]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
d068b506 | 14 | * through a control-dependency in io_get_cqe (smp_store_release to |
1e84b97b SB |
15 | * store head will do). Failure to do so could lead to reading invalid |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
4ce8ad95 | 92 | #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8 |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
101 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
102 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 103 | |
2d091d62 PB |
104 | #define IO_RSRC_TAG_TABLE_SHIFT 9 |
105 | #define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT) | |
106 | #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1) | |
107 | ||
489809e2 PB |
108 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
109 | ||
b16fed66 PB |
110 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
111 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
112 | IOSQE_BUFFER_SELECT) | |
c854357b PB |
113 | #define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \ |
114 | REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS) | |
b16fed66 | 115 | |
09899b19 PB |
116 | #define IO_TCTX_REFS_CACHE_NR (1U << 10) |
117 | ||
2b188cc1 JA |
118 | struct io_uring { |
119 | u32 head ____cacheline_aligned_in_smp; | |
120 | u32 tail ____cacheline_aligned_in_smp; | |
121 | }; | |
122 | ||
1e84b97b | 123 | /* |
75b28aff HV |
124 | * This data is shared with the application through the mmap at offsets |
125 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
126 | * |
127 | * The offsets to the member fields are published through struct | |
128 | * io_sqring_offsets when calling io_uring_setup. | |
129 | */ | |
75b28aff | 130 | struct io_rings { |
1e84b97b SB |
131 | /* |
132 | * Head and tail offsets into the ring; the offsets need to be | |
133 | * masked to get valid indices. | |
134 | * | |
75b28aff HV |
135 | * The kernel controls head of the sq ring and the tail of the cq ring, |
136 | * and the application controls tail of the sq ring and the head of the | |
137 | * cq ring. | |
1e84b97b | 138 | */ |
75b28aff | 139 | struct io_uring sq, cq; |
1e84b97b | 140 | /* |
75b28aff | 141 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
142 | * ring_entries - 1) |
143 | */ | |
75b28aff HV |
144 | u32 sq_ring_mask, cq_ring_mask; |
145 | /* Ring sizes (constant, power of 2) */ | |
146 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
147 | /* |
148 | * Number of invalid entries dropped by the kernel due to | |
149 | * invalid index stored in array | |
150 | * | |
151 | * Written by the kernel, shouldn't be modified by the | |
152 | * application (i.e. get number of "new events" by comparing to | |
153 | * cached value). | |
154 | * | |
155 | * After a new SQ head value was read by the application this | |
156 | * counter includes all submissions that were dropped reaching | |
157 | * the new SQ head (and possibly more). | |
158 | */ | |
75b28aff | 159 | u32 sq_dropped; |
1e84b97b | 160 | /* |
0d9b5b3a | 161 | * Runtime SQ flags |
1e84b97b SB |
162 | * |
163 | * Written by the kernel, shouldn't be modified by the | |
164 | * application. | |
165 | * | |
166 | * The application needs a full memory barrier before checking | |
167 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
168 | */ | |
75b28aff | 169 | u32 sq_flags; |
0d9b5b3a SG |
170 | /* |
171 | * Runtime CQ flags | |
172 | * | |
173 | * Written by the application, shouldn't be modified by the | |
174 | * kernel. | |
175 | */ | |
fe7e3257 | 176 | u32 cq_flags; |
1e84b97b SB |
177 | /* |
178 | * Number of completion events lost because the queue was full; | |
179 | * this should be avoided by the application by making sure | |
0b4295b5 | 180 | * there are not more requests pending than there is space in |
1e84b97b SB |
181 | * the completion queue. |
182 | * | |
183 | * Written by the kernel, shouldn't be modified by the | |
184 | * application (i.e. get number of "new events" by comparing to | |
185 | * cached value). | |
186 | * | |
187 | * As completion events come in out of order this counter is not | |
188 | * ordered with any other data. | |
189 | */ | |
75b28aff | 190 | u32 cq_overflow; |
1e84b97b SB |
191 | /* |
192 | * Ring buffer of completion events. | |
193 | * | |
194 | * The kernel writes completion events fresh every time they are | |
195 | * produced, so the application is allowed to modify pending | |
196 | * entries. | |
197 | */ | |
75b28aff | 198 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
199 | }; |
200 | ||
45d189c6 PB |
201 | enum io_uring_cmd_flags { |
202 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 203 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
204 | }; |
205 | ||
edafccee JA |
206 | struct io_mapped_ubuf { |
207 | u64 ubuf; | |
4751f53d | 208 | u64 ubuf_end; |
edafccee | 209 | unsigned int nr_bvecs; |
de293938 | 210 | unsigned long acct_pages; |
41edf1a5 | 211 | struct bio_vec bvec[]; |
edafccee JA |
212 | }; |
213 | ||
50238531 BM |
214 | struct io_ring_ctx; |
215 | ||
6c2450ae PB |
216 | struct io_overflow_cqe { |
217 | struct io_uring_cqe cqe; | |
218 | struct list_head list; | |
219 | }; | |
220 | ||
a04b0ac0 PB |
221 | struct io_fixed_file { |
222 | /* file * with additional FFS_* flags */ | |
223 | unsigned long file_ptr; | |
224 | }; | |
225 | ||
269bbe5f BM |
226 | struct io_rsrc_put { |
227 | struct list_head list; | |
b60c8dce | 228 | u64 tag; |
50238531 BM |
229 | union { |
230 | void *rsrc; | |
231 | struct file *file; | |
bd54b6fe | 232 | struct io_mapped_ubuf *buf; |
50238531 | 233 | }; |
269bbe5f BM |
234 | }; |
235 | ||
aeca241b PB |
236 | struct io_file_table { |
237 | /* two level table */ | |
238 | struct io_fixed_file **files; | |
31b51510 JA |
239 | }; |
240 | ||
b895c9a6 | 241 | struct io_rsrc_node { |
05589553 XW |
242 | struct percpu_ref refs; |
243 | struct list_head node; | |
269bbe5f | 244 | struct list_head rsrc_list; |
b895c9a6 | 245 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 246 | struct llist_node llist; |
e297822b | 247 | bool done; |
05589553 XW |
248 | }; |
249 | ||
40ae0ff7 PB |
250 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
251 | ||
b895c9a6 | 252 | struct io_rsrc_data { |
05f3fb3c JA |
253 | struct io_ring_ctx *ctx; |
254 | ||
2d091d62 PB |
255 | u64 **tags; |
256 | unsigned int nr; | |
40ae0ff7 | 257 | rsrc_put_fn *do_put; |
3e942498 | 258 | atomic_t refs; |
05f3fb3c | 259 | struct completion done; |
8bad28d8 | 260 | bool quiesce; |
05f3fb3c JA |
261 | }; |
262 | ||
5a2e745d JA |
263 | struct io_buffer { |
264 | struct list_head list; | |
265 | __u64 addr; | |
d1f82808 | 266 | __u32 len; |
5a2e745d JA |
267 | __u16 bid; |
268 | }; | |
269 | ||
21b55dbc SG |
270 | struct io_restriction { |
271 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
272 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
273 | u8 sqe_flags_allowed; | |
274 | u8 sqe_flags_required; | |
7e84e1c7 | 275 | bool registered; |
21b55dbc SG |
276 | }; |
277 | ||
37d1e2e3 JA |
278 | enum { |
279 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
280 | IO_SQ_THREAD_SHOULD_PARK, | |
281 | }; | |
282 | ||
534ca6d6 JA |
283 | struct io_sq_data { |
284 | refcount_t refs; | |
9e138a48 | 285 | atomic_t park_pending; |
09a6f4ef | 286 | struct mutex lock; |
69fb2131 JA |
287 | |
288 | /* ctx's that are using this sqd */ | |
289 | struct list_head ctx_list; | |
69fb2131 | 290 | |
534ca6d6 JA |
291 | struct task_struct *thread; |
292 | struct wait_queue_head wait; | |
08369246 XW |
293 | |
294 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
295 | int sq_cpu; |
296 | pid_t task_pid; | |
5c2469e0 | 297 | pid_t task_tgid; |
37d1e2e3 JA |
298 | |
299 | unsigned long state; | |
37d1e2e3 | 300 | struct completion exited; |
534ca6d6 JA |
301 | }; |
302 | ||
258b29a9 | 303 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 304 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 305 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 306 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
307 | |
308 | struct io_comp_state { | |
6dd0be1e | 309 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 310 | unsigned int nr; |
c7dae4ba | 311 | /* inline/task_work completion list, under ->uring_lock */ |
1b4c351f | 312 | struct list_head free_list; |
258b29a9 PB |
313 | }; |
314 | ||
a1ab7b35 PB |
315 | struct io_submit_link { |
316 | struct io_kiocb *head; | |
317 | struct io_kiocb *last; | |
318 | }; | |
319 | ||
258b29a9 PB |
320 | struct io_submit_state { |
321 | struct blk_plug plug; | |
a1ab7b35 | 322 | struct io_submit_link link; |
258b29a9 PB |
323 | |
324 | /* | |
325 | * io_kiocb alloc cache | |
326 | */ | |
bf019da7 | 327 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
328 | unsigned int free_reqs; |
329 | ||
330 | bool plug_started; | |
331 | ||
332 | /* | |
333 | * Batch completion logic | |
334 | */ | |
335 | struct io_comp_state comp; | |
336 | ||
337 | /* | |
338 | * File reference cache | |
339 | */ | |
340 | struct file *file; | |
341 | unsigned int fd; | |
342 | unsigned int file_refs; | |
343 | unsigned int ios_left; | |
344 | }; | |
345 | ||
2b188cc1 | 346 | struct io_ring_ctx { |
b52ecf8c | 347 | /* const or read-mostly hot data */ |
2b188cc1 JA |
348 | struct { |
349 | struct percpu_ref refs; | |
2b188cc1 | 350 | |
b52ecf8c | 351 | struct io_rings *rings; |
2b188cc1 | 352 | unsigned int flags; |
e1d85334 | 353 | unsigned int compat: 1; |
e1d85334 RD |
354 | unsigned int drain_next: 1; |
355 | unsigned int eventfd_async: 1; | |
21b55dbc | 356 | unsigned int restricted: 1; |
f18ee4cf | 357 | unsigned int off_timeout_used: 1; |
10c66904 | 358 | unsigned int drain_active: 1; |
b52ecf8c | 359 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 360 | |
7f1129d2 | 361 | /* submission data */ |
b52ecf8c | 362 | struct { |
0499e582 PB |
363 | struct mutex uring_lock; |
364 | ||
75b28aff HV |
365 | /* |
366 | * Ring buffer of indices into array of io_uring_sqe, which is | |
367 | * mmapped by the application using the IORING_OFF_SQES offset. | |
368 | * | |
369 | * This indirection could e.g. be used to assign fixed | |
370 | * io_uring_sqe entries to operations and only submit them to | |
371 | * the queue when needed. | |
372 | * | |
373 | * The kernel modifies neither the indices array nor the entries | |
374 | * array. | |
375 | */ | |
376 | u32 *sq_array; | |
c7af47cf | 377 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
378 | unsigned cached_sq_head; |
379 | unsigned sq_entries; | |
de0617e4 | 380 | struct list_head defer_list; |
7f1129d2 PB |
381 | |
382 | /* | |
383 | * Fixed resources fast path, should be accessed only under | |
384 | * uring_lock, and updated through io_uring_register(2) | |
385 | */ | |
386 | struct io_rsrc_node *rsrc_node; | |
387 | struct io_file_table file_table; | |
388 | unsigned nr_user_files; | |
389 | unsigned nr_user_bufs; | |
