ref_put_and_test(req)) { |
c7dae4ba JA |
1574 | struct io_comp_state *cs = &ctx->submit_state.comp; |
1575 | ||
7a612350 PB |
1576 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1577 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1578 | io_disarm_next(req); | |
1579 | if (req->link) { | |
1580 | io_req_task_queue(req->link); | |
1581 | req->link = NULL; | |
1582 | } | |
1583 | } | |
c7dae4ba JA |
1584 | io_dismantle_req(req); |
1585 | io_put_task(req->task, 1); | |
1586 | list_add(&req->compl.list, &cs->locked_free_list); | |
1587 | cs->locked_free_nr++; | |
180f829f PB |
1588 | } else { |
1589 | if (!percpu_ref_tryget(&ctx->refs)) | |
1590 | req = NULL; | |
1591 | } | |
7a612350 | 1592 | io_commit_cqring(ctx); |
2b188cc1 | 1593 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1594 | |
180f829f PB |
1595 | if (req) { |
1596 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1597 | percpu_ref_put(&ctx->refs); |
180f829f | 1598 | } |
229a7b63 JA |
1599 | } |
1600 | ||
4e3d9ff9 JA |
1601 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1602 | { | |
75652a30 JA |
1603 | return req->flags & (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | |
1604 | REQ_F_POLLED); | |
4e3d9ff9 JA |
1605 | } |
1606 | ||
a38d68db | 1607 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1608 | unsigned int cflags) |
229a7b63 | 1609 | { |
4e3d9ff9 | 1610 | if (io_req_needs_clean(req)) |
68fb8979 | 1611 | io_clean_op(req); |
a38d68db PB |
1612 | req->result = res; |
1613 | req->compl.cflags = cflags; | |
e342c807 | 1614 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1615 | } |
1616 | ||
889fca73 PB |
1617 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1618 | long res, unsigned cflags) | |
bcda7baa | 1619 | { |
889fca73 PB |
1620 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1621 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1622 | else |
c7dae4ba | 1623 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1624 | } |
1625 | ||
a38d68db | 1626 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1627 | { |
889fca73 | 1628 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1629 | } |
1630 | ||
f41db273 PB |
1631 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1632 | { | |
1633 | req_set_fail_links(req); | |
1634 | io_put_req(req); | |
1635 | io_req_complete_post(req, res, 0); | |
1636 | } | |
1637 | ||
dac7a098 PB |
1638 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1639 | struct io_comp_state *cs) | |
1640 | { | |
1641 | spin_lock_irq(&ctx->completion_lock); | |
1642 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1643 | cs->locked_free_nr = 0; | |
1644 | spin_unlock_irq(&ctx->completion_lock); | |
1645 | } | |
1646 | ||
dd78f492 | 1647 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1648 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1649 | { |
c7dae4ba JA |
1650 | struct io_submit_state *state = &ctx->submit_state; |
1651 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1652 | int nr; |
0ddf92e8 | 1653 | |
c7dae4ba JA |
1654 | /* |
1655 | * If we have more than a batch's worth of requests in our IRQ side | |
1656 | * locked cache, grab the lock and move them over to our submission | |
1657 | * side cache. | |
1658 | */ | |
dac7a098 PB |
1659 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) |
1660 | io_flush_cached_locked_reqs(ctx, cs); | |
0ddf92e8 | 1661 | |
dd78f492 | 1662 | nr = state->free_reqs; |
c7dae4ba | 1663 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1664 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1665 | struct io_kiocb, compl.list); | |
1666 | ||
1b4c351f | 1667 | list_del(&req->compl.list); |
dd78f492 PB |
1668 | state->reqs[nr++] = req; |
1669 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1670 | break; |
1b4c351f JA |
1671 | } |
1672 | ||
dd78f492 PB |
1673 | state->free_reqs = nr; |
1674 | return nr != 0; | |
0ddf92e8 JA |
1675 | } |
1676 | ||
e5d1bc0a | 1677 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1678 | { |
e5d1bc0a PB |
1679 | struct io_submit_state *state = &ctx->submit_state; |
1680 | ||
1681 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1682 | ||
f6b6c7d6 | 1683 | if (!state->free_reqs) { |
291b2821 | 1684 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1685 | int ret; |
1686 | ||
c7dae4ba | 1687 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1688 | goto got_req; |
1689 | ||
bf019da7 PB |
1690 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1691 | state->reqs); | |
fd6fab2c JA |
1692 | |
1693 | /* | |
1694 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1695 | * retry single alloc to be on the safe side. | |
1696 | */ | |
1697 | if (unlikely(ret <= 0)) { | |
1698 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1699 | if (!state->reqs[0]) | |
3893f39f | 1700 | return NULL; |
fd6fab2c JA |
1701 | ret = 1; |
1702 | } | |
291b2821 | 1703 | state->free_reqs = ret; |
2b188cc1 | 1704 | } |
e5d1bc0a | 1705 | got_req: |
291b2821 PB |
1706 | state->free_reqs--; |
1707 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1708 | } |
1709 | ||
e1d767f0 | 1710 | static inline void io_put_file(struct file *file) |
8da11c19 | 1711 | { |
e1d767f0 | 1712 | if (file) |
8da11c19 PB |
1713 | fput(file); |
1714 | } | |
1715 | ||
4edf20f9 | 1716 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1717 | { |
094bae49 PB |
1718 | unsigned int flags = req->flags; |
1719 | ||
e1d767f0 PB |
1720 | if (!(flags & REQ_F_FIXED_FILE)) |
1721 | io_put_file(req->file); | |
4e3d9ff9 | 1722 | if (io_req_needs_clean(req) || (req->flags & REQ_F_INFLIGHT)) { |
094bae49 | 1723 | io_clean_op(req); |
094bae49 | 1724 | if (req->flags & REQ_F_INFLIGHT) { |
b303fe2e | 1725 | struct io_uring_task *tctx = req->task->io_uring; |
094bae49 | 1726 | |
b303fe2e | 1727 | atomic_dec(&tctx->inflight_tracked); |
094bae49 PB |
1728 | req->flags &= ~REQ_F_INFLIGHT; |
1729 | } | |
1730 | } | |
269bbe5f BM |
1731 | if (req->fixed_rsrc_refs) |
1732 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1733 | if (req->async_data) |
1734 | kfree(req->async_data); | |
003e8dcc JA |
1735 | if (req->work.creds) { |
1736 | put_cred(req->work.creds); | |
1737 | req->work.creds = NULL; | |
1738 | } | |
e65ef56d JA |
1739 | } |
1740 | ||
b23fcf47 | 1741 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1742 | static inline void io_put_task(struct task_struct *task, int nr) |
1743 | { | |
1744 | struct io_uring_task *tctx = task->io_uring; | |
1745 | ||
1746 | percpu_counter_sub(&tctx->inflight, nr); | |
1747 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1748 | wake_up(&tctx->wait); | |
1749 | put_task_struct_many(task, nr); | |
1750 | } | |
1751 | ||
216578e5 | 1752 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1753 | { |
51a4cc11 | 1754 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1755 | |
216578e5 | 1756 | io_dismantle_req(req); |
7c660731 | 1757 | io_put_task(req->task, 1); |
c6ca97b3 | 1758 | |
3893f39f | 1759 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1760 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1761 | } |
1762 | ||
f2f87370 PB |
1763 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1764 | { | |
1765 | struct io_kiocb *nxt = req->link; | |
1766 | ||
1767 | req->link = nxt->link; | |
1768 | nxt->link = NULL; | |
1769 | } | |
1770 | ||
33cc89a9 PB |
1771 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1772 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1773 | { |
33cc89a9 | 1774 | struct io_kiocb *link = req->link; |
f2f87370 | 1775 | |
900fad45 PB |
1776 | /* |
1777 | * Can happen if a linked timeout fired and link had been like | |
1778 | * req -> link t-out -> link t-out [-> ...] | |
1779 | */ | |
c9abd7ad PB |
1780 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1781 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1782 | |
f2f87370 | 1783 | io_remove_next_linked(req); |
90cd7e42 | 1784 | link->timeout.head = NULL; |
fd9c7bc5 | 1785 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ff642164 | 1786 | io_cqring_fill_event(link, -ECANCELED, 0); |
33cc89a9 | 1787 | io_put_req_deferred(link, 1); |
d4729fbd | 1788 | return true; |
c9abd7ad PB |
1789 | } |
1790 | } | |
d4729fbd | 1791 | return false; |
7c86ffee PB |
1792 | } |
1793 | ||
d148ca4b | 1794 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1795 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1796 | { |
33cc89a9 | 1797 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1798 | |
f2f87370 | 1799 | req->link = NULL; |
f2f87370 PB |
1800 | while (link) { |
1801 | nxt = link->link; | |
1802 | link->link = NULL; | |
2665abfd | 1803 | |
f2f87370 | 1804 | trace_io_uring_fail_link(req, link); |
ff642164 | 1805 | io_cqring_fill_event(link, -ECANCELED, 0); |
1575f21a | 1806 | io_put_req_deferred(link, 2); |
f2f87370 | 1807 | link = nxt; |
9e645e11 | 1808 | } |
33cc89a9 | 1809 | } |
9e645e11 | 1810 | |
33cc89a9 PB |
1811 | static bool io_disarm_next(struct io_kiocb *req) |
1812 | __must_hold(&req->ctx->completion_lock) | |
1813 | { | |
1814 | bool posted = false; | |
1815 | ||
1816 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1817 | posted = io_kill_linked_timeout(req); | |
e4335ed3 PB |
1818 | if (unlikely((req->flags & REQ_F_FAIL_LINK) && |
1819 | !(req->flags & REQ_F_HARDLINK))) { | |
33cc89a9 PB |
1820 | posted |= (req->link != NULL); |
1821 | io_fail_links(req); | |
1822 | } | |
1823 | return posted; | |
9e645e11 JA |
1824 | } |
1825 | ||
3fa5e0f3 | 1826 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1827 | { |
33cc89a9 | 1828 | struct io_kiocb *nxt; |
944e58bf | 1829 | |
9e645e11 JA |
1830 | /* |
1831 | * If LINK is set, we have dependent requests in this chain. If we | |
1832 | * didn't fail this request, queue the first one up, moving any other | |
1833 | * dependencies to the next request. In case of failure, fail the rest | |
1834 | * of the chain. | |
1835 | */ | |
33cc89a9 PB |
1836 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1837 | struct io_ring_ctx *ctx = req->ctx; | |
1838 | unsigned long flags; | |
1839 | bool posted; | |
1840 | ||
1841 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1842 | posted = io_disarm_next(req); | |
1843 | if (posted) | |
1844 | io_commit_cqring(req->ctx); | |
1845 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1846 | if (posted) | |
1847 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1848 | } |
33cc89a9 PB |
1849 | nxt = req->link; |
1850 | req->link = NULL; | |
1851 | return nxt; | |
4d7dd462 | 1852 | } |
9e645e11 | 1853 | |
f2f87370 | 1854 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1855 | { |
cdbff982 | 1856 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1857 | return NULL; |
1858 | return __io_req_find_next(req); | |
1859 | } | |
1860 | ||
2c32395d PB |
1861 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1862 | { | |
1863 | if (!ctx) | |
1864 | return; | |
1865 | if (ctx->submit_state.comp.nr) { | |
1866 | mutex_lock(&ctx->uring_lock); | |
1867 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1868 | mutex_unlock(&ctx->uring_lock); | |
1869 | } | |
1870 | percpu_ref_put(&ctx->refs); | |
1871 | } | |
1872 | ||
7cbf1722 | 1873 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1874 | { |
65453d1e | 1875 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1876 | struct io_wq_work_list list; |
1877 | struct io_wq_work_node *node; | |
c2c4c83c | 1878 | |
7cbf1722 JA |
1879 | if (wq_list_empty(&tctx->task_list)) |
1880 | return false; | |
6200b0ae | 1881 | |
0b81e80c | 1882 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1883 | list = tctx->task_list; |
1884 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1885 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1886 | |
7cbf1722 JA |
1887 | node = list.first; |
1888 | while (node) { | |
1889 | struct io_wq_work_node *next = node->next; | |
1890 | struct io_kiocb *req; | |
0ba9c9ed | 1891 | |
7cbf1722 | 1892 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1893 | if (req->ctx != ctx) { |
1894 | ctx_flush_and_put(ctx); | |
1895 | ctx = req->ctx; | |
1896 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1897 | } |
65453d1e | 1898 | |
2c32395d PB |
1899 | req->task_work.func(&req->task_work); |
1900 | node = next; | |
7cbf1722 JA |
1901 | } |
1902 | ||
2c32395d | 1903 | ctx_flush_and_put(ctx); |
7cbf1722 | 1904 | return list.first != NULL; |
c2c4c83c JA |
1905 | } |
1906 | ||
7cbf1722 | 1907 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1908 | { |
7cbf1722 | 1909 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1910 | |
1d5f360d JA |
1911 | clear_bit(0, &tctx->task_state); |
1912 | ||
7cbf1722 JA |
1913 | while (__tctx_task_work(tctx)) |
1914 | cond_resched(); | |
7cbf1722 JA |
1915 | } |
1916 | ||
c15b79de | 1917 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1918 | { |
c15b79de | 1919 | struct task_struct *tsk = req->task; |
7cbf1722 | 1920 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1921 | enum task_work_notify_mode notify; |
7cbf1722 | 1922 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1923 | unsigned long flags; |
c15b79de PB |
1924 | int ret = 0; |
1925 | ||
1926 | if (unlikely(tsk->flags & PF_EXITING)) | |
1927 | return -ESRCH; | |
7cbf1722 JA |
1928 | |
1929 | WARN_ON_ONCE(!tctx); | |
1930 | ||
0b81e80c | 1931 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1932 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1933 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1934 | |
1935 | /* task_work already pending, we're done */ | |
1936 | if (test_bit(0, &tctx->task_state) || | |
1937 | test_and_set_bit(0, &tctx->task_state)) | |
1938 | return 0; | |
1939 | ||
c15b79de PB |
1940 | /* |
1941 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1942 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1943 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1944 | * will do the job. | |
1945 | */ | |
1946 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1947 | ||
1948 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1949 | wake_up_process(tsk); | |
7cbf1722 | 1950 | return 0; |
c15b79de | 1951 | } |
7cbf1722 JA |
1952 | |
1953 | /* | |
1954 | * Slow path - we failed, find and delete work. if the work is not | |
1955 | * in the list, it got run and we're fine. | |
1956 | */ | |
0b81e80c | 1957 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1958 | wq_list_for_each(node, prev, &tctx->task_list) { |
1959 | if (&req->io_task_work.node == node) { | |
1960 | wq_list_del(&tctx->task_list, node, prev); | |
1961 | ret = 1; | |
1962 | break; | |
1963 | } | |
1964 | } | |
0b81e80c | 1965 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1966 | clear_bit(0, &tctx->task_state); |
1967 | return ret; | |
1968 | } | |
1969 | ||
9b465711 PB |
1970 | static bool io_run_task_work_head(struct callback_head **work_head) |
1971 | { | |
1972 | struct callback_head *work, *next; | |
1973 | bool executed = false; | |
1974 | ||
1975 | do { | |
1976 | work = xchg(work_head, NULL); | |
1977 | if (!work) | |
1978 | break; | |
1979 | ||
1980 | do { | |
1981 | next = work->next; | |
1982 | work->func(work); | |
1983 | work = next; | |
1984 | cond_resched(); | |
1985 | } while (work); | |
1986 | executed = true; | |
1987 | } while (1); | |
1988 | ||
1989 | return executed; | |
1990 | } | |
1991 | ||
1992 | static void io_task_work_add_head(struct callback_head **work_head, | |
1993 | struct callback_head *task_work) | |
eab30c4d | 1994 | { |
7c25c0d1 | 1995 | struct callback_head *head; |
eab30c4d | 1996 | |
7c25c0d1 | 1997 | do { |
9b465711 PB |
1998 | head = READ_ONCE(*work_head); |
1999 | task_work->next = head; | |
2000 | } while (cmpxchg(work_head, head, task_work) != head); | |
2001 | } | |
2002 | ||
2003 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2004 | task_work_func_t cb) | |
2005 | { | |
2006 | init_task_work(&req->task_work, cb); | |
2007 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2008 | } |
2009 | ||
c40f6379 JA |
2010 | static void io_req_task_cancel(struct callback_head *cb) |
2011 | { | |
2012 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2013 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2014 | |
e83acd7d | 2015 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2016 | mutex_lock(&ctx->uring_lock); |
2593553a | 2017 | io_req_complete_failed(req, req->result); |
792bb6eb | 2018 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2019 | } |
2020 | ||
2021 | static void __io_req_task_submit(struct io_kiocb *req) | |
2022 | { | |
2023 | struct io_ring_ctx *ctx = req->ctx; | |
2024 | ||
04fc6c80 | 2025 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2026 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2027 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2028 | __io_queue_sqe(req); |
81b6d05c | 2029 | else |
2593553a | 2030 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2031 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2032 | } |
2033 | ||
2034 | static void io_req_task_submit(struct callback_head *cb) | |
2035 | { | |
2036 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2037 | ||
2038 | __io_req_task_submit(req); | |
2039 | } | |
2040 | ||
2c4b8eb6 | 2041 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2042 | { |
2c4b8eb6 PB |
2043 | req->result = ret; |
2044 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2045 | |
2c4b8eb6 | 2046 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2047 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2048 | } |
2049 | ||
2c4b8eb6 | 2050 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2051 | { |
2c4b8eb6 | 2052 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2053 | |
2054 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2055 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2056 | } |
2057 | ||
f2f87370 | 2058 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2059 | { |
9b5f7bd9 | 2060 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2061 | |
2062 | if (nxt) | |
906a8c3f | 2063 | io_req_task_queue(nxt); |
c69f8dbe JL |
2064 | } |
2065 | ||
c3524383 | 2066 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2067 | { |
c3524383 PB |
2068 | io_queue_next(req); |
2069 | __io_free_req(req); | |
2070 | } | |
8766dd51 | 2071 | |
2d6500d4 | 2072 | struct req_batch { |
5af1d13e PB |
2073 | struct task_struct *task; |
2074 | int task_refs; | |
1b4c351f | 2075 | int ctx_refs; |
2d6500d4 PB |
2076 | }; |
2077 | ||
5af1d13e PB |
2078 | static inline void io_init_req_batch(struct req_batch *rb) |
2079 | { | |
5af1d13e | 2080 | rb->task_refs = 0; |
9ae72463 | 2081 | rb->ctx_refs = 0; |
5af1d13e PB |
2082 | rb->task = NULL; |
2083 | } | |
2084 | ||
2d6500d4 PB |
2085 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2086 | struct req_batch *rb) | |
2087 | { | |
6e833d53 | 2088 | if (rb->task) |
7c660731 | 2089 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2090 | if (rb->ctx_refs) |
2091 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2092 | } |
2093 | ||
6ff119a6 PB |
2094 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2095 | struct io_submit_state *state) | |
2d6500d4 | 2096 | { |
f2f87370 | 2097 | io_queue_next(req); |
96670657 | 2098 | io_dismantle_req(req); |
2d6500d4 | 2099 | |
e3bc8e9d | 2100 | if (req->task != rb->task) { |
7c660731 PB |
2101 | if (rb->task) |
2102 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2103 | rb->task = req->task; |
2104 | rb->task_refs = 0; | |
5af1d13e | 2105 | } |
e3bc8e9d | 2106 | rb->task_refs++; |
9ae72463 | 2107 | rb->ctx_refs++; |
5af1d13e | 2108 | |
bd759045 | 2109 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2110 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2111 | else |
2112 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2113 | } |
2114 | ||
905c172f PB |
2115 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2116 | struct io_ring_ctx *ctx) | |
2117 | { | |
2118 | int i, nr = cs->nr; | |
2119 | struct io_kiocb *req; | |
2120 | struct req_batch rb; | |
2121 | ||
2122 | io_init_req_batch(&rb); | |
2123 | spin_lock_irq(&ctx->completion_lock); | |
2124 | for (i = 0; i < nr; i++) { | |
2125 | req = cs->reqs[i]; | |
8d13326e | 2126 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
905c172f PB |
2127 | } |
2128 | io_commit_cqring(ctx); | |
2129 | spin_unlock_irq(&ctx->completion_lock); | |
2130 | ||
2131 | io_cqring_ev_posted(ctx); | |
2132 | for (i = 0; i < nr; i++) { | |
2133 | req = cs->reqs[i]; | |
2134 | ||
2135 | /* submission and completion refs */ | |
de9b4cca | 2136 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2137 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2138 | } |
2139 | ||
2140 | io_req_free_batch_finish(ctx, &rb); | |
2141 | cs->nr = 0; | |
7a743e22 PB |
2142 | } |
2143 | ||
ba816ad6 JA |
2144 | /* |
2145 | * Drop reference to request, return next in chain (if there is one) if this | |
2146 | * was the last reference to this request. | |
2147 | */ | |
0d85035a | 2148 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2149 | { |
9b5f7bd9 PB |
2150 | struct io_kiocb *nxt = NULL; |
2151 | ||
de9b4cca | 2152 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2153 | nxt = io_req_find_next(req); |
4d7dd462 | 2154 | __io_free_req(req); |
2a44f467 | 2155 | } |
9b5f7bd9 | 2156 | return nxt; |
2b188cc1 JA |
2157 | } |
2158 | ||
0d85035a | 2159 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2160 | { |
de9b4cca | 2161 | if (req_ref_put_and_test(req)) |
e65ef56d | 2162 | io_free_req(req); |
2b188cc1 JA |
2163 | } |
2164 | ||
216578e5 PB |
2165 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2166 | { | |
2167 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2168 | ||
2169 | io_free_req(req); | |
2170 | } | |
2171 | ||
2172 | static void io_free_req_deferred(struct io_kiocb *req) | |
2173 | { | |
7cbf1722 | 2174 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2175 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2176 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2177 | } |
2178 | ||
2179 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2180 | { | |
de9b4cca | 2181 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2182 | io_free_req_deferred(req); |
2183 | } | |
2184 | ||
6c503150 | 2185 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2186 | { |
2187 | /* See comment at the top of this file */ | |
2188 | smp_rmb(); | |
e23de15f | 2189 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2190 | } |
2191 | ||
fb5ccc98 PB |
2192 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2193 | { | |
2194 | struct io_rings *rings = ctx->rings; | |
2195 | ||
2196 | /* make sure SQ entry isn't read before tail */ | |
2197 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2198 | } | |
2199 | ||
8ff069bf | 2200 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2201 | { |
8ff069bf | 2202 | unsigned int cflags; |
e94f141b | 2203 | |
bcda7baa JA |
2204 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2205 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2206 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2207 | kfree(kbuf); |
2208 | return cflags; | |
e94f141b JA |
2209 | } |
2210 | ||
8ff069bf | 2211 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2212 | { |
4d954c25 | 2213 | struct io_buffer *kbuf; |
bcda7baa | 2214 | |
4d954c25 | 2215 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2216 | return io_put_kbuf(req, kbuf); |
2217 | } | |
2218 | ||
4c6e277c JA |
2219 | static inline bool io_run_task_work(void) |
2220 | { | |
6200b0ae JA |
2221 | /* |
2222 | * Not safe to run on exiting task, and the task_work handling will | |
2223 | * not add work to such a task. | |
2224 | */ | |
2225 | if (unlikely(current->flags & PF_EXITING)) | |
2226 | return false; | |
4c6e277c JA |
2227 | if (current->task_works) { |
2228 | __set_current_state(TASK_RUNNING); | |
2229 | task_work_run(); | |
2230 | return true; | |
2231 | } | |
2232 | ||
2233 | return false; | |
bcda7baa JA |
2234 | } |
2235 | ||
def596e9 JA |
2236 | /* |
2237 | * Find and free completed poll iocbs | |
2238 | */ | |
2239 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2240 | struct list_head *done) | |
2241 | { | |
8237e045 | 2242 | struct req_batch rb; |
def596e9 | 2243 | struct io_kiocb *req; |
bbde017a XW |
2244 | |
2245 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2246 | smp_rmb(); | |
def596e9 | 2247 | |
5af1d13e | 2248 | io_init_req_batch(&rb); |
def596e9 | 2249 | while (!list_empty(done)) { |
bcda7baa JA |
2250 | int cflags = 0; |
2251 | ||
d21ffe7e | 2252 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2253 | list_del(&req->inflight_entry); |
2254 | ||
8c130827 PB |
2255 | if (READ_ONCE(req->result) == -EAGAIN && |
2256 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2257 | req->iopoll_completed = 0; |
8c130827 PB |
2258 | req_ref_get(req); |
2259 | io_queue_async_work(req); | |
2260 | continue; | |
bbde017a | 2261 | } |
def596e9 | 2262 | |
bcda7baa | 2263 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2264 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2265 | |
8d13326e | 2266 | __io_cqring_fill_event(req, req->result, cflags); |
def596e9 JA |
2267 | (*nr_events)++; |
2268 | ||
de9b4cca | 2269 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2270 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2271 | } |
def596e9 | 2272 | |
09bb8394 | 2273 | io_commit_cqring(ctx); |
80c18e4a | 2274 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2275 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2276 | } |
2277 | ||
def596e9 JA |
2278 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2279 | long min) | |
2280 | { | |
2281 | struct io_kiocb *req, *tmp; | |
2282 | LIST_HEAD(done); | |
2283 | bool spin; | |
2284 | int ret; | |
2285 | ||
2286 | /* | |
2287 | * Only spin for completions if we don't have multiple devices hanging | |
2288 | * off our complete list, and we're under the requested amount. | |
2289 | */ | |
2290 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2291 | ||
2292 | ret = 0; | |
d21ffe7e | 2293 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2294 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2295 | |
2296 | /* | |
581f9810 BM |
2297 | * Move completed and retryable entries to our local lists. |
2298 | * If we find a request that requires polling, break out | |
2299 | * and complete those lists first, if we have entries there. | |
def596e9 | 2300 | */ |
65a6543d | 2301 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2302 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2303 | continue; |
2304 | } | |
2305 | if (!list_empty(&done)) | |
2306 | break; | |
2307 | ||
2308 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2309 | if (ret < 0) | |
2310 | break; | |
2311 | ||
3aadc23e PB |
2312 | /* iopoll may have completed current req */ |
2313 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2314 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2315 | |
def596e9 JA |
2316 | if (ret && spin) |
2317 | spin = false; | |
2318 | ret = 0; | |
2319 | } | |
2320 | ||
2321 | if (!list_empty(&done)) | |
2322 | io_iopoll_complete(ctx, nr_events, &done); | |
2323 | ||
2324 | return ret; | |
2325 | } | |
2326 | ||
def596e9 JA |
2327 | /* |
2328 | * We can't just wait for polled events to come to us, we have to actively | |
2329 | * find and complete them. | |
2330 | */ | |
b2edc0a7 | 2331 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2332 | { |
2333 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2334 | return; | |
2335 | ||
2336 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2337 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2338 | unsigned int nr_events = 0; |
2339 | ||
b2edc0a7 | 2340 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2341 | |
b2edc0a7 PB |
2342 | /* let it sleep and repeat later if can't complete a request */ |
2343 | if (nr_events == 0) | |
2344 | break; | |
08f5439f JA |
2345 | /* |
2346 | * Ensure we allow local-to-the-cpu processing to take place, | |
2347 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2348 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2349 | */ |
3fcee5a6 PB |
2350 | if (need_resched()) { |
2351 | mutex_unlock(&ctx->uring_lock); | |
2352 | cond_resched(); | |
2353 | mutex_lock(&ctx->uring_lock); | |
2354 | } | |
def596e9 JA |
2355 | } |
2356 | mutex_unlock(&ctx->uring_lock); | |
2357 | } | |
2358 | ||
7668b92a | 2359 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2360 | { |
7668b92a | 2361 | unsigned int nr_events = 0; |
e9979b36 | 2362 | int ret = 0; |
500f9fba | 2363 | |
c7849be9 XW |
2364 | /* |
2365 | * We disallow the app entering submit/complete with polling, but we | |
2366 | * still need to lock the ring to prevent racing with polled issue | |
2367 | * that got punted to a workqueue. | |
2368 | */ | |
2369 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2370 | /* |
2371 | * Don't enter poll loop if we already have events pending. | |
2372 | * If we do, we can potentially be spinning for commands that | |
2373 | * already triggered a CQE (eg in error). | |
2374 | */ | |
2375 | if (test_bit(0, &ctx->cq_check_overflow)) | |
2376 | __io_cqring_overflow_flush(ctx, false); | |
2377 | if (io_cqring_events(ctx)) | |
2378 | goto out; | |
def596e9 | 2379 | do { |
500f9fba JA |
2380 | /* |
2381 | * If a submit got punted to a workqueue, we can have the | |
2382 | * application entering polling for a command before it gets | |
2383 | * issued. That app will hold the uring_lock for the duration | |
2384 | * of the poll right here, so we need to take a breather every | |
2385 | * now and then to ensure that the issue has a chance to add | |
2386 | * the poll to the issued list. Otherwise we can spin here | |
2387 | * forever, while the workqueue is stuck trying to acquire the | |
2388 | * very same mutex. | |
2389 | */ | |
e9979b36 | 2390 | if (list_empty(&ctx->iopoll_list)) { |
500f9fba | 2391 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2392 | io_run_task_work(); |
500f9fba | 2393 | mutex_lock(&ctx->uring_lock); |
e9979b36 PB |
2394 | |
2395 | if (list_empty(&ctx->iopoll_list)) | |
2396 | break; | |
500f9fba | 2397 | } |
f39c8a5b PB |
2398 | ret = io_do_iopoll(ctx, &nr_events, min); |
2399 | } while (!ret && nr_events < min && !need_resched()); | |
2400 | out: | |
500f9fba | 2401 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2402 | return ret; |
2403 | } | |
2404 | ||
491381ce | 2405 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2406 | { |
491381ce JA |
2407 | /* |
2408 | * Tell lockdep we inherited freeze protection from submission | |
2409 | * thread. | |
2410 | */ | |
2411 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2412 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2413 | |
1c98679d PB |
2414 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2415 | sb_end_write(sb); | |
2b188cc1 JA |
2416 | } |
2417 | } | |
2418 | ||
b63534c4 | 2419 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2420 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2421 | { |
ab454438 PB |
2422 | struct io_async_rw *rw = req->async_data; |
2423 | ||
2424 | if (!rw) | |
2425 | return !io_req_prep_async(req); | |
2426 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2427 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2428 | return true; | |
b63534c4 | 2429 | } |
b63534c4 | 2430 | |
3e6a0d3c | 2431 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2432 | { |
355afaeb | 2433 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2434 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2435 | |
355afaeb JA |
2436 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2437 | return false; | |
3e6a0d3c JA |
2438 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2439 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2440 | return false; |
7c977a58 JA |
2441 | /* |
2442 | * If ref is dying, we might be running poll reap from the exit work. | |
2443 | * Don't attempt to reissue from that path, just let it fail with | |
2444 | * -EAGAIN. | |
2445 | */ | |
3e6a0d3c JA |
2446 | if (percpu_ref_is_dying(&ctx->refs)) |
2447 | return false; | |
2448 | return true; | |
2449 | } | |
e82ad485 | 2450 | #else |
a1ff1e3f JA |
2451 | static bool io_resubmit_prep(struct io_kiocb *req) |
2452 | { | |
2453 | return false; | |
2454 | } | |
e82ad485 JA |
2455 | static bool io_rw_should_reissue(struct io_kiocb *req) |
2456 | { | |
2457 | return false; | |
2458 | } | |
3e6a0d3c JA |
2459 | #endif |
2460 | ||
a1d7c393 | 2461 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2462 | unsigned int issue_flags) |
a1d7c393 | 2463 | { |
2f8e45f1 PB |
2464 | int cflags = 0; |
2465 | ||
b65c128f PB |
2466 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2467 | kiocb_end_write(req); | |
9532b99b PB |
2468 | if (res != req->result) { |
2469 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2470 | io_rw_should_reissue(req)) { | |
2471 | req->flags |= REQ_F_REISSUE; | |
2472 | return; | |
2473 | } | |
2f8e45f1 | 2474 | req_set_fail_links(req); |
9532b99b | 2475 | } |
2f8e45f1 PB |
2476 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2477 | cflags = io_put_rw_kbuf(req); | |
2478 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2479 | } |
2480 | ||
2481 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2482 | { | |
9adbd45d | 2483 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2484 | |
889fca73 | 2485 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2486 | } |
2487 | ||
def596e9 JA |
2488 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2489 | { | |
9adbd45d | 2490 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2491 | |
491381ce JA |
2492 | if (kiocb->ki_flags & IOCB_WRITE) |
2493 | kiocb_end_write(req); | |
9532b99b | 2494 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2495 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2496 | io_resubmit_prep(req))) { | |
9532b99b PB |
2497 | req_set_fail_links(req); |
2498 | req->flags |= REQ_F_DONT_REISSUE; | |
2499 | } | |
8c130827 | 2500 | } |
bbde017a XW |
2501 | |
2502 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2503 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2504 | smp_wmb(); |
2505 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2506 | } |
2507 | ||
2508 | /* | |
2509 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2510 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2511 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2512 | * accessing the kiocb cookie. |
2513 | */ | |
2e9dbe90 | 2514 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2515 | { |
2516 | struct io_ring_ctx *ctx = req->ctx; | |
2517 | ||
2518 | /* | |
2519 | * Track whether we have multiple files in our lists. This will impact | |
2520 | * how we do polling eventually, not spinning if we're on potentially | |
2521 | * different devices. | |
2522 | */ | |
540e32a0 | 2523 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2524 | ctx->poll_multi_file = false; |
2525 | } else if (!ctx->poll_multi_file) { | |
2526 | struct io_kiocb *list_req; | |
2527 | ||
540e32a0 | 2528 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2529 | inflight_entry); |
9adbd45d | 2530 | if (list_req->file != req->file) |
def596e9 JA |
2531 | ctx->poll_multi_file = true; |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * For fast devices, IO may have already completed. If it has, add | |
2536 | * it to the front so we find it first. | |
2537 | */ | |
65a6543d | 2538 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2539 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2540 | else |
d21ffe7e | 2541 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2542 | |
2e9dbe90 XW |
2543 | /* |
2544 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2545 | * task context or in io worker task context. If current task context is | |
2546 | * sq thread, we don't need to check whether should wake up sq thread. | |
2547 | */ | |
2548 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2549 | wq_has_sleeper(&ctx->sq_data->wait)) |
2550 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2551 | } |
2552 | ||
9f13c35b PB |
2553 | static inline void io_state_file_put(struct io_submit_state *state) |
2554 | { | |
02b23a9a PB |
2555 | if (state->file_refs) { |
2556 | fput_many(state->file, state->file_refs); | |
2557 | state->file_refs = 0; | |
2558 | } | |
9a56a232 JA |
2559 | } |
2560 | ||
2561 | /* | |
2562 | * Get as many references to a file as we have IOs left in this submission, | |
2563 | * assuming most submissions are for one file, or at least that each file | |
2564 | * has more than one submission. | |
2565 | */ | |
8da11c19 | 2566 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2567 | { |
2568 | if (!state) | |
2569 | return fget(fd); | |
2570 | ||
6e1271e6 | 2571 | if (state->file_refs) { |
9a56a232 | 2572 | if (state->fd == fd) { |
6e1271e6 | 2573 | state->file_refs--; |
9a56a232 JA |
2574 | return state->file; |
2575 | } | |
02b23a9a | 2576 | io_state_file_put(state); |
9a56a232 JA |
2577 | } |
2578 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2579 | if (unlikely(!state->file)) |
9a56a232 JA |
2580 | return NULL; |
2581 | ||
2582 | state->fd = fd; | |
6e1271e6 | 2583 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2584 | return state->file; |
2585 | } | |
2586 | ||
4503b767 JA |
2587 | static bool io_bdev_nowait(struct block_device *bdev) |
2588 | { | |
9ba0d0c8 | 2589 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2590 | } |
2591 | ||
2b188cc1 JA |
2592 | /* |
2593 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2594 | * any file. For now, just ensure that anything potentially problematic is done | |
2595 | * inline. | |
2596 | */ | |
7b29f92d | 2597 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2598 | { |
2599 | umode_t mode = file_inode(file)->i_mode; | |
2600 | ||
4503b767 | 2601 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2602 | if (IS_ENABLED(CONFIG_BLOCK) && |
2603 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2604 | return true; |
2605 | return false; | |
2606 | } | |
2607 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2608 | return true; |
4503b767 | 2609 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2610 | if (IS_ENABLED(CONFIG_BLOCK) && |
2611 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2612 | file->f_op != &io_uring_fops) |
2613 | return true; | |
2614 | return false; | |
2615 | } | |
2b188cc1 | 2616 | |
c5b85625 JA |