390 | struct io_mapped_ubuf **user_bufs; | |
391 | ||
392 | struct io_submit_state submit_state; | |
5262f567 | 393 | struct list_head timeout_list; |
1d7bb1d5 | 394 | struct list_head cq_overflow_list; |
7f1129d2 PB |
395 | struct xarray io_buffers; |
396 | struct xarray personalities; | |
397 | u32 pers_next; | |
398 | unsigned sq_thread_idle; | |
2b188cc1 JA |
399 | } ____cacheline_aligned_in_smp; |
400 | ||
d0acdee2 PB |
401 | /* IRQ completion list, under ->completion_lock */ |
402 | struct list_head locked_free_list; | |
403 | unsigned int locked_free_nr; | |
3c1a2ead | 404 | |
7c30f36a | 405 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
406 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
407 | ||
90554200 | 408 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 409 | struct list_head sqd_list; |
75b28aff | 410 | |
5ed7a37d PB |
411 | unsigned long check_cq_overflow; |
412 | ||
206aefde JA |
413 | struct { |
414 | unsigned cached_cq_tail; | |
415 | unsigned cq_entries; | |
0499e582 | 416 | struct eventfd_ctx *cq_ev_fd; |
311997b3 | 417 | struct wait_queue_head poll_wait; |
0499e582 PB |
418 | struct wait_queue_head cq_wait; |
419 | unsigned cq_extra; | |
420 | atomic_t cq_timeouts; | |
206aefde | 421 | struct fasync_struct *cq_fasync; |
0499e582 | 422 | unsigned cq_last_tm_flush; |
206aefde | 423 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 424 | |
2b188cc1 JA |
425 | struct { |
426 | spinlock_t completion_lock; | |
e94f141b | 427 | |
def596e9 | 428 | /* |
540e32a0 | 429 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
430 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
431 | * For SQPOLL, only the single threaded io_sq_thread() will | |
432 | * manipulate the list, hence no extra locking is needed there. | |
433 | */ | |
540e32a0 | 434 | struct list_head iopoll_list; |
78076bb6 JA |
435 | struct hlist_head *cancel_hash; |
436 | unsigned cancel_hash_bits; | |
e94f141b | 437 | bool poll_multi_file; |
2b188cc1 | 438 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 439 | |
21b55dbc | 440 | struct io_restriction restrictions; |
3c1a2ead | 441 | |
b13a8918 PB |
442 | /* slow path rsrc auxilary data, used by update/register */ |
443 | struct { | |
444 | struct io_rsrc_node *rsrc_backup_node; | |
445 | struct io_mapped_ubuf *dummy_ubuf; | |
446 | struct io_rsrc_data *file_data; | |
447 | struct io_rsrc_data *buf_data; | |
448 | ||
449 | struct delayed_work rsrc_put_work; | |
450 | struct llist_head rsrc_put_llist; | |
451 | struct list_head rsrc_ref_list; | |
452 | spinlock_t rsrc_ref_lock; | |
453 | }; | |
454 | ||
3c1a2ead | 455 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
456 | struct { |
457 | #if defined(CONFIG_UNIX) | |
458 | struct socket *ring_sock; | |
459 | #endif | |
460 | /* hashed buffered write serialization */ | |
461 | struct io_wq_hash *hash_map; | |
462 | ||
463 | /* Only used for accounting purposes */ | |
464 | struct user_struct *user; | |
465 | struct mm_struct *mm_account; | |
466 | ||
467 | /* ctx exit and cancelation */ | |
468 | struct callback_head *exit_task_work; | |
469 | struct work_struct exit_work; | |
470 | struct list_head tctx_list; | |
471 | struct completion ref_comp; | |
472 | }; | |
2b188cc1 JA |
473 | }; |
474 | ||
53e043b2 SM |
475 | struct io_uring_task { |
476 | /* submission side */ | |
09899b19 | 477 | int cached_refs; |
53e043b2 SM |
478 | struct xarray xa; |
479 | struct wait_queue_head wait; | |
ee53fb2b SM |
480 | const struct io_ring_ctx *last; |
481 | struct io_wq *io_wq; | |
53e043b2 | 482 | struct percpu_counter inflight; |
b303fe2e | 483 | atomic_t inflight_tracked; |
53e043b2 | 484 | atomic_t in_idle; |
53e043b2 SM |
485 | |
486 | spinlock_t task_lock; | |
487 | struct io_wq_work_list task_list; | |
488 | unsigned long task_state; | |
489 | struct callback_head task_work; | |
490 | }; | |
491 | ||
09bb8394 JA |
492 | /* |
493 | * First field must be the file pointer in all the | |
494 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
495 | */ | |
221c5eb2 JA |
496 | struct io_poll_iocb { |
497 | struct file *file; | |
018043be | 498 | struct wait_queue_head *head; |
221c5eb2 | 499 | __poll_t events; |
8c838788 | 500 | bool done; |
221c5eb2 | 501 | bool canceled; |
392edb45 | 502 | struct wait_queue_entry wait; |
221c5eb2 JA |
503 | }; |
504 | ||
9d805892 | 505 | struct io_poll_update { |
018043be | 506 | struct file *file; |
9d805892 PB |
507 | u64 old_user_data; |
508 | u64 new_user_data; | |
509 | __poll_t events; | |
b69de288 JA |
510 | bool update_events; |
511 | bool update_user_data; | |
018043be PB |
512 | }; |
513 | ||
b5dba59e JA |
514 | struct io_close { |
515 | struct file *file; | |
b5dba59e JA |
516 | int fd; |
517 | }; | |
518 | ||
ad8a48ac JA |
519 | struct io_timeout_data { |
520 | struct io_kiocb *req; | |
521 | struct hrtimer timer; | |
522 | struct timespec64 ts; | |
523 | enum hrtimer_mode mode; | |
524 | }; | |
525 | ||
8ed8d3c3 JA |
526 | struct io_accept { |
527 | struct file *file; | |
528 | struct sockaddr __user *addr; | |
529 | int __user *addr_len; | |
530 | int flags; | |
09952e3e | 531 | unsigned long nofile; |
8ed8d3c3 JA |
532 | }; |
533 | ||
534 | struct io_sync { | |
535 | struct file *file; | |
536 | loff_t len; | |
537 | loff_t off; | |
538 | int flags; | |
d63d1b5e | 539 | int mode; |
8ed8d3c3 JA |
540 | }; |
541 | ||
fbf23849 JA |
542 | struct io_cancel { |
543 | struct file *file; | |
544 | u64 addr; | |
545 | }; | |
546 | ||
b29472ee JA |
547 | struct io_timeout { |
548 | struct file *file; | |
bfe68a22 PB |
549 | u32 off; |
550 | u32 target_seq; | |
135fcde8 | 551 | struct list_head list; |
90cd7e42 PB |
552 | /* head of the link, used by linked timeouts only */ |
553 | struct io_kiocb *head; | |
b29472ee JA |
554 | }; |
555 | ||
0bdf7a2d PB |
556 | struct io_timeout_rem { |
557 | struct file *file; | |
558 | u64 addr; | |
9c8e11b3 PB |
559 | |
560 | /* timeout update */ | |
561 | struct timespec64 ts; | |
562 | u32 flags; | |
0bdf7a2d PB |
563 | }; |
564 | ||
9adbd45d JA |
565 | struct io_rw { |
566 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
567 | struct kiocb kiocb; | |
568 | u64 addr; | |
569 | u64 len; | |
570 | }; | |
571 | ||
3fbb51c1 JA |
572 | struct io_connect { |
573 | struct file *file; | |
574 | struct sockaddr __user *addr; | |
575 | int addr_len; | |
576 | }; | |
577 | ||
e47293fd JA |
578 | struct io_sr_msg { |
579 | struct file *file; | |
fddaface | 580 | union { |
4af3417a PB |
581 | struct compat_msghdr __user *umsg_compat; |
582 | struct user_msghdr __user *umsg; | |
583 | void __user *buf; | |
fddaface | 584 | }; |
e47293fd | 585 | int msg_flags; |
bcda7baa | 586 | int bgid; |
fddaface | 587 | size_t len; |
bcda7baa | 588 | struct io_buffer *kbuf; |
e47293fd JA |
589 | }; |
590 | ||
15b71abe JA |
591 | struct io_open { |
592 | struct file *file; | |
593 | int dfd; | |
15b71abe | 594 | struct filename *filename; |
c12cedf2 | 595 | struct open_how how; |
4022e7af | 596 | unsigned long nofile; |
15b71abe JA |
597 | }; |
598 | ||
269bbe5f | 599 | struct io_rsrc_update { |
05f3fb3c JA |
600 | struct file *file; |
601 | u64 arg; | |
602 | u32 nr_args; | |
603 | u32 offset; | |
604 | }; | |
605 | ||
4840e418 JA |
606 | struct io_fadvise { |
607 | struct file *file; | |
608 | u64 offset; | |
609 | u32 len; | |
610 | u32 advice; | |
611 | }; | |
612 | ||
c1ca757b JA |
613 | struct io_madvise { |
614 | struct file *file; | |
615 | u64 addr; | |
616 | u32 len; | |
617 | u32 advice; | |
618 | }; | |
619 | ||
3e4827b0 JA |
620 | struct io_epoll { |
621 | struct file *file; | |
622 | int epfd; | |
623 | int op; | |
624 | int fd; | |
625 | struct epoll_event event; | |
e47293fd JA |
626 | }; |
627 | ||
7d67af2c PB |
628 | struct io_splice { |
629 | struct file *file_out; | |
630 | struct file *file_in; | |
631 | loff_t off_out; | |
632 | loff_t off_in; | |
633 | u64 len; | |
634 | unsigned int flags; | |
635 | }; | |
636 | ||
ddf0322d JA |
637 | struct io_provide_buf { |
638 | struct file *file; | |
639 | __u64 addr; | |
38134ada | 640 | __u32 len; |
ddf0322d JA |
641 | __u32 bgid; |
642 | __u16 nbufs; | |
643 | __u16 bid; | |
644 | }; | |
645 | ||
1d9e1288 BM |
646 | struct io_statx { |
647 | struct file *file; | |
648 | int dfd; | |
649 | unsigned int mask; | |
650 | unsigned int flags; | |
e62753e4 | 651 | const char __user *filename; |
1d9e1288 BM |
652 | struct statx __user *buffer; |
653 | }; | |
654 | ||
36f4fa68 JA |
655 | struct io_shutdown { |
656 | struct file *file; | |
657 | int how; | |
658 | }; | |
659 | ||
80a261fd JA |
660 | struct io_rename { |
661 | struct file *file; | |
662 | int old_dfd; | |
663 | int new_dfd; | |
664 | struct filename *oldpath; | |
665 | struct filename *newpath; | |
666 | int flags; | |
667 | }; | |
668 | ||
14a1143b JA |
669 | struct io_unlink { |
670 | struct file *file; | |
671 | int dfd; | |
672 | int flags; | |
673 | struct filename *filename; | |
674 | }; | |
675 | ||
3ca405eb PB |
676 | struct io_completion { |
677 | struct file *file; | |
678 | struct list_head list; | |
8c3f9cd1 | 679 | u32 cflags; |
3ca405eb PB |
680 | }; |
681 | ||
f499a021 JA |
682 | struct io_async_connect { |
683 | struct sockaddr_storage address; | |
684 | }; | |
685 | ||
03b1230c JA |
686 | struct io_async_msghdr { |
687 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
688 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
689 | struct iovec *free_iov; | |
03b1230c JA |
690 | struct sockaddr __user *uaddr; |
691 | struct msghdr msg; | |
b537916c | 692 | struct sockaddr_storage addr; |
03b1230c JA |
693 | }; |
694 | ||
f67676d1 JA |
695 | struct io_async_rw { |
696 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
697 | const struct iovec *free_iovec; |
698 | struct iov_iter iter; | |
227c0c96 | 699 | size_t bytes_done; |
bcf5a063 | 700 | struct wait_page_queue wpq; |
f67676d1 JA |
701 | }; |
702 | ||
6b47ee6e PB |
703 | enum { |
704 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
705 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
706 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
707 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
708 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 709 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 710 | |
dddca226 | 711 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 712 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
713 | REQ_F_INFLIGHT_BIT, |
714 | REQ_F_CUR_POS_BIT, | |
715 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 716 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 717 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 718 | REQ_F_POLLED_BIT, |
bcda7baa | 719 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 720 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 721 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 722 | REQ_F_REISSUE_BIT, |
8c130827 | 723 | REQ_F_DONT_REISSUE_BIT, |
b8e64b53 | 724 | REQ_F_CREDS_BIT, |
7b29f92d JA |
725 | /* keep async read/write and isreg together and in order */ |
726 | REQ_F_ASYNC_READ_BIT, | |
727 | REQ_F_ASYNC_WRITE_BIT, | |
728 | REQ_F_ISREG_BIT, | |
84557871 JA |
729 | |
730 | /* not a real bit, just to check we're not overflowing the space */ | |
731 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
732 | }; |
733 | ||
734 | enum { | |
735 | /* ctx owns file */ | |
736 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
737 | /* drain existing IO first */ | |
738 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
739 | /* linked sqes */ | |
740 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
741 | /* doesn't sever on completion < 0 */ | |
742 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
743 | /* IOSQE_ASYNC */ | |
744 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
745 | /* IOSQE_BUFFER_SELECT */ |
746 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 747 | |
6b47ee6e | 748 | /* fail rest of links */ |
93d2bcd2 | 749 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 750 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
751 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
752 | /* read/write uses file position */ | |
753 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
754 | /* must not punt to workers */ | |
755 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 756 | /* has or had linked timeout */ |
6b47ee6e | 757 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
758 | /* needs cleanup */ |
759 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
760 | /* already went through poll handler */ |
761 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
762 | /* buffer already selected */ |
763 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
764 | /* linked timeout is active, i.e. prepared by link's head */ |
765 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
766 | /* completion is deferred through io_comp_state */ |
767 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
768 | /* caller should reissue async */ |
769 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
770 | /* don't attempt request reissue, see io_rw_reissue() */ |
771 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
772 | /* supports async reads */ |
773 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
774 | /* supports async writes */ | |
775 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
776 | /* regular file */ | |
777 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
778 | /* has creds assigned */ |
779 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
d7718a9d JA |
780 | }; |
781 | ||
782 | struct async_poll { | |
783 | struct io_poll_iocb poll; | |
807abcb0 | 784 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
785 | }; |
786 | ||
7cbf1722 JA |
787 | struct io_task_work { |
788 | struct io_wq_work_node node; | |
789 | task_work_func_t func; | |
790 | }; | |
791 | ||
992da01a PB |
792 | enum { |
793 | IORING_RSRC_FILE = 0, | |
794 | IORING_RSRC_BUFFER = 1, | |
795 | }; | |
796 | ||
09bb8394 JA |
797 | /* |
798 | * NOTE! Each of the iocb union members has the file pointer | |
799 | * as the first entry in their struct definition. So you can | |
800 | * access the file pointer through any of the sub-structs, | |
801 | * or directly as just 'ki_filp' in this struct. | |
802 | */ | |
2b188cc1 | 803 | struct io_kiocb { |
221c5eb2 | 804 | union { |
09bb8394 | 805 | struct file *file; |
9adbd45d | 806 | struct io_rw rw; |
221c5eb2 | 807 | struct io_poll_iocb poll; |
9d805892 | 808 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
809 | struct io_accept accept; |
810 | struct io_sync sync; | |
fbf23849 | 811 | struct io_cancel cancel; |
b29472ee | 812 | struct io_timeout timeout; |
0bdf7a2d | 813 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 814 | struct io_connect connect; |
e47293fd | 815 | struct io_sr_msg sr_msg; |
15b71abe | 816 | struct io_open open; |
b5dba59e | 817 | struct io_close close; |
269bbe5f | 818 | struct io_rsrc_update rsrc_update; |
4840e418 | 819 | struct io_fadvise fadvise; |
c1ca757b | 820 | struct io_madvise madvise; |
3e4827b0 | 821 | struct io_epoll epoll; |
7d67af2c | 822 | struct io_splice splice; |
ddf0322d | 823 | struct io_provide_buf pbuf; |
1d9e1288 | 824 | struct io_statx statx; |
36f4fa68 | 825 | struct io_shutdown shutdown; |
80a261fd | 826 | struct io_rename rename; |
14a1143b | 827 | struct io_unlink unlink; |
3ca405eb PB |
828 | /* use only after cleaning per-op data, see io_clean_op() */ |
829 | struct io_completion compl; | |
221c5eb2 | 830 | }; |
2b188cc1 | 831 | |
e8c2bc1f JA |
832 | /* opcode allocated if it needs to store data for async defer */ |
833 | void *async_data; | |
d625c6ee | 834 | u8 opcode; |
65a6543d XW |
835 | /* polled IO has completed */ |
836 | u8 iopoll_completed; | |
2b188cc1 | 837 | |
4f4eeba8 | 838 | u16 buf_index; |
9cf7c104 | 839 | u32 result; |
4f4eeba8 | 840 | |
010e8e6b PB |
841 | struct io_ring_ctx *ctx; |
842 | unsigned int flags; | |
abc54d63 | 843 | atomic_t refs; |
010e8e6b PB |
844 | struct task_struct *task; |
845 | u64 user_data; | |
d7718a9d | 846 | |
f2f87370 | 847 | struct io_kiocb *link; |
269bbe5f | 848 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 849 | |
b303fe2e | 850 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 851 | struct list_head inflight_entry; |
7cbf1722 JA |
852 | union { |
853 | struct io_task_work io_task_work; | |
854 | struct callback_head task_work; | |
855 | }; | |
010e8e6b PB |
856 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
857 | struct hlist_node hash_node; | |
858 | struct async_poll *apoll; | |
859 | struct io_wq_work work; | |
fe7e3257 | 860 | const struct cred *creds; |
c10d1f98 | 861 | |
eae071c9 PB |
862 | /* store used ubuf, so we can prevent reloading */ |
863 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 864 | }; |
05589553 | 865 | |
13bf43f5 PB |
866 | struct io_tctx_node { |
867 | struct list_head ctx_node; | |
868 | struct task_struct *task; | |
13bf43f5 PB |
869 | struct io_ring_ctx *ctx; |
870 | }; | |
871 | ||
27dc8338 PB |
872 | struct io_defer_entry { |
873 | struct list_head list; | |
874 | struct io_kiocb *req; | |
9cf7c104 | 875 | u32 seq; |
2b188cc1 JA |
876 | }; |
877 | ||
d3656344 | 878 | struct io_op_def { |
d3656344 JA |
879 | /* needs req->file assigned */ |
880 | unsigned needs_file : 1; | |
d3656344 JA |
881 | /* hash wq insertion if file is a regular file */ |
882 | unsigned hash_reg_file : 1; | |
883 | /* unbound wq insertion if file is a non-regular file */ | |
884 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
885 | /* opcode is not supported by this kernel */ |
886 | unsigned not_supported : 1; | |
8a72758c JA |
887 | /* set if opcode supports polled "wait" */ |
888 | unsigned pollin : 1; | |
889 | unsigned pollout : 1; | |
bcda7baa JA |
890 | /* op supports buffer selection */ |
891 | unsigned buffer_select : 1; | |
26f0505a PB |
892 | /* do prep async if is going to be punted */ |
893 | unsigned needs_async_setup : 1; | |
27926b68 JA |
894 | /* should block plug */ |
895 | unsigned plug : 1; | |
e8c2bc1f JA |
896 | /* size of async data needed, if any */ |
897 | unsigned short async_size; | |
d3656344 JA |
898 | }; |
899 | ||
0918682b | 900 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
901 | [IORING_OP_NOP] = {}, |
902 | [IORING_OP_READV] = { | |
d3656344 JA |
903 | .needs_file = 1, |
904 | .unbound_nonreg_file = 1, | |
8a72758c | 905 | .pollin = 1, |
4d954c25 | 906 | .buffer_select = 1, |
26f0505a | 907 | .needs_async_setup = 1, |
27926b68 | 908 | .plug = 1, |
e8c2bc1f | 909 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 910 | }, |
0463b6c5 | 911 | [IORING_OP_WRITEV] = { |
d3656344 JA |
912 | .needs_file = 1, |
913 | .hash_reg_file = 1, | |
914 | .unbound_nonreg_file = 1, | |
8a72758c | 915 | .pollout = 1, |
26f0505a | 916 | .needs_async_setup = 1, |
27926b68 | 917 | .plug = 1, |
e8c2bc1f | 918 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 919 | }, |
0463b6c5 | 920 | [IORING_OP_FSYNC] = { |
d3656344 JA |
921 | .needs_file = 1, |
922 | }, | |
0463b6c5 | 923 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
924 | .needs_file = 1, |
925 | .unbound_nonreg_file = 1, | |
8a72758c | 926 | .pollin = 1, |
27926b68 | 927 | .plug = 1, |
e8c2bc1f | 928 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 929 | }, |
0463b6c5 | 930 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
931 | .needs_file = 1, |
932 | .hash_reg_file = 1, | |
933 | .unbound_nonreg_file = 1, | |
8a72758c | 934 | .pollout = 1, |
27926b68 | 935 | .plug = 1, |
e8c2bc1f | 936 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 937 | }, |
0463b6c5 | 938 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
939 | .needs_file = 1, |
940 | .unbound_nonreg_file = 1, | |
941 | }, | |
0463b6c5 PB |
942 | [IORING_OP_POLL_REMOVE] = {}, |
943 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
944 | .needs_file = 1, |
945 | }, | |
0463b6c5 | 946 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
947 | .needs_file = 1, |
948 | .unbound_nonreg_file = 1, | |
8a72758c | 949 | .pollout = 1, |
26f0505a | 950 | .needs_async_setup = 1, |
e8c2bc1f | 951 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 952 | }, |
0463b6c5 | 953 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
954 | .needs_file = 1, |
955 | .unbound_nonreg_file = 1, | |
8a72758c | 956 | .pollin = 1, |
52de1fe1 | 957 | .buffer_select = 1, |
26f0505a | 958 | .needs_async_setup = 1, |
e8c2bc1f | 959 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 960 | }, |
0463b6c5 | 961 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 962 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 963 | }, |
9c8e11b3 PB |
964 | [IORING_OP_TIMEOUT_REMOVE] = { |
965 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 966 | }, |
0463b6c5 | 967 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
968 | .needs_file = 1, |
969 | .unbound_nonreg_file = 1, | |
8a72758c | 970 | .pollin = 1, |
d3656344 | 971 | }, |
0463b6c5 PB |
972 | [IORING_OP_ASYNC_CANCEL] = {}, |
973 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 974 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 975 | }, |
0463b6c5 | 976 | [IORING_OP_CONNECT] = { |
d3656344 JA |
977 | .needs_file = 1, |
978 | .unbound_nonreg_file = 1, | |
8a72758c | 979 | .pollout = 1, |
26f0505a | 980 | .needs_async_setup = 1, |
e8c2bc1f | 981 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 982 | }, |
0463b6c5 | 983 | [IORING_OP_FALLOCATE] = { |
d3656344 | 984 | .needs_file = 1, |
d3656344 | 985 | }, |
44526bed JA |
986 | [IORING_OP_OPENAT] = {}, |
987 | [IORING_OP_CLOSE] = {}, | |
988 | [IORING_OP_FILES_UPDATE] = {}, | |
989 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 990 | [IORING_OP_READ] = { |
3a6820f2 JA |
991 | .needs_file = 1, |
992 | .unbound_nonreg_file = 1, | |
8a72758c | 993 | .pollin = 1, |
bcda7baa | 994 | .buffer_select = 1, |