2617 | /* any ->read/write should understand O_NONBLOCK */ |
2618 | if (file->f_flags & O_NONBLOCK) | |
2619 | return true; | |
2620 | ||
af197f50 JA |
2621 | if (!(file->f_mode & FMODE_NOWAIT)) |
2622 | return false; | |
2623 | ||
2624 | if (rw == READ) | |
2625 | return file->f_op->read_iter != NULL; | |
2626 | ||
2627 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2628 | } |
2629 | ||
7b29f92d JA |
2630 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2631 | { | |
2632 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2633 | return true; | |
2634 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2635 | return true; | |
2636 | ||
2637 | return __io_file_supports_async(req->file, rw); | |
2638 | } | |
2639 | ||
a88fc400 | 2640 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2641 | { |
def596e9 | 2642 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2643 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2644 | struct file *file = req->file; |
09bb8394 JA |
2645 | unsigned ioprio; |
2646 | int ret; | |
2b188cc1 | 2647 | |
7b29f92d | 2648 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2649 | req->flags |= REQ_F_ISREG; |
2650 | ||
2b188cc1 | 2651 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2652 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2653 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2654 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2655 | } |
2b188cc1 | 2656 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2657 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2658 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2659 | if (unlikely(ret)) | |
2660 | return ret; | |
2b188cc1 | 2661 | |
75c668cd PB |
2662 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2663 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2664 | req->flags |= REQ_F_NOWAIT; | |
2665 | ||
2b188cc1 JA |
2666 | ioprio = READ_ONCE(sqe->ioprio); |
2667 | if (ioprio) { | |
2668 | ret = ioprio_check_cap(ioprio); | |
2669 | if (ret) | |
09bb8394 | 2670 | return ret; |
2b188cc1 JA |
2671 | |
2672 | kiocb->ki_ioprio = ioprio; | |
2673 | } else | |
2674 | kiocb->ki_ioprio = get_current_ioprio(); | |
2675 | ||
def596e9 | 2676 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2677 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2678 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2679 | return -EOPNOTSUPP; |
2b188cc1 | 2680 | |
def596e9 JA |
2681 | kiocb->ki_flags |= IOCB_HIPRI; |
2682 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2683 | req->iopoll_completed = 0; |
def596e9 | 2684 | } else { |
09bb8394 JA |
2685 | if (kiocb->ki_flags & IOCB_HIPRI) |
2686 | return -EINVAL; | |
def596e9 JA |
2687 | kiocb->ki_complete = io_complete_rw; |
2688 | } | |
9adbd45d | 2689 | |
3529d8c2 JA |
2690 | req->rw.addr = READ_ONCE(sqe->addr); |
2691 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2692 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2693 | return 0; |
2b188cc1 JA |
2694 | } |
2695 | ||
2696 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2697 | { | |
2698 | switch (ret) { | |
2699 | case -EIOCBQUEUED: | |
2700 | break; | |
2701 | case -ERESTARTSYS: | |
2702 | case -ERESTARTNOINTR: | |
2703 | case -ERESTARTNOHAND: | |
2704 | case -ERESTART_RESTARTBLOCK: | |
2705 | /* | |
2706 | * We can't just restart the syscall, since previously | |
2707 | * submitted sqes may already be in progress. Just fail this | |
2708 | * IO with EINTR. | |
2709 | */ | |
2710 | ret = -EINTR; | |
df561f66 | 2711 | fallthrough; |
2b188cc1 JA |
2712 | default: |
2713 | kiocb->ki_complete(kiocb, ret, 0); | |
2714 | } | |
2715 | } | |
2716 | ||
a1d7c393 | 2717 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2718 | unsigned int issue_flags) |
ba816ad6 | 2719 | { |
ba04291e | 2720 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2721 | struct io_async_rw *io = req->async_data; |
97284637 | 2722 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2723 | |
227c0c96 | 2724 | /* add previously done IO, if any */ |
e8c2bc1f | 2725 | if (io && io->bytes_done > 0) { |
227c0c96 | 2726 | if (ret < 0) |
e8c2bc1f | 2727 | ret = io->bytes_done; |
227c0c96 | 2728 | else |
e8c2bc1f | 2729 | ret += io->bytes_done; |
227c0c96 JA |
2730 | } |
2731 | ||
ba04291e JA |
2732 | if (req->flags & REQ_F_CUR_POS) |
2733 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2734 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2735 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2736 | else |
2737 | io_rw_done(kiocb, ret); | |
97284637 PB |
2738 | |
2739 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2740 | req->flags &= ~REQ_F_REISSUE; | |
a7be7c23 | 2741 | if (io_resubmit_prep(req)) { |
8c130827 PB |
2742 | req_ref_get(req); |
2743 | io_queue_async_work(req); | |
2744 | } else { | |
97284637 PB |
2745 | int cflags = 0; |
2746 | ||
2747 | req_set_fail_links(req); | |
2748 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2749 | cflags = io_put_rw_kbuf(req); | |
2750 | __io_req_complete(req, issue_flags, ret, cflags); | |
2751 | } | |
2752 | } | |
ba816ad6 JA |
2753 | } |
2754 | ||
847595de | 2755 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2756 | { |
9adbd45d JA |
2757 | struct io_ring_ctx *ctx = req->ctx; |
2758 | size_t len = req->rw.len; | |
edafccee | 2759 | struct io_mapped_ubuf *imu; |
4be1c615 | 2760 | u16 index, buf_index = req->buf_index; |
75769e3f | 2761 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2762 | size_t offset; |
edafccee | 2763 | |
edafccee JA |
2764 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2765 | return -EFAULT; | |
edafccee JA |
2766 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2767 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2768 | buf_addr = req->rw.addr; |
edafccee | 2769 | |
75769e3f | 2770 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2771 | return -EFAULT; |
2772 | /* not inside the mapped region */ | |
4751f53d | 2773 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2774 | return -EFAULT; |
2775 | ||
2776 | /* | |
2777 | * May not be a start of buffer, set size appropriately | |
2778 | * and advance us to the beginning. | |
2779 | */ | |
2780 | offset = buf_addr - imu->ubuf; | |
2781 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2782 | |
2783 | if (offset) { | |
2784 | /* | |
2785 | * Don't use iov_iter_advance() here, as it's really slow for | |
2786 | * using the latter parts of a big fixed buffer - it iterates | |
2787 | * over each segment manually. We can cheat a bit here, because | |
2788 | * we know that: | |
2789 | * | |
2790 | * 1) it's a BVEC iter, we set it up | |
2791 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2792 | * first and last bvec | |
2793 | * | |
2794 | * So just find our index, and adjust the iterator afterwards. | |
2795 | * If the offset is within the first bvec (or the whole first | |
2796 | * bvec, just use iov_iter_advance(). This makes it easier | |
2797 | * since we can just skip the first segment, which may not | |
2798 | * be PAGE_SIZE aligned. | |
2799 | */ | |
2800 | const struct bio_vec *bvec = imu->bvec; | |
2801 | ||
2802 | if (offset <= bvec->bv_len) { | |
2803 | iov_iter_advance(iter, offset); | |
2804 | } else { | |
2805 | unsigned long seg_skip; | |
2806 | ||
2807 | /* skip first vec */ | |
2808 | offset -= bvec->bv_len; | |
2809 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2810 | ||
2811 | iter->bvec = bvec + seg_skip; | |
2812 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2813 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2814 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2815 | } |
2816 | } | |
2817 | ||
847595de | 2818 | return 0; |
edafccee JA |
2819 | } |
2820 | ||
bcda7baa JA |
2821 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2822 | { | |
2823 | if (needs_lock) | |
2824 | mutex_unlock(&ctx->uring_lock); | |
2825 | } | |
2826 | ||
2827 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2828 | { | |
2829 | /* | |
2830 | * "Normal" inline submissions always hold the uring_lock, since we | |
2831 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2832 | * The only exception is when we've detached the request and issue it | |
2833 | * from an async worker thread, grab the lock for that case. | |
2834 | */ | |
2835 | if (needs_lock) | |
2836 | mutex_lock(&ctx->uring_lock); | |
2837 | } | |
2838 | ||
2839 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2840 | int bgid, struct io_buffer *kbuf, | |
2841 | bool needs_lock) | |
2842 | { | |
2843 | struct io_buffer *head; | |
2844 | ||
2845 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2846 | return kbuf; | |
2847 | ||
2848 | io_ring_submit_lock(req->ctx, needs_lock); | |
2849 | ||
2850 | lockdep_assert_held(&req->ctx->uring_lock); | |
2851 | ||
9e15c3a0 | 2852 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2853 | if (head) { |
2854 | if (!list_empty(&head->list)) { | |
2855 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2856 | list); | |
2857 | list_del(&kbuf->list); | |
2858 | } else { | |
2859 | kbuf = head; | |
9e15c3a0 | 2860 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2861 | } |
2862 | if (*len > kbuf->len) | |
2863 | *len = kbuf->len; | |
2864 | } else { | |
2865 | kbuf = ERR_PTR(-ENOBUFS); | |
2866 | } | |
2867 | ||
2868 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2869 | ||
2870 | return kbuf; | |
2871 | } | |
2872 | ||
4d954c25 JA |
2873 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2874 | bool needs_lock) | |
2875 | { | |
2876 | struct io_buffer *kbuf; | |
4f4eeba8 | 2877 | u16 bgid; |
4d954c25 JA |
2878 | |
2879 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2880 | bgid = req->buf_index; |
4d954c25 JA |
2881 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2882 | if (IS_ERR(kbuf)) | |
2883 | return kbuf; | |
2884 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2885 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2886 | return u64_to_user_ptr(kbuf->addr); | |
2887 | } | |
2888 | ||
2889 | #ifdef CONFIG_COMPAT | |
2890 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2891 | bool needs_lock) | |
2892 | { | |
2893 | struct compat_iovec __user *uiov; | |
2894 | compat_ssize_t clen; | |
2895 | void __user *buf; | |
2896 | ssize_t len; | |
2897 | ||
2898 | uiov = u64_to_user_ptr(req->rw.addr); | |
2899 | if (!access_ok(uiov, sizeof(*uiov))) | |
2900 | return -EFAULT; | |
2901 | if (__get_user(clen, &uiov->iov_len)) | |
2902 | return -EFAULT; | |
2903 | if (clen < 0) | |
2904 | return -EINVAL; | |
2905 | ||
2906 | len = clen; | |
2907 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2908 | if (IS_ERR(buf)) | |
2909 | return PTR_ERR(buf); | |
2910 | iov[0].iov_base = buf; | |
2911 | iov[0].iov_len = (compat_size_t) len; | |
2912 | return 0; | |
2913 | } | |
2914 | #endif | |
2915 | ||
2916 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2917 | bool needs_lock) | |
2918 | { | |
2919 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2920 | void __user *buf; | |
2921 | ssize_t len; | |
2922 | ||
2923 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2924 | return -EFAULT; | |
2925 | ||
2926 | len = iov[0].iov_len; | |
2927 | if (len < 0) | |
2928 | return -EINVAL; | |
2929 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2930 | if (IS_ERR(buf)) | |
2931 | return PTR_ERR(buf); | |
2932 | iov[0].iov_base = buf; | |
2933 | iov[0].iov_len = len; | |
2934 | return 0; | |
2935 | } | |
2936 | ||
2937 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2938 | bool needs_lock) | |
2939 | { | |
dddb3e26 JA |
2940 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2941 | struct io_buffer *kbuf; | |
2942 | ||
2943 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2944 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2945 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2946 | return 0; |
dddb3e26 | 2947 | } |
dd201662 | 2948 | if (req->rw.len != 1) |
4d954c25 JA |
2949 | return -EINVAL; |
2950 | ||
2951 | #ifdef CONFIG_COMPAT | |
2952 | if (req->ctx->compat) | |
2953 | return io_compat_import(req, iov, needs_lock); | |
2954 | #endif | |
2955 | ||
2956 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2957 | } | |
2958 | ||
847595de PB |
2959 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2960 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2961 | { |
9adbd45d JA |
2962 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2963 | size_t sqe_len = req->rw.len; | |
847595de | 2964 | u8 opcode = req->opcode; |
4d954c25 | 2965 | ssize_t ret; |
edafccee | 2966 | |
7d009165 | 2967 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2968 | *iovec = NULL; |
9adbd45d | 2969 | return io_import_fixed(req, rw, iter); |
edafccee | 2970 | } |
2b188cc1 | 2971 | |
bcda7baa | 2972 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2973 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2974 | return -EINVAL; |
2975 | ||
3a6820f2 | 2976 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2977 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2978 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2979 | if (IS_ERR(buf)) |
4d954c25 | 2980 | return PTR_ERR(buf); |
3f9d6441 | 2981 | req->rw.len = sqe_len; |
bcda7baa JA |
2982 | } |
2983 | ||
3a6820f2 JA |
2984 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2985 | *iovec = NULL; | |
10fc72e4 | 2986 | return ret; |
3a6820f2 JA |
2987 | } |
2988 | ||
4d954c25 JA |
2989 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2990 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2991 | if (!ret) |
2992 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2993 | *iovec = NULL; |
2994 | return ret; | |
2995 | } | |
2996 | ||
89cd35c5 CH |
2997 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2998 | req->ctx->compat); | |
2b188cc1 JA |
2999 | } |
3000 | ||
0fef9483 JA |
3001 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3002 | { | |
5b09e37e | 3003 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3004 | } |
3005 | ||
31b51510 | 3006 | /* |
32960613 JA |
3007 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3008 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3009 | */ |
4017eb91 | 3010 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3011 | { |
4017eb91 JA |
3012 | struct kiocb *kiocb = &req->rw.kiocb; |
3013 | struct file *file = req->file; | |
32960613 JA |
3014 | ssize_t ret = 0; |
3015 | ||
3016 | /* | |
3017 | * Don't support polled IO through this interface, and we can't | |
3018 | * support non-blocking either. For the latter, this just causes | |
3019 | * the kiocb to be handled from an async context. | |
3020 | */ | |
3021 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3022 | return -EOPNOTSUPP; | |
3023 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3024 | return -EAGAIN; | |
3025 | ||
3026 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3027 | struct iovec iovec; |
32960613 JA |
3028 | ssize_t nr; |
3029 | ||
311ae9e1 PB |
3030 | if (!iov_iter_is_bvec(iter)) { |
3031 | iovec = iov_iter_iovec(iter); | |
3032 | } else { | |
4017eb91 JA |
3033 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3034 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3035 | } |
3036 | ||
32960613 JA |
3037 | if (rw == READ) { |
3038 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3039 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3040 | } else { |
3041 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3042 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3043 | } |
3044 | ||
3045 | if (nr < 0) { | |
3046 | if (!ret) | |
3047 | ret = nr; | |
3048 | break; | |
3049 | } | |
3050 | ret += nr; | |
3051 | if (nr != iovec.iov_len) | |
3052 | break; | |
4017eb91 JA |
3053 | req->rw.len -= nr; |
3054 | req->rw.addr += nr; | |
32960613 JA |
3055 | iov_iter_advance(iter, nr); |
3056 | } | |
3057 | ||
3058 | return ret; | |
3059 | } | |
3060 | ||
ff6165b2 JA |
3061 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3062 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3063 | { |
e8c2bc1f | 3064 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3065 | |
ff6165b2 | 3066 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3067 | rw->free_iovec = iovec; |
227c0c96 | 3068 | rw->bytes_done = 0; |
ff6165b2 | 3069 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3070 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3071 | return; |
b64e3444 | 3072 | if (!iovec) { |
ff6165b2 JA |
3073 | unsigned iov_off = 0; |
3074 | ||
3075 | rw->iter.iov = rw->fast_iov; | |
3076 | if (iter->iov != fast_iov) { | |
3077 | iov_off = iter->iov - fast_iov; | |
3078 | rw->iter.iov += iov_off; | |
3079 | } | |
3080 | if (rw->fast_iov != fast_iov) | |
3081 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3082 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3083 | } else { |
3084 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3085 | } |
3086 | } | |
3087 | ||
6cb78689 | 3088 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3089 | { |
e8c2bc1f JA |
3090 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3091 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3092 | return req->async_data == NULL; | |
3d9932a8 XW |
3093 | } |
3094 | ||
ff6165b2 JA |
3095 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3096 | const struct iovec *fast_iov, | |
227c0c96 | 3097 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3098 | { |
26f0505a | 3099 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3100 | return 0; |
e8c2bc1f | 3101 | if (!req->async_data) { |
6cb78689 | 3102 | if (io_alloc_async_data(req)) { |
6bf985dc | 3103 | kfree(iovec); |
5d204bcf | 3104 | return -ENOMEM; |
6bf985dc | 3105 | } |
b7bb4f7d | 3106 | |
ff6165b2 | 3107 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3108 | } |
b7bb4f7d | 3109 | return 0; |
f67676d1 JA |
3110 | } |
3111 | ||
73debe68 | 3112 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3113 | { |
e8c2bc1f | 3114 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3115 | struct iovec *iov = iorw->fast_iov; |
847595de | 3116 | int ret; |
c3e330a4 | 3117 | |
2846c481 | 3118 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3119 | if (unlikely(ret < 0)) |
3120 | return ret; | |
3121 | ||
ab0b196c PB |
3122 | iorw->bytes_done = 0; |
3123 | iorw->free_iovec = iov; | |
3124 | if (iov) | |
3125 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3126 | return 0; |
3127 | } | |
3128 | ||
73debe68 | 3129 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3130 | { |
3529d8c2 JA |
3131 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3132 | return -EBADF; | |
93642ef8 | 3133 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3134 | } |
3135 | ||
c1dd91d1 JA |
3136 | /* |
3137 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3138 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3139 | * This gets called when the page is unlocked, and we generally expect that to | |
3140 | * happen when the page IO is completed and the page is now uptodate. This will | |
3141 | * queue a task_work based retry of the operation, attempting to copy the data | |
3142 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3143 | * do a thread based blocking retry of the operation. That's the unexpected | |
3144 | * slow path. | |
3145 | */ | |
bcf5a063 JA |
3146 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3147 | int sync, void *arg) | |
3148 | { | |
3149 | struct wait_page_queue *wpq; | |
3150 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3151 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3152 | |
3153 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3154 | ||
cdc8fcb4 LT |
3155 | if (!wake_page_match(wpq, key)) |
3156 | return 0; | |
3157 | ||
c8d317aa | 3158 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3159 | list_del_init(&wait->entry); |
3160 | ||
bcf5a063 | 3161 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3162 | req_ref_get(req); |
921b9054 | 3163 | io_req_task_queue(req); |
bcf5a063 JA |
3164 | return 1; |
3165 | } | |
3166 | ||
c1dd91d1 JA |
3167 | /* |
3168 | * This controls whether a given IO request should be armed for async page | |
3169 | * based retry. If we return false here, the request is handed to the async | |
3170 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3171 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3172 | * will either succeed because the page is now uptodate and unlocked, or it | |
3173 | * will register a callback when the page is unlocked at IO completion. Through | |
3174 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3175 | * That retry will attempt the buffered read again. The retry will generally | |
3176 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3177 | * async worker threads for a blocking retry. | |
3178 | */ | |
227c0c96 | 3179 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3180 | { |
e8c2bc1f JA |
3181 | struct io_async_rw *rw = req->async_data; |
3182 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3183 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3184 | |
bcf5a063 JA |
3185 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3186 | if (req->flags & REQ_F_NOWAIT) | |
3187 | return false; | |
f67676d1 | 3188 | |
227c0c96 | 3189 | /* Only for buffered IO */ |
3b2a4439 | 3190 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3191 | return false; |
3b2a4439 | 3192 | |
bcf5a063 JA |
3193 | /* |
3194 | * just use poll if we can, and don't attempt if the fs doesn't | |
3195 | * support callback based unlocks | |
3196 | */ | |
3197 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3198 | return false; | |
f67676d1 | 3199 | |
3b2a4439 JA |
3200 | wait->wait.func = io_async_buf_func; |
3201 | wait->wait.private = req; | |
3202 | wait->wait.flags = 0; | |
3203 | INIT_LIST_HEAD(&wait->wait.entry); | |
3204 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3205 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3206 | kiocb->ki_waitq = wait; |
3b2a4439 | 3207 | return true; |
bcf5a063 JA |
3208 | } |
3209 | ||
3210 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3211 | { | |
3212 | if (req->file->f_op->read_iter) | |
3213 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3214 | else if (req->file->f_op->read) |
4017eb91 | 3215 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3216 | else |
3217 | return -EINVAL; | |
f67676d1 JA |
3218 | } |
3219 | ||
889fca73 | 3220 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3221 | { |
3222 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3223 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3224 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3225 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3226 | ssize_t io_size, ret, ret2; |
45d189c6 | 3227 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3228 | |
2846c481 | 3229 | if (rw) { |
e8c2bc1f | 3230 | iter = &rw->iter; |
2846c481 PB |
3231 | iovec = NULL; |
3232 | } else { | |
3233 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3234 | if (ret < 0) | |
3235 | return ret; | |
3236 | } | |
632546c4 | 3237 | io_size = iov_iter_count(iter); |
fa15bafb | 3238 | req->result = io_size; |
2b188cc1 | 3239 | |
fd6c2e4c JA |
3240 | /* Ensure we clear previously set non-block flag */ |
3241 | if (!force_nonblock) | |
29de5f6a | 3242 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3243 | else |
3244 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3245 | ||
24c74678 | 3246 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3247 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3248 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3249 | return ret ?: -EAGAIN; |
6713e7a6 | 3250 | } |
9e645e11 | 3251 | |
632546c4 | 3252 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3253 | if (unlikely(ret)) { |
3254 | kfree(iovec); | |
3255 | return ret; | |
3256 | } | |
2b188cc1 | 3257 | |
227c0c96 | 3258 | ret = io_iter_do_read(req, iter); |
32960613 | 3259 | |
230d50d4 | 3260 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3261 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3262 | /* IOPOLL retry should happen for io-wq threads */ |
3263 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3264 | goto done; |
75c668cd PB |
3265 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3266 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3267 | goto done; |
84216315 | 3268 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3269 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3270 | ret = 0; |
230d50d4 JA |
3271 | } else if (ret == -EIOCBQUEUED) { |
3272 | goto out_free; | |
7335e3bf | 3273 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3274 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3275 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3276 | goto done; |
227c0c96 JA |
3277 | } |
3278 | ||
227c0c96 | 3279 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3280 | if (ret2) |
3281 | return ret2; | |
3282 | ||
fe1cdd55 | 3283 | iovec = NULL; |
e8c2bc1f | 3284 | rw = req->async_data; |
227c0c96 | 3285 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3286 | iter = &rw->iter; |
227c0c96 | 3287 | |
b23df91b PB |
3288 | do { |
3289 | io_size -= ret; | |
3290 | rw->bytes_done += ret; | |
3291 | /* if we can retry, do so with the callbacks armed */ | |
3292 | if (!io_rw_should_retry(req)) { | |
3293 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3294 | return -EAGAIN; | |
3295 | } | |
3296 | ||
3297 | /* | |
3298 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3299 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3300 | * desired page gets unlocked. We can also get a partial read | |
3301 | * here, and if we do, then just retry at the new offset. | |
3302 | */ | |
3303 | ret = io_iter_do_read(req, iter); | |
3304 | if (ret == -EIOCBQUEUED) | |
3305 | return 0; | |
227c0c96 | 3306 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3307 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3308 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3309 | done: |
889fca73 | 3310 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3311 | out_free: |
3312 | /* it's faster to check here then delegate to kfree */ | |
3313 | if (iovec) | |
3314 | kfree(iovec); | |
5ea5dd45 | 3315 | return 0; |
2b188cc1 JA |
3316 | } |
3317 | ||
73debe68 | 3318 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3319 | { |
3529d8c2 JA |
3320 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3321 | return -EBADF; | |
93642ef8 | 3322 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3323 | } |
3324 | ||
889fca73 | 3325 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3326 | { |
3327 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3328 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3329 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3330 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3331 | ssize_t ret, ret2, io_size; |
45d189c6 | 3332 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3333 | |
2846c481 | 3334 | if (rw) { |
e8c2bc1f | 3335 | iter = &rw->iter; |
2846c481 PB |
3336 | iovec = NULL; |
3337 | } else { | |
3338 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3339 | if (ret < 0) | |
3340 | return ret; | |
3341 | } | |
632546c4 | 3342 | io_size = iov_iter_count(iter); |
fa15bafb | 3343 | req->result = io_size; |
2b188cc1 | 3344 | |
fd6c2e4c JA |
3345 | /* Ensure we clear previously set non-block flag */ |
3346 | if (!force_nonblock) | |
a88fc400 PB |
3347 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3348 | else | |
3349 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3350 | |
24c74678 | 3351 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3352 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3353 | goto copy_iov; |
31b51510 | 3354 | |
10d59345 JA |
3355 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3356 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3357 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3358 | goto copy_iov; |
31b51510 | 3359 | |
632546c4 | 3360 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3361 | if (unlikely(ret)) |
3362 | goto out_free; | |
4ed734b0 | 3363 | |
fa15bafb PB |
3364 | /* |
3365 | * Open-code file_start_write here to grab freeze protection, | |
3366 | * which will be released by another thread in | |
3367 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3368 | * released so that it doesn't complain about the held lock when | |
3369 | * we return to userspace. | |
3370 | */ | |
3371 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3372 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3373 | __sb_writers_release(file_inode(req->file)->i_sb, |
3374 | SB_FREEZE_WRITE); | |
3375 | } | |
3376 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3377 | |
fa15bafb | 3378 | if (req->file->f_op->write_iter) |
ff6165b2 | 3379 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3380 | else if (req->file->f_op->write) |
4017eb91 | 3381 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3382 | else |
3383 | ret2 = -EINVAL; | |
4ed734b0 | 3384 | |
6ad7f233 PB |
3385 | if (req->flags & REQ_F_REISSUE) { |
3386 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3387 | ret2 = -EAGAIN; |
6ad7f233 | 3388 | } |
230d50d4 | 3389 | |
fa15bafb PB |
3390 | /* |
3391 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3392 | * retry them without IOCB_NOWAIT. | |
3393 | */ | |
3394 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3395 | ret2 = -EAGAIN; | |
75c668cd PB |
3396 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3397 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3398 | goto done; |
fa15bafb | 3399 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3400 | /* IOPOLL retry should happen for io-wq threads */ |
3401 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3402 | goto copy_iov; | |
355afaeb | 3403 | done: |
889fca73 | 3404 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3405 | } else { |
f67676d1 | 3406 | copy_iov: |
84216315 | 3407 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3408 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3409 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3410 | return ret ?: -EAGAIN; |
2b188cc1 | 3411 | } |
31b51510 | 3412 | out_free: |
f261c168 | 3413 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3414 | if (iovec) |
6f2cc166 | 3415 | kfree(iovec); |
2b188cc1 JA |
3416 | return ret; |
3417 | } | |
3418 | ||
80a261fd JA |
3419 | static int io_renameat_prep(struct io_kiocb *req, |
3420 | const struct io_uring_sqe *sqe) | |
3421 | { | |
3422 | struct io_rename *ren = &req->rename; | |
3423 | const char __user *oldf, *newf; | |
3424 | ||
3425 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3426 | return -EBADF; | |
3427 | ||
3428 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3429 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3430 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3431 | ren->new_dfd = READ_ONCE(sqe->len); | |
3432 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3433 | ||
3434 | ren->oldpath = getname(oldf); | |
3435 | if (IS_ERR(ren->oldpath)) | |
3436 | return PTR_ERR(ren->oldpath); | |
3437 | ||
3438 | ren->newpath = getname(newf); | |
3439 | if (IS_ERR(ren->newpath)) { | |
3440 | putname(ren->oldpath); | |
3441 | return PTR_ERR(ren->newpath); | |
3442 | } | |
3443 | ||
3444 | req->flags |= REQ_F_NEED_CLEANUP; | |
3445 | return 0; | |
3446 | } | |
3447 | ||
45d189c6 | 3448 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3449 | { |
3450 | struct io_rename *ren = &req->rename; | |
3451 | int ret; | |
3452 | ||
45d189c6 | 3453 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3454 | return -EAGAIN; |
3455 | ||
3456 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3457 | ren->newpath, ren->flags); | |
3458 | ||
3459 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3460 | if (ret < 0) | |
3461 | req_set_fail_links(req); | |
3462 | io_req_complete(req, ret); | |
3463 | return 0; | |
3464 | } | |
3465 | ||
14a1143b JA |
3466 | static int io_unlinkat_prep(struct io_kiocb *req, |
3467 | const struct io_uring_sqe *sqe) | |
3468 | { | |
3469 | struct io_unlink *un = &req->unlink; | |
3470 | const char __user *fname; | |
3471 | ||
3472 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3473 | return -EBADF; | |
3474 | ||
3475 | un->dfd = READ_ONCE(sqe->fd); | |
3476 | ||
3477 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3478 | if (un->flags & ~AT_REMOVEDIR) | |
3479 | return -EINVAL; | |
3480 | ||
3481 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3482 | un->filename = getname(fname); | |
3483 | if (IS_ERR(un->filename)) | |
3484 | return PTR_ERR(un->filename); | |
3485 | ||
3486 | req->flags |= REQ_F_NEED_CLEANUP; | |
3487 | return 0; | |
3488 | } | |
3489 | ||
45d189c6 | 3490 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3491 | { |
3492 | struct io_unlink *un = &req->unlink; | |
3493 | int ret; | |
3494 | ||
45d189c6 | 3495 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3496 | return -EAGAIN; |
3497 | ||
3498 | if (un->flags & AT_REMOVEDIR) | |
3499 | ret = do_rmdir(un->dfd, un->filename); | |
3500 | else | |
3501 | ret = do_unlinkat(un->dfd, un->filename); | |
3502 | ||
3503 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3504 | if (ret < 0) | |
3505 | req_set_fail_links(req); | |
3506 | io_req_complete(req, ret); | |
3507 | return 0; | |
3508 | } | |
3509 | ||
36f4fa68 JA |
3510 | static int io_shutdown_prep(struct io_kiocb *req, |
3511 | const struct io_uring_sqe *sqe) | |
3512 | { | |
3513 | #if defined(CONFIG_NET) | |
3514 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3515 | return -EINVAL; | |
3516 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3517 | sqe->buf_index) | |
3518 | return -EINVAL; | |
3519 | ||
3520 | req->shutdown.how = READ_ONCE(sqe->len); | |
3521 | return 0; | |
3522 | #else | |
3523 | return -EOPNOTSUPP; | |
3524 | #endif | |
3525 | } | |
3526 | ||
45d189c6 | 3527 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3528 | { |
3529 | #if defined(CONFIG_NET) | |
3530 | struct socket *sock; | |
3531 | int ret; | |
3532 | ||
45d189c6 | 3533 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3534 | return -EAGAIN; |
3535 | ||
48aba79b | 3536 | sock = sock_from_file(req->file); |
36f4fa68 | 3537 | if (unlikely(!sock)) |
48aba79b | 3538 | return -ENOTSOCK; |
36f4fa68 JA |
3539 | |
3540 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3541 | if (ret < 0) |
3542 | req_set_fail_links(req); | |
36f4fa68 JA |
3543 | io_req_complete(req, ret); |
3544 | return 0; | |
3545 | #else | |
3546 | return -EOPNOTSUPP; | |
3547 | #endif | |
3548 | } | |
3549 | ||
f2a8d5c7 PB |
3550 | static int __io_splice_prep(struct io_kiocb *req, |
3551 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3552 | { |
3553 | struct io_splice* sp = &req->splice; | |
3554 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3555 | |
3232dd02 PB |
3556 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3557 | return -EINVAL; | |
7d67af2c PB |
3558 | |
3559 | sp->file_in = NULL; | |
7d67af2c PB |
3560 | sp->len = READ_ONCE(sqe->len); |
3561 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3562 | ||
3563 | if (unlikely(sp->flags & ~valid_flags)) | |
3564 | return -EINVAL; | |
3565 | ||
8371adf5 PB |
3566 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3567 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3568 | if (!sp->file_in) | |
3569 | return -EBADF; | |
7d67af2c | 3570 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3571 | return 0; |
3572 | } | |
3573 | ||
f2a8d5c7 PB |
3574 | static int io_tee_prep(struct io_kiocb *req, |
3575 | const struct io_uring_sqe *sqe) | |
3576 | { | |
3577 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3578 | return -EINVAL; | |
3579 | return __io_splice_prep(req, sqe); | |
3580 | } | |
3581 | ||
45d189c6 | 3582 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3583 | { |
3584 | struct io_splice *sp = &req->splice; | |
3585 | struct file *in = sp->file_in; | |
3586 | struct file *out = sp->file_out; | |
3587 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3588 | long ret = 0; | |
3589 | ||
45d189c6 | 3590 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3591 | return -EAGAIN; |
3592 | if (sp->len) | |
3593 | ret = do_tee(in, out, sp->len, flags); | |
3594 | ||
e1d767f0 PB |
3595 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3596 | io_put_file(in); | |
f2a8d5c7 PB |
3597 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3598 | ||
f2a8d5c7 PB |
3599 | if (ret != sp->len) |
3600 | req_set_fail_links(req); | |
e1e16097 | 3601 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3602 | return 0; |
3603 | } | |
3604 | ||
3605 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3606 | { | |
3607 | struct io_splice* sp = &req->splice; | |
3608 | ||
3609 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3610 | sp->off_out = READ_ONCE(sqe->off); | |
3611 | return __io_splice_prep(req, sqe); | |
3612 | } | |
3613 | ||
45d189c6 | 3614 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3615 | { |
3616 | struct io_splice *sp = &req->splice; | |
3617 | struct file *in = sp->file_in; | |
3618 | struct file *out = sp->file_out; | |
3619 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3620 | loff_t *poff_in, *poff_out; | |
c9687426 | 3621 | long ret = 0; |
7d67af2c | 3622 | |
45d189c6 | 3623 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3624 | return -EAGAIN; |
7d67af2c PB |
3625 | |
3626 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3627 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3628 | |
948a7749 | 3629 | if (sp->len) |
c9687426 | 3630 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3631 | |
e1d767f0 PB |
3632 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3633 | io_put_file(in); | |
7d67af2c PB |
3634 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3635 | ||
7d67af2c PB |
3636 | if (ret != sp->len) |
3637 | req_set_fail_links(req); | |
e1e16097 | 3638 | io_req_complete(req, ret); |
7d67af2c PB |
3639 | return 0; |
3640 | } | |
3641 | ||
2b188cc1 JA |
3642 | /* |
3643 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3644 | */ | |
889fca73 | 3645 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3646 | { |
3647 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3648 | |
def596e9 JA |
3649 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3650 | return -EINVAL; | |