27926b68 | 995 | .plug = 1, |
e8c2bc1f | 996 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 997 | }, |
0463b6c5 | 998 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
999 | .needs_file = 1, |
1000 | .unbound_nonreg_file = 1, | |
8a72758c | 1001 | .pollout = 1, |
27926b68 | 1002 | .plug = 1, |
e8c2bc1f | 1003 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1004 | }, |
0463b6c5 | 1005 | [IORING_OP_FADVISE] = { |
4840e418 | 1006 | .needs_file = 1, |
c1ca757b | 1007 | }, |
44526bed | 1008 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1009 | [IORING_OP_SEND] = { |
fddaface JA |
1010 | .needs_file = 1, |
1011 | .unbound_nonreg_file = 1, | |
8a72758c | 1012 | .pollout = 1, |
fddaface | 1013 | }, |
0463b6c5 | 1014 | [IORING_OP_RECV] = { |
fddaface JA |
1015 | .needs_file = 1, |
1016 | .unbound_nonreg_file = 1, | |
8a72758c | 1017 | .pollin = 1, |
bcda7baa | 1018 | .buffer_select = 1, |
fddaface | 1019 | }, |
0463b6c5 | 1020 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1021 | }, |
3e4827b0 JA |
1022 | [IORING_OP_EPOLL_CTL] = { |
1023 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1024 | }, |
7d67af2c PB |
1025 | [IORING_OP_SPLICE] = { |
1026 | .needs_file = 1, | |
1027 | .hash_reg_file = 1, | |
1028 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1029 | }, |
1030 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1031 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1032 | [IORING_OP_TEE] = { |
1033 | .needs_file = 1, | |
1034 | .hash_reg_file = 1, | |
1035 | .unbound_nonreg_file = 1, | |
1036 | }, | |
36f4fa68 JA |
1037 | [IORING_OP_SHUTDOWN] = { |
1038 | .needs_file = 1, | |
1039 | }, | |
44526bed JA |
1040 | [IORING_OP_RENAMEAT] = {}, |
1041 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1042 | }; |
1043 | ||
7a612350 | 1044 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1045 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1046 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1047 | struct task_struct *task, | |
3dd0c97a | 1048 | bool cancel_all); |
78cc687b | 1049 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
b895c9a6 | 1050 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1051 | |
d4d19c19 PB |
1052 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1053 | long res, unsigned int cflags); | |
ec9c02ad | 1054 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1055 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1056 | static void io_dismantle_req(struct io_kiocb *req); |
1057 | static void io_put_task(struct task_struct *task, int nr); | |
94ae5e77 JA |
1058 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1059 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
fdecb662 | 1060 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1061 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1062 | unsigned nr_args); |
68fb8979 | 1063 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1064 | static struct file *io_file_get(struct io_submit_state *state, |
1065 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1066 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1067 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1068 | |
907d1df3 | 1069 | static void io_req_task_queue(struct io_kiocb *req); |
2a2758f2 | 1070 | static void io_submit_flush_completions(struct io_ring_ctx *ctx); |
5082620f | 1071 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1072 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1073 | |
2b188cc1 JA |
1074 | static struct kmem_cache *req_cachep; |
1075 | ||
0918682b | 1076 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1077 | |
1078 | struct sock *io_uring_get_socket(struct file *file) | |
1079 | { | |
1080 | #if defined(CONFIG_UNIX) | |
1081 | if (file->f_op == &io_uring_fops) { | |
1082 | struct io_ring_ctx *ctx = file->private_data; | |
1083 | ||
1084 | return ctx->ring_sock->sk; | |
1085 | } | |
1086 | #endif | |
1087 | return NULL; | |
1088 | } | |
1089 | EXPORT_SYMBOL(io_uring_get_socket); | |
1090 | ||
f2f87370 PB |
1091 | #define io_for_each_link(pos, head) \ |
1092 | for (pos = (head); pos; pos = pos->link) | |
1093 | ||
b895c9a6 | 1094 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1095 | { |
1096 | struct io_ring_ctx *ctx = req->ctx; | |
1097 | ||
269bbe5f | 1098 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1099 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1100 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1101 | } |
1102 | } | |
1103 | ||
f70865db PB |
1104 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1105 | { | |
1106 | bool got = percpu_ref_tryget(ref); | |
1107 | ||
1108 | /* already at zero, wait for ->release() */ | |
1109 | if (!got) | |
1110 | wait_for_completion(compl); | |
1111 | percpu_ref_resurrect(ref); | |
1112 | if (got) | |
1113 | percpu_ref_put(ref); | |
1114 | } | |
1115 | ||
3dd0c97a PB |
1116 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1117 | bool cancel_all) | |
08d23634 PB |
1118 | { |
1119 | struct io_kiocb *req; | |
1120 | ||
68207680 | 1121 | if (task && head->task != task) |
08d23634 | 1122 | return false; |
3dd0c97a | 1123 | if (cancel_all) |
08d23634 PB |
1124 | return true; |
1125 | ||
1126 | io_for_each_link(req, head) { | |
b05a1bcd | 1127 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1128 | return true; |
08d23634 PB |
1129 | } |
1130 | return false; | |
1131 | } | |
1132 | ||
93d2bcd2 | 1133 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1134 | { |
93d2bcd2 | 1135 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1136 | } |
4a38aed2 | 1137 | |
2b188cc1 JA |
1138 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1139 | { | |
1140 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1141 | ||
0f158b4c | 1142 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1143 | } |
1144 | ||
8eb7e2d0 PB |
1145 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1146 | { | |
1147 | return !req->timeout.off; | |
1148 | } | |
1149 | ||
2b188cc1 JA |
1150 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1151 | { | |
1152 | struct io_ring_ctx *ctx; | |
78076bb6 | 1153 | int hash_bits; |
2b188cc1 JA |
1154 | |
1155 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1156 | if (!ctx) | |
1157 | return NULL; | |
1158 | ||
78076bb6 JA |
1159 | /* |
1160 | * Use 5 bits less than the max cq entries, that should give us around | |
1161 | * 32 entries per hash list if totally full and uniformly spread. | |
1162 | */ | |
1163 | hash_bits = ilog2(p->cq_entries); | |
1164 | hash_bits -= 5; | |
1165 | if (hash_bits <= 0) | |
1166 | hash_bits = 1; | |
1167 | ctx->cancel_hash_bits = hash_bits; | |
1168 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1169 | GFP_KERNEL); | |
1170 | if (!ctx->cancel_hash) | |
1171 | goto err; | |
1172 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1173 | ||
6224843d PB |
1174 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1175 | if (!ctx->dummy_ubuf) | |
1176 | goto err; | |
1177 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1178 | ctx->dummy_ubuf->ubuf = -1UL; | |
1179 | ||
21482896 | 1180 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1181 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1182 | goto err; | |
2b188cc1 JA |
1183 | |
1184 | ctx->flags = p->flags; | |
90554200 | 1185 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1186 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1187 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1188 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1189 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1190 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1191 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1192 | mutex_init(&ctx->uring_lock); |
311997b3 | 1193 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1194 | spin_lock_init(&ctx->completion_lock); |
540e32a0 | 1195 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1196 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1197 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1198 | spin_lock_init(&ctx->rsrc_ref_lock); |
1199 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1200 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1201 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1202 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1203 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
d0acdee2 | 1204 | INIT_LIST_HEAD(&ctx->locked_free_list); |
2b188cc1 | 1205 | return ctx; |
206aefde | 1206 | err: |
6224843d | 1207 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1208 | kfree(ctx->cancel_hash); |
206aefde JA |
1209 | kfree(ctx); |
1210 | return NULL; | |
2b188cc1 JA |
1211 | } |
1212 | ||
8f6ed49a PB |
1213 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1214 | { | |
1215 | struct io_rings *r = ctx->rings; | |
1216 | ||
1217 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1218 | ctx->cq_extra--; | |
1219 | } | |
1220 | ||
9cf7c104 | 1221 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1222 | { |
2bc9930e JA |
1223 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1224 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1225 | |
8f6ed49a | 1226 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1227 | } |
de0617e4 | 1228 | |
9d858b21 | 1229 | return false; |
de0617e4 JA |
1230 | } |
1231 | ||
ce3d5aae PB |
1232 | static void io_req_track_inflight(struct io_kiocb *req) |
1233 | { | |
ce3d5aae | 1234 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1235 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1236 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1237 | } |
1238 | } | |
1239 | ||
1e6fa521 JA |
1240 | static void io_prep_async_work(struct io_kiocb *req) |
1241 | { | |
1242 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1243 | struct io_ring_ctx *ctx = req->ctx; |
1244 | ||
b8e64b53 PB |
1245 | if (!(req->flags & REQ_F_CREDS)) { |
1246 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1247 | req->creds = get_current_cred(); |
b8e64b53 | 1248 | } |
003e8dcc | 1249 | |
e1d675df PB |
1250 | req->work.list.next = NULL; |
1251 | req->work.flags = 0; | |
feaadc4f PB |
1252 | if (req->flags & REQ_F_FORCE_ASYNC) |
1253 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1254 | ||
1e6fa521 JA |
1255 | if (req->flags & REQ_F_ISREG) { |
1256 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1257 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1258 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1259 | if (def->unbound_nonreg_file) |
1260 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1261 | } | |
e1d675df PB |
1262 | |
1263 | switch (req->opcode) { | |
1264 | case IORING_OP_SPLICE: | |
1265 | case IORING_OP_TEE: | |
e1d675df PB |
1266 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1267 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1268 | break; | |
1269 | } | |
561fb04a | 1270 | } |
cccf0ee8 | 1271 | |
cbdcb435 | 1272 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1273 | { |
cbdcb435 | 1274 | struct io_kiocb *cur; |
54a91f3b | 1275 | |
f2f87370 PB |
1276 | io_for_each_link(cur, req) |
1277 | io_prep_async_work(cur); | |
561fb04a JA |
1278 | } |
1279 | ||
ebf93667 | 1280 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1281 | { |
a197f664 | 1282 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1283 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1284 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1285 | |
3bfe6106 JA |
1286 | BUG_ON(!tctx); |
1287 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1288 | |
cbdcb435 PB |
1289 | /* init ->work of the whole link before punting */ |
1290 | io_prep_async_link(req); | |
d07f1e8a PB |
1291 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1292 | &req->work, req->flags); | |
ebf93667 | 1293 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1294 | if (link) |
1295 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1296 | } |
1297 | ||
1ee4160c | 1298 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1299 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1300 | { |
e8c2bc1f | 1301 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1302 | |
fd9c7bc5 | 1303 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1304 | atomic_set(&req->ctx->cq_timeouts, |
1305 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1306 | list_del_init(&req->timeout.list); |
d4d19c19 | 1307 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
216578e5 | 1308 | io_put_req_deferred(req, 1); |
5262f567 JA |
1309 | } |
1310 | } | |
1311 | ||
441b8a78 | 1312 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1313 | { |
441b8a78 | 1314 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1315 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1316 | struct io_defer_entry, list); | |
de0617e4 | 1317 | |
9cf7c104 | 1318 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1319 | break; |
27dc8338 | 1320 | list_del_init(&de->list); |
907d1df3 | 1321 | io_req_task_queue(de->req); |
27dc8338 | 1322 | kfree(de); |
441b8a78 | 1323 | } |
04518945 PB |
1324 | } |
1325 | ||
360428f8 | 1326 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1327 | { |
441b8a78 | 1328 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1329 | |
f18ee4cf | 1330 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1331 | u32 events_needed, events_got; |
360428f8 | 1332 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1333 | struct io_kiocb, timeout.list); |
de0617e4 | 1334 | |
8eb7e2d0 | 1335 | if (io_is_timeout_noseq(req)) |
360428f8 | 1336 | break; |
f010505b MDG |
1337 | |
1338 | /* | |
1339 | * Since seq can easily wrap around over time, subtract | |
1340 | * the last seq at which timeouts were flushed before comparing. | |
1341 | * Assuming not more than 2^31-1 events have happened since, | |
1342 | * these subtractions won't have wrapped, so we can check if | |
1343 | * target is in [last_seq, current_seq] by comparing the two. | |
1344 | */ | |
1345 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1346 | events_got = seq - ctx->cq_last_tm_flush; | |
1347 | if (events_got < events_needed) | |
360428f8 | 1348 | break; |
bfe68a22 | 1349 | |
135fcde8 | 1350 | list_del_init(&req->timeout.list); |
1ee4160c | 1351 | io_kill_timeout(req, 0); |
f18ee4cf | 1352 | } |
f010505b | 1353 | ctx->cq_last_tm_flush = seq; |
360428f8 | 1354 | } |
5262f567 | 1355 | |
2335f6f5 | 1356 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1357 | { |
2335f6f5 PB |
1358 | if (ctx->off_timeout_used) |
1359 | io_flush_timeouts(ctx); | |
1360 | if (ctx->drain_active) | |
1361 | io_queue_deferred(ctx); | |
1362 | } | |
1363 | ||
1364 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) | |
1365 | { | |
1366 | if (unlikely(ctx->off_timeout_used || ctx->drain_active)) | |
1367 | __io_commit_cqring_flush(ctx); | |
ec30e04b PB |
1368 | /* order cqe stores with ring update */ |
1369 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1370 | } |
1371 | ||
90554200 JA |
1372 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1373 | { | |
1374 | struct io_rings *r = ctx->rings; | |
1375 | ||
a566c556 | 1376 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1377 | } |
1378 | ||
888aae2e PB |
1379 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1380 | { | |
1381 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1382 | } | |
1383 | ||
d068b506 | 1384 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1385 | { |
75b28aff | 1386 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1387 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1388 | |
115e12e5 SB |
1389 | /* |
1390 | * writes to the cq entry need to come after reading head; the | |
1391 | * control dependency is enough as we're using WRITE_ONCE to | |
1392 | * fill the cq entry | |
1393 | */ | |
a566c556 | 1394 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1395 | return NULL; |
1396 | ||
888aae2e | 1397 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1398 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1399 | } |
1400 | ||
f2842ab5 JA |
1401 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1402 | { | |
44c769de | 1403 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1404 | return false; |
7e55a19c SG |
1405 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1406 | return false; | |
44c769de | 1407 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1408 | } |
1409 | ||
b41e9852 | 1410 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1411 | { |
b1445e59 PB |
1412 | /* see waitqueue_active() comment */ |
1413 | smp_mb(); | |
1414 | ||
311997b3 PB |
1415 | if (waitqueue_active(&ctx->cq_wait)) |
1416 | wake_up(&ctx->cq_wait); | |
534ca6d6 JA |
1417 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1418 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1419 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1420 | eventfd_signal(ctx->cq_ev_fd, 1); |
311997b3 PB |
1421 | if (waitqueue_active(&ctx->poll_wait)) { |
1422 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1423 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1424 | } | |
1d7bb1d5 JA |
1425 | } |
1426 | ||
80c18e4a PB |
1427 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1428 | { | |
b1445e59 PB |
1429 | /* see waitqueue_active() comment */ |
1430 | smp_mb(); | |
1431 | ||
80c18e4a | 1432 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
311997b3 PB |
1433 | if (waitqueue_active(&ctx->cq_wait)) |
1434 | wake_up(&ctx->cq_wait); | |
80c18e4a PB |
1435 | } |
1436 | if (io_should_trigger_evfd(ctx)) | |
1437 | eventfd_signal(ctx->cq_ev_fd, 1); | |
311997b3 PB |
1438 | if (waitqueue_active(&ctx->poll_wait)) { |
1439 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1440 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1441 | } | |
80c18e4a PB |
1442 | } |
1443 | ||
c4a2ed72 | 1444 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1445 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1446 | { |
1d7bb1d5 | 1447 | unsigned long flags; |
b18032bb | 1448 | bool all_flushed, posted; |
1d7bb1d5 | 1449 | |
a566c556 | 1450 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1451 | return false; |
1d7bb1d5 | 1452 | |
b18032bb | 1453 | posted = false; |
1d7bb1d5 | 1454 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae | 1455 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1456 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1457 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1458 | |
1d7bb1d5 JA |
1459 | if (!cqe && !force) |
1460 | break; | |
6c2450ae PB |
1461 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1462 | struct io_overflow_cqe, list); | |
1463 | if (cqe) | |
1464 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1465 | else | |
8f6ed49a PB |
1466 | io_account_cq_overflow(ctx); |
1467 | ||
b18032bb | 1468 | posted = true; |
6c2450ae PB |
1469 | list_del(&ocqe->list); |
1470 | kfree(ocqe); | |
1d7bb1d5 JA |
1471 | } |
1472 | ||
09e88404 PB |
1473 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1474 | if (all_flushed) { | |
5ed7a37d | 1475 | clear_bit(0, &ctx->check_cq_overflow); |
09e88404 PB |
1476 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; |
1477 | } | |
46930143 | 1478 | |
b18032bb JA |
1479 | if (posted) |
1480 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1481 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1482 | if (posted) |
1483 | io_cqring_ev_posted(ctx); | |
09e88404 | 1484 | return all_flushed; |
1d7bb1d5 JA |
1485 | } |
1486 | ||
6c2450ae | 1487 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1488 | { |
ca0a2651 JA |
1489 | bool ret = true; |
1490 | ||
5ed7a37d | 1491 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1492 | /* iopoll syncs against uring_lock, not completion_lock */ |
1493 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1494 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1495 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1496 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1497 | mutex_unlock(&ctx->uring_lock); | |
1498 | } | |
ca0a2651 JA |
1499 | |
1500 | return ret; | |
6c503150 PB |
1501 | } |
1502 | ||
abc54d63 JA |
1503 | /* |
1504 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1505 | * see commit f958d7b528b1 for details. | |
1506 | */ | |
1507 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1508 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1509 | ||
de9b4cca JA |
1510 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1511 | { | |
abc54d63 | 1512 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1513 | } |
1514 | ||
1515 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1516 | { | |
abc54d63 JA |
1517 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1518 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1519 | } |
1520 | ||
1521 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1522 | { | |
abc54d63 JA |
1523 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1524 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1525 | } |
1526 | ||
1527 | static inline void req_ref_put(struct io_kiocb *req) | |
1528 | { | |
abc54d63 | 1529 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1530 | } |
1531 | ||
1532 | static inline void req_ref_get(struct io_kiocb *req) | |
1533 | { | |
abc54d63 JA |
1534 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1535 | atomic_inc(&req->refs); | |
de9b4cca JA |
1536 | } |
1537 | ||
d4d19c19 PB |
1538 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1539 | long res, unsigned int cflags) | |
2b188cc1 | 1540 | { |
cce4b8b0 | 1541 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1542 | |
cce4b8b0 PB |
1543 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1544 | if (!ocqe) { | |
1545 | /* | |
1546 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1547 | * or cannot allocate an overflow entry, then we need to drop it | |