3651 | ||
889fca73 | 3652 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3653 | return 0; |
3654 | } | |
3655 | ||
1155c76a | 3656 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3657 | { |
6b06314c | 3658 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3659 | |
09bb8394 JA |
3660 | if (!req->file) |
3661 | return -EBADF; | |
c992fe29 | 3662 | |
6b06314c | 3663 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3664 | return -EINVAL; |
edafccee | 3665 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3666 | return -EINVAL; |
3667 | ||
8ed8d3c3 JA |
3668 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3669 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3670 | return -EINVAL; | |
3671 | ||
3672 | req->sync.off = READ_ONCE(sqe->off); | |
3673 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3674 | return 0; |
3675 | } | |
3676 | ||
45d189c6 | 3677 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3678 | { |
8ed8d3c3 | 3679 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3680 | int ret; |
3681 | ||
ac45abc0 | 3682 | /* fsync always requires a blocking context */ |
45d189c6 | 3683 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3684 | return -EAGAIN; |
3685 | ||
9adbd45d | 3686 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3687 | end > 0 ? end : LLONG_MAX, |
3688 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3689 | if (ret < 0) | |
3690 | req_set_fail_links(req); | |
e1e16097 | 3691 | io_req_complete(req, ret); |
c992fe29 CH |
3692 | return 0; |
3693 | } | |
3694 | ||
d63d1b5e JA |
3695 | static int io_fallocate_prep(struct io_kiocb *req, |
3696 | const struct io_uring_sqe *sqe) | |
3697 | { | |
3698 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3699 | return -EINVAL; | |
3232dd02 PB |
3700 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3701 | return -EINVAL; | |
d63d1b5e JA |
3702 | |
3703 | req->sync.off = READ_ONCE(sqe->off); | |
3704 | req->sync.len = READ_ONCE(sqe->addr); | |
3705 | req->sync.mode = READ_ONCE(sqe->len); | |
3706 | return 0; | |
3707 | } | |
3708 | ||
45d189c6 | 3709 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3710 | { |
ac45abc0 PB |
3711 | int ret; |
3712 | ||
d63d1b5e | 3713 | /* fallocate always requiring blocking context */ |
45d189c6 | 3714 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3715 | return -EAGAIN; |
ac45abc0 PB |
3716 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3717 | req->sync.len); | |
ac45abc0 PB |
3718 | if (ret < 0) |
3719 | req_set_fail_links(req); | |
e1e16097 | 3720 | io_req_complete(req, ret); |
5d17b4a4 JA |
3721 | return 0; |
3722 | } | |
3723 | ||
ec65fea5 | 3724 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3725 | { |
f8748881 | 3726 | const char __user *fname; |
15b71abe | 3727 | int ret; |
b7bb4f7d | 3728 | |
ec65fea5 | 3729 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3730 | return -EINVAL; |
ec65fea5 | 3731 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3732 | return -EBADF; |
03b1230c | 3733 | |
ec65fea5 PB |
3734 | /* open.how should be already initialised */ |
3735 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3736 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3737 | |
25e72d10 PB |
3738 | req->open.dfd = READ_ONCE(sqe->fd); |
3739 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3740 | req->open.filename = getname(fname); |
15b71abe JA |
3741 | if (IS_ERR(req->open.filename)) { |
3742 | ret = PTR_ERR(req->open.filename); | |
3743 | req->open.filename = NULL; | |
3744 | return ret; | |
3745 | } | |
4022e7af | 3746 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3747 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3748 | return 0; |
03b1230c JA |
3749 | } |
3750 | ||
ec65fea5 PB |
3751 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3752 | { | |
3753 | u64 flags, mode; | |
3754 | ||
14587a46 | 3755 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3756 | return -EINVAL; |
ec65fea5 PB |
3757 | mode = READ_ONCE(sqe->len); |
3758 | flags = READ_ONCE(sqe->open_flags); | |
3759 | req->open.how = build_open_how(flags, mode); | |
3760 | return __io_openat_prep(req, sqe); | |
3761 | } | |
3762 | ||
cebdb986 | 3763 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3764 | { |
cebdb986 | 3765 | struct open_how __user *how; |
cebdb986 | 3766 | size_t len; |
0fa03c62 JA |
3767 | int ret; |
3768 | ||
14587a46 | 3769 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3770 | return -EINVAL; |
cebdb986 JA |
3771 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3772 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3773 | if (len < OPEN_HOW_SIZE_VER0) |
3774 | return -EINVAL; | |
3529d8c2 | 3775 | |
cebdb986 JA |
3776 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3777 | len); | |
3778 | if (ret) | |
3779 | return ret; | |
3529d8c2 | 3780 | |
ec65fea5 | 3781 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3782 | } |
3783 | ||
45d189c6 | 3784 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3785 | { |
3786 | struct open_flags op; | |
15b71abe | 3787 | struct file *file; |
3a81fd02 JA |
3788 | bool nonblock_set; |
3789 | bool resolve_nonblock; | |
15b71abe JA |
3790 | int ret; |
3791 | ||
cebdb986 | 3792 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3793 | if (ret) |
3794 | goto err; | |
3a81fd02 JA |
3795 | nonblock_set = op.open_flag & O_NONBLOCK; |
3796 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3797 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3798 | /* |
3799 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3800 | * it'll always -EAGAIN | |
3801 | */ | |
3802 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3803 | return -EAGAIN; | |
3804 | op.lookup_flags |= LOOKUP_CACHED; | |
3805 | op.open_flag |= O_NONBLOCK; | |
3806 | } | |
15b71abe | 3807 | |
4022e7af | 3808 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3809 | if (ret < 0) |
3810 | goto err; | |
3811 | ||
3812 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3813 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3814 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3815 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3816 | /* |
3a81fd02 JA |
3817 | * We could hang on to this 'fd', but seems like marginal |
3818 | * gain for something that is now known to be a slower path. | |
3819 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3820 | */ |
3a81fd02 JA |
3821 | put_unused_fd(ret); |
3822 | return -EAGAIN; | |
3823 | } | |
3824 | ||
15b71abe JA |
3825 | if (IS_ERR(file)) { |
3826 | put_unused_fd(ret); | |
3827 | ret = PTR_ERR(file); | |
3828 | } else { | |
45d189c6 | 3829 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3830 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3831 | fsnotify_open(file); |
3832 | fd_install(ret, file); | |
3833 | } | |
3834 | err: | |
3835 | putname(req->open.filename); | |
8fef80bf | 3836 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3837 | if (ret < 0) |
3838 | req_set_fail_links(req); | |
0bdf3398 | 3839 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3840 | return 0; |
3841 | } | |
3842 | ||
45d189c6 | 3843 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3844 | { |
e45cff58 | 3845 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3846 | } |
3847 | ||
067524e9 JA |
3848 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3849 | const struct io_uring_sqe *sqe) | |
3850 | { | |
3851 | struct io_provide_buf *p = &req->pbuf; | |
3852 | u64 tmp; | |
3853 | ||
3854 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3855 | return -EINVAL; | |
3856 | ||
3857 | tmp = READ_ONCE(sqe->fd); | |
3858 | if (!tmp || tmp > USHRT_MAX) | |
3859 | return -EINVAL; | |
3860 | ||
3861 | memset(p, 0, sizeof(*p)); | |
3862 | p->nbufs = tmp; | |
3863 | p->bgid = READ_ONCE(sqe->buf_group); | |
3864 | return 0; | |
3865 | } | |
3866 | ||
3867 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3868 | int bgid, unsigned nbufs) | |
3869 | { | |
3870 | unsigned i = 0; | |
3871 | ||
3872 | /* shouldn't happen */ | |
3873 | if (!nbufs) | |
3874 | return 0; | |
3875 | ||
3876 | /* the head kbuf is the list itself */ | |
3877 | while (!list_empty(&buf->list)) { | |
3878 | struct io_buffer *nxt; | |
3879 | ||
3880 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3881 | list_del(&nxt->list); | |
3882 | kfree(nxt); | |
3883 | if (++i == nbufs) | |
3884 | return i; | |
3885 | } | |
3886 | i++; | |
3887 | kfree(buf); | |
9e15c3a0 | 3888 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3889 | |
3890 | return i; | |
3891 | } | |
3892 | ||
889fca73 | 3893 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3894 | { |
3895 | struct io_provide_buf *p = &req->pbuf; | |
3896 | struct io_ring_ctx *ctx = req->ctx; | |
3897 | struct io_buffer *head; | |
3898 | int ret = 0; | |
45d189c6 | 3899 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3900 | |
3901 | io_ring_submit_lock(ctx, !force_nonblock); | |
3902 | ||
3903 | lockdep_assert_held(&ctx->uring_lock); | |
3904 | ||
3905 | ret = -ENOENT; | |
9e15c3a0 | 3906 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3907 | if (head) |
3908 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3909 | if (ret < 0) |
3910 | req_set_fail_links(req); | |
067524e9 | 3911 | |
9fb8cb49 PB |
3912 | /* complete before unlock, IOPOLL may need the lock */ |
3913 | __io_req_complete(req, issue_flags, ret, 0); | |
3914 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3915 | return 0; |
3916 | } | |
3917 | ||
ddf0322d JA |
3918 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3919 | const struct io_uring_sqe *sqe) | |
3920 | { | |
38134ada | 3921 | unsigned long size, tmp_check; |
ddf0322d JA |
3922 | struct io_provide_buf *p = &req->pbuf; |
3923 | u64 tmp; | |
3924 | ||
3925 | if (sqe->ioprio || sqe->rw_flags) | |
3926 | return -EINVAL; | |
3927 | ||
3928 | tmp = READ_ONCE(sqe->fd); | |
3929 | if (!tmp || tmp > USHRT_MAX) | |
3930 | return -E2BIG; | |
3931 | p->nbufs = tmp; | |
3932 | p->addr = READ_ONCE(sqe->addr); | |
3933 | p->len = READ_ONCE(sqe->len); | |
3934 | ||
38134ada PB |
3935 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
3936 | &size)) | |
3937 | return -EOVERFLOW; | |
3938 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
3939 | return -EOVERFLOW; | |
3940 | ||
d81269fe PB |
3941 | size = (unsigned long)p->len * p->nbufs; |
3942 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3943 | return -EFAULT; |
3944 | ||
3945 | p->bgid = READ_ONCE(sqe->buf_group); | |
3946 | tmp = READ_ONCE(sqe->off); | |
3947 | if (tmp > USHRT_MAX) | |
3948 | return -E2BIG; | |
3949 | p->bid = tmp; | |
3950 | return 0; | |
3951 | } | |
3952 | ||
3953 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3954 | { | |
3955 | struct io_buffer *buf; | |
3956 | u64 addr = pbuf->addr; | |
3957 | int i, bid = pbuf->bid; | |
3958 | ||
3959 | for (i = 0; i < pbuf->nbufs; i++) { | |
3960 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3961 | if (!buf) | |
3962 | break; | |
3963 | ||
3964 | buf->addr = addr; | |
3965 | buf->len = pbuf->len; | |
3966 | buf->bid = bid; | |
3967 | addr += pbuf->len; | |
3968 | bid++; | |
3969 | if (!*head) { | |
3970 | INIT_LIST_HEAD(&buf->list); | |
3971 | *head = buf; | |
3972 | } else { | |
3973 | list_add_tail(&buf->list, &(*head)->list); | |
3974 | } | |
3975 | } | |
3976 | ||
3977 | return i ? i : -ENOMEM; | |
3978 | } | |
3979 | ||
889fca73 | 3980 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3981 | { |
3982 | struct io_provide_buf *p = &req->pbuf; | |
3983 | struct io_ring_ctx *ctx = req->ctx; | |
3984 | struct io_buffer *head, *list; | |
3985 | int ret = 0; | |
45d189c6 | 3986 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3987 | |
3988 | io_ring_submit_lock(ctx, !force_nonblock); | |
3989 | ||
3990 | lockdep_assert_held(&ctx->uring_lock); | |
3991 | ||
9e15c3a0 | 3992 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
3993 | |
3994 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
3995 | if (ret >= 0 && !list) { |
3996 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
3997 | if (ret < 0) | |
067524e9 | 3998 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 3999 | } |
ddf0322d JA |
4000 | if (ret < 0) |
4001 | req_set_fail_links(req); | |
9fb8cb49 PB |
4002 | /* complete before unlock, IOPOLL may need the lock */ |
4003 | __io_req_complete(req, issue_flags, ret, 0); | |
4004 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4005 | return 0; |
cebdb986 JA |
4006 | } |
4007 | ||
3e4827b0 JA |
4008 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4009 | const struct io_uring_sqe *sqe) | |
4010 | { | |
4011 | #if defined(CONFIG_EPOLL) | |
4012 | if (sqe->ioprio || sqe->buf_index) | |
4013 | return -EINVAL; | |
6ca56f84 | 4014 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4015 | return -EINVAL; |
3e4827b0 JA |
4016 | |
4017 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4018 | req->epoll.op = READ_ONCE(sqe->len); | |
4019 | req->epoll.fd = READ_ONCE(sqe->off); | |
4020 | ||
4021 | if (ep_op_has_event(req->epoll.op)) { | |
4022 | struct epoll_event __user *ev; | |
4023 | ||
4024 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4025 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4026 | return -EFAULT; | |
4027 | } | |
4028 | ||
4029 | return 0; | |
4030 | #else | |
4031 | return -EOPNOTSUPP; | |
4032 | #endif | |
4033 | } | |
4034 | ||
889fca73 | 4035 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4036 | { |
4037 | #if defined(CONFIG_EPOLL) | |
4038 | struct io_epoll *ie = &req->epoll; | |
4039 | int ret; | |
45d189c6 | 4040 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4041 | |
4042 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4043 | if (force_nonblock && ret == -EAGAIN) | |
4044 | return -EAGAIN; | |
4045 | ||
4046 | if (ret < 0) | |
4047 | req_set_fail_links(req); | |
889fca73 | 4048 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4049 | return 0; |
4050 | #else | |
4051 | return -EOPNOTSUPP; | |
4052 | #endif | |
4053 | } | |
4054 | ||
c1ca757b JA |
4055 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4056 | { | |
4057 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4058 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4059 | return -EINVAL; | |
3232dd02 PB |
4060 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4061 | return -EINVAL; | |
c1ca757b JA |
4062 | |
4063 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4064 | req->madvise.len = READ_ONCE(sqe->len); | |
4065 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4066 | return 0; | |
4067 | #else | |
4068 | return -EOPNOTSUPP; | |
4069 | #endif | |
4070 | } | |
4071 | ||
45d189c6 | 4072 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4073 | { |
4074 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4075 | struct io_madvise *ma = &req->madvise; | |
4076 | int ret; | |
4077 | ||
45d189c6 | 4078 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4079 | return -EAGAIN; |
4080 | ||
0726b01e | 4081 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4082 | if (ret < 0) |
4083 | req_set_fail_links(req); | |
e1e16097 | 4084 | io_req_complete(req, ret); |
c1ca757b JA |
4085 | return 0; |
4086 | #else | |
4087 | return -EOPNOTSUPP; | |
4088 | #endif | |
4089 | } | |
4090 | ||
4840e418 JA |
4091 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4092 | { | |
4093 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4094 | return -EINVAL; | |
3232dd02 PB |
4095 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4096 | return -EINVAL; | |
4840e418 JA |
4097 | |
4098 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4099 | req->fadvise.len = READ_ONCE(sqe->len); | |
4100 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4101 | return 0; | |
4102 | } | |
4103 | ||
45d189c6 | 4104 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4105 | { |
4106 | struct io_fadvise *fa = &req->fadvise; | |
4107 | int ret; | |
4108 | ||
45d189c6 | 4109 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4110 | switch (fa->advice) { |
4111 | case POSIX_FADV_NORMAL: | |
4112 | case POSIX_FADV_RANDOM: | |
4113 | case POSIX_FADV_SEQUENTIAL: | |
4114 | break; | |
4115 | default: | |
4116 | return -EAGAIN; | |
4117 | } | |
4118 | } | |
4840e418 JA |
4119 | |
4120 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4121 | if (ret < 0) | |
4122 | req_set_fail_links(req); | |
0bdf3398 | 4123 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4124 | return 0; |
4125 | } | |
4126 | ||
eddc7ef5 JA |
4127 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4128 | { | |
6ca56f84 | 4129 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4130 | return -EINVAL; |
eddc7ef5 JA |
4131 | if (sqe->ioprio || sqe->buf_index) |
4132 | return -EINVAL; | |
9c280f90 | 4133 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4134 | return -EBADF; |
eddc7ef5 | 4135 | |
1d9e1288 BM |
4136 | req->statx.dfd = READ_ONCE(sqe->fd); |
4137 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4138 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4139 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4140 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4141 | |
4142 | return 0; | |
4143 | } | |
4144 | ||
45d189c6 | 4145 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4146 | { |
1d9e1288 | 4147 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4148 | int ret; |
4149 | ||
59d70013 | 4150 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4151 | return -EAGAIN; |
4152 | ||
e62753e4 BM |
4153 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4154 | ctx->buffer); | |
eddc7ef5 | 4155 | |
eddc7ef5 JA |
4156 | if (ret < 0) |
4157 | req_set_fail_links(req); | |
e1e16097 | 4158 | io_req_complete(req, ret); |
eddc7ef5 JA |
4159 | return 0; |
4160 | } | |
4161 | ||
b5dba59e JA |
4162 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4163 | { | |
14587a46 | 4164 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4165 | return -EINVAL; |
b5dba59e JA |
4166 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4167 | sqe->rw_flags || sqe->buf_index) | |
4168 | return -EINVAL; | |
9c280f90 | 4169 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4170 | return -EBADF; |
b5dba59e JA |
4171 | |
4172 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4173 | return 0; |
b5dba59e JA |
4174 | } |
4175 | ||
889fca73 | 4176 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4177 | { |
9eac1904 | 4178 | struct files_struct *files = current->files; |
3af73b28 | 4179 | struct io_close *close = &req->close; |
9eac1904 | 4180 | struct fdtable *fdt; |
a1fde923 PB |
4181 | struct file *file = NULL; |
4182 | int ret = -EBADF; | |
b5dba59e | 4183 | |
9eac1904 JA |
4184 | spin_lock(&files->file_lock); |
4185 | fdt = files_fdtable(files); | |
4186 | if (close->fd >= fdt->max_fds) { | |
4187 | spin_unlock(&files->file_lock); | |
4188 | goto err; | |
4189 | } | |
4190 | file = fdt->fd[close->fd]; | |
a1fde923 | 4191 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4192 | spin_unlock(&files->file_lock); |
4193 | file = NULL; | |
4194 | goto err; | |
3af73b28 | 4195 | } |
b5dba59e JA |
4196 | |
4197 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4198 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4199 | spin_unlock(&files->file_lock); |
0bf0eefd | 4200 | return -EAGAIN; |
a2100672 | 4201 | } |
b5dba59e | 4202 | |
9eac1904 JA |
4203 | ret = __close_fd_get_file(close->fd, &file); |
4204 | spin_unlock(&files->file_lock); | |
4205 | if (ret < 0) { | |
4206 | if (ret == -ENOENT) | |
4207 | ret = -EBADF; | |
4208 | goto err; | |
4209 | } | |
4210 | ||
3af73b28 | 4211 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4212 | ret = filp_close(file, current->files); |
4213 | err: | |
3af73b28 PB |
4214 | if (ret < 0) |
4215 | req_set_fail_links(req); | |
9eac1904 JA |
4216 | if (file) |
4217 | fput(file); | |
889fca73 | 4218 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4219 | return 0; |
b5dba59e JA |
4220 | } |
4221 | ||
1155c76a | 4222 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4223 | { |
4224 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4225 | |
5d17b4a4 JA |
4226 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4227 | return -EINVAL; | |
4228 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4229 | return -EINVAL; | |
4230 | ||
8ed8d3c3 JA |
4231 | req->sync.off = READ_ONCE(sqe->off); |
4232 | req->sync.len = READ_ONCE(sqe->len); | |
4233 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4234 | return 0; |
4235 | } | |
4236 | ||
45d189c6 | 4237 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4238 | { |
8ed8d3c3 JA |
4239 | int ret; |
4240 | ||
ac45abc0 | 4241 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4242 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4243 | return -EAGAIN; |
4244 | ||
9adbd45d | 4245 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4246 | req->sync.flags); |
4247 | if (ret < 0) | |
4248 | req_set_fail_links(req); | |
e1e16097 | 4249 | io_req_complete(req, ret); |
5d17b4a4 JA |
4250 | return 0; |
4251 | } | |
4252 | ||
469956e8 | 4253 | #if defined(CONFIG_NET) |
02d27d89 PB |
4254 | static int io_setup_async_msg(struct io_kiocb *req, |
4255 | struct io_async_msghdr *kmsg) | |
4256 | { | |
e8c2bc1f JA |
4257 | struct io_async_msghdr *async_msg = req->async_data; |
4258 | ||
4259 | if (async_msg) | |
02d27d89 | 4260 | return -EAGAIN; |
e8c2bc1f | 4261 | if (io_alloc_async_data(req)) { |
257e84a5 | 4262 | kfree(kmsg->free_iov); |
02d27d89 PB |
4263 | return -ENOMEM; |
4264 | } | |
e8c2bc1f | 4265 | async_msg = req->async_data; |
02d27d89 | 4266 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4267 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4268 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4269 | /* if were using fast_iov, set it to the new one */ |
4270 | if (!async_msg->free_iov) | |
4271 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4272 | ||
02d27d89 PB |
4273 | return -EAGAIN; |
4274 | } | |
4275 | ||
2ae523ed PB |
4276 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4277 | struct io_async_msghdr *iomsg) | |
4278 | { | |
2ae523ed | 4279 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4280 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4281 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4282 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4283 | } |
4284 | ||
93642ef8 PB |
4285 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4286 | { | |
4287 | int ret; | |
4288 | ||
93642ef8 PB |
4289 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4290 | if (!ret) | |
4291 | req->flags |= REQ_F_NEED_CLEANUP; | |
4292 | return ret; | |
4293 | } | |
4294 | ||
3529d8c2 | 4295 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4296 | { |
e47293fd | 4297 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4298 | |
d2b6f48b PB |
4299 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4300 | return -EINVAL; | |
4301 | ||
270a5940 | 4302 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4303 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4304 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4305 | if (sr->msg_flags & MSG_DONTWAIT) | |
4306 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4307 | |
d8768362 JA |
4308 | #ifdef CONFIG_COMPAT |
4309 | if (req->ctx->compat) | |
4310 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4311 | #endif | |
93642ef8 | 4312 | return 0; |
03b1230c JA |
4313 | } |
4314 | ||
889fca73 | 4315 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4316 | { |
6b754c8b | 4317 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4318 | struct socket *sock; |
7a7cacba | 4319 | unsigned flags; |
0031275d | 4320 | int min_ret = 0; |
0fa03c62 JA |
4321 | int ret; |
4322 | ||
dba4a925 | 4323 | sock = sock_from_file(req->file); |
7a7cacba | 4324 | if (unlikely(!sock)) |
dba4a925 | 4325 | return -ENOTSOCK; |
3529d8c2 | 4326 | |
257e84a5 PB |
4327 | kmsg = req->async_data; |
4328 | if (!kmsg) { | |
7a7cacba PB |
4329 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4330 | if (ret) | |
4331 | return ret; | |
4332 | kmsg = &iomsg; | |
0fa03c62 | 4333 | } |
0fa03c62 | 4334 | |
04411806 PB |
4335 | flags = req->sr_msg.msg_flags; |
4336 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4337 | flags |= MSG_DONTWAIT; |
0031275d SM |
4338 | if (flags & MSG_WAITALL) |
4339 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4340 | ||
7a7cacba | 4341 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4342 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4343 | return io_setup_async_msg(req, kmsg); |
4344 | if (ret == -ERESTARTSYS) | |
4345 | ret = -EINTR; | |
0fa03c62 | 4346 | |
257e84a5 PB |
4347 | /* fast path, check for non-NULL to avoid function call */ |
4348 | if (kmsg->free_iov) | |
4349 | kfree(kmsg->free_iov); | |
99bc4c38 | 4350 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4351 | if (ret < min_ret) |
4e88d6e7 | 4352 | req_set_fail_links(req); |
889fca73 | 4353 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4354 | return 0; |
03b1230c | 4355 | } |
aa1fa28f | 4356 | |
889fca73 | 4357 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4358 | { |
7a7cacba PB |
4359 | struct io_sr_msg *sr = &req->sr_msg; |
4360 | struct msghdr msg; | |
4361 | struct iovec iov; | |
fddaface | 4362 | struct socket *sock; |
7a7cacba | 4363 | unsigned flags; |
0031275d | 4364 | int min_ret = 0; |
fddaface JA |
4365 | int ret; |
4366 | ||
dba4a925 | 4367 | sock = sock_from_file(req->file); |
7a7cacba | 4368 | if (unlikely(!sock)) |
dba4a925 | 4369 | return -ENOTSOCK; |
fddaface | 4370 | |
7a7cacba PB |
4371 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4372 | if (unlikely(ret)) | |
14db8411 | 4373 | return ret; |
fddaface | 4374 | |
7a7cacba PB |
4375 | msg.msg_name = NULL; |
4376 | msg.msg_control = NULL; | |
4377 | msg.msg_controllen = 0; | |
4378 | msg.msg_namelen = 0; | |
fddaface | 4379 | |
04411806 PB |
4380 | flags = req->sr_msg.msg_flags; |
4381 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4382 | flags |= MSG_DONTWAIT; |
0031275d SM |
4383 | if (flags & MSG_WAITALL) |
4384 | min_ret = iov_iter_count(&msg.msg_iter); | |
4385 | ||
7a7cacba PB |
4386 | msg.msg_flags = flags; |
4387 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4388 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4389 | return -EAGAIN; |
4390 | if (ret == -ERESTARTSYS) | |
4391 | ret = -EINTR; | |
fddaface | 4392 | |
0031275d | 4393 | if (ret < min_ret) |
fddaface | 4394 | req_set_fail_links(req); |
889fca73 | 4395 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4396 | return 0; |
fddaface JA |
4397 | } |
4398 | ||
1400e697 PB |
4399 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4400 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4401 | { |
4402 | struct io_sr_msg *sr = &req->sr_msg; | |
4403 | struct iovec __user *uiov; | |
4404 | size_t iov_len; | |
4405 | int ret; | |
4406 | ||
1400e697 PB |
4407 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4408 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4409 | if (ret) |
4410 | return ret; | |
4411 | ||
4412 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4413 | if (iov_len > 1) | |
4414 | return -EINVAL; | |
5476dfed | 4415 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4416 | return -EFAULT; |
5476dfed | 4417 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4418 | iomsg->free_iov = NULL; |
52de1fe1 | 4419 | } else { |
257e84a5 | 4420 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4421 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4422 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4423 | false); |
52de1fe1 JA |
4424 | if (ret > 0) |
4425 | ret = 0; | |
4426 | } | |
4427 | ||
4428 | return ret; | |
4429 | } | |
4430 | ||
4431 | #ifdef CONFIG_COMPAT | |
4432 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4433 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4434 | { |
52de1fe1 JA |
4435 | struct io_sr_msg *sr = &req->sr_msg; |
4436 | struct compat_iovec __user *uiov; | |
4437 | compat_uptr_t ptr; | |
4438 | compat_size_t len; | |
4439 | int ret; | |
4440 | ||
4af3417a PB |
4441 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4442 | &ptr, &len); | |
52de1fe1 JA |
4443 | if (ret) |
4444 | return ret; | |
4445 | ||
4446 | uiov = compat_ptr(ptr); | |
4447 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4448 | compat_ssize_t clen; | |
4449 | ||
4450 | if (len > 1) | |
4451 | return -EINVAL; | |
4452 | if (!access_ok(uiov, sizeof(*uiov))) | |
4453 | return -EFAULT; | |
4454 | if (__get_user(clen, &uiov->iov_len)) | |
4455 | return -EFAULT; | |
4456 | if (clen < 0) | |
4457 | return -EINVAL; | |
2d280bc8 | 4458 | sr->len = clen; |
257e84a5 | 4459 | iomsg->free_iov = NULL; |
52de1fe1 | 4460 | } else { |
257e84a5 | 4461 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4462 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4463 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4464 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4465 | if (ret < 0) |
4466 | return ret; | |
4467 | } | |
4468 | ||
4469 | return 0; | |
4470 | } | |
4471 | #endif | |
4472 | ||
1400e697 PB |
4473 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4474 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4475 | { |
1400e697 | 4476 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4477 | |
4478 | #ifdef CONFIG_COMPAT | |
4479 | if (req->ctx->compat) | |
1400e697 | 4480 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4481 | #endif |
52de1fe1 | 4482 | |
1400e697 | 4483 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4484 | } |
4485 | ||
bcda7baa | 4486 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4487 | bool needs_lock) |
bcda7baa JA |
4488 | { |
4489 | struct io_sr_msg *sr = &req->sr_msg; | |
4490 | struct io_buffer *kbuf; | |
4491 | ||
bcda7baa JA |
4492 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4493 | if (IS_ERR(kbuf)) | |
4494 | return kbuf; | |
4495 | ||
4496 | sr->kbuf = kbuf; | |
4497 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4498 | return kbuf; |
fddaface JA |
4499 | } |
4500 | ||
7fbb1b54 PB |
4501 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4502 | { | |
4503 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4504 | } | |
4505 | ||
93642ef8 | 4506 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4507 | { |
99bc4c38 | 4508 | int ret; |
3529d8c2 | 4509 | |
93642ef8 PB |
4510 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4511 | if (!ret) | |
4512 | req->flags |= REQ_F_NEED_CLEANUP; | |
4513 | return ret; | |
4514 | } | |
4515 | ||
4516 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4517 | { | |
4518 | struct io_sr_msg *sr = &req->sr_msg; | |
4519 | ||
d2b6f48b PB |
4520 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4521 | return -EINVAL; | |
4522 | ||
270a5940 | 4523 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4524 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4525 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4526 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4527 | if (sr->msg_flags & MSG_DONTWAIT) | |
4528 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4529 | |
d8768362 JA |
4530 | #ifdef CONFIG_COMPAT |
4531 | if (req->ctx->compat) | |
4532 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4533 | #endif | |
93642ef8 | 4534 | return 0; |
aa1fa28f JA |
4535 | } |
4536 | ||
889fca73 | 4537 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4538 | { |
6b754c8b | 4539 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4540 | struct socket *sock; |
7fbb1b54 | 4541 | struct io_buffer *kbuf; |
7a7cacba | 4542 | unsigned flags; |
0031275d | 4543 | int min_ret = 0; |
52de1fe1 | 4544 | int ret, cflags = 0; |
45d189c6 | 4545 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4546 | |
dba4a925 | 4547 | sock = sock_from_file(req->file); |
7a7cacba | 4548 | if (unlikely(!sock)) |
dba4a925 | 4549 | return -ENOTSOCK; |
3529d8c2 | 4550 | |
257e84a5 PB |
4551 | kmsg = req->async_data; |
4552 | if (!kmsg) { | |
7a7cacba PB |
4553 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4554 | if (ret) | |
681fda8d | 4555 | return ret; |
7a7cacba PB |
4556 | kmsg = &iomsg; |
4557 | } | |
03b1230c | 4558 | |
bc02ef33 | 4559 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4560 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4561 | if (IS_ERR(kbuf)) |
52de1fe1 | 4562 | return PTR_ERR(kbuf); |
7a7cacba | 4563 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4564 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4565 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4566 | 1, req->sr_msg.len); |
4567 | } | |
52de1fe1 | 4568 | |
04411806 PB |
4569 | flags = req->sr_msg.msg_flags; |
4570 | if (force_nonblock) | |
7a7cacba | 4571 | flags |= MSG_DONTWAIT; |
0031275d SM |
4572 | if (flags & MSG_WAITALL) |
4573 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4574 | ||
7a7cacba PB |
4575 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4576 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4577 | if (force_nonblock && ret == -EAGAIN) |
4578 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4579 | if (ret == -ERESTARTSYS) |
4580 | ret = -EINTR; | |
03b1230c | 4581 | |
7fbb1b54 PB |
4582 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4583 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4584 | /* fast path, check for non-NULL to avoid function call */ |
4585 | if (kmsg->free_iov) | |
4586 | kfree(kmsg->free_iov); | |
99bc4c38 | 4587 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4588 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
4e88d6e7 | 4589 | req_set_fail_links(req); |
889fca73 | 4590 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4591 | return 0; |
0fa03c62 | 4592 | } |
5d17b4a4 | 4593 | |
889fca73 | 4594 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4595 | { |
6b754c8b | 4596 | struct io_buffer *kbuf; |
7a7cacba PB |
4597 | struct io_sr_msg *sr = &req->sr_msg; |
4598 | struct msghdr msg; | |
4599 | void __user *buf = sr->buf; | |
fddaface | 4600 | struct socket *sock; |
7a7cacba PB |
4601 | struct iovec iov; |
4602 | unsigned flags; | |
0031275d | 4603 | int min_ret = 0; |
bcda7baa | 4604 | int ret, cflags = 0; |
45d189c6 | 4605 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4606 | |
dba4a925 | 4607 | sock = sock_from_file(req->file); |
7a7cacba | 4608 | if (unlikely(!sock)) |
dba4a925 | 4609 | return -ENOTSOCK; |
fddaface | 4610 | |
bc02ef33 | 4611 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4612 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4613 | if (IS_ERR(kbuf)) |
4614 | return PTR_ERR(kbuf); | |
7a7cacba | 4615 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4616 | } |
bcda7baa | 4617 | |
7a7cacba | 4618 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4619 | if (unlikely(ret)) |
4620 | goto out_free; | |
fddaface | 4621 | |
7a7cacba PB |
4622 | msg.msg_name = NULL; |
4623 | msg.msg_control = NULL; | |
4624 | msg.msg_controllen = 0; | |
4625 | msg.msg_namelen = 0; | |
4626 | msg.msg_iocb = NULL; | |
4627 | msg.msg_flags = 0; | |
fddaface | 4628 | |
04411806 PB |
4629 | flags = req->sr_msg.msg_flags; |
4630 | if (force_nonblock) | |
7a7cacba | 4631 | flags |= MSG_DONTWAIT; |
0031275d SM |
4632 | if (flags & MSG_WAITALL) |
4633 | min_ret = iov_iter_count(&msg.msg_iter); | |
4634 | ||
7a7cacba PB |
4635 | ret = sock_recvmsg(sock, &msg, flags); |
4636 | if (force_nonblock && ret == -EAGAIN) | |
4637 | return -EAGAIN; | |
4638 | if (ret == -ERESTARTSYS) | |
4639 | ret = -EINTR; | |
14c32eee | 4640 | out_free: |
7fbb1b54 PB |
4641 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4642 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4643 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
fddaface | 4644 | req_set_fail_links(req); |
889fca73 | 4645 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4646 | return 0; |
fddaface JA |
4647 | } |
4648 | ||
3529d8c2 | 4649 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4650 | { |
8ed8d3c3 JA |
4651 | struct io_accept *accept = &req->accept; |
4652 | ||
14587a46 | 4653 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4654 | return -EINVAL; |
8042d6ce | 4655 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4656 | return -EINVAL; |
4657 | ||
d55e5f5b JA |
4658 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4659 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4660 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4661 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4662 | return 0; |
8ed8d3c3 | 4663 | } |
17f2fe35 | 4664 | |
889fca73 | 4665 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4666 | { |
4667 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4668 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4669 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4670 | int ret; |
4671 | ||
e697deed JX |
4672 | if (req->file->f_flags & O_NONBLOCK) |
4673 | req->flags |= REQ_F_NOWAIT; | |
4674 | ||
8ed8d3c3 | 4675 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4676 | accept->addr_len, accept->flags, |
4677 | accept->nofile); | |
8ed8d3c3 | 4678 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4679 | return -EAGAIN; |
ac45abc0 PB |
4680 | if (ret < 0) { |
4681 | if (ret == -ERESTARTSYS) | |
4682 | ret = -EINTR; | |
4e88d6e7 | 4683 | req_set_fail_links(req); |
ac45abc0 | 4684 | } |
889fca73 | 4685 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4686 | return 0; |
8ed8d3c3 JA |
4687 | } |
4688 | ||
93642ef8 PB |
4689 | static int io_connect_prep_async(struct io_kiocb *req) |
4690 | { | |
4691 | struct io_async_connect *io = req->async_data; | |
4692 | struct io_connect *conn = &req->connect; | |
4693 | ||
4694 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4695 | } | |
4696 | ||
3529d8c2 | 4697 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4698 | { |
3529d8c2 | 4699 | struct io_connect *conn = &req->connect; |
f499a021 | 4700 | |
14587a46 | 4701 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4702 | return -EINVAL; |
4703 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4704 | return -EINVAL; | |
4705 | ||
3529d8c2 JA |
4706 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4707 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4708 | return 0; |
f499a021 JA |
4709 | } |
4710 | ||
889fca73 | 4711 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4712 | { |
e8c2bc1f | 4713 | struct io_async_connect __io, *io; |
f8e85cf2 | 4714 | unsigned file_flags; |
3fbb51c1 | 4715 | int ret; |
45d189c6 | 4716 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4717 | |
e8c2bc1f JA |
4718 | if (req->async_data) { |
4719 | io = req->async_data; | |
f499a021 | 4720 | } else { |
3529d8c2 JA |
4721 | ret = move_addr_to_kernel(req->connect.addr, |
4722 | req->connect.addr_len, | |
e8c2bc1f | 4723 | &__io.address); |
f499a021 JA |
4724 | if (ret) |
4725 | goto out; | |
4726 | io = &__io; | |
4727 | } | |
4728 | ||
3fbb51c1 JA |
4729 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4730 | ||
e8c2bc1f | 4731 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4732 | req->connect.addr_len, file_flags); |
87f80d62 | 4733 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4734 | if (req->async_data) |
b7bb4f7d | 4735 | return -EAGAIN; |
e8c2bc1f | 4736 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4737 | ret = -ENOMEM; |
4738 | goto out; | |
4739 | } | |
e8c2bc1f | 4740 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4741 | return -EAGAIN; |
f499a021 | 4742 | } |
f8e85cf2 JA |
4743 | if (ret == -ERESTARTSYS) |
4744 | ret = -EINTR; | |
f499a021 | 4745 | out: |
4e88d6e7 JA |
4746 | if (ret < 0) |
4747 | req_set_fail_links(req); | |
889fca73 | 4748 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4749 | return 0; |
469956e8 Y |
4750 | } |
4751 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4752 | #define IO_NETOP_FN(op) \ |
4753 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4754 | { \ | |
4755 | return -EOPNOTSUPP; \ | |
4756 | } | |
4757 | ||
4758 | #define IO_NETOP_PREP(op) \ | |
4759 | IO_NETOP_FN(op) \ | |
4760 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4761 | { \ | |
4762 | return -EOPNOTSUPP; \ | |
4763 | } \ | |
4764 | ||
4765 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4766 | IO_NETOP_PREP(op) \ | |
4767 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4768 | { \ | |
4769 | return -EOPNOTSUPP; \ | |
4770 | } | |
4771 | ||
4772 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4773 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4774 | IO_NETOP_PREP_ASYNC(connect); | |
4775 | IO_NETOP_PREP(accept); | |
4776 | IO_NETOP_FN(send); | |
4777 | IO_NETOP_FN(recv); | |
469956e8 | 4778 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4779 | |
d7718a9d JA |
4780 | struct io_poll_table { |
4781 | struct poll_table_struct pt; | |
4782 | struct io_kiocb *req; | |
4783 | int error; | |
4784 | }; | |
ce593a6c | 4785 | |
d7718a9d JA |
4786 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4787 | __poll_t mask, task_work_func_t func) | |
4788 | { | |
aa96bf8a | 4789 | int ret; |
d7718a9d JA |
4790 | |
4791 | /* for instances that support it check for an event match first: */ | |
4792 | if (mask && !(mask & poll->events)) | |
4793 | return 0; | |
4794 | ||
4795 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4796 | ||
4797 | list_del_init(&poll->wait.entry); | |
4798 | ||
d7718a9d | 4799 | req->result = mask; |
7cbf1722 | 4800 | req->task_work.func = func; |
6d816e08 | 4801 | |
d7718a9d | 4802 | /* |
e3aabf95 JA |
4803 | * If this fails, then the task is exiting. When a task exits, the |
4804 | * work gets canceled, so just cancel this request as well instead | |
4805 | * of executing it. We can't safely execute it anyway, as we may not | |
4806 | * have the needed state needed for it anyway. | |
d7718a9d | 4807 | */ |
355fb9e2 | 4808 | ret = io_req_task_work_add(req); |
aa96bf8a | 4809 | if (unlikely(ret)) { |
e3aabf95 | 4810 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4811 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4812 | } |
d7718a9d JA |
4813 | return 1; |
4814 | } | |
4815 | ||
74ce6ce4 JA |
4816 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4817 | __acquires(&req->ctx->completion_lock) | |
4818 | { | |
4819 | struct io_ring_ctx *ctx = req->ctx; | |
4820 | ||
4821 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4822 | struct poll_table_struct pt = { ._key = poll->events }; | |
4823 | ||
4824 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4825 | } | |
4826 | ||
4827 | spin_lock_irq(&ctx->completion_lock); | |
4828 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4829 | add_wait_queue(poll->head, &poll->wait); | |
4830 | return true; | |
4831 | } | |
4832 | ||
4833 | return false; | |
4834 | } | |
4835 | ||
d4e7cd36 | 4836 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4837 | { |
e8c2bc1f | 4838 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4839 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4840 | return req->async_data; |
d4e7cd36 JA |
4841 | return req->apoll->double_poll; |
4842 | } | |
4843 | ||
4844 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4845 | { | |
4846 | if (req->opcode == IORING_OP_POLL_ADD) | |
4847 | return &req->poll; | |
4848 | return &req->apoll->poll; | |
4849 | } | |
4850 | ||
4851 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4852 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4853 | { |
4854 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4855 | |
4856 | lockdep_assert_held(&req->ctx->completion_lock); | |
4857 | ||
4858 | if (poll && poll->head) { | |
4859 | struct wait_queue_head *head = poll->head; | |
4860 | ||
4861 | spin_lock(&head->lock); | |
4862 | list_del_init(&poll->wait.entry); | |
4863 | if (poll->wait.private) | |
de9b4cca | 4864 | req_ref_put(req); |
18bceab1 JA |
4865 | poll->head = NULL; |
4866 | spin_unlock(&head->lock); | |
4867 | } | |
4868 | } | |
4869 | ||
e27414be | 4870 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4871 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4872 | { |
4873 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4874 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4875 | int error; |
18bceab1 | 4876 | |
e27414be | 4877 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4878 | error = -ECANCELED; |
88e41cf9 | 4879 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4880 | } else { |
5082620f | 4881 | error = mangle_poll(mask); |
e27414be | 4882 | } |
b69de288 JA |
4883 | if (req->poll.events & EPOLLONESHOT) |
4884 | flags = 0; | |
ff642164 | 4885 | if (!io_cqring_fill_event(req, error, flags)) { |
5082620f | 4886 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4887 | req->poll.done = true; |
4888 | flags = 0; | |
4889 | } | |
18bceab1 | 4890 | io_commit_cqring(ctx); |
88e41cf9 | 4891 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4892 | } |
4893 | ||
dd221f46 | 4894 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4895 | { |
dd221f46 | 4896 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4897 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4898 | struct io_kiocb *nxt; |
18bceab1 JA |
4899 | |
4900 | if (io_poll_rewait(req, &req->poll)) { | |
4901 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4902 | } else { |
f40b964a | 4903 | bool done; |
88e41cf9 | 4904 | |
e27414be | 4905 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4906 | if (done) { |
4907 | hash_del(&req->hash_node); | |
f40b964a | 4908 | } else { |
88e41cf9 JA |
4909 | req->result = 0; |
4910 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4911 | } | |
dd221f46 | 4912 | spin_unlock_irq(&ctx->completion_lock); |
f40b964a | 4913 | io_cqring_ev_posted(ctx); |
18bceab1 | 4914 | |
88e41cf9 JA |
4915 | if (done) { |
4916 | nxt = io_put_req_find_next(req); | |
4917 | if (nxt) | |
4918 | __io_req_task_submit(nxt); | |
4919 | } | |
dd221f46 | 4920 | } |
18bceab1 JA |
4921 | } |
4922 | ||
4923 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4924 | int sync, void *key) | |
4925 | { | |
4926 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4927 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4928 | __poll_t mask = key_to_poll(key); |
4929 | ||
4930 | /* for instances that support it check for an event match first: */ | |
4931 | if (mask && !(mask & poll->events)) | |
4932 | return 0; | |
88e41cf9 JA |
4933 | if (!(poll->events & EPOLLONESHOT)) |
4934 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4935 | |
8706e04e JA |
4936 | list_del_init(&wait->entry); |
4937 | ||
807abcb0 | 4938 | if (poll && poll->head) { |
18bceab1 JA |
4939 | bool done; |
4940 | ||
807abcb0 JA |
4941 | spin_lock(&poll->head->lock); |
4942 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4943 | if (!done) |
807abcb0 | 4944 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4945 | /* make sure double remove sees this as being gone */ |
4946 | wait->private = NULL; | |
807abcb0 | 4947 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4948 | if (!done) { |
4949 | /* use wait func handler, so it matches the rq type */ | |
4950 | poll->wait.func(&poll->wait, mode, sync, key); | |
4951 | } | |
18bceab1 | 4952 | } |
de9b4cca | 4953 | req_ref_put(req); |
18bceab1 JA |
4954 | return 1; |
4955 | } | |
4956 | ||
4957 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4958 | wait_queue_func_t wake_func) | |
4959 | { | |
4960 | poll->head = NULL; | |
4961 | poll->done = false; | |
4962 | poll->canceled = false; | |
464dca61 JA |
4963 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
4964 | /* mask in events that we always want/need */ | |
4965 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
4966 | INIT_LIST_HEAD(&poll->wait.entry); |
4967 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4968 | } | |
4969 | ||
4970 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4971 | struct wait_queue_head *head, |
4972 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4973 | { |
4974 | struct io_kiocb *req = pt->req; | |
4975 | ||
4976 | /* | |
4977 | * If poll->head is already set, it's because the file being polled | |
4978 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4979 | * for write). Setup a separate io_poll_iocb if this happens. | |
4980 | */ | |
4981 | if (unlikely(poll->head)) { | |
58852d4d PB |
4982 | struct io_poll_iocb *poll_one = poll; |
4983 | ||
18bceab1 | 4984 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4985 | if (*poll_ptr) { |
18bceab1 JA |
4986 | pt->error = -EINVAL; |
4987 | return; | |
4988 | } | |
ea6a693d JA |
4989 | /* |
4990 | * Can't handle multishot for double wait for now, turn it | |
4991 | * into one-shot mode. | |
4992 | */ | |
4993 | if (!(req->poll.events & EPOLLONESHOT)) | |
4994 | req->poll.events |= EPOLLONESHOT; | |
1c3b3e65 JA |
4995 | /* double add on the same waitqueue head, ignore */ |
4996 | if (poll->head == head) | |
4997 | return; | |
18bceab1 JA |
4998 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
4999 | if (!poll) { | |
5000 | pt->error = -ENOMEM; | |
5001 | return; | |
5002 | } | |
58852d4d | 5003 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5004 | req_ref_get(req); |
18bceab1 | 5005 | poll->wait.private = req; |
807abcb0 | 5006 | *poll_ptr = poll; |
18bceab1 JA |
5007 | } |
5008 | ||
5009 | pt->error = 0; | |
5010 | poll->head = head; | |
a31eb4a2 JX |
5011 | |
5012 | if (poll->events & EPOLLEXCLUSIVE) | |
5013 | add_wait_queue_exclusive(head, &poll->wait); | |
5014 | else | |
5015 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5016 | } |
5017 | ||
5018 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5019 | struct poll_table_struct *p) | |
5020 | { | |
5021 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5022 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5023 | |
807abcb0 | 5024 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5025 | } |
5026 | ||
d7718a9d JA |
5027 | static void io_async_task_func(struct callback_head *cb) |
5028 | { | |
5029 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5030 | struct async_poll *apoll = req->apoll; | |
5031 | struct io_ring_ctx *ctx = req->ctx; | |
5032 | ||
5033 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5034 | ||
74ce6ce4 | 5035 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5036 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5037 | return; |
d7718a9d JA |
5038 | } |
5039 | ||
0ea13b44 | 5040 | hash_del(&req->hash_node); |
d4e7cd36 | 5041 | io_poll_remove_double(req); |
74ce6ce4 JA |
5042 | spin_unlock_irq(&ctx->completion_lock); |
5043 | ||
0be0b0e3 PB |
5044 | if (!READ_ONCE(apoll->poll.canceled)) |
5045 | __io_req_task_submit(req); | |
5046 | else | |
2593553a | 5047 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5048 | } |
5049 | ||
5050 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5051 | void *key) | |
5052 | { | |
5053 | struct io_kiocb *req = wait->private; | |
5054 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5055 | ||
5056 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5057 | key_to_poll(key)); | |
5058 | ||
5059 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5060 | } | |
5061 | ||
5062 | static void io_poll_req_insert(struct io_kiocb *req) | |
5063 | { | |
5064 | struct io_ring_ctx *ctx = req->ctx; | |
5065 | struct hlist_head *list; | |
5066 | ||
5067 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5068 | hlist_add_head(&req->hash_node, list); | |
5069 | } | |
5070 | ||
5071 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5072 | struct io_poll_iocb *poll, | |
5073 | struct io_poll_table *ipt, __poll_t mask, | |
5074 | wait_queue_func_t wake_func) | |
5075 | __acquires(&ctx->completion_lock) | |
5076 | { | |
5077 | struct io_ring_ctx *ctx = req->ctx; | |
5078 | bool cancel = false; | |
5079 | ||
4d52f338 | 5080 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5081 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5082 | poll->file = req->file; |
18bceab1 | 5083 | poll->wait.private = req; |
d7718a9d JA |
5084 | |
5085 | ipt->pt._key = mask; | |
5086 | ipt->req = req; | |
5087 | ipt->error = -EINVAL; | |
5088 | ||
d7718a9d JA |
5089 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5090 | ||
5091 | spin_lock_irq(&ctx->completion_lock); | |
5092 | if (likely(poll->head)) { | |
5093 | spin_lock(&poll->head->lock); | |
5094 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5095 | if (ipt->error) | |
5096 | cancel = true; | |
5097 | ipt->error = 0; | |
5098 | mask = 0; | |
5099 | } | |
88e41cf9 | 5100 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5101 | list_del_init(&poll->wait.entry); |
5102 | else if (cancel) | |
5103 | WRITE_ONCE(poll->canceled, true); | |
5104 | else if (!poll->done) /* actually waiting for an event */ | |
5105 | io_poll_req_insert(req); | |
5106 | spin_unlock(&poll->head->lock); | |
5107 | } | |
5108 | ||
5109 | return mask; | |
5110 | } | |
5111 | ||
5112 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5113 | { | |
5114 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5115 | struct io_ring_ctx *ctx = req->ctx; | |
5116 | struct async_poll *apoll; | |
5117 | struct io_poll_table ipt; | |
5118 | __poll_t mask, ret; | |
9dab14b8 | 5119 | int rw; |
d7718a9d JA |
5120 | |
5121 | if (!req->file || !file_can_poll(req->file)) | |
5122 | return false; | |
24c74678 | 5123 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5124 | return false; |
9dab14b8 JA |
5125 | if (def->pollin) |
5126 | rw = READ; | |
5127 | else if (def->pollout) | |
5128 | rw = WRITE; | |
5129 | else | |
5130 | return false; | |
5131 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5132 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5133 | return false; |
5134 | ||
5135 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5136 | if (unlikely(!apoll)) | |
5137 | return false; | |
807abcb0 | 5138 | apoll->double_poll = NULL; |
d7718a9d JA |
5139 | |
5140 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5141 | req->apoll = apoll; |
d7718a9d | 5142 | |
88e41cf9 | 5143 | mask = EPOLLONESHOT; |
d7718a9d | 5144 | if (def->pollin) |
8755d97a | 5145 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5146 | if (def->pollout) |
5147 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5148 | |
5149 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5150 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5151 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5152 | mask &= ~POLLIN; | |
5153 | ||
d7718a9d JA |
5154 | mask |= POLLERR | POLLPRI; |
5155 | ||
5156 | ipt.pt._qproc = io_async_queue_proc; | |
5157 | ||
5158 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5159 | io_async_wake); | |
a36da65c | 5160 | if (ret || ipt.error) { |
d4e7cd36 | 5161 | io_poll_remove_double(req); |
d7718a9d | 5162 | spin_unlock_irq(&ctx->completion_lock); |
d7718a9d JA |
5163 | return false; |
5164 | } | |
5165 | spin_unlock_irq(&ctx->completion_lock); | |
5166 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5167 | apoll->poll.events); | |
5168 | return true; | |
5169 | } | |
5170 | ||
5171 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5172 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5173 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5174 | { |
b41e9852 | 5175 | bool do_complete = false; |
221c5eb2 | 5176 | |
5082620f JA |
5177 | if (!poll->head) |
5178 | return false; | |
221c5eb2 | 5179 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5180 | if (do_cancel) |
5181 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5182 | if (!list_empty(&poll->wait.entry)) { |
5183 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5184 | do_complete = true; |
221c5eb2 JA |
5185 | } |
5186 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5187 | hash_del(&req->hash_node); |
d7718a9d JA |
5188 | return do_complete; |
5189 | } | |
5190 | ||
b2c3f7e1 | 5191 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5192 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5193 | { |
5194 | bool do_complete; | |
5195 | ||
d4e7cd36 | 5196 | io_poll_remove_double(req); |
e31001a3 | 5197 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5198 | |
e31001a3 | 5199 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
d7718a9d | 5200 | /* non-poll requests have submit ref still */ |
e31001a3 | 5201 | req_ref_put(req); |
b1f573bd | 5202 | } |
b2c3f7e1 JA |
5203 | return do_complete; |
5204 | } | |
5205 | ||
5206 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5207 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5208 | { |
5209 | bool do_complete; | |
5210 | ||
5211 | do_complete = io_poll_remove_waitqs(req); | |
b41e9852 | 5212 | if (do_complete) { |
ff642164 | 5213 | io_cqring_fill_event(req, -ECANCELED, 0); |
b41e9852 | 5214 | io_commit_cqring(req->ctx); |
f254ac04 | 5215 | req_set_fail_links(req); |
216578e5 | 5216 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5217 | } |
5218 | ||
5219 | return do_complete; | |
221c5eb2 JA |
5220 | } |
5221 | ||
76e1b642 JA |
5222 | /* |
5223 | * Returns true if we found and killed one or more poll requests | |
5224 | */ | |
6b81928d PB |
5225 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5226 | struct files_struct *files) | |
221c5eb2 | 5227 | { |
78076bb6 | 5228 | struct hlist_node *tmp; |
221c5eb2 | 5229 | struct io_kiocb *req; |
8e2e1faf | 5230 | int posted = 0, i; |
221c5eb2 JA |
5231 | |
5232 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5233 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5234 | struct hlist_head *list; | |
5235 | ||
5236 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5237 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5238 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5239 | posted += io_poll_remove_one(req); |
5240 | } | |
221c5eb2 JA |
5241 | } |
5242 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5243 | |
8e2e1faf JA |
5244 | if (posted) |
5245 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5246 | |
5247 | return posted != 0; | |
221c5eb2 JA |
5248 | } |
5249 | ||
9ba5fac8 PB |
5250 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5251 | bool poll_only) | |
e07785b0 | 5252 | __must_hold(&ctx->completion_lock) |
47f46768 | 5253 | { |
78076bb6 | 5254 | struct hlist_head *list; |
47f46768 JA |
5255 | struct io_kiocb *req; |
5256 | ||
78076bb6 JA |
5257 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5258 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5259 | if (sqe_addr != req->user_data) |
5260 | continue; | |
9ba5fac8 PB |
5261 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5262 | continue; | |
b2cb805f | 5263 | return req; |
47f46768 | 5264 | } |
b2cb805f JA |
5265 | return NULL; |
5266 | } | |
5267 | ||
9ba5fac8 PB |
5268 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5269 | bool poll_only) | |
e07785b0 | 5270 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5271 | { |
5272 | struct io_kiocb *req; | |
5273 | ||
9ba5fac8 | 5274 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5275 | if (!req) |
5276 | return -ENOENT; | |
5277 | if (io_poll_remove_one(req)) | |
5278 | return 0; | |
5279 | ||
5280 | return -EALREADY; | |
47f46768 JA |
5281 | } |
5282 | ||
9096af3e PB |
5283 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5284 | unsigned int flags) | |
5285 | { | |
5286 | u32 events; | |
5287 | ||
5288 | events = READ_ONCE(sqe->poll32_events); | |
5289 | #ifdef __BIG_ENDIAN | |
5290 | events = swahw32(events); | |
5291 | #endif | |
5292 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5293 | events |= EPOLLONESHOT; | |
5294 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
5295 | } | |
5296 | ||
c5de0036 | 5297 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5298 | const struct io_uring_sqe *sqe) |
0969e783 | 5299 | { |
c5de0036 PB |
5300 | struct io_poll_update *upd = &req->poll_update; |
5301 | u32 flags; | |
5302 | ||
0969e783 JA |
5303 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5304 | return -EINVAL; | |
c5de0036 PB |
5305 | if (sqe->ioprio || sqe->buf_index) |
5306 | return -EINVAL; | |
5307 | flags = READ_ONCE(sqe->len); | |
5308 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5309 | IORING_POLL_ADD_MULTI)) | |
5310 | return -EINVAL; | |
5311 | /* meaningless without update */ | |
5312 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5313 | return -EINVAL; |
5314 | ||
c5de0036 PB |
5315 | upd->old_user_data = READ_ONCE(sqe->addr); |
5316 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5317 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5318 | |
c5de0036 PB |
5319 | upd->new_user_data = READ_ONCE(sqe->off); |
5320 | if (!upd->update_user_data && upd->new_user_data) | |
5321 | return -EINVAL; | |
5322 | if (upd->update_events) | |
5323 | upd->events = io_poll_parse_events(sqe, flags); | |
5324 | else if (sqe->poll32_events) | |
5325 | return -EINVAL; | |
221c5eb2 | 5326 | |
221c5eb2 JA |
5327 | return 0; |
5328 | } | |
5329 | ||
221c5eb2 JA |
5330 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5331 | void *key) | |
5332 | { | |
c2f2eb7d JA |
5333 | struct io_kiocb *req = wait->private; |
5334 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5335 | |
d7718a9d | 5336 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5337 | } |
5338 | ||
221c5eb2 JA |
5339 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5340 | struct poll_table_struct *p) | |
5341 | { | |
5342 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5343 | ||
e8c2bc1f | 5344 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5345 | } |
5346 | ||
3529d8c2 | 5347 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 | 5348 | { |
c5de0036 PB |
5349 | struct io_poll_iocb *poll = &req->poll; |
5350 | u32 flags; | |
221c5eb2 JA |
5351 | |
5352 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5353 | return -EINVAL; | |
c5de0036 | 5354 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5355 | return -EINVAL; |
5356 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5357 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 | 5358 | return -EINVAL; |
9096af3e | 5359 | |
c5de0036 | 5360 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5361 | return 0; |
5362 | } | |
5363 | ||
c5de0036 | 5364 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5365 | { |
5366 | struct io_poll_iocb *poll = &req->poll; | |
5367 | struct io_ring_ctx *ctx = req->ctx; | |
5368 | struct io_poll_table ipt; | |
0969e783 | 5369 | __poll_t mask; |
0969e783 | 5370 | |
d7718a9d | 5371 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5372 | |
d7718a9d JA |
5373 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5374 | io_poll_wake); | |
221c5eb2 | 5375 | |
8c838788 | 5376 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5377 | ipt.error = 0; |
e27414be | 5378 | io_poll_complete(req, mask); |
221c5eb2 | 5379 | } |
221c5eb2 JA |
5380 | spin_unlock_irq(&ctx->completion_lock); |
5381 | ||
8c838788 JA |
5382 | if (mask) { |
5383 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5384 | if (poll->events & EPOLLONESHOT) |
5385 | io_put_req(req); | |
221c5eb2 | 5386 | } |
8c838788 | 5387 | return ipt.error; |
221c5eb2 JA |
5388 | } |
5389 | ||
c5de0036 | 5390 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5391 | { |
5392 | struct io_ring_ctx *ctx = req->ctx; | |
5393 | struct io_kiocb *preq; | |
cb3b200e | 5394 | bool completing; |
b69de288 JA |
5395 | int ret; |
5396 | ||
5397 | spin_lock_irq(&ctx->completion_lock); | |
9ba5fac8 | 5398 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5399 | if (!preq) { |
5400 | ret = -ENOENT; | |
5401 | goto err; | |
b69de288 | 5402 | } |
cb3b200e | 5403 | |
c5de0036 PB |
5404 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5405 | completing = true; | |
5406 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5407 | goto err; | |
5408 | } | |
5409 | ||
cb3b200e JA |
5410 | /* |
5411 | * Don't allow racy completion with singleshot, as we cannot safely | |
5412 | * update those. For multishot, if we're racing with completion, just | |
5413 | * let completion re-add it. | |
5414 | */ | |
5415 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5416 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5417 | ret = -EALREADY; | |
5418 | goto err; | |
b69de288 JA |
5419 | } |
5420 | /* we now have a detached poll request. reissue. */ | |
5421 | ret = 0; | |
5422 | err: | |
b69de288 | 5423 | if (ret < 0) { |
cb3b200e | 5424 | spin_unlock_irq(&ctx->completion_lock); |
b69de288 JA |
5425 | req_set_fail_links(req); |
5426 | io_req_complete(req, ret); | |
5427 | return 0; | |
5428 | } | |
5429 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5430 | if (req->poll_update.update_events) { |
b69de288 | 5431 | preq->poll.events &= ~0xffff; |
9d805892 | 5432 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5433 | preq->poll.events |= IO_POLL_UNMASK; |
5434 | } | |
9d805892 PB |
5435 | if (req->poll_update.update_user_data) |
5436 | preq->user_data = req->poll_update.new_user_data; | |
cb3b200e JA |
5437 | spin_unlock_irq(&ctx->completion_lock); |
5438 | ||
b69de288 JA |
5439 | /* complete update request, we're done with it */ |
5440 | io_req_complete(req, ret); | |
5441 | ||
cb3b200e | 5442 | if (!completing) { |
c5de0036 | 5443 | ret = io_poll_add(preq, issue_flags); |
cb3b200e JA |
5444 | if (ret < 0) { |
5445 | req_set_fail_links(preq); | |
5446 | io_req_complete(preq, ret); | |
5447 | } | |
b69de288 JA |
5448 | } |
5449 | return 0; | |
5450 | } | |
5451 | ||
5262f567 JA |
5452 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5453 | { | |
ad8a48ac JA |
5454 | struct io_timeout_data *data = container_of(timer, |
5455 | struct io_timeout_data, timer); | |
5456 | struct io_kiocb *req = data->req; | |
5457 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5458 | unsigned long flags; |
5459 | ||
5262f567 | 5460 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5461 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5462 | atomic_set(&req->ctx->cq_timeouts, |
5463 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5464 | ||
ff642164 | 5465 | io_cqring_fill_event(req, -ETIME, 0); |
5262f567 JA |
5466 | io_commit_cqring(ctx); |
5467 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5468 | ||
5469 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5470 | req_set_fail_links(req); |
5262f567 JA |
5471 | io_put_req(req); |
5472 | return HRTIMER_NORESTART; | |
5473 | } | |
5474 | ||
fbd15848 PB |
5475 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5476 | __u64 user_data) | |
e07785b0 | 5477 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5478 | { |
fbd15848 | 5479 | struct io_timeout_data *io; |
47f46768 | 5480 | struct io_kiocb *req; |
fd9c7bc5 | 5481 | bool found = false; |
f254ac04 | 5482 | |
135fcde8 | 5483 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5484 | found = user_data == req->user_data; |
5485 | if (found) | |
47f46768 | 5486 | break; |
47f46768 | 5487 | } |
fd9c7bc5 PB |
5488 | if (!found) |
5489 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5490 | |
5491 | io = req->async_data; | |
fd9c7bc5 | 5492 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5493 | return ERR_PTR(-EALREADY); |
a71976f3 | 5494 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5495 | return req; |
5496 | } | |
47f46768 | 5497 | |
fbd15848 | 5498 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5499 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5500 | { |
5501 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5502 | ||
5503 | if (IS_ERR(req)) | |
5504 | return PTR_ERR(req); | |
f254ac04 JA |
5505 | |
5506 | req_set_fail_links(req); | |
ff642164 | 5507 | io_cqring_fill_event(req, -ECANCELED, 0); |
216578e5 | 5508 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5509 | return 0; |
5510 | } | |
5511 | ||
9c8e11b3 PB |
5512 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5513 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5514 | __must_hold(&ctx->completion_lock) |
47f46768 | 5515 | { |
9c8e11b3 PB |
5516 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5517 | struct io_timeout_data *data; | |
47f46768 | 5518 | |
9c8e11b3 PB |
5519 | if (IS_ERR(req)) |
5520 | return PTR_ERR(req); | |
47f46768 | 5521 | |
9c8e11b3 PB |
5522 | req->timeout.off = 0; /* noseq */ |
5523 | data = req->async_data; | |
5524 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5525 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5526 | data->timer.function = io_timeout_fn; | |
5527 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5528 | return 0; | |
47f46768 JA |
5529 | } |
5530 | ||
3529d8c2 JA |
5531 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5532 | const struct io_uring_sqe *sqe) | |
b29472ee | 5533 | { |
9c8e11b3 PB |
5534 | struct io_timeout_rem *tr = &req->timeout_rem; |
5535 | ||
b29472ee JA |
5536 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5537 | return -EINVAL; | |
61710e43 DA |
5538 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5539 | return -EINVAL; | |
9c8e11b3 | 5540 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5541 | return -EINVAL; |
5542 | ||
9c8e11b3 PB |
5543 | tr->addr = READ_ONCE(sqe->addr); |
5544 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5545 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5546 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5547 | return -EINVAL; | |
5548 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5549 | return -EFAULT; | |
5550 | } else if (tr->flags) { | |
5551 | /* timeout removal doesn't support flags */ | |
b29472ee | 5552 | return -EINVAL; |
9c8e11b3 | 5553 | } |
b29472ee | 5554 | |
b29472ee JA |
5555 | return 0; |
5556 | } | |
5557 | ||
8662daec PB |
5558 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5559 | { | |
5560 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5561 | : HRTIMER_MODE_REL; | |
5562 | } | |
5563 | ||
11365043 JA |
5564 | /* |
5565 | * Remove or update an existing timeout command | |
5566 | */ | |
61e98203 | 5567 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5568 | { |
9c8e11b3 | 5569 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5570 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5571 | int ret; |
11365043 | 5572 | |
11365043 | 5573 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5574 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5575 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5576 | else |
5577 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5578 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5579 | |
ff642164 | 5580 | io_cqring_fill_event(req, ret, 0); |
11365043 JA |
5581 | io_commit_cqring(ctx); |
5582 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5583 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5584 | if (ret < 0) |
5585 | req_set_fail_links(req); | |
ec9c02ad | 5586 | io_put_req(req); |
11365043 | 5587 | return 0; |
5262f567 JA |
5588 | } |
5589 | ||
3529d8c2 | 5590 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5591 | bool is_timeout_link) |
5262f567 | 5592 | { |
ad8a48ac | 5593 | struct io_timeout_data *data; |
a41525ab | 5594 | unsigned flags; |
56080b02 | 5595 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5596 | |
ad8a48ac | 5597 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5598 | return -EINVAL; |
ad8a48ac | 5599 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5600 | return -EINVAL; |
56080b02 | 5601 | if (off && is_timeout_link) |
2d28390a | 5602 | return -EINVAL; |
a41525ab JA |
5603 | flags = READ_ONCE(sqe->timeout_flags); |
5604 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5605 | return -EINVAL; |
bdf20073 | 5606 | |
bfe68a22 | 5607 | req->timeout.off = off; |
26a61679 | 5608 | |
e8c2bc1f | 5609 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5610 | return -ENOMEM; |
5611 | ||
e8c2bc1f | 5612 | data = req->async_data; |
ad8a48ac | 5613 | data->req = req; |
ad8a48ac JA |
5614 | |
5615 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5616 | return -EFAULT; |
5617 | ||
8662daec | 5618 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5619 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5620 | if (is_timeout_link) |
5621 | io_req_track_inflight(req); | |
ad8a48ac JA |
5622 | return 0; |
5623 | } | |
5624 | ||
61e98203 | 5625 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5626 | { |
ad8a48ac | 5627 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5628 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5629 | struct list_head *entry; |
bfe68a22 | 5630 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5631 | |
733f5c95 | 5632 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5633 | |
5262f567 JA |
5634 | /* |
5635 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5636 | * timeout event to be satisfied. If it isn't set, then this is |
5637 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5638 | */ |
8eb7e2d0 | 5639 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5640 | entry = ctx->timeout_list.prev; |
5641 | goto add; | |
5642 | } | |
5262f567 | 5643 | |
bfe68a22 PB |
5644 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5645 | req->timeout.target_seq = tail + off; | |
5262f567 | 5646 | |
f010505b MDG |
5647 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5648 | * This is safe because ->completion_lock is held, and submissions | |
5649 | * and completions are never mixed in the same ->completion_lock section. | |
5650 | */ | |
5651 | ctx->cq_last_tm_flush = tail; | |
5652 | ||
5262f567 JA |
5653 | /* |
5654 | * Insertion sort, ensuring the first entry in the list is always | |
5655 | * the one we need first. | |
5656 | */ | |
5262f567 | 5657 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5658 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5659 | timeout.list); | |
5262f567 | 5660 | |
8eb7e2d0 | 5661 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5662 | continue; |
bfe68a22 PB |
5663 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5664 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5665 | break; |
5666 | } | |
93bd25bb | 5667 | add: |
135fcde8 | 5668 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5669 | data->timer.function = io_timeout_fn; |
5670 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5671 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5672 | return 0; |
5673 | } | |
5262f567 | 5674 | |
f458dd84 PB |
5675 | struct io_cancel_data { |
5676 | struct io_ring_ctx *ctx; | |
5677 | u64 user_data; | |
5678 | }; | |
5679 | ||
62755e35 JA |
5680 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5681 | { | |
5682 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5683 | struct io_cancel_data *cd = data; |
62755e35 | 5684 | |
f458dd84 | 5685 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5686 | } |
5687 | ||
f458dd84 PB |
5688 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5689 | struct io_ring_ctx *ctx) | |
62755e35 | 5690 | { |
f458dd84 | 5691 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5692 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5693 | int ret = 0; |
5694 | ||
f458dd84 | 5695 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5696 | return -ENOENT; |
5697 | ||
f458dd84 | 5698 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5699 | switch (cancel_ret) { |
5700 | case IO_WQ_CANCEL_OK: | |
5701 | ret = 0; | |
5702 | break; | |
5703 | case IO_WQ_CANCEL_RUNNING: | |
5704 | ret = -EALREADY; | |
5705 | break; | |
5706 | case IO_WQ_CANCEL_NOTFOUND: | |
5707 | ret = -ENOENT; | |
5708 | break; | |
5709 | } | |
5710 | ||
e977d6d3 JA |
5711 | return ret; |
5712 | } | |
5713 | ||
47f46768 JA |
5714 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5715 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5716 | int success_ret) |
47f46768 JA |
5717 | { |
5718 | unsigned long flags; | |
5719 | int ret; | |
5720 | ||
f458dd84 | 5721 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5722 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5723 | if (ret != -ENOENT) |
5724 | goto done; | |
47f46768 JA |
5725 | ret = io_timeout_cancel(ctx, sqe_addr); |
5726 | if (ret != -ENOENT) | |
5727 | goto done; | |
9ba5fac8 | 5728 | ret = io_poll_cancel(ctx, sqe_addr, false); |
47f46768 | 5729 | done: |
b0dd8a41 JA |
5730 | if (!ret) |
5731 | ret = success_ret; | |
ff642164 | 5732 | io_cqring_fill_event(req, ret, 0); |
47f46768 JA |
5733 | io_commit_cqring(ctx); |
5734 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5735 | io_cqring_ev_posted(ctx); | |
5736 | ||
4e88d6e7 JA |
5737 | if (ret < 0) |
5738 | req_set_fail_links(req); | |
47f46768 JA |
5739 | } |
5740 | ||
3529d8c2 JA |
5741 | static int io_async_cancel_prep(struct io_kiocb *req, |
5742 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5743 | { |
fbf23849 | 5744 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5745 | return -EINVAL; |