1548 | * on the floor. | |
1549 | */ | |
8f6ed49a | 1550 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1551 | return false; |
2b188cc1 | 1552 | } |
cce4b8b0 | 1553 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1554 | set_bit(0, &ctx->check_cq_overflow); |
cce4b8b0 PB |
1555 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
1556 | } | |
d4d19c19 | 1557 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1558 | ocqe->cqe.res = res; |
1559 | ocqe->cqe.flags = cflags; | |
1560 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1561 | return true; | |
2b188cc1 JA |
1562 | } |
1563 | ||
d4d19c19 PB |
1564 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1565 | long res, unsigned int cflags) | |
2b188cc1 JA |
1566 | { |
1567 | struct io_uring_cqe *cqe; | |
1568 | ||
d4d19c19 | 1569 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1570 | |
2b188cc1 JA |
1571 | /* |
1572 | * If we can't get a cq entry, userspace overflowed the | |
1573 | * submission (by quite a lot). Increment the overflow count in | |
1574 | * the ring. | |
1575 | */ | |
d068b506 | 1576 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1577 | if (likely(cqe)) { |
d4d19c19 | 1578 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1579 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1580 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1581 | return true; |
2b188cc1 | 1582 | } |
d4d19c19 | 1583 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1584 | } |
1585 | ||
8d13326e | 1586 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1587 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1588 | long res, unsigned int cflags) | |
bcda7baa | 1589 | { |
d4d19c19 | 1590 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1591 | } |
1592 | ||
7a612350 PB |
1593 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1594 | unsigned int cflags) | |
2b188cc1 | 1595 | { |
78e19bbe | 1596 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1597 | unsigned long flags; |
1598 | ||
1599 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
d4d19c19 | 1600 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1601 | /* |
1602 | * If we're the last reference to this request, add to our locked | |
1603 | * free_list cache. | |
1604 | */ | |
de9b4cca | 1605 | if (req_ref_put_and_test(req)) { |
7a612350 | 1606 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
93d2bcd2 | 1607 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) |
7a612350 PB |
1608 | io_disarm_next(req); |
1609 | if (req->link) { | |
1610 | io_req_task_queue(req->link); | |
1611 | req->link = NULL; | |
1612 | } | |
1613 | } | |
c7dae4ba JA |
1614 | io_dismantle_req(req); |
1615 | io_put_task(req->task, 1); | |
d0acdee2 PB |
1616 | list_add(&req->compl.list, &ctx->locked_free_list); |
1617 | ctx->locked_free_nr++; | |
180f829f PB |
1618 | } else { |
1619 | if (!percpu_ref_tryget(&ctx->refs)) | |
1620 | req = NULL; | |
1621 | } | |
7a612350 | 1622 | io_commit_cqring(ctx); |
2b188cc1 | 1623 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1624 | |
180f829f PB |
1625 | if (req) { |
1626 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1627 | percpu_ref_put(&ctx->refs); |
180f829f | 1628 | } |
229a7b63 JA |
1629 | } |
1630 | ||
4e3d9ff9 JA |
1631 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1632 | { | |
c854357b | 1633 | return req->flags & IO_REQ_CLEAN_FLAGS; |
4e3d9ff9 JA |
1634 | } |
1635 | ||
a38d68db | 1636 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1637 | unsigned int cflags) |
229a7b63 | 1638 | { |
4e3d9ff9 | 1639 | if (io_req_needs_clean(req)) |
68fb8979 | 1640 | io_clean_op(req); |
a38d68db PB |
1641 | req->result = res; |
1642 | req->compl.cflags = cflags; | |
e342c807 | 1643 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1644 | } |
1645 | ||
889fca73 PB |
1646 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1647 | long res, unsigned cflags) | |
bcda7baa | 1648 | { |
889fca73 PB |
1649 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1650 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1651 | else |
c7dae4ba | 1652 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1653 | } |
1654 | ||
a38d68db | 1655 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1656 | { |
889fca73 | 1657 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1658 | } |
1659 | ||
f41db273 PB |
1660 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1661 | { | |
93d2bcd2 | 1662 | req_set_fail(req); |
f41db273 PB |
1663 | io_put_req(req); |
1664 | io_req_complete_post(req, res, 0); | |
1665 | } | |
1666 | ||
dac7a098 PB |
1667 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1668 | struct io_comp_state *cs) | |
1669 | { | |
1670 | spin_lock_irq(&ctx->completion_lock); | |
d0acdee2 PB |
1671 | list_splice_init(&ctx->locked_free_list, &cs->free_list); |
1672 | ctx->locked_free_nr = 0; | |
dac7a098 PB |
1673 | spin_unlock_irq(&ctx->completion_lock); |
1674 | } | |
1675 | ||
dd78f492 | 1676 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1677 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1678 | { |
c7dae4ba JA |
1679 | struct io_submit_state *state = &ctx->submit_state; |
1680 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1681 | int nr; |
0ddf92e8 | 1682 | |
c7dae4ba JA |
1683 | /* |
1684 | * If we have more than a batch's worth of requests in our IRQ side | |
1685 | * locked cache, grab the lock and move them over to our submission | |
1686 | * side cache. | |
1687 | */ | |
d0acdee2 | 1688 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
dac7a098 | 1689 | io_flush_cached_locked_reqs(ctx, cs); |
0ddf92e8 | 1690 | |
dd78f492 | 1691 | nr = state->free_reqs; |
c7dae4ba | 1692 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1693 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1694 | struct io_kiocb, compl.list); | |
1695 | ||
1b4c351f | 1696 | list_del(&req->compl.list); |
dd78f492 PB |
1697 | state->reqs[nr++] = req; |
1698 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1699 | break; |
1b4c351f JA |
1700 | } |
1701 | ||
dd78f492 PB |
1702 | state->free_reqs = nr; |
1703 | return nr != 0; | |
0ddf92e8 JA |
1704 | } |
1705 | ||
e5d1bc0a | 1706 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1707 | { |
e5d1bc0a PB |
1708 | struct io_submit_state *state = &ctx->submit_state; |
1709 | ||
fe7e3257 | 1710 | BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); |
e5d1bc0a | 1711 | |
f6b6c7d6 | 1712 | if (!state->free_reqs) { |
291b2821 | 1713 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1714 | int ret; |
1715 | ||
c7dae4ba | 1716 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1717 | goto got_req; |
1718 | ||
bf019da7 PB |
1719 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1720 | state->reqs); | |
fd6fab2c JA |
1721 | |
1722 | /* | |
1723 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1724 | * retry single alloc to be on the safe side. | |
1725 | */ | |
1726 | if (unlikely(ret <= 0)) { | |
1727 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1728 | if (!state->reqs[0]) | |
3893f39f | 1729 | return NULL; |
fd6fab2c JA |
1730 | ret = 1; |
1731 | } | |
291b2821 | 1732 | state->free_reqs = ret; |
2b188cc1 | 1733 | } |
e5d1bc0a | 1734 | got_req: |
291b2821 PB |
1735 | state->free_reqs--; |
1736 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1737 | } |
1738 | ||
e1d767f0 | 1739 | static inline void io_put_file(struct file *file) |
8da11c19 | 1740 | { |
e1d767f0 | 1741 | if (file) |
8da11c19 PB |
1742 | fput(file); |
1743 | } | |
1744 | ||
4edf20f9 | 1745 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1746 | { |
094bae49 | 1747 | unsigned int flags = req->flags; |
929a3af9 | 1748 | |
3a0a6902 PB |
1749 | if (io_req_needs_clean(req)) |
1750 | io_clean_op(req); | |
e1d767f0 PB |
1751 | if (!(flags & REQ_F_FIXED_FILE)) |
1752 | io_put_file(req->file); | |
269bbe5f BM |
1753 | if (req->fixed_rsrc_refs) |
1754 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1755 | if (req->async_data) |
1756 | kfree(req->async_data); | |
e65ef56d JA |
1757 | } |
1758 | ||
b23fcf47 | 1759 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1760 | static inline void io_put_task(struct task_struct *task, int nr) |
1761 | { | |
1762 | struct io_uring_task *tctx = task->io_uring; | |
1763 | ||
1764 | percpu_counter_sub(&tctx->inflight, nr); | |
1765 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1766 | wake_up(&tctx->wait); | |
1767 | put_task_struct_many(task, nr); | |
1768 | } | |
1769 | ||
216578e5 | 1770 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1771 | { |
51a4cc11 | 1772 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1773 | |
216578e5 | 1774 | io_dismantle_req(req); |
7c660731 | 1775 | io_put_task(req->task, 1); |
c6ca97b3 | 1776 | |
3893f39f | 1777 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1778 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1779 | } |
1780 | ||
f2f87370 PB |
1781 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1782 | { | |
1783 | struct io_kiocb *nxt = req->link; | |
1784 | ||
1785 | req->link = nxt->link; | |
1786 | nxt->link = NULL; | |
1787 | } | |
1788 | ||
33cc89a9 PB |
1789 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1790 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1791 | { |
33cc89a9 | 1792 | struct io_kiocb *link = req->link; |
f2f87370 | 1793 | |
900fad45 PB |
1794 | /* |
1795 | * Can happen if a linked timeout fired and link had been like | |
1796 | * req -> link t-out -> link t-out [-> ...] | |
1797 | */ | |
c9abd7ad PB |
1798 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1799 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1800 | |
f2f87370 | 1801 | io_remove_next_linked(req); |
90cd7e42 | 1802 | link->timeout.head = NULL; |
fd9c7bc5 | 1803 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
d4d19c19 PB |
1804 | io_cqring_fill_event(link->ctx, link->user_data, |
1805 | -ECANCELED, 0); | |
33cc89a9 | 1806 | io_put_req_deferred(link, 1); |
d4729fbd | 1807 | return true; |
c9abd7ad PB |
1808 | } |
1809 | } | |
d4729fbd | 1810 | return false; |
7c86ffee PB |
1811 | } |
1812 | ||
d148ca4b | 1813 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1814 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1815 | { |
33cc89a9 | 1816 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1817 | |
f2f87370 | 1818 | req->link = NULL; |
f2f87370 PB |
1819 | while (link) { |
1820 | nxt = link->link; | |
1821 | link->link = NULL; | |
2665abfd | 1822 | |
f2f87370 | 1823 | trace_io_uring_fail_link(req, link); |
d4d19c19 | 1824 | io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0); |
1575f21a | 1825 | io_put_req_deferred(link, 2); |
f2f87370 | 1826 | link = nxt; |
9e645e11 | 1827 | } |
33cc89a9 | 1828 | } |
9e645e11 | 1829 | |
33cc89a9 PB |
1830 | static bool io_disarm_next(struct io_kiocb *req) |
1831 | __must_hold(&req->ctx->completion_lock) | |
1832 | { | |
1833 | bool posted = false; | |