61710e43 DA |
5746 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5747 | return -EINVAL; | |
5748 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5749 | return -EINVAL; |
5750 | ||
fbf23849 JA |
5751 | req->cancel.addr = READ_ONCE(sqe->addr); |
5752 | return 0; | |
5753 | } | |
5754 | ||
61e98203 | 5755 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5756 | { |
5757 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5758 | u64 sqe_addr = req->cancel.addr; |
5759 | struct io_tctx_node *node; | |
5760 | int ret; | |
5761 | ||
5762 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5763 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5764 | spin_lock_irq(&ctx->completion_lock); | |
5765 | if (ret != -ENOENT) | |
5766 | goto done; | |
5767 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5768 | if (ret != -ENOENT) | |
5769 | goto done; | |
9ba5fac8 | 5770 | ret = io_poll_cancel(ctx, sqe_addr, false); |
58f99373 PB |
5771 | if (ret != -ENOENT) |
5772 | goto done; | |
5773 | spin_unlock_irq(&ctx->completion_lock); | |
5774 | ||
5775 | /* slow path, try all io-wq's */ | |
5776 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5777 | ret = -ENOENT; | |
5778 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5779 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5780 | |
58f99373 PB |
5781 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5782 | if (ret != -ENOENT) | |
5783 | break; | |
5784 | } | |
5785 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5786 | ||
5787 | spin_lock_irq(&ctx->completion_lock); | |
5788 | done: | |
ff642164 | 5789 | io_cqring_fill_event(req, ret, 0); |
58f99373 PB |
5790 | io_commit_cqring(ctx); |
5791 | spin_unlock_irq(&ctx->completion_lock); | |
5792 | io_cqring_ev_posted(ctx); | |
5793 | ||
5794 | if (ret < 0) | |
5795 | req_set_fail_links(req); | |
5796 | io_put_req(req); | |
5262f567 JA |
5797 | return 0; |
5798 | } | |
5799 | ||
269bbe5f | 5800 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5801 | const struct io_uring_sqe *sqe) |
5802 | { | |
6ca56f84 JA |
5803 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5804 | return -EINVAL; | |
61710e43 DA |
5805 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5806 | return -EINVAL; | |
5807 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5808 | return -EINVAL; |
5809 | ||
269bbe5f BM |
5810 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5811 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5812 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5813 | return -EINVAL; |
269bbe5f | 5814 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5815 | return 0; |
5816 | } | |
5817 | ||
889fca73 | 5818 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5819 | { |
5820 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5821 | struct io_uring_rsrc_update up; |
05f3fb3c | 5822 | int ret; |
fbf23849 | 5823 | |
45d189c6 | 5824 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5825 | return -EAGAIN; |
05f3fb3c | 5826 | |
269bbe5f BM |
5827 | up.offset = req->rsrc_update.offset; |
5828 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5829 | |
5830 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5831 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5832 | mutex_unlock(&ctx->uring_lock); |
5833 | ||
5834 | if (ret < 0) | |
5835 | req_set_fail_links(req); | |
889fca73 | 5836 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5837 | return 0; |
5838 | } | |
5839 | ||
bfe76559 | 5840 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5841 | { |
d625c6ee | 5842 | switch (req->opcode) { |
e781573e | 5843 | case IORING_OP_NOP: |
bfe76559 | 5844 | return 0; |
f67676d1 JA |
5845 | case IORING_OP_READV: |
5846 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5847 | case IORING_OP_READ: |
bfe76559 | 5848 | return io_read_prep(req, sqe); |
f67676d1 JA |
5849 | case IORING_OP_WRITEV: |
5850 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5851 | case IORING_OP_WRITE: |
bfe76559 | 5852 | return io_write_prep(req, sqe); |
0969e783 | 5853 | case IORING_OP_POLL_ADD: |
bfe76559 | 5854 | return io_poll_add_prep(req, sqe); |
0969e783 | 5855 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5856 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5857 | case IORING_OP_FSYNC: |
1155c76a | 5858 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5859 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5860 | return io_sfr_prep(req, sqe); |
03b1230c | 5861 | case IORING_OP_SENDMSG: |
fddaface | 5862 | case IORING_OP_SEND: |
bfe76559 | 5863 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5864 | case IORING_OP_RECVMSG: |
fddaface | 5865 | case IORING_OP_RECV: |
bfe76559 | 5866 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5867 | case IORING_OP_CONNECT: |
bfe76559 | 5868 | return io_connect_prep(req, sqe); |
2d28390a | 5869 | case IORING_OP_TIMEOUT: |
bfe76559 | 5870 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5871 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5872 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5873 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5874 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5875 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5876 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5877 | case IORING_OP_ACCEPT: |
bfe76559 | 5878 | return io_accept_prep(req, sqe); |
d63d1b5e | 5879 | case IORING_OP_FALLOCATE: |
bfe76559 | 5880 | return io_fallocate_prep(req, sqe); |
15b71abe | 5881 | case IORING_OP_OPENAT: |
bfe76559 | 5882 | return io_openat_prep(req, sqe); |
b5dba59e | 5883 | case IORING_OP_CLOSE: |
bfe76559 | 5884 | return io_close_prep(req, sqe); |
05f3fb3c | 5885 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5886 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5887 | case IORING_OP_STATX: |
bfe76559 | 5888 | return io_statx_prep(req, sqe); |
4840e418 | 5889 | case IORING_OP_FADVISE: |
bfe76559 | 5890 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5891 | case IORING_OP_MADVISE: |
bfe76559 | 5892 | return io_madvise_prep(req, sqe); |
cebdb986 | 5893 | case IORING_OP_OPENAT2: |
bfe76559 | 5894 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5895 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5896 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5897 | case IORING_OP_SPLICE: |
bfe76559 | 5898 | return io_splice_prep(req, sqe); |
ddf0322d | 5899 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5900 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5901 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5902 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5903 | case IORING_OP_TEE: |
bfe76559 | 5904 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5905 | case IORING_OP_SHUTDOWN: |
5906 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5907 | case IORING_OP_RENAMEAT: |
5908 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5909 | case IORING_OP_UNLINKAT: |
5910 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5911 | } |
5912 | ||
bfe76559 PB |
5913 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5914 | req->opcode); | |
5915 | return-EINVAL; | |
5916 | } | |
5917 | ||
93642ef8 | 5918 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5919 | { |
b7e298d2 PB |
5920 | if (!io_op_defs[req->opcode].needs_async_setup) |
5921 | return 0; | |
5922 | if (WARN_ON_ONCE(req->async_data)) | |
5923 | return -EFAULT; | |
5924 | if (io_alloc_async_data(req)) | |
5925 | return -EAGAIN; | |
5926 | ||
93642ef8 PB |
5927 | switch (req->opcode) { |
5928 | case IORING_OP_READV: | |
93642ef8 PB |
5929 | return io_rw_prep_async(req, READ); |
5930 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5931 | return io_rw_prep_async(req, WRITE); |
5932 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5933 | return io_sendmsg_prep_async(req); |
5934 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5935 | return io_recvmsg_prep_async(req); |
5936 | case IORING_OP_CONNECT: | |
5937 | return io_connect_prep_async(req); | |
5938 | } | |
b7e298d2 PB |
5939 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5940 | req->opcode); | |
5941 | return -EFAULT; | |
f67676d1 JA |
5942 | } |
5943 | ||
9cf7c104 PB |
5944 | static u32 io_get_sequence(struct io_kiocb *req) |
5945 | { | |
5946 | struct io_kiocb *pos; | |
5947 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5948 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5949 | |
f2f87370 PB |
5950 | io_for_each_link(pos, req) |
5951 | nr_reqs++; | |
9cf7c104 PB |
5952 | |
5953 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5954 | return total_submitted - nr_reqs; | |
5955 | } | |
5956 | ||
be7053b7 | 5957 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5958 | { |
a197f664 | 5959 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5960 | struct io_defer_entry *de; |
f67676d1 | 5961 | int ret; |
9cf7c104 | 5962 | u32 seq; |
de0617e4 | 5963 | |
9d858b21 | 5964 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5965 | if (likely(list_empty_careful(&ctx->defer_list) && |
5966 | !(req->flags & REQ_F_IO_DRAIN))) | |
5967 | return 0; | |
5968 | ||
5969 | seq = io_get_sequence(req); | |
5970 | /* Still a chance to pass the sequence check */ | |
5971 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5972 | return 0; |
5973 | ||
b7e298d2 | 5974 | ret = io_req_prep_async(req); |
be7053b7 PB |
5975 | if (ret) |
5976 | return ret; | |
cbdcb435 | 5977 | io_prep_async_link(req); |
27dc8338 PB |
5978 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5979 | if (!de) | |
5980 | return -ENOMEM; | |
2d28390a | 5981 | |
de0617e4 | 5982 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5983 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5984 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5985 | kfree(de); |
ae34817b PB |
5986 | io_queue_async_work(req); |
5987 | return -EIOCBQUEUED; | |
de0617e4 JA |
5988 | } |
5989 | ||
915967f6 | 5990 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5991 | de->req = req; |
9cf7c104 | 5992 | de->seq = seq; |
27dc8338 | 5993 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5994 | spin_unlock_irq(&ctx->completion_lock); |
5995 | return -EIOCBQUEUED; | |
5996 | } | |
5997 | ||
68fb8979 | 5998 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5999 | { |
0e1b6fe3 PB |
6000 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6001 | switch (req->opcode) { | |
6002 | case IORING_OP_READV: | |
6003 | case IORING_OP_READ_FIXED: | |
6004 | case IORING_OP_READ: | |
bcda7baa | 6005 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6006 | break; |
6007 | case IORING_OP_RECVMSG: | |
6008 | case IORING_OP_RECV: | |
bcda7baa | 6009 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6010 | break; |
6011 | } | |
6012 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6013 | } |
6014 | ||
0e1b6fe3 PB |
6015 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6016 | switch (req->opcode) { | |
6017 | case IORING_OP_READV: | |
6018 | case IORING_OP_READ_FIXED: | |
6019 | case IORING_OP_READ: | |
6020 | case IORING_OP_WRITEV: | |
6021 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6022 | case IORING_OP_WRITE: { |
6023 | struct io_async_rw *io = req->async_data; | |
6024 | if (io->free_iovec) | |
6025 | kfree(io->free_iovec); | |
0e1b6fe3 | 6026 | break; |
e8c2bc1f | 6027 | } |
0e1b6fe3 | 6028 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6029 | case IORING_OP_SENDMSG: { |
6030 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6031 | |
6032 | kfree(io->free_iov); | |
0e1b6fe3 | 6033 | break; |
e8c2bc1f | 6034 | } |
0e1b6fe3 PB |
6035 | case IORING_OP_SPLICE: |
6036 | case IORING_OP_TEE: | |
e1d767f0 PB |
6037 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6038 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6039 | break; |
f3cd4850 JA |
6040 | case IORING_OP_OPENAT: |
6041 | case IORING_OP_OPENAT2: | |
6042 | if (req->open.filename) | |
6043 | putname(req->open.filename); | |
6044 | break; | |
80a261fd JA |
6045 | case IORING_OP_RENAMEAT: |
6046 | putname(req->rename.oldpath); | |
6047 | putname(req->rename.newpath); | |
6048 | break; | |
14a1143b JA |
6049 | case IORING_OP_UNLINKAT: |
6050 | putname(req->unlink.filename); | |
6051 | break; | |
0e1b6fe3 PB |
6052 | } |
6053 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6054 | } |
75652a30 JA |
6055 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6056 | kfree(req->apoll->double_poll); | |
6057 | kfree(req->apoll); | |
6058 | req->apoll = NULL; | |
6059 | } | |
99bc4c38 PB |
6060 | } |
6061 | ||
889fca73 | 6062 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6063 | { |
a197f664 | 6064 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6065 | const struct cred *creds = NULL; |
d625c6ee | 6066 | int ret; |
2b188cc1 | 6067 | |
003e8dcc JA |
6068 | if (req->work.creds && req->work.creds != current_cred()) |
6069 | creds = override_creds(req->work.creds); | |
5730b27e | 6070 | |
d625c6ee | 6071 | switch (req->opcode) { |
2b188cc1 | 6072 | case IORING_OP_NOP: |
889fca73 | 6073 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6074 | break; |
6075 | case IORING_OP_READV: | |
edafccee | 6076 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6077 | case IORING_OP_READ: |
889fca73 | 6078 | ret = io_read(req, issue_flags); |
edafccee | 6079 | break; |
3529d8c2 | 6080 | case IORING_OP_WRITEV: |
edafccee | 6081 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6082 | case IORING_OP_WRITE: |
889fca73 | 6083 | ret = io_write(req, issue_flags); |
2b188cc1 | 6084 | break; |
c992fe29 | 6085 | case IORING_OP_FSYNC: |
45d189c6 | 6086 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6087 | break; |
221c5eb2 | 6088 | case IORING_OP_POLL_ADD: |
61e98203 | 6089 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6090 | break; |
6091 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6092 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6093 | break; |
5d17b4a4 | 6094 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6095 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6096 | break; |
0fa03c62 | 6097 | case IORING_OP_SENDMSG: |
889fca73 | 6098 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6099 | break; |
fddaface | 6100 | case IORING_OP_SEND: |
889fca73 | 6101 | ret = io_send(req, issue_flags); |
0fa03c62 | 6102 | break; |
aa1fa28f | 6103 | case IORING_OP_RECVMSG: |
889fca73 | 6104 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6105 | break; |
fddaface | 6106 | case IORING_OP_RECV: |
889fca73 | 6107 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6108 | break; |
5262f567 | 6109 | case IORING_OP_TIMEOUT: |
61e98203 | 6110 | ret = io_timeout(req, issue_flags); |
5262f567 | 6111 | break; |
11365043 | 6112 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6113 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6114 | break; |
17f2fe35 | 6115 | case IORING_OP_ACCEPT: |
889fca73 | 6116 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6117 | break; |
f8e85cf2 | 6118 | case IORING_OP_CONNECT: |
889fca73 | 6119 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6120 | break; |
62755e35 | 6121 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6122 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6123 | break; |
d63d1b5e | 6124 | case IORING_OP_FALLOCATE: |
45d189c6 | 6125 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6126 | break; |
15b71abe | 6127 | case IORING_OP_OPENAT: |
45d189c6 | 6128 | ret = io_openat(req, issue_flags); |
15b71abe | 6129 | break; |
b5dba59e | 6130 | case IORING_OP_CLOSE: |
889fca73 | 6131 | ret = io_close(req, issue_flags); |
b5dba59e | 6132 | break; |
05f3fb3c | 6133 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6134 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6135 | break; |
eddc7ef5 | 6136 | case IORING_OP_STATX: |
45d189c6 | 6137 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6138 | break; |
4840e418 | 6139 | case IORING_OP_FADVISE: |
45d189c6 | 6140 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6141 | break; |
c1ca757b | 6142 | case IORING_OP_MADVISE: |
45d189c6 | 6143 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6144 | break; |
cebdb986 | 6145 | case IORING_OP_OPENAT2: |
45d189c6 | 6146 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6147 | break; |
3e4827b0 | 6148 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6149 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6150 | break; |
7d67af2c | 6151 | case IORING_OP_SPLICE: |
45d189c6 | 6152 | ret = io_splice(req, issue_flags); |
7d67af2c | 6153 | break; |
ddf0322d | 6154 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6155 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6156 | break; |
067524e9 | 6157 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6158 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6159 | break; |
f2a8d5c7 | 6160 | case IORING_OP_TEE: |
45d189c6 | 6161 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6162 | break; |
36f4fa68 | 6163 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6164 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6165 | break; |
80a261fd | 6166 | case IORING_OP_RENAMEAT: |
45d189c6 | 6167 | ret = io_renameat(req, issue_flags); |
80a261fd | 6168 | break; |
14a1143b | 6169 | case IORING_OP_UNLINKAT: |
45d189c6 | 6170 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6171 | break; |
2b188cc1 JA |
6172 | default: |
6173 | ret = -EINVAL; | |
6174 | break; | |
6175 | } | |
6176 | ||
5730b27e JA |
6177 | if (creds) |
6178 | revert_creds(creds); | |
6179 | ||
def596e9 JA |
6180 | if (ret) |
6181 | return ret; | |
6182 | ||
b532576e JA |
6183 | /* If the op doesn't have a file, we're not polling for it */ |
6184 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6185 | const bool in_async = io_wq_current_is_worker(); |
6186 | ||
11ba820b JA |
6187 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6188 | if (in_async) | |
6189 | mutex_lock(&ctx->uring_lock); | |
6190 | ||
2e9dbe90 | 6191 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6192 | |
6193 | if (in_async) | |
6194 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6195 | } |
6196 | ||
6197 | return 0; | |
2b188cc1 JA |
6198 | } |
6199 | ||
5280f7e5 | 6200 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6201 | { |
6202 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6203 | struct io_kiocb *timeout; |
561fb04a | 6204 | int ret = 0; |
2b188cc1 | 6205 | |
6df1db6b PB |
6206 | timeout = io_prep_linked_timeout(req); |
6207 | if (timeout) | |
6208 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6209 | |
4014d943 | 6210 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6211 | ret = -ECANCELED; |
31b51510 | 6212 | |
561fb04a | 6213 | if (!ret) { |
561fb04a | 6214 | do { |
889fca73 | 6215 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6216 | /* |
6217 | * We can get EAGAIN for polled IO even though we're | |
6218 | * forcing a sync submission from here, since we can't | |
6219 | * wait for request slots on the block side. | |
6220 | */ | |
6221 | if (ret != -EAGAIN) | |
6222 | break; | |
6223 | cond_resched(); | |
6224 | } while (1); | |
6225 | } | |
31b51510 | 6226 | |
a3df7698 | 6227 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6228 | if (ret) { |
a3df7698 | 6229 | /* io-wq is going to take one down */ |
de9b4cca | 6230 | req_ref_get(req); |
a3df7698 | 6231 | io_req_task_queue_fail(req, ret); |
edafccee | 6232 | } |
2b188cc1 JA |
6233 | } |
6234 | ||
7b29f92d JA |
6235 | #define FFS_ASYNC_READ 0x1UL |
6236 | #define FFS_ASYNC_WRITE 0x2UL | |
6237 | #ifdef CONFIG_64BIT | |
6238 | #define FFS_ISREG 0x4UL | |
6239 | #else | |
6240 | #define FFS_ISREG 0x0UL | |
6241 | #endif | |
6242 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6243 | ||
aeca241b | 6244 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6245 | unsigned i) |
65e19f54 | 6246 | { |
aeca241b | 6247 | struct io_fixed_file *table_l2; |
65e19f54 | 6248 | |
aeca241b PB |
6249 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6250 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6251 | } |
6252 | ||
6253 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, | |
6254 | int index) | |
6255 | { | |
aeca241b | 6256 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
7b29f92d | 6257 | |
a04b0ac0 | 6258 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6259 | } |
6260 | ||
a04b0ac0 | 6261 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6262 | { |
6263 | unsigned long file_ptr = (unsigned long) file; | |
6264 | ||
6265 | if (__io_file_supports_async(file, READ)) | |
6266 | file_ptr |= FFS_ASYNC_READ; | |
6267 | if (__io_file_supports_async(file, WRITE)) | |
6268 | file_ptr |= FFS_ASYNC_WRITE; | |
6269 | if (S_ISREG(file_inode(file)->i_mode)) | |
6270 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6271 | file_slot->file_ptr = file_ptr; |
9a321c98 PB |
6272 | } |
6273 | ||
8371adf5 PB |
6274 | static struct file *io_file_get(struct io_submit_state *state, |
6275 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6276 | { |
a197f664 | 6277 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6278 | struct file *file; |
09bb8394 | 6279 | |
8da11c19 | 6280 | if (fixed) { |
7b29f92d JA |
6281 | unsigned long file_ptr; |
6282 | ||
479f517b | 6283 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6284 | return NULL; |
b7620121 | 6285 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6286 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6287 | file = (struct file *) (file_ptr & FFS_MASK); |
6288 | file_ptr &= ~FFS_MASK; | |
6289 | /* mask in overlapping REQ_F and FFS bits */ | |
6290 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6291 | io_req_set_rsrc_node(req); |
09bb8394 | 6292 | } else { |
c826bd7a | 6293 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6294 | file = __io_file_get(state, fd); |
d44f554e JA |
6295 | |
6296 | /* we don't allow fixed io_uring files */ | |
6297 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6298 | io_req_track_inflight(req); | |
09bb8394 JA |
6299 | } |
6300 | ||
8371adf5 | 6301 | return file; |
09bb8394 JA |
6302 | } |
6303 | ||
2665abfd | 6304 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6305 | { |
ad8a48ac JA |
6306 | struct io_timeout_data *data = container_of(timer, |
6307 | struct io_timeout_data, timer); | |
90cd7e42 | 6308 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6309 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6310 | unsigned long flags; |
2665abfd JA |
6311 | |
6312 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6313 | prev = req->timeout.head; |
6314 | req->timeout.head = NULL; | |
2665abfd JA |
6315 | |
6316 | /* | |
6317 | * We don't expect the list to be empty, that will only happen if we | |
6318 | * race with the completion of the linked work. | |
6319 | */ | |
de9b4cca | 6320 | if (prev && req_ref_inc_not_zero(prev)) |
f2f87370 | 6321 | io_remove_next_linked(prev); |
90cd7e42 PB |
6322 | else |
6323 | prev = NULL; | |
2665abfd JA |
6324 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6325 | ||
6326 | if (prev) { | |
014db007 | 6327 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6328 | io_put_req_deferred(prev, 1); |
47f46768 | 6329 | } else { |
9ae1f8dd | 6330 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6331 | } |
df9727af | 6332 | io_put_req_deferred(req, 1); |
2665abfd JA |
6333 | return HRTIMER_NORESTART; |
6334 | } | |
6335 | ||
de968c18 | 6336 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6337 | { |
de968c18 PB |
6338 | struct io_ring_ctx *ctx = req->ctx; |
6339 | ||
6340 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6341 | /* |
f2f87370 PB |
6342 | * If the back reference is NULL, then our linked request finished |
6343 | * before we got a chance to setup the timer | |
76a46e06 | 6344 | */ |
90cd7e42 | 6345 | if (req->timeout.head) { |
e8c2bc1f | 6346 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6347 | |
ad8a48ac JA |
6348 | data->timer.function = io_link_timeout_fn; |
6349 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6350 | data->mode); | |
2665abfd | 6351 | } |
76a46e06 | 6352 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6353 | /* drop submission reference */ |
76a46e06 JA |
6354 | io_put_req(req); |
6355 | } | |
2665abfd | 6356 | |
ad8a48ac | 6357 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6358 | { |
f2f87370 | 6359 | struct io_kiocb *nxt = req->link; |
2665abfd | 6360 | |
f2f87370 PB |
6361 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6362 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6363 | return NULL; |
2665abfd | 6364 | |
90cd7e42 | 6365 | nxt->timeout.head = req; |
900fad45 | 6366 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6367 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6368 | return nxt; |
2665abfd JA |
6369 | } |
6370 | ||
c5eef2b9 | 6371 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6372 | { |
d3d7298d | 6373 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6374 | int ret; |
2b188cc1 | 6375 | |
c5eef2b9 | 6376 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6377 | |
491381ce JA |
6378 | /* |
6379 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6380 | * doesn't support non-blocking read/write attempts | |
6381 | */ | |
1840038e | 6382 | if (likely(!ret)) { |
0d63c148 | 6383 | /* drop submission reference */ |
e342c807 | 6384 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6385 | struct io_ring_ctx *ctx = req->ctx; |
6386 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6387 | |
6dd0be1e | 6388 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6389 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6390 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6391 | } else { |
d3d7298d | 6392 | io_put_req(req); |
0d63c148 | 6393 | } |
1840038e PB |
6394 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6395 | if (!io_arm_poll_handler(req)) { | |
6396 | /* | |
6397 | * Queued up for async execution, worker will release | |
6398 | * submit reference when the iocb is actually submitted. | |
6399 | */ | |
6400 | io_queue_async_work(req); | |
6401 | } | |
0d63c148 | 6402 | } else { |
f41db273 | 6403 | io_req_complete_failed(req, ret); |
9e645e11 | 6404 | } |
d3d7298d PB |
6405 | if (linked_timeout) |
6406 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6407 | } |
6408 | ||
be7053b7 | 6409 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6410 | { |
6411 | int ret; | |
6412 | ||
be7053b7 | 6413 | ret = io_req_defer(req); |
4fe2c963 JL |
6414 | if (ret) { |
6415 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6416 | fail_req: |
f41db273 | 6417 | io_req_complete_failed(req, ret); |
4fe2c963 | 6418 | } |
2550878f | 6419 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
b7e298d2 | 6420 | ret = io_req_prep_async(req); |
be7053b7 PB |
6421 | if (unlikely(ret)) |
6422 | goto fail_req; | |
ce35a47a JA |
6423 | io_queue_async_work(req); |
6424 | } else { | |
c5eef2b9 | 6425 | __io_queue_sqe(req); |
ce35a47a | 6426 | } |
4fe2c963 JL |
6427 | } |
6428 | ||
b16fed66 PB |
6429 | /* |
6430 | * Check SQE restrictions (opcode and flags). | |
6431 | * | |
6432 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6433 | */ | |
6434 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6435 | struct io_kiocb *req, | |
6436 | unsigned int sqe_flags) | |
4fe2c963 | 6437 | { |
b16fed66 PB |
6438 | if (!ctx->restricted) |
6439 | return true; | |
6440 | ||
6441 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6442 | return false; | |
6443 | ||
6444 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6445 | ctx->restrictions.sqe_flags_required) | |
6446 | return false; | |
6447 | ||
6448 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6449 | ctx->restrictions.sqe_flags_required)) | |
6450 | return false; | |
6451 | ||
6452 | return true; | |
4fe2c963 JL |
6453 | } |
6454 | ||
b16fed66 PB |
6455 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6456 | const struct io_uring_sqe *sqe) | |
6457 | { | |
6458 | struct io_submit_state *state; | |
6459 | unsigned int sqe_flags; | |
003e8dcc | 6460 | int personality, ret = 0; |
b16fed66 PB |
6461 | |
6462 | req->opcode = READ_ONCE(sqe->opcode); | |
6463 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6464 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6465 | req->user_data = READ_ONCE(sqe->user_data); | |
6466 | req->async_data = NULL; | |
6467 | req->file = NULL; | |
6468 | req->ctx = ctx; | |
6469 | req->link = NULL; | |
6470 | req->fixed_rsrc_refs = NULL; | |
6471 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6472 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6473 | req->task = current; |
6474 | req->result = 0; | |
93e68e03 | 6475 | req->work.creds = NULL; |
b16fed66 PB |
6476 | |
6477 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6478 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6479 | req->flags = 0; | |
b16fed66 | 6480 | return -EINVAL; |
ebf4a5db | 6481 | } |
b16fed66 PB |
6482 | |
6483 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6484 | return -EINVAL; | |
6485 | ||
b16fed66 PB |
6486 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6487 | return -EACCES; | |
6488 | ||
6489 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6490 | !io_op_defs[req->opcode].buffer_select) | |
6491 | return -EOPNOTSUPP; | |
863e0560 | 6492 | |
003e8dcc JA |
6493 | personality = READ_ONCE(sqe->personality); |
6494 | if (personality) { | |
61cf9370 | 6495 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6496 | if (!req->work.creds) |
6497 | return -EINVAL; | |
6498 | get_cred(req->work.creds); | |
003e8dcc | 6499 | } |
b16fed66 PB |
6500 | state = &ctx->submit_state; |
6501 | ||
6502 | /* | |
6503 | * Plug now if we have more than 1 IO left after this, and the target | |
6504 | * is potentially a read/write to block based storage. | |
6505 | */ | |
6506 | if (!state->plug_started && state->ios_left > 1 && | |
6507 | io_op_defs[req->opcode].plug) { | |
6508 | blk_start_plug(&state->plug); | |
6509 | state->plug_started = true; | |
6510 | } | |
6511 | ||
6512 | if (io_op_defs[req->opcode].needs_file) { | |
6513 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6514 | ||
6515 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6516 | if (unlikely(!req->file)) | |
6517 | ret = -EBADF; | |
6518 | } | |
6519 | ||
6520 | state->ios_left--; | |
6521 | return ret; | |
6522 | } | |
6523 | ||
a6b8cadc | 6524 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6525 | const struct io_uring_sqe *sqe) |
9e645e11 | 6526 | { |
a1ab7b35 | 6527 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6528 | int ret; |
9e645e11 | 6529 | |
a6b8cadc PB |
6530 | ret = io_init_req(ctx, req, sqe); |
6531 | if (unlikely(ret)) { | |
6532 | fail_req: | |
de59bc10 PB |
6533 | if (link->head) { |
6534 | /* fail even hard links since we don't submit */ | |
cf109604 | 6535 | link->head->flags |= REQ_F_FAIL_LINK; |
f41db273 | 6536 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6537 | link->head = NULL; |
6538 | } | |
f41db273 | 6539 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6540 | return ret; |
6541 | } | |
be7053b7 PB |
6542 | ret = io_req_prep(req, sqe); |
6543 | if (unlikely(ret)) | |
6544 | goto fail_req; | |
a6b8cadc | 6545 | |
be7053b7 | 6546 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6547 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6548 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6549 | ||
9e645e11 JA |
6550 | /* |
6551 | * If we already have a head request, queue this one for async | |
6552 | * submittal once the head completes. If we don't have a head but | |
6553 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6554 | * submitted sync once the chain is complete. If none of those | |
6555 | * conditions are true (normal request), then just queue it. | |
6556 | */ | |
863e0560 PB |
6557 | if (link->head) { |
6558 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6559 | |
8cdf2193 PB |
6560 | /* |
6561 | * Taking sequential execution of a link, draining both sides | |
6562 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6563 | * requests in the link. So, it drains the head and the | |
6564 | * next after the link request. The last one is done via | |
6565 | * drain_next flag to persist the effect across calls. | |
6566 | */ | |
ef4ff581 | 6567 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6568 | head->flags |= REQ_F_IO_DRAIN; |
6569 | ctx->drain_next = 1; | |
6570 | } | |
b7e298d2 | 6571 | ret = io_req_prep_async(req); |
cf109604 | 6572 | if (unlikely(ret)) |
a6b8cadc | 6573 | goto fail_req; |
9d76377f | 6574 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6575 | link->last->link = req; |
863e0560 | 6576 | link->last = req; |
32fe525b PB |
6577 | |
6578 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6579 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6580 | io_queue_sqe(head); |
863e0560 | 6581 | link->head = NULL; |
32fe525b | 6582 | } |
9e645e11 | 6583 | } else { |
711be031 PB |
6584 | if (unlikely(ctx->drain_next)) { |
6585 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6586 | ctx->drain_next = 0; |
711be031 | 6587 | } |
ef4ff581 | 6588 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6589 | link->head = req; |
6590 | link->last = req; | |
711be031 | 6591 | } else { |
be7053b7 | 6592 | io_queue_sqe(req); |
711be031 | 6593 | } |
9e645e11 | 6594 | } |
2e6e1fde | 6595 | |
1d4240cc | 6596 | return 0; |
9e645e11 JA |
6597 | } |
6598 | ||
9a56a232 JA |
6599 | /* |
6600 | * Batched submission is done, ensure local IO is flushed out. | |
6601 | */ | |
ba88ff11 PB |
6602 | static void io_submit_state_end(struct io_submit_state *state, |
6603 | struct io_ring_ctx *ctx) | |
9a56a232 | 6604 | { |
a1ab7b35 | 6605 | if (state->link.head) |
de59bc10 | 6606 | io_queue_sqe(state->link.head); |
6dd0be1e | 6607 | if (state->comp.nr) |
ba88ff11 | 6608 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6609 | if (state->plug_started) |
6610 | blk_finish_plug(&state->plug); | |
9f13c35b | 6611 | io_state_file_put(state); |
9a56a232 JA |
6612 | } |
6613 | ||
6614 | /* | |
6615 | * Start submission side cache. | |
6616 | */ | |
6617 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6618 | unsigned int max_ios) |
9a56a232 | 6619 | { |
27926b68 | 6620 | state->plug_started = false; |
9a56a232 | 6621 | state->ios_left = max_ios; |
a1ab7b35 PB |
6622 | /* set only head, no need to init link_last in advance */ |
6623 | state->link.head = NULL; | |
9a56a232 JA |
6624 | } |
6625 | ||
2b188cc1 JA |
6626 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6627 | { | |
75b28aff | 6628 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6629 | |
caf582c6 PB |
6630 | /* |
6631 | * Ensure any loads from the SQEs are done at this point, | |
6632 | * since once we write the new head, the application could | |
6633 | * write new data to them. | |
6634 | */ | |
6635 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6636 | } |
6637 | ||
2b188cc1 | 6638 | /* |
3529d8c2 | 6639 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6640 | * that is mapped by userspace. This means that care needs to be taken to |
6641 | * ensure that reads are stable, as we cannot rely on userspace always | |
6642 | * being a good citizen. If members of the sqe are validated and then later | |
6643 | * used, it's important that those reads are done through READ_ONCE() to | |
6644 | * prevent a re-load down the line. | |
6645 | */ | |
709b302f | 6646 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6647 | { |
75b28aff | 6648 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6649 | unsigned head; |
6650 | ||
6651 | /* | |
6652 | * The cached sq head (or cq tail) serves two purposes: | |
6653 | * | |
6654 | * 1) allows us to batch the cost of updating the user visible | |
6655 | * head updates. | |
6656 | * 2) allows the kernel side to track the head on its own, even | |
6657 | * though the application is the one updating it. | |
6658 | */ | |
4fccfcbb | 6659 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6660 | if (likely(head < ctx->sq_entries)) |
6661 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6662 | |
6663 | /* drop invalid entries */ | |
498ccd9e | 6664 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6665 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6666 | return NULL; |
6667 | } | |
6668 | ||
0f212204 | 6669 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6670 | { |
46c4e16a | 6671 | int submitted = 0; |
6c271ce2 | 6672 | |
ee7d46d9 PB |
6673 | /* make sure SQ entry isn't read before tail */ |
6674 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6675 | |
2b85edfc PB |
6676 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6677 | return -EAGAIN; | |
6c271ce2 | 6678 | |
d8a6df10 | 6679 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6680 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6681 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6682 | |
46c4e16a | 6683 | while (submitted < nr) { |
3529d8c2 | 6684 | const struct io_uring_sqe *sqe; |
196be95c | 6685 | struct io_kiocb *req; |
fb5ccc98 | 6686 | |
258b29a9 | 6687 | req = io_alloc_req(ctx); |
196be95c PB |
6688 | if (unlikely(!req)) { |
6689 | if (!submitted) | |
6690 | submitted = -EAGAIN; | |
fb5ccc98 | 6691 | break; |
196be95c | 6692 | } |
4fccfcbb PB |
6693 | sqe = io_get_sqe(ctx); |
6694 | if (unlikely(!sqe)) { | |
6695 | kmem_cache_free(req_cachep, req); | |
6696 | break; | |
6697 | } | |
d3656344 JA |
6698 | /* will complete beyond this point, count as submitted */ |
6699 | submitted++; | |
a1ab7b35 | 6700 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6701 | break; |
6c271ce2 JA |
6702 | } |
6703 | ||
9466f437 PB |
6704 | if (unlikely(submitted != nr)) { |
6705 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6706 | struct io_uring_task *tctx = current->io_uring; |
6707 | int unused = nr - ref_used; | |
9466f437 | 6708 | |
d8a6df10 JA |
6709 | percpu_ref_put_many(&ctx->refs, unused); |
6710 | percpu_counter_sub(&tctx->inflight, unused); | |
6711 | put_task_struct_many(current, unused); | |
9466f437 | 6712 | } |
6c271ce2 | 6713 | |
a1ab7b35 | 6714 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6715 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6716 | io_commit_sqring(ctx); | |
6717 | ||
6c271ce2 JA |
6718 | return submitted; |
6719 | } | |
6720 | ||
23b3628e XW |