1834 | ||
1835 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1836 | posted = io_kill_linked_timeout(req); | |
93d2bcd2 | 1837 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 1838 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
1839 | posted |= (req->link != NULL); |
1840 | io_fail_links(req); | |
1841 | } | |
1842 | return posted; | |
9e645e11 JA |
1843 | } |
1844 | ||
3fa5e0f3 | 1845 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1846 | { |
33cc89a9 | 1847 | struct io_kiocb *nxt; |
944e58bf | 1848 | |
9e645e11 JA |
1849 | /* |
1850 | * If LINK is set, we have dependent requests in this chain. If we | |
1851 | * didn't fail this request, queue the first one up, moving any other | |
1852 | * dependencies to the next request. In case of failure, fail the rest | |
1853 | * of the chain. | |
1854 | */ | |
93d2bcd2 | 1855 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) { |
33cc89a9 PB |
1856 | struct io_ring_ctx *ctx = req->ctx; |
1857 | unsigned long flags; | |
1858 | bool posted; | |
1859 | ||
1860 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1861 | posted = io_disarm_next(req); | |
1862 | if (posted) | |
1863 | io_commit_cqring(req->ctx); | |
1864 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1865 | if (posted) | |
1866 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1867 | } |
33cc89a9 PB |
1868 | nxt = req->link; |
1869 | req->link = NULL; | |
1870 | return nxt; | |
4d7dd462 | 1871 | } |
9e645e11 | 1872 | |
f2f87370 | 1873 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1874 | { |
cdbff982 | 1875 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1876 | return NULL; |
1877 | return __io_req_find_next(req); | |
1878 | } | |
1879 | ||
2c32395d PB |
1880 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1881 | { | |
1882 | if (!ctx) | |
1883 | return; | |
1884 | if (ctx->submit_state.comp.nr) { | |
1885 | mutex_lock(&ctx->uring_lock); | |
2a2758f2 | 1886 | io_submit_flush_completions(ctx); |
2c32395d PB |
1887 | mutex_unlock(&ctx->uring_lock); |
1888 | } | |
1889 | percpu_ref_put(&ctx->refs); | |
1890 | } | |
1891 | ||
7cbf1722 | 1892 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1893 | { |
ebd0df2e | 1894 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
1895 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
1896 | task_work); | |
c40f6379 | 1897 | |
16f72070 | 1898 | while (1) { |
3f18407d PB |
1899 | struct io_wq_work_node *node; |
1900 | ||
1901 | spin_lock_irq(&tctx->task_lock); | |
c6538be9 | 1902 | node = tctx->task_list.first; |
3f18407d PB |
1903 | INIT_WQ_LIST(&tctx->task_list); |
1904 | spin_unlock_irq(&tctx->task_lock); | |
1905 | ||
3f18407d PB |
1906 | while (node) { |
1907 | struct io_wq_work_node *next = node->next; | |
1908 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
1909 | io_task_work.node); | |
1910 | ||
1911 | if (req->ctx != ctx) { | |
1912 | ctx_flush_and_put(ctx); | |
1913 | ctx = req->ctx; | |
1914 | percpu_ref_get(&ctx->refs); | |
1915 | } | |
1916 | req->task_work.func(&req->task_work); | |
1917 | node = next; | |
1918 | } | |
7a778f9d PB |
1919 | if (wq_list_empty(&tctx->task_list)) { |
1920 | clear_bit(0, &tctx->task_state); | |
1921 | if (wq_list_empty(&tctx->task_list)) | |
1922 | break; | |
1923 | /* another tctx_task_work() is enqueued, yield */ | |
1924 | if (test_and_set_bit(0, &tctx->task_state)) | |
1925 | break; | |
1926 | } | |
7cbf1722 | 1927 | cond_resched(); |
3f18407d | 1928 | } |
ebd0df2e PB |
1929 | |
1930 | ctx_flush_and_put(ctx); | |
7cbf1722 JA |
1931 | } |
1932 | ||
c15b79de | 1933 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1934 | { |
c15b79de | 1935 | struct task_struct *tsk = req->task; |
7cbf1722 | 1936 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1937 | enum task_work_notify_mode notify; |
7cbf1722 | 1938 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1939 | unsigned long flags; |
c15b79de PB |
1940 | int ret = 0; |
1941 | ||
1942 | if (unlikely(tsk->flags & PF_EXITING)) | |
1943 | return -ESRCH; | |
7cbf1722 JA |
1944 | |
1945 | WARN_ON_ONCE(!tctx); | |
1946 | ||
0b81e80c | 1947 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1948 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1949 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1950 | |
1951 | /* task_work already pending, we're done */ | |
1952 | if (test_bit(0, &tctx->task_state) || | |
1953 | test_and_set_bit(0, &tctx->task_state)) | |
1954 | return 0; | |
1955 | ||
c15b79de PB |
1956 | /* |
1957 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1958 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1959 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1960 | * will do the job. | |
1961 | */ | |
1962 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1963 | ||
1964 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1965 | wake_up_process(tsk); | |
7cbf1722 | 1966 | return 0; |
c15b79de | 1967 | } |
7cbf1722 JA |
1968 | |
1969 | /* | |
1970 | * Slow path - we failed, find and delete work. if the work is not | |
1971 | * in the list, it got run and we're fine. | |
1972 | */ | |
0b81e80c | 1973 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1974 | wq_list_for_each(node, prev, &tctx->task_list) { |
1975 | if (&req->io_task_work.node == node) { | |
1976 | wq_list_del(&tctx->task_list, node, prev); | |
1977 | ret = 1; | |
1978 | break; | |
1979 | } | |
1980 | } | |
0b81e80c | 1981 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1982 | clear_bit(0, &tctx->task_state); |
1983 | return ret; | |
1984 | } | |
1985 | ||
9b465711 PB |
1986 | static bool io_run_task_work_head(struct callback_head **work_head) |
1987 | { | |
1988 | struct callback_head *work, *next; | |
1989 | bool executed = false; | |
1990 | ||
1991 | do { | |
1992 | work = xchg(work_head, NULL); | |
1993 | if (!work) | |
1994 | break; | |
1995 | ||
1996 | do { | |
1997 | next = work->next; | |
1998 | work->func(work); | |
1999 | work = next; | |
2000 | cond_resched(); | |
2001 | } while (work); | |
2002 | executed = true; | |
2003 | } while (1); | |
2004 | ||
2005 | return executed; | |
2006 | } | |
2007 | ||
2008 | static void io_task_work_add_head(struct callback_head **work_head, | |
2009 | struct callback_head *task_work) | |
eab30c4d | 2010 | { |
7c25c0d1 | 2011 | struct callback_head *head; |
eab30c4d | 2012 | |
7c25c0d1 | 2013 | do { |
9b465711 PB |
2014 | head = READ_ONCE(*work_head); |
2015 | task_work->next = head; | |
2016 | } while (cmpxchg(work_head, head, task_work) != head); | |
2017 | } | |
2018 | ||
2019 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2020 | task_work_func_t cb) | |
2021 | { | |
2022 | init_task_work(&req->task_work, cb); | |
2023 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2024 | } |
2025 | ||
c40f6379 JA |
2026 | static void io_req_task_cancel(struct callback_head *cb) |
2027 | { | |
2028 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2029 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2030 | |
e83acd7d | 2031 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2032 | mutex_lock(&ctx->uring_lock); |
2593553a | 2033 | io_req_complete_failed(req, req->result); |
792bb6eb | 2034 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2035 | } |
2036 | ||
2037 | static void __io_req_task_submit(struct io_kiocb *req) | |
2038 | { | |
2039 | struct io_ring_ctx *ctx = req->ctx; | |
2040 | ||
04fc6c80 | 2041 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2042 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2043 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2044 | __io_queue_sqe(req); |
81b6d05c | 2045 | else |
2593553a | 2046 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2047 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2048 | } |
2049 | ||
2050 | static void io_req_task_submit(struct callback_head *cb) | |
2051 | { | |
2052 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2053 | ||
2054 | __io_req_task_submit(req); | |
2055 | } | |
2056 | ||
2c4b8eb6 | 2057 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2058 | { |
2c4b8eb6 PB |
2059 | req->result = ret; |
2060 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2061 | |
2c4b8eb6 | 2062 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2063 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2064 | } |
2065 | ||
2c4b8eb6 | 2066 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2067 | { |
2c4b8eb6 | 2068 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2069 | |
2070 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2071 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2072 | } |
2073 | ||
f2f87370 | 2074 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2075 | { |
9b5f7bd9 | 2076 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2077 | |
2078 | if (nxt) | |
906a8c3f | 2079 | io_req_task_queue(nxt); |
c69f8dbe JL |
2080 | } |
2081 | ||
c3524383 | 2082 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2083 | { |
c3524383 PB |
2084 | io_queue_next(req); |
2085 | __io_free_req(req); | |
2086 | } | |
8766dd51 | 2087 | |
2d6500d4 | 2088 | struct req_batch { |
5af1d13e PB |
2089 | struct task_struct *task; |
2090 | int task_refs; | |
1b4c351f | 2091 | int ctx_refs; |
2d6500d4 PB |
2092 | }; |
2093 | ||
5af1d13e PB |
2094 | static inline void io_init_req_batch(struct req_batch *rb) |
2095 | { | |
5af1d13e | 2096 | rb->task_refs = 0; |
9ae72463 | 2097 | rb->ctx_refs = 0; |
5af1d13e PB |
2098 | rb->task = NULL; |
2099 | } | |
2100 | ||
2d6500d4 PB |
2101 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2102 | struct req_batch *rb) | |
2103 | { | |
6e833d53 | 2104 | if (rb->task) |
7c660731 | 2105 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2106 | if (rb->ctx_refs) |
2107 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2108 | } |
2109 | ||
6ff119a6 PB |
2110 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2111 | struct io_submit_state *state) | |
2d6500d4 | 2112 | { |
f2f87370 | 2113 | io_queue_next(req); |
96670657 | 2114 | io_dismantle_req(req); |
2d6500d4 | 2115 | |
e3bc8e9d | 2116 | if (req->task != rb->task) { |
7c660731 PB |
2117 | if (rb->task) |
2118 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2119 | rb->task = req->task; |
2120 | rb->task_refs = 0; | |
5af1d13e | 2121 | } |
e3bc8e9d | 2122 | rb->task_refs++; |
9ae72463 | 2123 | rb->ctx_refs++; |
5af1d13e | 2124 | |
bd759045 | 2125 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2126 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2127 | else |
2128 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2129 | } |
2130 | ||
2a2758f2 | 2131 | static void io_submit_flush_completions(struct io_ring_ctx *ctx) |
905c172f | 2132 | { |
2a2758f2 | 2133 | struct io_comp_state *cs = &ctx->submit_state.comp; |
905c172f PB |
2134 | int i, nr = cs->nr; |
2135 | struct io_kiocb *req; | |
2136 | struct req_batch rb; | |
2137 | ||
2138 | io_init_req_batch(&rb); | |