6721 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6722 | { | |
6723 | /* Tell userspace we may need a wakeup call */ | |
6724 | spin_lock_irq(&ctx->completion_lock); | |
6725 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6726 | spin_unlock_irq(&ctx->completion_lock); | |
6727 | } | |
6728 | ||
6729 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6730 | { | |
6731 | spin_lock_irq(&ctx->completion_lock); | |
6732 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6733 | spin_unlock_irq(&ctx->completion_lock); | |
6734 | } | |
6735 | ||
08369246 | 6736 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6737 | { |
c8d1ba58 | 6738 | unsigned int to_submit; |
bdcd3eab | 6739 | int ret = 0; |
6c271ce2 | 6740 | |
c8d1ba58 | 6741 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6742 | /* if we're handling multiple rings, cap submit size for fairness */ |
6743 | if (cap_entries && to_submit > 8) | |
6744 | to_submit = 8; | |
6745 | ||
906a3c6f | 6746 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6747 | unsigned nr_events = 0; |
a4c0b3de | 6748 | |
c8d1ba58 | 6749 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6750 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6751 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6752 | |
0298ef96 PB |
6753 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6754 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6755 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6756 | mutex_unlock(&ctx->uring_lock); |
6757 | } | |
6c271ce2 | 6758 | |
90554200 JA |
6759 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6760 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6761 | |
08369246 XW |
6762 | return ret; |
6763 | } | |
6c271ce2 | 6764 | |
08369246 XW |
6765 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6766 | { | |
6767 | struct io_ring_ctx *ctx; | |
6768 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6769 | |
c9dca27d PB |
6770 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6771 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6772 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6773 | } |
6c271ce2 | 6774 | |
c8d1ba58 JA |
6775 | static int io_sq_thread(void *data) |
6776 | { | |
69fb2131 JA |
6777 | struct io_sq_data *sqd = data; |
6778 | struct io_ring_ctx *ctx; | |
a0d9205f | 6779 | unsigned long timeout = 0; |
37d1e2e3 | 6780 | char buf[TASK_COMM_LEN]; |
08369246 | 6781 | DEFINE_WAIT(wait); |
6c271ce2 | 6782 | |
696ee88a | 6783 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6784 | set_task_comm(current, buf); |
37d1e2e3 JA |
6785 | current->pf_io_worker = NULL; |
6786 | ||
6787 | if (sqd->sq_cpu != -1) | |
6788 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6789 | else | |
6790 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6791 | current->flags |= PF_NO_SETAFFINITY; | |
6792 | ||
09a6f4ef | 6793 | mutex_lock(&sqd->lock); |
05962f95 | 6794 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6795 | int ret; |
6796 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6797 | |
82734c5b JA |
6798 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || |
6799 | signal_pending(current)) { | |
6800 | bool did_sig = false; | |
6801 | ||
09a6f4ef | 6802 | mutex_unlock(&sqd->lock); |
82734c5b JA |
6803 | if (signal_pending(current)) { |
6804 | struct ksignal ksig; | |
6805 | ||
6806 | did_sig = get_signal(&ksig); | |
6807 | } | |
05962f95 | 6808 | cond_resched(); |
09a6f4ef | 6809 | mutex_lock(&sqd->lock); |
82734c5b JA |
6810 | if (did_sig) |
6811 | break; | |
521d6a73 | 6812 | io_run_task_work(); |
b7f5a0bf | 6813 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6814 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6815 | continue; |
08369246 | 6816 | } |
08369246 | 6817 | sqt_spin = false; |
e95eee2d | 6818 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6819 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6820 | const struct cred *creds = NULL; |
6821 | ||
6822 | if (ctx->sq_creds != current_cred()) | |
6823 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6824 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6825 | if (creds) |
6826 | revert_creds(creds); | |
08369246 XW |
6827 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6828 | sqt_spin = true; | |
69fb2131 | 6829 | } |
6c271ce2 | 6830 | |
08369246 | 6831 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6832 | io_run_task_work(); |
6833 | cond_resched(); | |
08369246 XW |
6834 | if (sqt_spin) |
6835 | timeout = jiffies + sqd->sq_thread_idle; | |
6836 | continue; | |
6837 | } | |
6838 | ||
08369246 XW |
6839 | needs_sched = true; |
6840 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6841 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6842 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6843 | !list_empty_careful(&ctx->iopoll_list)) { | |
6844 | needs_sched = false; | |
6845 | break; | |
6846 | } | |
6847 | if (io_sqring_entries(ctx)) { | |
6848 | needs_sched = false; | |
6849 | break; | |
6850 | } | |
6851 | } | |
6852 | ||
05962f95 | 6853 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6854 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6855 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6856 | |
09a6f4ef | 6857 | mutex_unlock(&sqd->lock); |
69fb2131 | 6858 | schedule(); |
09a6f4ef | 6859 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6860 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6861 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6862 | } |
08369246 XW |
6863 | |
6864 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6865 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6866 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6867 | } |
28cea78a | 6868 | |
521d6a73 PB |
6869 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6870 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6871 | sqd->thread = NULL; |
05962f95 | 6872 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6873 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6874 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6875 | |
6876 | io_run_task_work(); | |
b7f5a0bf | 6877 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6878 | complete(&sqd->exited); |
6879 | do_exit(0); | |
6c271ce2 JA |
6880 | } |
6881 | ||
bda52162 JA |
6882 | struct io_wait_queue { |
6883 | struct wait_queue_entry wq; | |
6884 | struct io_ring_ctx *ctx; | |
6885 | unsigned to_wait; | |
6886 | unsigned nr_timeouts; | |
6887 | }; | |
6888 | ||
6c503150 | 6889 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6890 | { |
6891 | struct io_ring_ctx *ctx = iowq->ctx; | |
6892 | ||
6893 | /* | |
d195a66e | 6894 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6895 | * started waiting. For timeouts, we always want to return to userspace, |
6896 | * regardless of event count. | |
6897 | */ | |
6c503150 | 6898 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6899 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6900 | } | |
6901 | ||
6902 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6903 | int wake_flags, void *key) | |
6904 | { | |
6905 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6906 | wq); | |
6907 | ||
6c503150 PB |
6908 | /* |
6909 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6910 | * the task, and the next invocation will do it. | |
6911 | */ | |
6912 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6913 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6914 | return -1; | |
bda52162 JA |
6915 | } |
6916 | ||
af9c1a44 JA |
6917 | static int io_run_task_work_sig(void) |
6918 | { | |
6919 | if (io_run_task_work()) | |
6920 | return 1; | |
6921 | if (!signal_pending(current)) | |
6922 | return 0; | |
0b8cfa97 | 6923 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6924 | return -ERESTARTSYS; |
af9c1a44 JA |
6925 | return -EINTR; |
6926 | } | |
6927 | ||
eeb60b9a PB |
6928 | /* when returns >0, the caller should retry */ |
6929 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6930 | struct io_wait_queue *iowq, | |
6931 | signed long *timeout) | |
6932 | { | |
6933 | int ret; | |
6934 | ||
6935 | /* make sure we run task_work before checking for signals */ | |
6936 | ret = io_run_task_work_sig(); | |
6937 | if (ret || io_should_wake(iowq)) | |
6938 | return ret; | |
6939 | /* let the caller flush overflows, retry */ | |
6940 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6941 | return 1; | |
6942 | ||
6943 | *timeout = schedule_timeout(*timeout); | |
6944 | return !*timeout ? -ETIME : 1; | |
6945 | } | |
6946 | ||
2b188cc1 JA |
6947 | /* |
6948 | * Wait until events become available, if we don't already have some. The | |
6949 | * application must reap them itself, as they reside on the shared cq ring. | |
6950 | */ | |
6951 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6952 | const sigset_t __user *sig, size_t sigsz, |
6953 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6954 | { |
bda52162 JA |
6955 | struct io_wait_queue iowq = { |
6956 | .wq = { | |
6957 | .private = current, | |
6958 | .func = io_wake_function, | |
6959 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6960 | }, | |
6961 | .ctx = ctx, | |
6962 | .to_wait = min_events, | |
6963 | }; | |
75b28aff | 6964 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6965 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6966 | int ret; | |
2b188cc1 | 6967 | |
b41e9852 | 6968 | do { |
6c2450ae | 6969 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 6970 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 6971 | return 0; |
4c6e277c | 6972 | if (!io_run_task_work()) |
b41e9852 | 6973 | break; |
b41e9852 | 6974 | } while (1); |
2b188cc1 JA |
6975 | |
6976 | if (sig) { | |
9e75ad5d AB |
6977 | #ifdef CONFIG_COMPAT |
6978 | if (in_compat_syscall()) | |
6979 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6980 | sigsz); |
9e75ad5d AB |
6981 | else |
6982 | #endif | |
b772434b | 6983 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6984 | |
2b188cc1 JA |
6985 | if (ret) |
6986 | return ret; | |
6987 | } | |
6988 | ||
c73ebb68 | 6989 | if (uts) { |
c1d5a224 PB |
6990 | struct timespec64 ts; |
6991 | ||
c73ebb68 HX |
6992 | if (get_timespec64(&ts, uts)) |
6993 | return -EFAULT; | |
6994 | timeout = timespec64_to_jiffies(&ts); | |
6995 | } | |
6996 | ||
bda52162 | 6997 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6998 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6999 | do { |
ca0a2651 | 7000 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7001 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7002 | ret = -EBUSY; |
7003 | break; | |
7004 | } | |
bda52162 JA |
7005 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7006 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7007 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7008 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 7009 | cond_resched(); |
eeb60b9a | 7010 | } while (ret > 0); |
bda52162 | 7011 | |
b7db41c9 | 7012 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7013 | |
75b28aff | 7014 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7015 | } |
7016 | ||
aeca241b | 7017 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) |
846a4ef2 PB |
7018 | { |
7019 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); | |
7020 | ||
7021 | for (i = 0; i < nr_tables; i++) | |
aeca241b PB |
7022 | kfree(table->files[i]); |
7023 | kfree(table->files); | |
7024 | table->files = NULL; | |
846a4ef2 PB |
7025 | } |
7026 | ||
6b06314c JA |
7027 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7028 | { | |
7029 | #if defined(CONFIG_UNIX) | |
7030 | if (ctx->ring_sock) { | |
7031 | struct sock *sock = ctx->ring_sock->sk; | |
7032 | struct sk_buff *skb; | |
7033 | ||
7034 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7035 | kfree_skb(skb); | |
7036 | } | |
7037 | #else | |
7038 | int i; | |
7039 | ||
65e19f54 JA |
7040 | for (i = 0; i < ctx->nr_user_files; i++) { |
7041 | struct file *file; | |
7042 | ||
7043 | file = io_file_from_index(ctx, i); | |
7044 | if (file) | |
7045 | fput(file); | |
7046 | } | |
6b06314c | 7047 | #endif |
08480400 PB |
7048 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); |
7049 | kfree(ctx->file_data); | |
7050 | ctx->file_data = NULL; | |
7051 | ctx->nr_user_files = 0; | |
6b06314c JA |
7052 | } |
7053 | ||
2a63b2d9 | 7054 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7055 | { |
2a63b2d9 | 7056 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7057 | } |
7058 | ||
2a63b2d9 | 7059 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7060 | { |
2a63b2d9 BM |
7061 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7062 | } | |
65e19f54 | 7063 | |
28a9fe25 PB |
7064 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
7065 | { | |
7066 | percpu_ref_exit(&ref_node->refs); | |
7067 | kfree(ref_node); | |
7068 | } | |
7069 | ||
a7f0ed5a PB |
7070 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7071 | struct io_rsrc_data *data_to_kill) | |
1642b445 | 7072 | { |
a7f0ed5a PB |
7073 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7074 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
82fbcfa9 | 7075 | |
a7f0ed5a PB |
7076 | if (data_to_kill) { |
7077 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7078 | |
a7f0ed5a PB |
7079 | rsrc_node->rsrc_data = data_to_kill; |
7080 | io_rsrc_ref_lock(ctx); | |
7081 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7082 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7083 | |
3e942498 | 7084 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7085 | percpu_ref_kill(&rsrc_node->refs); |
7086 | ctx->rsrc_node = NULL; | |
7087 | } | |
6b06314c | 7088 | |
a7f0ed5a PB |
7089 | if (!ctx->rsrc_node) { |
7090 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7091 | ctx->rsrc_backup_node = NULL; | |
7092 | } | |
8bad28d8 HX |
7093 | } |
7094 | ||
a7f0ed5a | 7095 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7096 | { |
7097 | if (ctx->rsrc_backup_node) | |
7098 | return 0; | |
b895c9a6 | 7099 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe PB |
7100 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
7101 | } | |
7102 | ||
40ae0ff7 | 7103 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7104 | { |
7105 | int ret; | |
05589553 | 7106 | |
215c3902 | 7107 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7108 | if (data->quiesce) |
7109 | return -ENXIO; | |
05589553 | 7110 | |
8bad28d8 | 7111 | data->quiesce = true; |
1ffc5422 | 7112 | do { |
a7f0ed5a | 7113 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7114 | if (ret) |
f2303b1f | 7115 | break; |
a7f0ed5a PB |
7116 | io_rsrc_node_switch(ctx, data); |
7117 | ||
3e942498 PB |
7118 | /* kill initial ref, already quiesced if zero */ |
7119 | if (atomic_dec_and_test(&data->refs)) | |
7120 | break; | |
8bad28d8 | 7121 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7122 | ret = wait_for_completion_interruptible(&data->done); |
7123 | if (!ret) | |
7124 | break; | |
8bad28d8 | 7125 | |
3e942498 PB |
7126 | atomic_inc(&data->refs); |
7127 | /* wait for all works potentially completing data->done */ | |
7128 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7129 | reinit_completion(&data->done); |
8dd03afe | 7130 | |
8bad28d8 | 7131 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7132 | ret = io_run_task_work_sig(); |
8bad28d8 | 7133 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7134 | } while (ret >= 0); |
8bad28d8 | 7135 | data->quiesce = false; |
05f3fb3c | 7136 | |
8bad28d8 | 7137 | return ret; |
d7954b2b BM |
7138 | } |
7139 | ||
40ae0ff7 PB |
7140 | static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx, |
7141 | rsrc_put_fn *do_put) | |
1ad555c6 | 7142 | { |
b895c9a6 | 7143 | struct io_rsrc_data *data; |
1ad555c6 BM |
7144 | |
7145 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7146 | if (!data) | |
7147 | return NULL; | |
7148 | ||
3e942498 | 7149 | atomic_set(&data->refs, 1); |
1ad555c6 | 7150 | data->ctx = ctx; |
40ae0ff7 | 7151 | data->do_put = do_put; |
1ad555c6 BM |
7152 | init_completion(&data->done); |
7153 | return data; | |
7154 | } | |
7155 | ||
d7954b2b BM |
7156 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7157 | { | |
d7954b2b BM |
7158 | int ret; |
7159 | ||
08480400 | 7160 | if (!ctx->file_data) |
d7954b2b | 7161 | return -ENXIO; |
08480400 PB |
7162 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7163 | if (!ret) | |
7164 | __io_sqe_files_unregister(ctx); | |
7165 | return ret; | |
6b06314c JA |
7166 | } |
7167 | ||
37d1e2e3 | 7168 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7169 | __releases(&sqd->lock) |
37d1e2e3 | 7170 | { |
521d6a73 PB |
7171 | WARN_ON_ONCE(sqd->thread == current); |
7172 | ||
9e138a48 PB |
7173 | /* |
7174 | * Do the dance but not conditional clear_bit() because it'd race with | |
7175 | * other threads incrementing park_pending and setting the bit. | |
7176 | */ | |
37d1e2e3 | 7177 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7178 | if (atomic_dec_return(&sqd->park_pending)) |
7179 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7180 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7181 | } |
7182 | ||
86e0d676 | 7183 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7184 | __acquires(&sqd->lock) |
37d1e2e3 | 7185 | { |
521d6a73 PB |
7186 | WARN_ON_ONCE(sqd->thread == current); |
7187 | ||
9e138a48 | 7188 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7189 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7190 | mutex_lock(&sqd->lock); |
05962f95 | 7191 | if (sqd->thread) |
86e0d676 | 7192 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7193 | } |
7194 | ||
7195 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7196 | { | |
521d6a73 | 7197 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7198 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7199 | |
05962f95 | 7200 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7201 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7202 | if (sqd->thread) |
7203 | wake_up_process(sqd->thread); | |
09a6f4ef | 7204 | mutex_unlock(&sqd->lock); |
05962f95 | 7205 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7206 | } |
7207 | ||
534ca6d6 | 7208 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7209 | { |
534ca6d6 | 7210 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7211 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7212 | ||
37d1e2e3 JA |
7213 | io_sq_thread_stop(sqd); |
7214 | kfree(sqd); | |
7215 | } | |
7216 | } | |
7217 | ||
7218 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7219 | { | |
7220 | struct io_sq_data *sqd = ctx->sq_data; | |
7221 | ||
7222 | if (sqd) { | |
05962f95 | 7223 | io_sq_thread_park(sqd); |
521d6a73 | 7224 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7225 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7226 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7227 | |
7228 | io_put_sq_data(sqd); | |
7229 | ctx->sq_data = NULL; | |
7c30f36a SM |
7230 | if (ctx->sq_creds) |
7231 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7232 | } |
7233 | } | |
7234 | ||
aa06165d JA |
7235 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7236 | { | |
7237 | struct io_ring_ctx *ctx_attach; | |
7238 | struct io_sq_data *sqd; | |
7239 | struct fd f; | |
7240 | ||
7241 | f = fdget(p->wq_fd); | |
7242 | if (!f.file) | |
7243 | return ERR_PTR(-ENXIO); | |
7244 | if (f.file->f_op != &io_uring_fops) { | |
7245 | fdput(f); | |
7246 | return ERR_PTR(-EINVAL); | |
7247 | } | |
7248 | ||
7249 | ctx_attach = f.file->private_data; | |
7250 | sqd = ctx_attach->sq_data; | |
7251 | if (!sqd) { | |
7252 | fdput(f); | |
7253 | return ERR_PTR(-EINVAL); | |
7254 | } | |
5c2469e0 JA |
7255 | if (sqd->task_tgid != current->tgid) { |
7256 | fdput(f); | |
7257 | return ERR_PTR(-EPERM); | |
7258 | } | |
aa06165d JA |
7259 | |
7260 | refcount_inc(&sqd->refs); | |
7261 | fdput(f); | |
7262 | return sqd; | |
7263 | } | |
7264 | ||
26984fbf PB |
7265 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7266 | bool *attached) | |
534ca6d6 JA |
7267 | { |
7268 | struct io_sq_data *sqd; | |
7269 | ||
26984fbf | 7270 | *attached = false; |
5c2469e0 JA |
7271 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7272 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7273 | if (!IS_ERR(sqd)) { |
7274 | *attached = true; | |
5c2469e0 | 7275 | return sqd; |
26984fbf | 7276 | } |
5c2469e0 JA |
7277 | /* fall through for EPERM case, setup new sqd/task */ |
7278 | if (PTR_ERR(sqd) != -EPERM) | |
7279 | return sqd; | |
7280 | } | |
aa06165d | 7281 | |
534ca6d6 JA |
7282 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7283 | if (!sqd) | |
7284 | return ERR_PTR(-ENOMEM); | |
7285 | ||
9e138a48 | 7286 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7287 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7288 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7289 | mutex_init(&sqd->lock); |
534ca6d6 | 7290 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7291 | init_completion(&sqd->exited); |
534ca6d6 JA |
7292 | return sqd; |
7293 | } | |
7294 | ||
6b06314c | 7295 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7296 | /* |
7297 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7298 | * the io_uring can be safely unregistered on process exit, even if we have | |
7299 | * loops in the file referencing. | |
7300 | */ | |
7301 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7302 | { | |
7303 | struct sock *sk = ctx->ring_sock->sk; | |
7304 | struct scm_fp_list *fpl; | |
7305 | struct sk_buff *skb; | |
08a45173 | 7306 | int i, nr_files; |
6b06314c | 7307 | |
6b06314c JA |
7308 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7309 | if (!fpl) | |
7310 | return -ENOMEM; | |
7311 | ||
7312 | skb = alloc_skb(0, GFP_KERNEL); | |
7313 | if (!skb) { | |
7314 | kfree(fpl); | |
7315 | return -ENOMEM; | |
7316 | } | |
7317 | ||
7318 | skb->sk = sk; | |
6b06314c | 7319 | |
08a45173 | 7320 | nr_files = 0; |
62e398be | 7321 | fpl->user = get_uid(current_user()); |
6b06314c | 7322 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7323 | struct file *file = io_file_from_index(ctx, i + offset); |
7324 | ||
7325 | if (!file) | |
08a45173 | 7326 | continue; |
65e19f54 | 7327 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7328 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7329 | nr_files++; | |
6b06314c JA |
7330 | } |
7331 | ||
08a45173 JA |
7332 | if (nr_files) { |
7333 | fpl->max = SCM_MAX_FD; | |
7334 | fpl->count = nr_files; | |
7335 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7336 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7337 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7338 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7339 | |
08a45173 JA |
7340 | for (i = 0; i < nr_files; i++) |
7341 | fput(fpl->fp[i]); | |
7342 | } else { | |
7343 | kfree_skb(skb); | |
7344 | kfree(fpl); | |
7345 | } | |
6b06314c JA |
7346 | |
7347 | return 0; | |
7348 | } | |
7349 | ||
7350 | /* | |
7351 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7352 | * causes regular reference counting to break down. We rely on the UNIX | |
7353 | * garbage collection to take care of this problem for us. | |
7354 | */ | |
7355 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7356 | { | |
7357 | unsigned left, total; | |
7358 | int ret = 0; | |
7359 | ||
7360 | total = 0; | |
7361 | left = ctx->nr_user_files; | |
7362 | while (left) { | |
7363 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7364 | |
7365 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7366 | if (ret) | |
7367 | break; | |
7368 | left -= this_files; | |
7369 | total += this_files; | |
7370 | } | |
7371 | ||
7372 | if (!ret) | |
7373 | return 0; | |
7374 | ||
7375 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7376 | struct file *file = io_file_from_index(ctx, total); |
7377 | ||
7378 | if (file) | |
7379 | fput(file); | |
6b06314c JA |
7380 | total++; |
7381 | } | |
7382 | ||
7383 | return ret; | |
7384 | } | |
7385 | #else | |
7386 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7387 | { | |
7388 | return 0; | |
7389 | } | |
7390 | #endif | |
7391 | ||
aeca241b | 7392 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
65e19f54 | 7393 | { |
846a4ef2 PB |
7394 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); |
7395 | ||
aeca241b PB |
7396 | table->files = kcalloc(nr_tables, sizeof(*table->files), GFP_KERNEL); |
7397 | if (!table->files) | |
846a4ef2 | 7398 | return false; |
65e19f54 JA |
7399 | |
7400 | for (i = 0; i < nr_tables; i++) { | |
846a4ef2 | 7401 | unsigned int this_files = min(nr_files, IORING_MAX_FILES_TABLE); |
65e19f54 | 7402 | |
aeca241b | 7403 | table->files[i] = kcalloc(this_files, sizeof(*table->files[i]), |
65e19f54 | 7404 | GFP_KERNEL); |
aeca241b | 7405 | if (!table->files[i]) |
65e19f54 JA |
7406 | break; |
7407 | nr_files -= this_files; | |
7408 | } | |
7409 | ||
7410 | if (i == nr_tables) | |
846a4ef2 | 7411 | return true; |
65e19f54 | 7412 | |
aeca241b | 7413 | io_free_file_tables(table, nr_tables * IORING_MAX_FILES_TABLE); |
846a4ef2 | 7414 | return false; |
65e19f54 JA |
7415 | } |
7416 | ||
47e90392 | 7417 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7418 | { |
50238531 | 7419 | struct file *file = prsrc->file; |
05f3fb3c JA |
7420 | #if defined(CONFIG_UNIX) |
7421 | struct sock *sock = ctx->ring_sock->sk; | |
7422 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7423 | struct sk_buff *skb; | |
7424 | int i; | |
7425 | ||
7426 | __skb_queue_head_init(&list); | |
7427 | ||
7428 | /* | |
7429 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7430 | * remove this entry and rearrange the file array. | |
7431 | */ | |
7432 | skb = skb_dequeue(head); | |
7433 | while (skb) { | |
7434 | struct scm_fp_list *fp; | |
7435 | ||
7436 | fp = UNIXCB(skb).fp; | |
7437 | for (i = 0; i < fp->count; i++) { | |
7438 | int left; | |
7439 | ||
7440 | if (fp->fp[i] != file) | |
7441 | continue; | |
7442 | ||
7443 | unix_notinflight(fp->user, fp->fp[i]); | |
7444 | left = fp->count - 1 - i; | |
7445 | if (left) { | |
7446 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7447 | left * sizeof(struct file *)); | |
7448 | } | |
7449 | fp->count--; | |
7450 | if (!fp->count) { | |
7451 | kfree_skb(skb); | |
7452 | skb = NULL; | |
7453 | } else { | |
7454 | __skb_queue_tail(&list, skb); | |
7455 | } | |
7456 | fput(file); | |
7457 | file = NULL; | |
7458 | break; | |
7459 | } | |
7460 | ||
7461 | if (!file) | |
7462 | break; | |
7463 | ||
7464 | __skb_queue_tail(&list, skb); | |
7465 | ||
7466 | skb = skb_dequeue(head); | |
7467 | } | |
7468 | ||
7469 | if (skb_peek(&list)) { | |
7470 | spin_lock_irq(&head->lock); | |
7471 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7472 | __skb_queue_tail(head, skb); | |
7473 | spin_unlock_irq(&head->lock); | |
7474 | } | |
7475 | #else | |
7476 | fput(file); | |
7477 | #endif | |
7478 | } | |
7479 | ||
b895c9a6 | 7480 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7481 | { |
b895c9a6 | 7482 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7483 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7484 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7485 | |
269bbe5f BM |
7486 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7487 | list_del(&prsrc->list); | |
40ae0ff7 | 7488 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7489 | kfree(prsrc); |
65e19f54 | 7490 | } |
05589553 | 7491 | |
28a9fe25 | 7492 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7493 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7494 | complete(&rsrc_data->done); | |
2faf852d | 7495 | } |
65e19f54 | 7496 | |
269bbe5f | 7497 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7498 | { |
7499 | struct io_ring_ctx *ctx; | |
7500 | struct llist_node *node; | |
7501 | ||
269bbe5f BM |
7502 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7503 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7504 | |
7505 | while (node) { | |
b895c9a6 | 7506 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7507 | struct llist_node *next = node->next; |
7508 | ||
b895c9a6 | 7509 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7510 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7511 | node = next; |
7512 | } | |
7513 | } | |
7514 | ||
00835dce | 7515 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7516 | { |
b895c9a6 | 7517 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7518 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7519 | bool first_add = false; |
65e19f54 | 7520 | |
2a63b2d9 | 7521 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7522 | node->done = true; |
e297822b | 7523 | |
d67d2263 | 7524 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7525 | node = list_first_entry(&ctx->rsrc_ref_list, |
7526 | struct io_rsrc_node, node); | |
e297822b | 7527 | /* recycle ref nodes in order */ |
b895c9a6 | 7528 | if (!node->done) |
e297822b | 7529 | break; |
b895c9a6 PB |
7530 | list_del(&node->node); |
7531 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7532 | } |
2a63b2d9 | 7533 | io_rsrc_ref_unlock(ctx); |
05589553 | 7534 | |
3e942498 PB |
7535 | if (first_add) |
7536 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7537 | } |
65e19f54 | 7538 | |
b895c9a6 | 7539 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7540 | { |
b895c9a6 | 7541 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7542 | |
05589553 XW |
7543 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7544 | if (!ref_node) | |
3e2224c5 | 7545 | return NULL; |
05f3fb3c | 7546 | |
00835dce | 7547 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7548 | 0, GFP_KERNEL)) { |
7549 | kfree(ref_node); | |
3e2224c5 | 7550 | return NULL; |
05589553 XW |
7551 | } |
7552 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7553 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7554 | ref_node->done = false; |
05589553 | 7555 | return ref_node; |
05589553 XW |
7556 | } |
7557 | ||
6b06314c JA |
7558 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7559 | unsigned nr_args) | |
7560 | { | |
7561 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7562 | struct file *file; |
f3baed39 | 7563 | int fd, ret; |
846a4ef2 | 7564 | unsigned i; |
b895c9a6 | 7565 | struct io_rsrc_data *file_data; |
6b06314c | 7566 | |
05f3fb3c | 7567 | if (ctx->file_data) |
6b06314c JA |
7568 | return -EBUSY; |
7569 | if (!nr_args) | |
7570 | return -EINVAL; | |
7571 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7572 | return -EMFILE; | |
a7f0ed5a | 7573 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7574 | if (ret) |
7575 | return ret; | |
6b06314c | 7576 | |
47e90392 | 7577 | file_data = io_rsrc_data_alloc(ctx, io_rsrc_file_put); |
5398ae69 | 7578 | if (!file_data) |
05f3fb3c | 7579 | return -ENOMEM; |
13770a71 | 7580 | ctx->file_data = file_data; |
f3baed39 | 7581 | ret = -ENOMEM; |
aeca241b | 7582 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7583 | goto out_free; |
65e19f54 | 7584 | |
08a45173 | 7585 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7586 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7587 | ret = -EFAULT; | |
7588 | goto out_fput; | |
7589 | } | |
08a45173 | 7590 | /* allow sparse sets */ |
600cf3f8 | 7591 | if (fd == -1) |
08a45173 | 7592 | continue; |
6b06314c | 7593 | |
05f3fb3c | 7594 | file = fget(fd); |
6b06314c | 7595 | ret = -EBADF; |
05f3fb3c | 7596 | if (!file) |
600cf3f8 | 7597 | goto out_fput; |
05f3fb3c | 7598 | |
6b06314c JA |
7599 | /* |
7600 | * Don't allow io_uring instances to be registered. If UNIX | |
7601 | * isn't enabled, then this causes a reference cycle and this | |
7602 | * instance can never get freed. If UNIX is enabled we'll | |
7603 | * handle it just fine, but there's still no point in allowing | |
7604 | * a ring fd as it doesn't support regular read/write anyway. | |
7605 | */ | |
05f3fb3c JA |
7606 | if (file->f_op == &io_uring_fops) { |
7607 | fput(file); | |
600cf3f8 | 7608 | goto out_fput; |
6b06314c | 7609 | } |
aeca241b | 7610 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7611 | } |
7612 | ||
6b06314c | 7613 | ret = io_sqe_files_scm(ctx); |
05589553 | 7614 | if (ret) { |
08480400 | 7615 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7616 | return ret; |
7617 | } | |
6b06314c | 7618 | |
a7f0ed5a | 7619 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7620 | return ret; |
600cf3f8 PB |
7621 | out_fput: |
7622 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7623 | file = io_file_from_index(ctx, i); | |
7624 | if (file) | |
7625 | fput(file); | |
7626 | } | |
aeca241b | 7627 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7628 | ctx->nr_user_files = 0; |
600cf3f8 | 7629 | out_free: |
3e942498 | 7630 | kfree(ctx->file_data); |
55cbc256 | 7631 | ctx->file_data = NULL; |
6b06314c JA |
7632 | return ret; |
7633 | } | |
7634 | ||
c3a31e60 JA |
7635 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7636 | int index) | |
7637 | { | |
7638 | #if defined(CONFIG_UNIX) | |
7639 | struct sock *sock = ctx->ring_sock->sk; | |
7640 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7641 | struct sk_buff *skb; | |
7642 | ||
7643 | /* | |
7644 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7645 | * file set. If there's no room, fall back to allocating a new skb | |
7646 | * and filling it in. | |
7647 | */ | |
7648 | spin_lock_irq(&head->lock); | |
7649 | skb = skb_peek(head); | |
7650 | if (skb) { | |
7651 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7652 | ||
7653 | if (fpl->count < SCM_MAX_FD) { | |
7654 | __skb_unlink(skb, head); | |
7655 | spin_unlock_irq(&head->lock); | |
7656 | fpl->fp[fpl->count] = get_file(file); | |
7657 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7658 | fpl->count++; | |
7659 | spin_lock_irq(&head->lock); | |
7660 | __skb_queue_head(head, skb); | |
7661 | } else { | |
7662 | skb = NULL; | |
7663 | } | |
7664 | } | |
7665 | spin_unlock_irq(&head->lock); | |
7666 | ||
7667 | if (skb) { | |
7668 | fput(file); | |
7669 | return 0; | |
7670 | } | |
7671 | ||
7672 | return __io_sqe_files_scm(ctx, 1, index); | |
7673 | #else | |
7674 | return 0; | |
7675 | #endif | |
7676 | } | |
7677 | ||
e7c78371 PB |
7678 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, |
7679 | struct io_rsrc_node *node, void *rsrc) | |
05f3fb3c | 7680 | { |
269bbe5f | 7681 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7682 | |
269bbe5f BM |
7683 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7684 | if (!prsrc) | |
a5318d3c | 7685 | return -ENOMEM; |
05f3fb3c | 7686 | |
50238531 | 7687 | prsrc->rsrc = rsrc; |
e7c78371 | 7688 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7689 | return 0; |
05f3fb3c JA |
7690 | } |
7691 | ||
7692 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
269bbe5f | 7693 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7694 | unsigned nr_args) |
7695 | { | |
b895c9a6 | 7696 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7697 | struct io_fixed_file *file_slot; |
7698 | struct file *file; | |
c3a31e60 JA |
7699 | __s32 __user *fds; |
7700 | int fd, i, err; | |
7701 | __u32 done; | |
05589553 | 7702 | bool needs_switch = false; |
c3a31e60 | 7703 | |
05f3fb3c | 7704 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7705 | return -EOVERFLOW; |
7706 | if (done > ctx->nr_user_files) | |
7707 | return -EINVAL; | |
a7f0ed5a | 7708 | err = io_rsrc_node_switch_start(ctx); |
8dd03afe PB |
7709 | if (err) |
7710 | return err; | |
05589553 | 7711 | |
269bbe5f | 7712 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7713 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7714 | err = 0; |
7715 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7716 | err = -EFAULT; | |
7717 | break; | |
7718 | } | |
4e0377a1 | 7719 | if (fd == IORING_REGISTER_FILES_SKIP) |
7720 | continue; | |
7721 | ||
67973b93 | 7722 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7723 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7724 | |
a04b0ac0 PB |
7725 | if (file_slot->file_ptr) { |
7726 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
a7f0ed5a | 7727 | err = io_queue_rsrc_removal(data, ctx->rsrc_node, file); |
a5318d3c HD |
7728 | if (err) |
7729 | break; | |
a04b0ac0 | 7730 | file_slot->file_ptr = 0; |
05589553 | 7731 | needs_switch = true; |
c3a31e60 JA |
7732 | } |
7733 | if (fd != -1) { | |
c3a31e60 JA |
7734 | file = fget(fd); |
7735 | if (!file) { | |
7736 | err = -EBADF; | |
7737 | break; | |
7738 | } | |
7739 | /* | |
7740 | * Don't allow io_uring instances to be registered. If | |
7741 | * UNIX isn't enabled, then this causes a reference | |
7742 | * cycle and this instance can never get freed. If UNIX | |
7743 | * is enabled we'll handle it just fine, but there's | |
7744 | * still no point in allowing a ring fd as it doesn't | |
7745 | * support regular read/write anyway. | |
7746 | */ | |
7747 | if (file->f_op == &io_uring_fops) { | |
7748 | fput(file); | |
7749 | err = -EBADF; | |
7750 | break; | |
7751 | } | |
9a321c98 | 7752 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7753 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7754 | if (err) { |
a04b0ac0 | 7755 | file_slot->file_ptr = 0; |
f3bd9dae | 7756 | fput(file); |
c3a31e60 | 7757 | break; |
f3bd9dae | 7758 | } |
c3a31e60 | 7759 | } |
05f3fb3c JA |
7760 | } |
7761 | ||
a7f0ed5a PB |
7762 | if (needs_switch) |
7763 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7764 | return done ? done : err; |
7765 | } | |
05589553 | 7766 | |
05f3fb3c JA |
7767 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7768 | unsigned nr_args) | |
7769 | { | |
269bbe5f | 7770 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7771 | |
7772 | if (!ctx->file_data) | |
7773 | return -ENXIO; | |
7774 | if (!nr_args) | |
7775 | return -EINVAL; | |
7776 | if (copy_from_user(&up, arg, sizeof(up))) | |
7777 | return -EFAULT; | |
7778 | if (up.resv) | |
7779 | return -EINVAL; | |
7780 | ||
7781 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7782 | } | |
c3a31e60 | 7783 | |
5280f7e5 | 7784 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7785 | { |
7786 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7787 | ||
5280f7e5 PB |
7788 | req = io_put_req_find_next(req); |
7789 | return req ? &req->work : NULL; | |
7d723065 JA |
7790 | } |
7791 | ||
685fe7fe JA |
7792 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7793 | struct task_struct *task) | |
24369c2e | 7794 | { |
e941894e | 7795 | struct io_wq_hash *hash; |
24369c2e | 7796 | struct io_wq_data data; |
24369c2e | 7797 | unsigned int concurrency; |
24369c2e | 7798 | |
e941894e JA |
7799 | hash = ctx->hash_map; |
7800 | if (!hash) { | |
7801 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7802 | if (!hash) | |
7803 | return ERR_PTR(-ENOMEM); | |
7804 | refcount_set(&hash->refs, 1); | |
7805 | init_waitqueue_head(&hash->wait); | |
7806 | ctx->hash_map = hash; | |
24369c2e PB |
7807 | } |
7808 | ||
e941894e | 7809 | data.hash = hash; |
685fe7fe | 7810 | data.task = task; |
e9fd9396 | 7811 | data.free_work = io_free_work; |
f5fa38c5 | 7812 | data.do_work = io_wq_submit_work; |
24369c2e | 7813 | |
d25e3a3d JA |
7814 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7815 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7816 | |
5aa75ed5 | 7817 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7818 | } |
7819 | ||
5aa75ed5 JA |
7820 | static int io_uring_alloc_task_context(struct task_struct *task, |
7821 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7822 | { |
7823 | struct io_uring_task *tctx; | |
d8a6df10 | 7824 | int ret; |
0f212204 JA |
7825 | |
7826 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7827 | if (unlikely(!tctx)) | |
7828 | return -ENOMEM; | |
7829 | ||
d8a6df10 JA |
7830 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7831 | if (unlikely(ret)) { | |
7832 | kfree(tctx); | |
7833 | return ret; | |
7834 | } | |
7835 | ||
685fe7fe | 7836 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7837 | if (IS_ERR(tctx->io_wq)) { |
7838 | ret = PTR_ERR(tctx->io_wq); | |
7839 | percpu_counter_destroy(&tctx->inflight); | |
7840 | kfree(tctx); | |
7841 | return ret; | |
7842 | } | |
7843 | ||
0f212204 JA |
7844 | xa_init(&tctx->xa); |
7845 | init_waitqueue_head(&tctx->wait); | |
7846 | tctx->last = NULL; | |
fdaf083c | 7847 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7848 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7849 | task->io_uring = tctx; |
7cbf1722 JA |
7850 | spin_lock_init(&tctx->task_lock); |
7851 | INIT_WQ_LIST(&tctx->task_list); | |
7852 | tctx->task_state = 0; | |
7853 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7854 | return 0; |
7855 | } | |
7856 | ||
7857 | void __io_uring_free(struct task_struct *tsk) | |
7858 | { | |
7859 | struct io_uring_task *tctx = tsk->io_uring; | |
7860 | ||
7861 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7862 | WARN_ON_ONCE(tctx->io_wq); |
7863 | ||
d8a6df10 | 7864 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7865 | kfree(tctx); |
7866 | tsk->io_uring = NULL; | |
7867 | } | |
7868 | ||
7e84e1c7 SG |
7869 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7870 | struct io_uring_params *p) | |
2b188cc1 JA |
7871 | { |
7872 | int ret; | |
7873 | ||
d25e3a3d JA |
7874 | /* Retain compatibility with failing for an invalid attach attempt */ |
7875 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7876 | IORING_SETUP_ATTACH_WQ) { | |
7877 | struct fd f; | |
7878 | ||
7879 | f = fdget(p->wq_fd); | |
7880 | if (!f.file) | |
7881 | return -ENXIO; | |
7882 | if (f.file->f_op != &io_uring_fops) { | |
7883 | fdput(f); | |
7884 | return -EINVAL; | |
7885 | } | |
7886 | fdput(f); | |
7887 | } | |
6c271ce2 | 7888 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7889 | struct task_struct *tsk; |
534ca6d6 | 7890 | struct io_sq_data *sqd; |
26984fbf | 7891 | bool attached; |
534ca6d6 | 7892 | |
26984fbf | 7893 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7894 | if (IS_ERR(sqd)) { |
7895 | ret = PTR_ERR(sqd); | |
7896 | goto err; | |
7897 | } | |
69fb2131 | 7898 | |
7c30f36a | 7899 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7900 | ctx->sq_data = sqd; |
917257da JA |
7901 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7902 | if (!ctx->sq_thread_idle) | |
7903 | ctx->sq_thread_idle = HZ; | |
7904 | ||
26984fbf | 7905 | ret = 0; |
78d7f6ba | 7906 | io_sq_thread_park(sqd); |
de75a3d3 PB |
7907 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
7908 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 7909 | /* don't attach to a dying SQPOLL thread, would be racy */ |
de75a3d3 | 7910 | if (attached && !sqd->thread) |
26984fbf | 7911 | ret = -ENXIO; |
78d7f6ba PB |
7912 | io_sq_thread_unpark(sqd); |
7913 | ||
de75a3d3 PB |
7914 | if (ret < 0) |
7915 | goto err; | |
7916 | if (attached) | |
5aa75ed5 | 7917 | return 0; |
aa06165d | 7918 | |
6c271ce2 | 7919 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7920 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7921 | |
917257da | 7922 | ret = -EINVAL; |
44a9bd18 | 7923 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7924 | goto err_sqpoll; |
7889f44d | 7925 | if (!cpu_online(cpu)) |
e8f98f24 | 7926 | goto err_sqpoll; |
917257da | 7927 | |
37d1e2e3 | 7928 | sqd->sq_cpu = cpu; |
6c271ce2 | 7929 | } else { |
37d1e2e3 | 7930 | sqd->sq_cpu = -1; |
6c271ce2 | 7931 | } |
37d1e2e3 JA |
7932 | |
7933 | sqd->task_pid = current->pid; | |
5c2469e0 | 7934 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7935 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7936 | if (IS_ERR(tsk)) { | |
7937 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7938 | goto err_sqpoll; |
6c271ce2 | 7939 | } |
97a73a0f | 7940 | |
46fe18b1 | 7941 | sqd->thread = tsk; |
97a73a0f | 7942 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7943 | wake_up_new_task(tsk); |
0f212204 JA |
7944 | if (ret) |
7945 | goto err; | |
6c271ce2 JA |
7946 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7947 | /* Can't have SQ_AFF without SQPOLL */ | |
7948 | ret = -EINVAL; | |
7949 | goto err; | |
7950 | } | |
7951 | ||
2b188cc1 JA |
7952 | return 0; |
7953 | err: | |
37d1e2e3 | 7954 | io_sq_thread_finish(ctx); |
2b188cc1 | 7955 | return ret; |
e8f98f24 JA |
7956 | err_sqpoll: |
7957 | complete(&ctx->sq_data->exited); | |
7958 | goto err; | |
2b188cc1 JA |
7959 | } |
7960 | ||
a087e2b5 BM |
7961 | static inline void __io_unaccount_mem(struct user_struct *user, |
7962 | unsigned long nr_pages) | |
2b188cc1 JA |
7963 | { |
7964 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7965 | } | |
7966 | ||
a087e2b5 BM |
7967 | static inline int __io_account_mem(struct user_struct *user, |
7968 | unsigned long nr_pages) | |
2b188cc1 JA |
7969 | { |
7970 | unsigned long page_limit, cur_pages, new_pages; | |
7971 | ||
7972 | /* Don't allow more pages than we can safely lock */ | |
7973 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7974 | ||
7975 | do { | |
7976 | cur_pages = atomic_long_read(&user->locked_vm); | |
7977 | new_pages = cur_pages + nr_pages; | |
7978 | if (new_pages > page_limit) | |
7979 | return -ENOMEM; | |
7980 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7981 | new_pages) != cur_pages); | |
7982 | ||
7983 | return 0; | |
7984 | } | |
7985 | ||
26bfa89e | 7986 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7987 | { |
62e398be | 7988 | if (ctx->user) |
a087e2b5 | 7989 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7990 | |
26bfa89e JA |
7991 | if (ctx->mm_account) |
7992 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7993 | } |
7994 | ||
26bfa89e | 7995 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7996 | { |
30975825 BM |
7997 | int ret; |
7998 | ||
62e398be | 7999 | if (ctx->user) { |
30975825 BM |
8000 | ret = __io_account_mem(ctx->user, nr_pages); |
8001 | if (ret) | |
8002 | return ret; | |
8003 | } | |
8004 | ||
26bfa89e JA |
8005 | if (ctx->mm_account) |
8006 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8007 | |
8008 | return 0; | |
8009 | } | |
8010 | ||
2b188cc1 JA |
8011 | static void io_mem_free(void *ptr) |
8012 | { | |
52e04ef4 MR |
8013 | struct page *page; |
8014 | ||
8015 | if (!ptr) | |
8016 | return; | |
2b188cc1 | 8017 | |
52e04ef4 | 8018 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8019 | if (put_page_testzero(page)) |
8020 | free_compound_page(page); | |
8021 | } | |
8022 | ||
8023 | static void *io_mem_alloc(size_t size) | |
8024 | { | |
8025 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8026 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8027 | |
8028 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8029 | } | |
8030 | ||
75b28aff HV |
8031 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8032 | size_t *sq_offset) | |
8033 | { | |
8034 | struct io_rings *rings; | |
8035 | size_t off, sq_array_size; | |
8036 | ||
8037 | off = struct_size(rings, cqes, cq_entries); | |
8038 | if (off == SIZE_MAX) | |
8039 | return SIZE_MAX; | |
8040 | ||
8041 | #ifdef CONFIG_SMP | |
8042 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8043 | if (off == 0) | |
8044 | return SIZE_MAX; | |
8045 | #endif | |
8046 | ||
b36200f5 DV |
8047 | if (sq_offset) |
8048 | *sq_offset = off; | |
8049 | ||
75b28aff HV |
8050 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8051 | if (sq_array_size == SIZE_MAX) | |
8052 | return SIZE_MAX; | |
8053 | ||
8054 | if (check_add_overflow(off, sq_array_size, &off)) | |
8055 | return SIZE_MAX; | |
8056 | ||
75b28aff HV |
8057 | return off; |
8058 | } | |
8059 | ||
7f61a1e9 PB |
8060 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf *imu) |
8061 | { | |
8062 | unsigned int i; | |
8063 | ||
8064 | for (i = 0; i < imu->nr_bvecs; i++) | |
8065 | unpin_user_page(imu->bvec[i].bv_page); | |
8066 | if (imu->acct_pages) | |
8067 | io_unaccount_mem(ctx, imu->acct_pages); | |
8068 | kvfree(imu->bvec); | |
8069 | imu->nr_bvecs = 0; | |
8070 | } | |
8071 | ||
0a96bbe4 | 8072 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8073 | { |
7f61a1e9 | 8074 | unsigned int i; |
edafccee JA |
8075 | |
8076 | if (!ctx->user_bufs) | |
8077 | return -ENXIO; | |
8078 | ||
7f61a1e9 PB |
8079 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8080 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee JA |
8081 | kfree(ctx->user_bufs); |
8082 | ctx->user_bufs = NULL; | |
8083 | ctx->nr_user_bufs = 0; | |
8084 | return 0; | |
8085 | } | |
8086 | ||
8087 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8088 | void __user *arg, unsigned index) | |
8089 | { | |
8090 | struct iovec __user *src; | |
8091 | ||
8092 | #ifdef CONFIG_COMPAT | |
8093 | if (ctx->compat) { | |
8094 | struct compat_iovec __user *ciovs; | |
8095 | struct compat_iovec ciov; | |
8096 | ||
8097 | ciovs = (struct compat_iovec __user *) arg; | |
8098 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8099 | return -EFAULT; | |
8100 | ||
d55e5f5b | 8101 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8102 | dst->iov_len = ciov.iov_len; |
8103 | return 0; | |
8104 | } | |
8105 | #endif | |
8106 | src = (struct iovec __user *) arg; | |
8107 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8108 | return -EFAULT; | |
8109 | return 0; | |
8110 | } | |
8111 | ||
de293938 JA |
8112 | /* |
8113 | * Not super efficient, but this is just a registration time. And we do cache | |
8114 | * the last compound head, so generally we'll only do a full search if we don't | |
8115 | * match that one. | |
8116 | * | |
8117 | * We check if the given compound head page has already been accounted, to | |
8118 | * avoid double accounting it. This allows us to account the full size of the | |
8119 | * page, not just the constituent pages of a huge page. | |
8120 | */ | |
8121 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8122 | int nr_pages, struct page *hpage) | |
8123 | { | |
8124 | int i, j; | |
8125 | ||
8126 | /* check current page array */ | |
8127 | for (i = 0; i < nr_pages; i++) { | |
8128 | if (!PageCompound(pages[i])) | |
8129 | continue; | |
8130 | if (compound_head(pages[i]) == hpage) | |
8131 | return true; | |
8132 | } | |
8133 | ||
8134 | /* check previously registered pages */ | |
8135 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8136 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8137 | ||
8138 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8139 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8140 | continue; | |
8141 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8142 | return true; | |
8143 | } | |
8144 | } | |
8145 | ||
8146 | return false; | |
8147 | } | |
8148 | ||
8149 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8150 | int nr_pages, struct io_mapped_ubuf *imu, | |
8151 | struct page **last_hpage) | |
8152 | { | |
8153 | int i, ret; | |
8154 | ||
8155 | for (i = 0; i < nr_pages; i++) { | |
8156 | if (!PageCompound(pages[i])) { | |
8157 | imu->acct_pages++; | |
8158 | } else { | |
8159 | struct page *hpage; | |
8160 | ||
8161 | hpage = compound_head(pages[i]); | |
8162 | if (hpage == *last_hpage) | |
8163 | continue; | |
8164 | *last_hpage = hpage; | |
8165 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8166 | continue; | |
8167 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8168 | } | |
8169 | } | |
8170 | ||
8171 | if (!imu->acct_pages) | |
8172 | return 0; | |
8173 | ||
26bfa89e | 8174 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8175 | if (ret) |
8176 | imu->acct_pages = 0; | |
8177 | return ret; | |
8178 | } | |
8179 | ||
0a96bbe4 BM |
8180 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8181 | struct io_mapped_ubuf *imu, | |
8182 | struct page **last_hpage) | |
edafccee JA |
8183 | { |
8184 | struct vm_area_struct **vmas = NULL; | |
8185 | struct page **pages = NULL; | |
0a96bbe4 BM |
8186 | unsigned long off, start, end, ubuf; |
8187 | size_t size; | |
8188 | int ret, pret, nr_pages, i; | |
8189 | ||
8190 | ubuf = (unsigned long) iov->iov_base; | |
8191 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8192 | start = ubuf >> PAGE_SHIFT; | |
8193 | nr_pages = end - start; | |
8194 | ||
8195 | ret = -ENOMEM; | |
8196 | ||
8197 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8198 | if (!pages) | |
8199 | goto done; | |
8200 | ||
8201 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8202 | GFP_KERNEL); | |
8203 | if (!vmas) | |
8204 | goto done; | |
edafccee | 8205 | |
0a96bbe4 BM |
8206 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8207 | GFP_KERNEL); | |
8208 | if (!imu->bvec) | |
8209 | goto done; | |
8210 | ||
8211 | ret = 0; | |
8212 | mmap_read_lock(current->mm); | |
8213 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8214 | pages, vmas); | |
8215 | if (pret == nr_pages) { | |
8216 | /* don't support file backed memory */ | |
8217 | for (i = 0; i < nr_pages; i++) { | |
8218 | struct vm_area_struct *vma = vmas[i]; | |
8219 | ||
8220 | if (vma->vm_file && | |
8221 | !is_file_hugepages(vma->vm_file)) { | |
8222 | ret = -EOPNOTSUPP; | |
8223 | break; | |
8224 | } | |
8225 | } | |
8226 | } else { | |
8227 | ret = pret < 0 ? pret : -EFAULT; | |
8228 | } | |
8229 | mmap_read_unlock(current->mm); | |
8230 | if (ret) { | |
8231 | /* | |
8232 | * if we did partial map, or found file backed vmas, | |
8233 | * release any pages we did get | |
8234 | */ | |
8235 | if (pret > 0) | |
8236 | unpin_user_pages(pages, pret); | |
8237 | kvfree(imu->bvec); | |
8238 | goto done; | |
8239 | } | |
8240 | ||
8241 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8242 | if (ret) { | |
8243 | unpin_user_pages(pages, pret); | |
8244 | kvfree(imu->bvec); | |
8245 | goto done; | |
8246 | } | |
8247 | ||
8248 | off = ubuf & ~PAGE_MASK; | |
8249 | size = iov->iov_len; | |
8250 | for (i = 0; i < nr_pages; i++) { | |
8251 | size_t vec_len; | |
8252 | ||
8253 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8254 | imu->bvec[i].bv_page = pages[i]; | |
8255 | imu->bvec[i].bv_len = vec_len; | |
8256 | imu->bvec[i].bv_offset = off; | |
8257 | off = 0; | |
8258 | size -= vec_len; | |
8259 | } | |
8260 | /* store original address for later verification */ | |
8261 | imu->ubuf = ubuf; | |
4751f53d | 8262 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 BM |
8263 | imu->nr_bvecs = nr_pages; |
8264 | ret = 0; | |
8265 | done: | |
8266 | kvfree(pages); | |
8267 | kvfree(vmas); | |
8268 | return ret; | |
8269 | } | |
8270 | ||
2b358604 | 8271 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8272 | { |
87094465 PB |
8273 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8274 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8275 | } |
edafccee | 8276 | |
2b358604 BM |
8277 | static int io_buffer_validate(struct iovec *iov) |
8278 | { | |
50e96989 PB |
8279 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8280 | ||
2b358604 BM |
8281 | /* |
8282 | * Don't impose further limits on the size and buffer | |
8283 | * constraints here, we'll -EINVAL later when IO is | |
8284 | * submitted if they are wrong. | |
8285 | */ | |
8286 | if (!iov->iov_base || !iov->iov_len) | |
8287 | return -EFAULT; | |
edafccee | 8288 | |
2b358604 BM |
8289 | /* arbitrary limit, but we need something */ |
8290 | if (iov->iov_len > SZ_1G) | |
8291 | return -EFAULT; | |
edafccee | 8292 | |
50e96989 PB |
8293 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8294 | return -EOVERFLOW; | |
8295 | ||
2b358604 BM |
8296 | return 0; |
8297 | } | |
edafccee | 8298 | |
2b358604 BM |
8299 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8300 | unsigned int nr_args) | |
8301 | { | |
8302 | int i, ret; | |
8303 | struct iovec iov; | |
8304 | struct page *last_hpage = NULL; | |
edafccee | 8305 | |
87094465 PB |
8306 | if (ctx->user_bufs) |
8307 | return -EBUSY; | |
8308 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8309 | return -EINVAL; | |
2b358604 BM |
8310 | ret = io_buffers_map_alloc(ctx, nr_args); |
8311 | if (ret) | |
8312 | return ret; | |
edafccee | 8313 | |
87094465 | 8314 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee | 8315 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; |
edafccee | 8316 | |
edafccee JA |
8317 | ret = io_copy_iov(ctx, &iov, arg, i); |
8318 | if (ret) | |
0a96bbe4 | 8319 | break; |
2b358604 BM |
8320 | ret = io_buffer_validate(&iov); |
8321 | if (ret) | |
0a96bbe4 | 8322 | break; |
0a96bbe4 BM |
8323 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8324 | if (ret) | |
8325 | break; | |
edafccee | 8326 | } |
0a96bbe4 BM |
8327 | |
8328 | if (ret) | |
8329 | io_sqe_buffers_unregister(ctx); | |
8330 | ||
edafccee JA |
8331 | return ret; |
8332 | } | |
8333 | ||
9b402849 JA |
8334 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8335 | { | |
8336 | __s32 __user *fds = arg; | |
8337 | int fd; | |
8338 | ||
8339 | if (ctx->cq_ev_fd) | |
8340 | return -EBUSY; | |
8341 | ||
8342 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8343 | return -EFAULT; | |
8344 | ||
8345 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8346 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8347 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8348 | ctx->cq_ev_fd = NULL; | |
8349 | return ret; | |
8350 | } | |
8351 | ||
8352 | return 0; | |
8353 | } | |
8354 | ||
8355 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8356 | { | |
8357 | if (ctx->cq_ev_fd) { | |
8358 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8359 | ctx->cq_ev_fd = NULL; | |
8360 | return 0; | |
8361 | } | |
8362 | ||
8363 | return -ENXIO; | |
8364 | } | |
8365 | ||
5a2e745d JA |
8366 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8367 | { | |
9e15c3a0 JA |
8368 | struct io_buffer *buf; |
8369 | unsigned long index; | |
8370 | ||
8371 | xa_for_each(&ctx->io_buffers, index, buf) | |
8372 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8373 | } |
8374 | ||
68e68ee6 | 8375 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8376 | { |
68e68ee6 | 8377 | struct io_kiocb *req, *nxt; |
1b4c351f | 8378 | |
68e68ee6 JA |
8379 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8380 | if (tsk && req->task != tsk) | |
8381 | continue; | |
1b4c351f JA |
8382 | list_del(&req->compl.list); |
8383 | kmem_cache_free(req_cachep, req); | |
8384 | } | |
8385 | } | |
8386 | ||
4010fec4 | 8387 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8388 | { |
bf019da7 | 8389 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8390 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8391 | |
9a4fdbd8 JA |
8392 | mutex_lock(&ctx->uring_lock); |
8393 | ||
8e5c66c4 | 8394 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8395 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8396 | submit_state->reqs); | |
8e5c66c4 PB |
8397 | submit_state->free_reqs = 0; |
8398 | } | |
9a4fdbd8 | 8399 | |
dac7a098 | 8400 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8401 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8402 | mutex_unlock(&ctx->uring_lock); |
8403 | } | |
8404 | ||
2b188cc1 JA |
8405 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8406 | { | |
37d1e2e3 | 8407 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8408 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8409 | |
37d1e2e3 | 8410 | if (ctx->mm_account) { |
2aede0e4 JA |
8411 | mmdrop(ctx->mm_account); |
8412 | ctx->mm_account = NULL; | |
30975825 | 8413 | } |
def596e9 | 8414 | |
8bad28d8 | 8415 | mutex_lock(&ctx->uring_lock); |
08480400 PB |
8416 | if (ctx->file_data) { |
8417 | if (!atomic_dec_and_test(&ctx->file_data->refs)) | |
8418 | wait_for_completion(&ctx->file_data->done); | |
8419 | __io_sqe_files_unregister(ctx); | |
8420 | } | |
c4ea060e PB |
8421 | if (ctx->rings) |
8422 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8423 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8424 | io_eventfd_unregister(ctx); |
5a2e745d | 8425 | io_destroy_buffers(ctx); |
def596e9 | 8426 | |
a7f0ed5a PB |
8427 | /* there are no registered resources left, nobody uses it */ |
8428 | if (ctx->rsrc_node) | |
8429 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8430 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8431 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8432 | flush_delayed_work(&ctx->rsrc_put_work); |
8433 | ||
8434 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8435 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
8dd03afe | 8436 | |
2b188cc1 | 8437 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8438 | if (ctx->ring_sock) { |
8439 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8440 | sock_release(ctx->ring_sock); |
355e8d26 | 8441 | } |
2b188cc1 JA |
8442 | #endif |
8443 | ||
75b28aff | 8444 | io_mem_free(ctx->rings); |
2b188cc1 | 8445 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8446 | |
8447 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8448 | free_uid(ctx->user); |
4010fec4 | 8449 | io_req_caches_free(ctx); |
e941894e JA |
8450 | if (ctx->hash_map) |
8451 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8452 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8453 | kfree(ctx); |
8454 | } | |
8455 | ||
8456 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8457 | { | |
8458 | struct io_ring_ctx *ctx = file->private_data; | |
8459 | __poll_t mask = 0; | |
8460 | ||
8461 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8462 | /* |
8463 | * synchronizes with barrier from wq_has_sleeper call in | |
8464 | * io_commit_cqring | |
8465 | */ | |
2b188cc1 | 8466 | smp_rmb(); |
90554200 | 8467 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8468 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8469 | |
8470 | /* | |
8471 | * Don't flush cqring overflow list here, just do a simple check. | |
8472 | * Otherwise there could possible be ABBA deadlock: | |
8473 | * CPU0 CPU1 | |
8474 | * ---- ---- | |
8475 | * lock(&ctx->uring_lock); | |
8476 | * lock(&ep->mtx); | |
8477 | * lock(&ctx->uring_lock); | |
8478 | * lock(&ep->mtx); | |
8479 | * | |
8480 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8481 | * pushs them to do the flush. | |
8482 | */ | |
8483 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8484 | mask |= EPOLLIN | EPOLLRDNORM; |
8485 | ||
8486 | return mask; | |
8487 | } | |
8488 | ||
8489 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8490 | { | |
8491 | struct io_ring_ctx *ctx = file->private_data; | |
8492 | ||
8493 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8494 | } | |
8495 | ||
0bead8cd | 8496 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8497 | { |
4379bf8b | 8498 | const struct cred *creds; |
071698e1 | 8499 | |
61cf9370 | 8500 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8501 | if (creds) { |
8502 | put_cred(creds); | |
0bead8cd | 8503 | return 0; |
1e6fa521 | 8504 | } |
0bead8cd YD |
8505 | |
8506 | return -EINVAL; | |
8507 | } | |
8508 | ||
9b465711 | 8509 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8510 | { |
9b465711 | 8511 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8512 | } |
8513 | ||
d56d938b PB |
8514 | struct io_tctx_exit { |
8515 | struct callback_head task_work; | |
8516 | struct completion completion; | |
baf186c4 | 8517 | struct io_ring_ctx *ctx; |
d56d938b PB |
8518 | }; |
8519 | ||
8520 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8521 | { | |
8522 | struct io_uring_task *tctx = current->io_uring; | |
8523 | struct io_tctx_exit *work; | |
8524 | ||
8525 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8526 | /* | |
8527 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8528 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8529 | */ | |
8530 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8531 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8532 | complete(&work->completion); |
8533 | } | |
8534 | ||
85faa7b8 JA |
8535 | static void io_ring_exit_work(struct work_struct *work) |
8536 | { | |
d56d938b | 8537 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8538 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8539 | struct io_tctx_exit exit; |
8540 | struct io_tctx_node *node; | |
8541 | int ret; | |
85faa7b8 | 8542 | |
a185f1db PB |
8543 | /* prevent SQPOLL from submitting new requests */ |
8544 | if (ctx->sq_data) { | |
8545 | io_sq_thread_park(ctx->sq_data); | |
8546 | list_del_init(&ctx->sqd_list); | |
8547 | io_sqd_update_thread_idle(ctx->sq_data); | |
8548 | io_sq_thread_unpark(ctx->sq_data); | |
8549 | } | |
8550 | ||
56952e91 JA |
8551 | /* |
8552 | * If we're doing polled IO and end up having requests being | |
8553 | * submitted async (out-of-line), then completions can come in while | |
8554 | * we're waiting for refs to drop. We need to reap these manually, | |
8555 | * as nobody else will be looking for them. | |
8556 | */ | |
b2edc0a7 | 8557 | do { |
9936c7c2 | 8558 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8559 | |
8560 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8561 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8562 | |
7f00651a PB |
8563 | init_completion(&exit.completion); |
8564 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8565 | exit.ctx = ctx; | |
89b5066e PB |
8566 | /* |
8567 | * Some may use context even when all refs and requests have been put, | |
8568 | * and they are free to do so while still holding uring_lock or | |
8569 | * completion_lock, see __io_req_task_submit(). Apart from other work, | |
8570 | * this lock/unlock section also waits them to finish. | |
8571 | */ | |
d56d938b PB |
8572 | mutex_lock(&ctx->uring_lock); |
8573 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8574 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8575 | ||
d56d938b PB |
8576 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8577 | ctx_node); | |
7f00651a PB |
8578 | /* don't spin on a single task if cancellation failed */ |
8579 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8580 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8581 | if (WARN_ON_ONCE(ret)) | |
8582 | continue; | |
8583 | wake_up_process(node->task); | |
8584 | ||
8585 | mutex_unlock(&ctx->uring_lock); | |
8586 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8587 | mutex_lock(&ctx->uring_lock); |
8588 | } | |
8589 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8590 | spin_lock_irq(&ctx->completion_lock); |
8591 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8592 | |
85faa7b8 JA |
8593 | io_ring_ctx_free(ctx); |
8594 | } | |
8595 | ||
80c4cbdb PB |
8596 | /* Returns true if we found and killed one or more timeouts */ |
8597 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
8598 | struct files_struct *files) | |
8599 | { | |
8600 | struct io_kiocb *req, *tmp; | |
8601 | int canceled = 0; | |
8602 | ||
8603 | spin_lock_irq(&ctx->completion_lock); | |
8604 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
8605 | if (io_match_task(req, tsk, files)) { | |
8606 | io_kill_timeout(req, -ECANCELED); | |
8607 | canceled++; | |
8608 | } | |
8609 | } | |
51520426 PB |
8610 | if (canceled != 0) |
8611 | io_commit_cqring(ctx); | |
80c4cbdb | 8612 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8613 | if (canceled != 0) |
8614 | io_cqring_ev_posted(ctx); | |
8615 | return canceled != 0; | |
8616 | } | |
8617 | ||
2b188cc1 JA |
8618 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8619 | { | |
61cf9370 MWO |
8620 | unsigned long index; |
8621 | struct creds *creds; | |
8622 | ||
2b188cc1 JA |
8623 | mutex_lock(&ctx->uring_lock); |
8624 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8625 | if (ctx->rings) |
6c2450ae | 8626 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8627 | xa_for_each(&ctx->personalities, index, creds) |
8628 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8629 | mutex_unlock(&ctx->uring_lock); |
8630 | ||
6b81928d PB |
8631 | io_kill_timeouts(ctx, NULL, NULL); |
8632 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8633 | |
15dff286 | 8634 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8635 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8636 | |
85faa7b8 | 8637 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8638 | /* |
8639 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8640 | * if we're exiting a ton of rings at the same time. It just adds | |
8641 | * noise and overhead, there's no discernable change in runtime | |
8642 | * over using system_wq. | |
8643 | */ | |
8644 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8645 | } |
8646 | ||
8647 | static int io_uring_release(struct inode *inode, struct file *file) | |
8648 | { | |
8649 | struct io_ring_ctx *ctx = file->private_data; | |
8650 | ||
8651 | file->private_data = NULL; | |
8652 | io_ring_ctx_wait_and_kill(ctx); | |
8653 | return 0; | |
8654 | } | |
8655 | ||
f6edbabb PB |
8656 | struct io_task_cancel { |
8657 | struct task_struct *task; | |
8658 | struct files_struct *files; | |
8659 | }; | |
f254ac04 | 8660 | |
f6edbabb | 8661 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8662 | { |
9a472ef7 | 8663 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8664 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8665 | bool ret; |
8666 | ||
f6edbabb | 8667 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8668 | unsigned long flags; |
8669 | struct io_ring_ctx *ctx = req->ctx; | |
8670 | ||
8671 | /* protect against races with linked timeouts */ | |
8672 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8673 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8674 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8675 | } else { | |
f6edbabb | 8676 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8677 | } |
8678 | return ret; | |
b711d4ea JA |
8679 | } |
8680 | ||
e1915f76 | 8681 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8682 | struct task_struct *task, |
b7ddce3c PB |
8683 | struct files_struct *files) |
8684 | { | |
e1915f76 | 8685 | struct io_defer_entry *de; |
b7ddce3c PB |
8686 | LIST_HEAD(list); |
8687 | ||
8688 | spin_lock_irq(&ctx->completion_lock); | |
8689 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8690 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8691 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8692 | break; | |
8693 | } | |
8694 | } | |
8695 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8696 | if (list_empty(&list)) |
8697 | return false; | |
b7ddce3c PB |
8698 | |
8699 | while (!list_empty(&list)) { | |
8700 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8701 | list_del_init(&de->list); | |
f41db273 | 8702 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8703 | kfree(de); |
8704 | } | |
e1915f76 | 8705 | return true; |
b7ddce3c PB |
8706 | } |
8707 | ||
1b00764f PB |
8708 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8709 | { | |
8710 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8711 | ||
8712 | return req->ctx == data; | |
8713 | } | |
8714 | ||
8715 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8716 | { | |
8717 | struct io_tctx_node *node; | |
8718 | enum io_wq_cancel cret; | |
8719 | bool ret = false; | |
8720 | ||
8721 | mutex_lock(&ctx->uring_lock); | |
8722 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8723 | struct io_uring_task *tctx = node->task->io_uring; | |
8724 | ||
8725 | /* | |
8726 | * io_wq will stay alive while we hold uring_lock, because it's | |
8727 | * killed after ctx nodes, which requires to take the lock. | |
8728 | */ | |
8729 | if (!tctx || !tctx->io_wq) | |
8730 | continue; | |
8731 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8732 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8733 | } | |
8734 | mutex_unlock(&ctx->uring_lock); | |
8735 | ||
8736 | return ret; | |
8737 | } | |
8738 | ||
9936c7c2 PB |
8739 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8740 | struct task_struct *task, | |
8741 | struct files_struct *files) | |
8742 | { | |
8743 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8744 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8745 | |
8746 | while (1) { | |
8747 | enum io_wq_cancel cret; | |
8748 | bool ret = false; | |
8749 | ||
1b00764f PB |
8750 | if (!task) { |
8751 | ret |= io_uring_try_cancel_iowq(ctx); | |
8752 | } else if (tctx && tctx->io_wq) { | |
8753 | /* | |
8754 | * Cancels requests of all rings, not only @ctx, but | |
8755 | * it's fine as the task is in exit/exec. | |
8756 | */ | |
5aa75ed5 | 8757 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8758 | &cancel, true); |
8759 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8760 | } | |
8761 | ||
8762 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8763 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8764 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8765 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8766 | io_iopoll_try_reap_events(ctx); | |
8767 | ret = true; | |
8768 | } | |
8769 | } | |
8770 | ||
e1915f76 | 8771 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8772 | ret |= io_poll_remove_all(ctx, task, files); |
8773 | ret |= io_kill_timeouts(ctx, task, files); | |
8774 | ret |= io_run_task_work(); | |
ba50a036 | 8775 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8776 | if (!ret) |
8777 | break; | |
8778 | cond_resched(); | |
8779 | } | |
8780 | } | |
8781 | ||
cf27f3b1 | 8782 | static int __io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8783 | { |
236434c3 | 8784 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8785 | struct io_tctx_node *node; |
a528b04e | 8786 | int ret; |
236434c3 MWO |
8787 | |
8788 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8789 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8790 | if (unlikely(ret)) |
8791 | return ret; | |
236434c3 | 8792 | tctx = current->io_uring; |
0f212204 | 8793 | } |
cf27f3b1 PB |
8794 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
8795 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
8796 | if (!node) | |
8797 | return -ENOMEM; | |
8798 | node->ctx = ctx; | |
8799 | node->task = current; | |
13bf43f5 | 8800 | |
cf27f3b1 PB |
8801 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
8802 | node, GFP_KERNEL)); | |
8803 | if (ret) { | |
8804 | kfree(node); | |
8805 | return ret; | |
0f212204 | 8806 | } |
cf27f3b1 PB |
8807 | |
8808 | mutex_lock(&ctx->uring_lock); | |
8809 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8810 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8811 | } |
cf27f3b1 | 8812 | tctx->last = ctx; |
0f212204 JA |
8813 | return 0; |
8814 | } | |
8815 | ||
cf27f3b1 PB |
8816 | /* |
8817 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8818 | */ | |
8819 | static inline int io_uring_add_task_file(struct io_ring_ctx *ctx) | |
8820 | { | |
8821 | struct io_uring_task *tctx = current->io_uring; | |
8822 | ||
8823 | if (likely(tctx && tctx->last == ctx)) | |
8824 | return 0; | |
8825 | return __io_uring_add_task_file(ctx); | |
8826 | } | |
8827 | ||
0f212204 JA |
8828 | /* |
8829 | * Remove this io_uring_file -> task mapping. | |
8830 | */ | |
2941267b | 8831 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8832 | { |
8833 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8834 | struct io_tctx_node *node; |
2941267b | 8835 | |
eebd2e37 PB |
8836 | if (!tctx) |
8837 | return; | |
13bf43f5 PB |
8838 | node = xa_erase(&tctx->xa, index); |
8839 | if (!node) | |
2941267b | 8840 | return; |
0f212204 | 8841 | |
13bf43f5 PB |
8842 | WARN_ON_ONCE(current != node->task); |
8843 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8844 | ||
8845 | mutex_lock(&node->ctx->uring_lock); | |
8846 | list_del(&node->ctx_node); | |
8847 | mutex_unlock(&node->ctx->uring_lock); | |
8848 | ||
baf186c4 | 8849 | if (tctx->last == node->ctx) |
0f212204 | 8850 | tctx->last = NULL; |
13bf43f5 | 8851 | kfree(node); |
0f212204 JA |
8852 | } |
8853 | ||
8452d4a6 | 8854 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8855 | { |
13bf43f5 | 8856 | struct io_tctx_node *node; |
de7f1d9e PB |
8857 | unsigned long index; |
8858 | ||
13bf43f5 | 8859 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8860 | io_uring_del_task_file(index); |
8452d4a6 PB |
8861 | if (tctx->io_wq) { |
8862 | io_wq_put_and_exit(tctx->io_wq); | |
8863 | tctx->io_wq = NULL; | |
8864 | } | |
de7f1d9e PB |
8865 | } |
8866 | ||
3f48cf18 | 8867 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 8868 | { |
3f48cf18 PB |
8869 | if (tracked) |
8870 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
8871 | return percpu_counter_sum(&tctx->inflight); |
8872 | } | |
8873 | ||
8874 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8875 | { | |
8876 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8877 | struct io_ring_ctx *ctx = work->ctx; | |
8878 | struct io_sq_data *sqd = ctx->sq_data; | |
8879 | ||
8880 | if (sqd->thread) | |
8881 | io_uring_cancel_sqpoll(ctx); | |
8882 | complete(&work->completion); | |
8883 | } | |
8884 | ||
8885 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8886 | { | |
8887 | struct io_sq_data *sqd = ctx->sq_data; | |
8888 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8889 | struct task_struct *task; | |
8890 | ||
8891 | io_sq_thread_park(sqd); | |
8892 | list_del_init(&ctx->sqd_list); | |
8893 | io_sqd_update_thread_idle(sqd); | |
8894 | task = sqd->thread; | |
8895 | if (task) { | |
8896 | init_completion(&work.completion); | |
8897 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8898 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
8899 | wake_up_process(task); |
8900 | } | |
8901 | io_sq_thread_unpark(sqd); | |
8902 | ||
8903 | if (task) | |
8904 | wait_for_completion(&work.completion); | |
8905 | } | |
8906 | ||
368b2080 | 8907 | static void io_uring_try_cancel(struct files_struct *files) |
0f212204 JA |
8908 | { |
8909 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8910 | struct io_tctx_node *node; |
ce765372 | 8911 | unsigned long index; |
0f212204 | 8912 | |
521d6a73 PB |
8913 | xa_for_each(&tctx->xa, index, node) { |
8914 | struct io_ring_ctx *ctx = node->ctx; | |
8915 | ||
8916 | if (ctx->sq_data) { | |
8917 | io_sqpoll_cancel_sync(ctx); | |
8918 | continue; | |
8919 | } | |
368b2080 | 8920 | io_uring_try_cancel_requests(ctx, current, files); |
521d6a73 | 8921 | } |
fdaf083c JA |
8922 | } |
8923 | ||
521d6a73 | 8924 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8925 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8926 | { | |
37d1e2e3 | 8927 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8928 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8929 | s64 inflight; |
8930 | DEFINE_WAIT(wait); | |
fdaf083c | 8931 | |
521d6a73 PB |
8932 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8933 | ||
0e9ddb39 PB |
8934 | atomic_inc(&tctx->in_idle); |
8935 | do { | |
8936 | /* read completions before cancelations */ | |
3f48cf18 | 8937 | inflight = tctx_inflight(tctx, false); |
0e9ddb39 PB |
8938 | if (!inflight) |
8939 | break; | |
521d6a73 | 8940 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 8941 | |
0e9ddb39 PB |
8942 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8943 | /* | |
8944 | * If we've seen completions, retry without waiting. This | |
8945 | * avoids a race where a completion comes in before we did | |
8946 | * prepare_to_wait(). | |
8947 | */ | |
3f48cf18 | 8948 | if (inflight == tctx_inflight(tctx, false)) |
0e9ddb39 PB |
8949 | schedule(); |
8950 | finish_wait(&tctx->wait, &wait); | |
8951 | } while (1); | |
8952 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
8953 | } |
8954 | ||
0f212204 JA |
8955 | /* |
8956 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8957 | * requests. | |
8958 | */ | |
3f48cf18 | 8959 | void __io_uring_cancel(struct files_struct *files) |
0f212204 JA |
8960 | { |
8961 | struct io_uring_task *tctx = current->io_uring; | |
8962 | DEFINE_WAIT(wait); | |
d8a6df10 | 8963 | s64 inflight; |
0f212204 JA |
8964 | |
8965 | /* make sure overflow events are dropped */ | |
fdaf083c | 8966 | atomic_inc(&tctx->in_idle); |
3f48cf18 | 8967 | io_uring_try_cancel(files); |
5a978dcf | 8968 | |
d8a6df10 | 8969 | do { |
0f212204 | 8970 | /* read completions before cancelations */ |
3f48cf18 | 8971 | inflight = tctx_inflight(tctx, !!files); |
d8a6df10 JA |
8972 | if (!inflight) |
8973 | break; | |
3f48cf18 | 8974 | io_uring_try_cancel(files); |
0f212204 JA |
8975 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8976 | ||
8977 | /* | |
a1bb3cd5 PB |
8978 | * If we've seen completions, retry without waiting. This |
8979 | * avoids a race where a completion comes in before we did | |
8980 | * prepare_to_wait(). | |
0f212204 | 8981 | */ |
3f48cf18 | 8982 | if (inflight == tctx_inflight(tctx, !!files)) |
a1bb3cd5 | 8983 | schedule(); |
f57555ed | 8984 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8985 | } while (1); |
fdaf083c | 8986 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8987 | |
8452d4a6 | 8988 | io_uring_clean_tctx(tctx); |
3f48cf18 PB |
8989 | if (!files) { |
8990 | /* for exec all current's requests should be gone, kill tctx */ | |
8991 | __io_uring_free(current); | |
8992 | } | |
44e728b8 PB |
8993 | } |
8994 | ||
6c5c240e RP |
8995 | static void *io_uring_validate_mmap_request(struct file *file, |
8996 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8997 | { |
2b188cc1 | 8998 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8999 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9000 | struct page *page; |
9001 | void *ptr; | |
9002 | ||
9003 | switch (offset) { | |
9004 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9005 | case IORING_OFF_CQ_RING: |
9006 | ptr = ctx->rings; | |
2b188cc1 JA |
9007 | break; |
9008 | case IORING_OFF_SQES: | |
9009 | ptr = ctx->sq_sqes; | |
9010 | break; | |
2b188cc1 | 9011 | default: |
6c5c240e | 9012 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9013 | } |
9014 | ||
9015 | page = virt_to_head_page(ptr); | |
a50b854e | 9016 | if (sz > page_size(page)) |
6c5c240e RP |
9017 | return ERR_PTR(-EINVAL); |
9018 | ||
9019 | return ptr; | |
9020 | } | |
9021 | ||
9022 | #ifdef CONFIG_MMU | |
9023 | ||
9024 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9025 | { | |
9026 | size_t sz = vma->vm_end - vma->vm_start; | |
9027 | unsigned long pfn; | |
9028 | void *ptr; | |
9029 | ||
9030 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9031 | if (IS_ERR(ptr)) | |
9032 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9033 | |
9034 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9035 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9036 | } | |
9037 | ||
6c5c240e RP |
9038 | #else /* !CONFIG_MMU */ |
9039 | ||
9040 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9041 | { | |
9042 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9043 | } | |
9044 | ||
9045 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9046 | { | |
9047 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9048 | } | |
9049 | ||
9050 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9051 | unsigned long addr, unsigned long len, | |
9052 | unsigned long pgoff, unsigned long flags) | |
9053 | { | |
9054 | void *ptr; | |
9055 | ||
9056 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9057 | if (IS_ERR(ptr)) | |
9058 | return PTR_ERR(ptr); | |
9059 | ||
9060 | return (unsigned long) ptr; | |
9061 | } | |
9062 | ||
9063 | #endif /* !CONFIG_MMU */ | |
9064 | ||
d9d05217 | 9065 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9066 | { |
9067 | DEFINE_WAIT(wait); | |
9068 | ||
9069 | do { | |
9070 | if (!io_sqring_full(ctx)) | |
9071 | break; | |
90554200 JA |
9072 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9073 | ||
9074 | if (!io_sqring_full(ctx)) | |
9075 | break; | |
90554200 JA |
9076 | schedule(); |
9077 | } while (!signal_pending(current)); | |
9078 | ||
9079 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9080 | return 0; |
90554200 JA |
9081 | } |
9082 | ||
c73ebb68 HX |
9083 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9084 | struct __kernel_timespec __user **ts, | |
9085 | const sigset_t __user **sig) | |
9086 | { | |
9087 | struct io_uring_getevents_arg arg; | |
9088 | ||
9089 | /* | |
9090 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9091 | * is just a pointer to the sigset_t. | |
9092 | */ | |
9093 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9094 | *sig = (const sigset_t __user *) argp; | |
9095 | *ts = NULL; | |
9096 | return 0; | |
9097 | } | |
9098 | ||
9099 | /* | |
9100 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9101 | * timespec and sigset_t pointers if good. | |
9102 | */ | |
9103 | if (*argsz != sizeof(arg)) | |
9104 | return -EINVAL; | |
9105 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9106 | return -EFAULT; | |
9107 | *sig = u64_to_user_ptr(arg.sigmask); | |
9108 | *argsz = arg.sigmask_sz; | |
9109 | *ts = u64_to_user_ptr(arg.ts); | |
9110 | return 0; | |
9111 | } | |
9112 | ||
2b188cc1 | 9113 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9114 | u32, min_complete, u32, flags, const void __user *, argp, |
9115 | size_t, argsz) | |
2b188cc1 JA |
9116 | { |
9117 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9118 | int submitted = 0; |
9119 | struct fd f; | |
33f993da | 9120 | long ret; |
2b188cc1 | 9121 | |
4c6e277c | 9122 | io_run_task_work(); |
b41e9852 | 9123 | |
33f993da PB |
9124 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9125 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9126 | return -EINVAL; |
9127 | ||
9128 | f = fdget(fd); | |
33f993da | 9129 | if (unlikely(!f.file)) |
2b188cc1 JA |
9130 | return -EBADF; |
9131 | ||
9132 | ret = -EOPNOTSUPP; | |
33f993da | 9133 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9134 | goto out_fput; |
9135 | ||
9136 | ret = -ENXIO; | |
9137 | ctx = f.file->private_data; | |
33f993da | 9138 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9139 | goto out_fput; |
9140 | ||
7e84e1c7 | 9141 | ret = -EBADFD; |
33f993da | 9142 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9143 | goto out; |
9144 | ||
6c271ce2 JA |
9145 | /* |
9146 | * For SQ polling, the thread will do all submissions and completions. | |
9147 | * Just return the requested submit count, and wake the thread if | |
9148 | * we were asked to. | |
9149 | */ | |
b2a9eada | 9150 | ret = 0; |
6c271ce2 | 9151 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9152 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9153 | |
d9d05217 | 9154 | ret = -EOWNERDEAD; |
04147488 SM |
9155 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9156 | goto out; | |
9157 | } | |
6c271ce2 | 9158 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9159 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9160 | if (flags & IORING_ENTER_SQ_WAIT) { |
9161 | ret = io_sqpoll_wait_sq(ctx); | |
9162 | if (ret) | |
9163 | goto out; | |
9164 | } | |
6c271ce2 | 9165 | submitted = to_submit; |
b2a9eada | 9166 | } else if (to_submit) { |
baf186c4 | 9167 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9168 | if (unlikely(ret)) |
9169 | goto out; | |
2b188cc1 | 9170 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9171 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9172 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9173 | |
9174 | if (submitted != to_submit) | |
9175 | goto out; | |
2b188cc1 JA |
9176 | } |
9177 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9178 | const sigset_t __user *sig; |
9179 | struct __kernel_timespec __user *ts; | |
9180 | ||
9181 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9182 | if (unlikely(ret)) | |
9183 | goto out; | |
9184 | ||
2b188cc1 JA |
9185 | min_complete = min(min_complete, ctx->cq_entries); |
9186 | ||
32b2244a XW |
9187 | /* |
9188 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9189 | * space applications don't need to do io completion events | |
9190 | * polling again, they can rely on io_sq_thread to do polling | |
9191 | * work, which can reduce cpu usage and uring_lock contention. | |
9192 | */ | |
9193 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9194 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9195 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9196 | } else { |
c73ebb68 | 9197 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9198 | } |
2b188cc1 JA |
9199 | } |
9200 | ||
7c504e65 | 9201 | out: |
6805b32e | 9202 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9203 | out_fput: |
9204 | fdput(f); | |
9205 | return submitted ? submitted : ret; | |
9206 | } | |
9207 | ||
bebdb65e | 9208 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9209 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9210 | const struct cred *cred) | |
87ce955b | 9211 | { |
87ce955b JA |
9212 | struct user_namespace *uns = seq_user_ns(m); |
9213 | struct group_info *gi; | |
9214 | kernel_cap_t cap; | |
9215 | unsigned __capi; | |
9216 | int g; | |
9217 | ||
9218 | seq_printf(m, "%5d\n", id); | |
9219 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9220 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9221 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9222 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9223 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9224 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9225 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9226 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9227 | seq_puts(m, "\n\tGroups:\t"); | |
9228 | gi = cred->group_info; | |
9229 | for (g = 0; g < gi->ngroups; g++) { | |
9230 | seq_put_decimal_ull(m, g ? " " : "", | |
9231 | from_kgid_munged(uns, gi->gid[g])); | |
9232 | } | |
9233 | seq_puts(m, "\n\tCapEff:\t"); | |
9234 | cap = cred->cap_effective; | |
9235 | CAP_FOR_EACH_U32(__capi) | |
9236 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9237 | seq_putc(m, '\n'); | |
9238 | return 0; | |
9239 | } | |
9240 | ||
9241 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9242 | { | |
dbbe9c64 | 9243 | struct io_sq_data *sq = NULL; |
fad8e0de | 9244 | bool has_lock; |
87ce955b JA |
9245 | int i; |
9246 | ||
fad8e0de JA |
9247 | /* |
9248 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9249 | * since fdinfo case grabs it in the opposite direction of normal use | |
9250 | * cases. If we fail to get the lock, we just don't iterate any | |
9251 | * structures that could be going away outside the io_uring mutex. | |
9252 | */ | |
9253 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9254 | ||
5f3f26f9 | 9255 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9256 | sq = ctx->sq_data; |
5f3f26f9 JA |
9257 | if (!sq->thread) |
9258 | sq = NULL; | |
9259 | } | |
dbbe9c64 JQ |
9260 | |
9261 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9262 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9263 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9264 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9265 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9266 | |
87ce955b JA |
9267 | if (f) |
9268 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9269 | else | |
9270 | seq_printf(m, "%5u: <none>\n", i); | |
9271 | } | |
9272 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9273 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b | 9274 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
4751f53d | 9275 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9276 | |
4751f53d | 9277 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9278 | } |
61cf9370 MWO |
9279 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9280 | unsigned long index; | |
9281 | const struct cred *cred; | |
9282 | ||
87ce955b | 9283 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9284 | xa_for_each(&ctx->personalities, index, cred) |
9285 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9286 | } |
d7718a9d JA |
9287 | seq_printf(m, "PollList:\n"); |
9288 | spin_lock_irq(&ctx->completion_lock); | |
9289 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9290 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9291 | struct io_kiocb *req; | |
9292 | ||
9293 | hlist_for_each_entry(req, list, hash_node) | |
9294 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9295 | req->task->task_works != NULL); | |
9296 | } | |
9297 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9298 | if (has_lock) |
9299 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9300 | } |
9301 | ||
9302 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9303 | { | |
9304 | struct io_ring_ctx *ctx = f->private_data; | |
9305 | ||
9306 | if (percpu_ref_tryget(&ctx->refs)) { | |
9307 | __io_uring_show_fdinfo(ctx, m); | |
9308 | percpu_ref_put(&ctx->refs); | |
9309 | } | |
9310 | } | |
bebdb65e | 9311 | #endif |
87ce955b | 9312 | |
2b188cc1 JA |
9313 | static const struct file_operations io_uring_fops = { |
9314 | .release = io_uring_release, | |
9315 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9316 | #ifndef CONFIG_MMU |
9317 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9318 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9319 | #endif | |
2b188cc1 JA |
9320 | .poll = io_uring_poll, |
9321 | .fasync = io_uring_fasync, | |
bebdb65e | 9322 | #ifdef CONFIG_PROC_FS |
87ce955b | 9323 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9324 | #endif |
2b188cc1 JA |
9325 | }; |
9326 | ||
9327 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9328 | struct io_uring_params *p) | |
9329 | { | |
75b28aff HV |
9330 | struct io_rings *rings; |
9331 | size_t size, sq_array_offset; | |
2b188cc1 | 9332 | |
bd740481 JA |
9333 | /* make sure these are sane, as we already accounted them */ |
9334 | ctx->sq_entries = p->sq_entries; | |
9335 | ctx->cq_entries = p->cq_entries; | |
9336 | ||
75b28aff HV |
9337 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9338 | if (size == SIZE_MAX) | |
9339 | return -EOVERFLOW; | |
9340 | ||
9341 | rings = io_mem_alloc(size); | |
9342 | if (!rings) | |
2b188cc1 JA |
9343 | return -ENOMEM; |
9344 | ||
75b28aff HV |
9345 | ctx->rings = rings; |
9346 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9347 | rings->sq_ring_mask = p->sq_entries - 1; | |
9348 | rings->cq_ring_mask = p->cq_entries - 1; | |
9349 | rings->sq_ring_entries = p->sq_entries; | |
9350 | rings->cq_ring_entries = p->cq_entries; | |
9351 | ctx->sq_mask = rings->sq_ring_mask; | |
9352 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9353 | |
9354 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9355 | if (size == SIZE_MAX) { |
9356 | io_mem_free(ctx->rings); | |
9357 | ctx->rings = NULL; | |
2b188cc1 | 9358 | return -EOVERFLOW; |
eb065d30 | 9359 | } |
2b188cc1 JA |
9360 | |
9361 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9362 | if (!ctx->sq_sqes) { |
9363 | io_mem_free(ctx->rings); | |
9364 | ctx->rings = NULL; | |
2b188cc1 | 9365 | return -ENOMEM; |
eb065d30 | 9366 | } |
2b188cc1 | 9367 | |
2b188cc1 JA |
9368 | return 0; |
9369 | } | |
9370 | ||
9faadcc8 PB |
9371 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9372 | { | |
9373 | int ret, fd; | |
9374 | ||
9375 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9376 | if (fd < 0) | |
9377 | return fd; | |
9378 | ||
baf186c4 | 9379 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9380 | if (ret) { |
9381 | put_unused_fd(fd); | |
9382 | return ret; | |
9383 | } | |
9384 | fd_install(fd, file); | |
9385 | return fd; | |
9386 | } | |
9387 | ||
2b188cc1 JA |
9388 | /* |
9389 | * Allocate an anonymous fd, this is what constitutes the application | |
9390 | * visible backing of an io_uring instance. The application mmaps this | |
9391 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9392 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9393 | */ | |
9faadcc8 | 9394 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9395 | { |
9396 | struct file *file; | |
9faadcc8 | 9397 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9398 | int ret; |
9399 | ||
2b188cc1 JA |
9400 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9401 | &ctx->ring_sock); | |
9402 | if (ret) | |
9faadcc8 | 9403 | return ERR_PTR(ret); |
2b188cc1 JA |
9404 | #endif |
9405 | ||
2b188cc1 JA |
9406 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9407 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9408 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9409 | if (IS_ERR(file)) { |
9410 | sock_release(ctx->ring_sock); | |
9411 | ctx->ring_sock = NULL; | |
9412 | } else { | |
9413 | ctx->ring_sock->file = file; | |
0f212204 | 9414 | } |
2b188cc1 | 9415 | #endif |
9faadcc8 | 9416 | return file; |
2b188cc1 JA |
9417 | } |
9418 | ||
7f13657d XW |
9419 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9420 | struct io_uring_params __user *params) | |
2b188cc1 | 9421 | { |
2b188cc1 | 9422 | struct io_ring_ctx *ctx; |
9faadcc8 | 9423 | struct file *file; |
2b188cc1 JA |
9424 | int ret; |
9425 | ||
8110c1a6 | 9426 | if (!entries) |
2b188cc1 | 9427 | return -EINVAL; |
8110c1a6 JA |
9428 | if (entries > IORING_MAX_ENTRIES) { |
9429 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9430 | return -EINVAL; | |
9431 | entries = IORING_MAX_ENTRIES; | |
9432 | } | |
2b188cc1 JA |
9433 | |
9434 | /* | |
9435 | * Use twice as many entries for the CQ ring. It's possible for the | |
9436 | * application to drive a higher depth than the size of the SQ ring, | |
9437 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9438 | * some flexibility in overcommitting a bit. If the application has |
9439 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9440 | * of CQ ring entries manually. | |
2b188cc1 JA |
9441 | */ |
9442 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9443 | if (p->flags & IORING_SETUP_CQSIZE) { |
9444 | /* | |
9445 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9446 | * to a power-of-two, if it isn't already. We do NOT impose | |
9447 | * any cq vs sq ring sizing. | |
9448 | */ | |
eb2667b3 | 9449 | if (!p->cq_entries) |
33a107f0 | 9450 | return -EINVAL; |
8110c1a6 JA |
9451 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9452 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9453 | return -EINVAL; | |
9454 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9455 | } | |
eb2667b3 JQ |
9456 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9457 | if (p->cq_entries < p->sq_entries) | |
9458 | return -EINVAL; | |
33a107f0 JA |
9459 | } else { |
9460 | p->cq_entries = 2 * p->sq_entries; | |
9461 | } | |
2b188cc1 | 9462 | |
2b188cc1 | 9463 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9464 | if (!ctx) |
2b188cc1 | 9465 | return -ENOMEM; |
2b188cc1 | 9466 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9467 | if (!capable(CAP_IPC_LOCK)) |
9468 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9469 | |
9470 | /* | |
9471 | * This is just grabbed for accounting purposes. When a process exits, | |
9472 | * the mm is exited and dropped before the files, hence we need to hang | |
9473 | * on to this mm purely for the purposes of being able to unaccount | |
9474 | * memory (locked/pinned vm). It's not used for anything else. | |
9475 | */ | |
6b7898eb | 9476 | mmgrab(current->mm); |
2aede0e4 | 9477 | ctx->mm_account = current->mm; |
6b7898eb | 9478 | |
2b188cc1 JA |
9479 | ret = io_allocate_scq_urings(ctx, p); |
9480 | if (ret) | |
9481 | goto err; | |
9482 | ||
7e84e1c7 | 9483 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9484 | if (ret) |
9485 | goto err; | |
9486 | ||
2b188cc1 | 9487 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9488 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9489 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9490 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9491 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9492 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9493 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9494 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9495 | |
9496 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9497 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9498 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9499 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9500 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9501 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9502 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9503 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9504 | |
7f13657d XW |
9505 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9506 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9507 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9508 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9509 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9510 | |
9511 | if (copy_to_user(params, p, sizeof(*p))) { | |
9512 | ret = -EFAULT; | |
9513 | goto err; | |
9514 | } | |
d1719f70 | 9515 | |
9faadcc8 PB |
9516 | file = io_uring_get_file(ctx); |
9517 | if (IS_ERR(file)) { | |
9518 | ret = PTR_ERR(file); | |
9519 | goto err; | |
9520 | } | |
9521 | ||
044c1ab3 JA |
9522 | /* |
9523 | * Install ring fd as the very last thing, so we don't risk someone | |
9524 | * having closed it before we finish setup | |
9525 | */ | |
9faadcc8 PB |
9526 | ret = io_uring_install_fd(ctx, file); |
9527 | if (ret < 0) { | |
9528 | /* fput will clean it up */ | |
9529 | fput(file); | |
9530 | return ret; | |
9531 | } | |
044c1ab3 | 9532 | |
c826bd7a | 9533 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9534 | return ret; |
9535 | err: | |
9536 | io_ring_ctx_wait_and_kill(ctx); | |
9537 | return ret; | |
9538 | } | |
9539 | ||
9540 | /* | |
9541 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9542 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9543 | * params structure passed in. | |
9544 | */ | |
9545 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9546 | { | |
9547 | struct io_uring_params p; | |
2b188cc1 JA |
9548 | int i; |
9549 | ||
9550 | if (copy_from_user(&p, params, sizeof(p))) | |
9551 | return -EFAULT; | |
9552 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9553 | if (p.resv[i]) | |
9554 | return -EINVAL; | |
9555 | } | |
9556 | ||
6c271ce2 | 9557 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9558 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9559 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9560 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9561 | return -EINVAL; |
9562 | ||
7f13657d | 9563 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9564 | } |
9565 | ||
9566 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9567 | struct io_uring_params __user *, params) | |
9568 | { | |
9569 | return io_uring_setup(entries, params); | |
9570 | } | |
9571 | ||
66f4af93 JA |
9572 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9573 | { | |
9574 | struct io_uring_probe *p; | |
9575 | size_t size; | |
9576 | int i, ret; | |
9577 | ||
9578 | size = struct_size(p, ops, nr_args); | |
9579 | if (size == SIZE_MAX) | |
9580 | return -EOVERFLOW; | |
9581 | p = kzalloc(size, GFP_KERNEL); | |
9582 | if (!p) | |
9583 | return -ENOMEM; | |
9584 | ||
9585 | ret = -EFAULT; | |
9586 | if (copy_from_user(p, arg, size)) | |
9587 | goto out; | |
9588 | ret = -EINVAL; | |
9589 | if (memchr_inv(p, 0, size)) | |
9590 | goto out; | |
9591 | ||
9592 | p->last_op = IORING_OP_LAST - 1; | |
9593 | if (nr_args > IORING_OP_LAST) | |
9594 | nr_args = IORING_OP_LAST; | |
9595 | ||
9596 | for (i = 0; i < nr_args; i++) { | |
9597 | p->ops[i].op = i; | |
9598 | if (!io_op_defs[i].not_supported) | |
9599 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9600 | } | |
9601 | p->ops_len = i; | |
9602 | ||
9603 | ret = 0; | |
9604 | if (copy_to_user(arg, p, size)) | |
9605 | ret = -EFAULT; | |
9606 | out: | |
9607 | kfree(p); | |
9608 | return ret; | |
9609 | } | |
9610 | ||
071698e1 JA |
9611 | static int io_register_personality(struct io_ring_ctx *ctx) |
9612 | { | |
4379bf8b | 9613 | const struct cred *creds; |
61cf9370 | 9614 | u32 id; |
1e6fa521 | 9615 | int ret; |
071698e1 | 9616 | |
4379bf8b | 9617 | creds = get_current_cred(); |
1e6fa521 | 9618 | |
61cf9370 MWO |
9619 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9620 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9621 | if (!ret) | |
9622 | return id; | |
9623 | put_cred(creds); | |
1e6fa521 | 9624 | return ret; |
071698e1 JA |
9625 | } |
9626 | ||
21b55dbc SG |
9627 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9628 | unsigned int nr_args) | |
9629 | { | |
9630 | struct io_uring_restriction *res; | |
9631 | size_t size; | |
9632 | int i, ret; | |
9633 | ||
7e84e1c7 SG |
9634 | /* Restrictions allowed only if rings started disabled */ |
9635 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9636 | return -EBADFD; | |
9637 | ||
21b55dbc | 9638 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9639 | if (ctx->restrictions.registered) |
21b55dbc SG |
9640 | return -EBUSY; |
9641 | ||
9642 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9643 | return -EINVAL; | |
9644 | ||
9645 | size = array_size(nr_args, sizeof(*res)); | |
9646 | if (size == SIZE_MAX) | |
9647 | return -EOVERFLOW; | |
9648 | ||
9649 | res = memdup_user(arg, size); | |
9650 | if (IS_ERR(res)) | |
9651 | return PTR_ERR(res); | |
9652 | ||
9653 | ret = 0; | |
9654 | ||
9655 | for (i = 0; i < nr_args; i++) { | |
9656 | switch (res[i].opcode) { | |
9657 | case IORING_RESTRICTION_REGISTER_OP: | |
9658 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9659 | ret = -EINVAL; | |
9660 | goto out; | |
9661 | } | |
9662 | ||
9663 | __set_bit(res[i].register_op, | |
9664 | ctx->restrictions.register_op); | |
9665 | break; | |
9666 | case IORING_RESTRICTION_SQE_OP: | |
9667 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9668 | ret = -EINVAL; | |
9669 | goto out; | |
9670 | } | |
9671 | ||
9672 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9673 | break; | |
9674 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9675 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9676 | break; | |
9677 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9678 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9679 | break; | |
9680 | default: | |
9681 | ret = -EINVAL; | |
9682 | goto out; | |
9683 | } | |
9684 | } | |
9685 | ||
9686 | out: | |
9687 | /* Reset all restrictions if an error happened */ | |
9688 | if (ret != 0) | |
9689 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9690 | else | |
7e84e1c7 | 9691 | ctx->restrictions.registered = true; |
21b55dbc SG |
9692 | |
9693 | kfree(res); | |
9694 | return ret; | |
9695 | } | |
9696 | ||
7e84e1c7 SG |
9697 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9698 | { | |
9699 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9700 | return -EBADFD; | |
9701 | ||
9702 | if (ctx->restrictions.registered) | |
9703 | ctx->restricted = 1; | |
9704 | ||
0298ef96 PB |
9705 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9706 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9707 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9708 | return 0; |
9709 | } | |
9710 | ||
071698e1 JA |
9711 | static bool io_register_op_must_quiesce(int op) |
9712 | { | |
9713 | switch (op) { | |
f4f7d21c | 9714 | case IORING_REGISTER_FILES: |
071698e1 JA |
9715 | case IORING_UNREGISTER_FILES: |
9716 | case IORING_REGISTER_FILES_UPDATE: | |
9717 | case IORING_REGISTER_PROBE: | |
9718 | case IORING_REGISTER_PERSONALITY: | |
9719 | case IORING_UNREGISTER_PERSONALITY: | |
9720 | return false; | |
9721 | default: | |
9722 | return true; | |
9723 | } | |
9724 | } | |
9725 | ||
edafccee JA |
9726 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9727 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9728 | __releases(ctx->uring_lock) |
9729 | __acquires(ctx->uring_lock) | |
edafccee JA |
9730 | { |
9731 | int ret; | |
9732 | ||
35fa71a0 JA |
9733 | /* |
9734 | * We're inside the ring mutex, if the ref is already dying, then | |
9735 | * someone else killed the ctx or is already going through | |
9736 | * io_uring_register(). | |
9737 | */ | |
9738 | if (percpu_ref_is_dying(&ctx->refs)) | |
9739 | return -ENXIO; | |
9740 | ||
071698e1 | 9741 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9742 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9743 | |
05f3fb3c JA |
9744 | /* |
9745 | * Drop uring mutex before waiting for references to exit. If | |
9746 | * another thread is currently inside io_uring_enter() it might | |
9747 | * need to grab the uring_lock to make progress. If we hold it | |
9748 | * here across the drain wait, then we can deadlock. It's safe | |
9749 | * to drop the mutex here, since no new references will come in | |
9750 | * after we've killed the percpu ref. | |
9751 | */ | |
9752 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9753 | do { |
9754 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9755 | if (!ret) | |
9756 | break; | |
ed6930c9 JA |
9757 | ret = io_run_task_work_sig(); |
9758 | if (ret < 0) | |
9759 | break; | |
af9c1a44 | 9760 | } while (1); |
05f3fb3c | 9761 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9762 | |
c150368b | 9763 | if (ret) { |
f70865db PB |
9764 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
9765 | return ret; | |
21b55dbc SG |
9766 | } |
9767 | } | |
9768 | ||
9769 | if (ctx->restricted) { | |
9770 | if (opcode >= IORING_REGISTER_LAST) { | |
9771 | ret = -EINVAL; | |
9772 | goto out; | |
9773 | } | |
9774 | ||
9775 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9776 | ret = -EACCES; | |
c150368b JA |
9777 | goto out; |
9778 | } | |
05f3fb3c | 9779 | } |
edafccee JA |
9780 | |
9781 | switch (opcode) { | |
9782 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9783 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9784 | break; |
9785 | case IORING_UNREGISTER_BUFFERS: | |
9786 | ret = -EINVAL; | |
9787 | if (arg || nr_args) | |
9788 | break; | |
0a96bbe4 | 9789 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9790 | break; |
6b06314c JA |
9791 | case IORING_REGISTER_FILES: |
9792 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9793 | break; | |
9794 | case IORING_UNREGISTER_FILES: | |
9795 | ret = -EINVAL; | |
9796 | if (arg || nr_args) | |
9797 | break; | |
9798 | ret = io_sqe_files_unregister(ctx); | |
9799 | break; | |
c3a31e60 JA |
9800 | case IORING_REGISTER_FILES_UPDATE: |
9801 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9802 | break; | |
9b402849 | 9803 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9804 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9805 | ret = -EINVAL; |
9806 | if (nr_args != 1) | |
9807 | break; | |
9808 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9809 | if (ret) |
9810 | break; | |
9811 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9812 | ctx->eventfd_async = 1; | |
9813 | else | |
9814 | ctx->eventfd_async = 0; | |
9b402849 JA |
9815 | break; |
9816 | case IORING_UNREGISTER_EVENTFD: | |
9817 | ret = -EINVAL; | |
9818 | if (arg || nr_args) | |
9819 | break; | |
9820 | ret = io_eventfd_unregister(ctx); | |
9821 | break; | |
66f4af93 JA |
9822 | case IORING_REGISTER_PROBE: |
9823 | ret = -EINVAL; | |
9824 | if (!arg || nr_args > 256) | |
9825 | break; | |
9826 | ret = io_probe(ctx, arg, nr_args); | |
9827 | break; | |
071698e1 JA |
9828 | case IORING_REGISTER_PERSONALITY: |
9829 | ret = -EINVAL; | |
9830 | if (arg || nr_args) | |
9831 | break; | |
9832 | ret = io_register_personality(ctx); | |
9833 | break; | |
9834 | case IORING_UNREGISTER_PERSONALITY: | |
9835 | ret = -EINVAL; | |
9836 | if (arg) | |
9837 | break; | |
9838 | ret = io_unregister_personality(ctx, nr_args); | |
9839 | break; | |
7e84e1c7 SG |
9840 | case IORING_REGISTER_ENABLE_RINGS: |
9841 | ret = -EINVAL; | |
9842 | if (arg || nr_args) | |
9843 | break; | |
9844 | ret = io_register_enable_rings(ctx); | |
9845 | break; | |
21b55dbc SG |
9846 | case IORING_REGISTER_RESTRICTIONS: |
9847 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9848 | break; | |
edafccee JA |
9849 | default: |
9850 | ret = -EINVAL; | |
9851 | break; | |
9852 | } | |
9853 | ||
21b55dbc | 9854 | out: |
071698e1 | 9855 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9856 | /* bring the ctx back to life */ |
05f3fb3c | 9857 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 9858 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9859 | } |
edafccee JA |
9860 | return ret; |
9861 | } | |
9862 | ||
9863 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9864 | void __user *, arg, unsigned int, nr_args) | |
9865 | { | |
9866 | struct io_ring_ctx *ctx; | |
9867 | long ret = -EBADF; | |
9868 | struct fd f; | |
9869 | ||
9870 | f = fdget(fd); | |
9871 | if (!f.file) | |
9872 | return -EBADF; | |
9873 | ||
9874 | ret = -EOPNOTSUPP; | |
9875 | if (f.file->f_op != &io_uring_fops) | |
9876 | goto out_fput; | |
9877 | ||
9878 | ctx = f.file->private_data; | |
9879 | ||
b6c23dd5 PB |
9880 | io_run_task_work(); |
9881 | ||
edafccee JA |
9882 | mutex_lock(&ctx->uring_lock); |
9883 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9884 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9885 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9886 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9887 | out_fput: |
9888 | fdput(f); | |
9889 | return ret; | |
9890 | } | |
9891 | ||
2b188cc1 JA |
9892 | static int __init io_uring_init(void) |
9893 | { | |
d7f62e82 SM |
9894 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9895 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9896 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9897 | } while (0) | |
9898 | ||
9899 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9900 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9901 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9902 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9903 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9904 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9905 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9906 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9907 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9908 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9909 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9910 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9911 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9912 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9913 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9914 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9915 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9916 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9917 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9918 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9919 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9920 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9921 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9922 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9923 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9924 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9925 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9926 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9927 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9928 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9929 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9930 | |
d3656344 | 9931 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9932 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9933 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9934 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9935 | return 0; |
9936 | }; | |
9937 | __initcall(io_uring_init); |