2139 | spin_lock_irq(&ctx->completion_lock); | |
2140 | for (i = 0; i < nr; i++) { | |
2141 | req = cs->reqs[i]; | |
d4d19c19 PB |
2142 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2143 | req->compl.cflags); | |
905c172f PB |
2144 | } |
2145 | io_commit_cqring(ctx); | |
2146 | spin_unlock_irq(&ctx->completion_lock); | |
2147 | ||
2148 | io_cqring_ev_posted(ctx); | |
2149 | for (i = 0; i < nr; i++) { | |
2150 | req = cs->reqs[i]; | |
2151 | ||
2152 | /* submission and completion refs */ | |
de9b4cca | 2153 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2154 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2155 | } |
2156 | ||
2157 | io_req_free_batch_finish(ctx, &rb); | |
2158 | cs->nr = 0; | |
7a743e22 PB |
2159 | } |
2160 | ||
ba816ad6 JA |
2161 | /* |
2162 | * Drop reference to request, return next in chain (if there is one) if this | |
2163 | * was the last reference to this request. | |
2164 | */ | |
0d85035a | 2165 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2166 | { |
9b5f7bd9 PB |
2167 | struct io_kiocb *nxt = NULL; |
2168 | ||
de9b4cca | 2169 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2170 | nxt = io_req_find_next(req); |
4d7dd462 | 2171 | __io_free_req(req); |
2a44f467 | 2172 | } |
9b5f7bd9 | 2173 | return nxt; |
2b188cc1 JA |
2174 | } |
2175 | ||
0d85035a | 2176 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2177 | { |
de9b4cca | 2178 | if (req_ref_put_and_test(req)) |
e65ef56d | 2179 | io_free_req(req); |
2b188cc1 JA |
2180 | } |
2181 | ||
216578e5 PB |
2182 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2183 | { | |
2184 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2185 | ||
2186 | io_free_req(req); | |
2187 | } | |
2188 | ||
2189 | static void io_free_req_deferred(struct io_kiocb *req) | |
2190 | { | |
7cbf1722 | 2191 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2192 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2193 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2194 | } |
2195 | ||
2196 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2197 | { | |
de9b4cca | 2198 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2199 | io_free_req_deferred(req); |
2200 | } | |
2201 | ||
6c503150 | 2202 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2203 | { |
2204 | /* See comment at the top of this file */ | |
2205 | smp_rmb(); | |
e23de15f | 2206 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2207 | } |
2208 | ||
fb5ccc98 PB |
2209 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2210 | { | |
2211 | struct io_rings *rings = ctx->rings; | |
2212 | ||
2213 | /* make sure SQ entry isn't read before tail */ | |
2214 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2215 | } | |
2216 | ||
8ff069bf | 2217 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2218 | { |
8ff069bf | 2219 | unsigned int cflags; |
e94f141b | 2220 | |
bcda7baa JA |
2221 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2222 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2223 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2224 | kfree(kbuf); |
2225 | return cflags; | |
e94f141b JA |
2226 | } |
2227 | ||
8ff069bf | 2228 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2229 | { |
4d954c25 | 2230 | struct io_buffer *kbuf; |
bcda7baa | 2231 | |
4d954c25 | 2232 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2233 | return io_put_kbuf(req, kbuf); |
2234 | } | |
2235 | ||
4c6e277c JA |
2236 | static inline bool io_run_task_work(void) |
2237 | { | |
6200b0ae JA |
2238 | /* |
2239 | * Not safe to run on exiting task, and the task_work handling will | |
2240 | * not add work to such a task. | |
2241 | */ | |
2242 | if (unlikely(current->flags & PF_EXITING)) | |
2243 | return false; | |
4c6e277c JA |
2244 | if (current->task_works) { |
2245 | __set_current_state(TASK_RUNNING); | |
2246 | task_work_run(); | |
2247 | return true; | |
2248 | } | |
2249 | ||
2250 | return false; | |
bcda7baa JA |
2251 | } |
2252 | ||
def596e9 JA |
2253 | /* |
2254 | * Find and free completed poll iocbs | |
2255 | */ | |
2256 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2257 | struct list_head *done) | |
2258 | { | |
8237e045 | 2259 | struct req_batch rb; |
def596e9 | 2260 | struct io_kiocb *req; |
bbde017a XW |
2261 | |
2262 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2263 | smp_rmb(); | |
def596e9 | 2264 | |
5af1d13e | 2265 | io_init_req_batch(&rb); |
def596e9 | 2266 | while (!list_empty(done)) { |
bcda7baa JA |
2267 | int cflags = 0; |
2268 | ||
d21ffe7e | 2269 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2270 | list_del(&req->inflight_entry); |
2271 | ||
8c130827 PB |
2272 | if (READ_ONCE(req->result) == -EAGAIN && |
2273 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2274 | req->iopoll_completed = 0; |
8c130827 PB |
2275 | req_ref_get(req); |
2276 | io_queue_async_work(req); | |
2277 | continue; | |
bbde017a | 2278 | } |
def596e9 | 2279 | |
bcda7baa | 2280 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2281 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2282 | |
d4d19c19 | 2283 | __io_cqring_fill_event(ctx, req->user_data, req->result, cflags); |
def596e9 JA |
2284 | (*nr_events)++; |
2285 | ||
de9b4cca | 2286 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2287 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2288 | } |
def596e9 | 2289 | |
09bb8394 | 2290 | io_commit_cqring(ctx); |
80c18e4a | 2291 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2292 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2293 | } |
2294 | ||
def596e9 JA |
2295 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2296 | long min) | |
2297 | { | |
2298 | struct io_kiocb *req, *tmp; | |
2299 | LIST_HEAD(done); | |
2300 | bool spin; | |
2301 | int ret; | |
2302 | ||
2303 | /* | |
2304 | * Only spin for completions if we don't have multiple devices hanging | |
2305 | * off our complete list, and we're under the requested amount. | |
2306 | */ | |
2307 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2308 | ||
2309 | ret = 0; | |
d21ffe7e | 2310 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2311 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2312 | |
2313 | /* | |
581f9810 BM |
2314 | * Move completed and retryable entries to our local lists. |
2315 | * If we find a request that requires polling, break out | |
2316 | * and complete those lists first, if we have entries there. | |
def596e9 | 2317 | */ |
65a6543d | 2318 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2319 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2320 | continue; |
2321 | } | |
2322 | if (!list_empty(&done)) | |
2323 | break; | |
2324 | ||
2325 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2326 | if (ret < 0) | |
2327 | break; | |
2328 | ||
3aadc23e PB |
2329 | /* iopoll may have completed current req */ |
2330 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2331 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2332 | |
def596e9 JA |
2333 | if (ret && spin) |
2334 | spin = false; | |
2335 | ret = 0; | |
2336 | } | |
2337 | ||
2338 | if (!list_empty(&done)) | |
2339 | io_iopoll_complete(ctx, nr_events, &done); | |
2340 | ||
2341 | return ret; | |
2342 | } | |
2343 | ||
def596e9 JA |
2344 | /* |
2345 | * We can't just wait for polled events to come to us, we have to actively | |
2346 | * find and complete them. | |
2347 | */ | |
b2edc0a7 | 2348 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2349 | { |
2350 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2351 | return; | |
2352 | ||
2353 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2354 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2355 | unsigned int nr_events = 0; |
2356 | ||
b2edc0a7 | 2357 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2358 | |
b2edc0a7 PB |
2359 | /* let it sleep and repeat later if can't complete a request */ |
2360 | if (nr_events == 0) | |
2361 | break; | |
08f5439f JA |
2362 | /* |
2363 | * Ensure we allow local-to-the-cpu processing to take place, | |
2364 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2365 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2366 | */ |
3fcee5a6 PB |
2367 | if (need_resched()) { |
2368 | mutex_unlock(&ctx->uring_lock); | |
2369 | cond_resched(); | |
2370 | mutex_lock(&ctx->uring_lock); | |
2371 | } | |
def596e9 JA |
2372 | } |
2373 | mutex_unlock(&ctx->uring_lock); | |
2374 | } | |
2375 | ||
7668b92a | 2376 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2377 | { |
7668b92a | 2378 | unsigned int nr_events = 0; |
e9979b36 | 2379 | int ret = 0; |
500f9fba | 2380 | |
c7849be9 XW |
2381 | /* |
2382 | * We disallow the app entering submit/complete with polling, but we | |
2383 | * still need to lock the ring to prevent racing with polled issue | |
2384 | * that got punted to a workqueue. | |
2385 | */ | |
2386 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2387 | /* |
2388 | * Don't enter poll loop if we already have events pending. | |
2389 | * If we do, we can potentially be spinning for commands that | |
2390 | * already triggered a CQE (eg in error). | |
2391 | */ | |
5ed7a37d | 2392 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2393 | __io_cqring_overflow_flush(ctx, false); |
2394 | if (io_cqring_events(ctx)) | |
2395 | goto out; | |
def596e9 | 2396 | do { |
500f9fba JA |
2397 | /* |
2398 | * If a submit got punted to a workqueue, we can have the | |
2399 | * application entering polling for a command before it gets | |
2400 | * issued. That app will hold the uring_lock for the duration | |
2401 | * of the poll right here, so we need to take a breather every | |
2402 | * now and then to ensure that the issue has a chance to add | |
2403 | * the poll to the issued list. Otherwise we can spin here | |
2404 | * forever, while the workqueue is stuck trying to acquire the | |
2405 | * very same mutex. | |
2406 | */ | |
e9979b36 | 2407 | if (list_empty(&ctx->iopoll_list)) { |
500f9fba | 2408 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2409 | io_run_task_work(); |
500f9fba | 2410 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2411 | |
e9979b36 PB |
2412 | if (list_empty(&ctx->iopoll_list)) |
2413 | break; | |
500f9fba | 2414 | } |
f39c8a5b PB |
2415 | ret = io_do_iopoll(ctx, &nr_events, min); |
2416 | } while (!ret && nr_events < min && !need_resched()); | |
2417 | out: | |
500f9fba | 2418 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2419 | return ret; |
2420 | } | |
2421 | ||
491381ce | 2422 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2423 | { |
491381ce JA |
2424 | /* |
2425 | * Tell lockdep we inherited freeze protection from submission | |
2426 | * thread. | |
2427 | */ | |
2428 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2429 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2430 | |
1c98679d PB |
2431 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2432 | sb_end_write(sb); | |
2b188cc1 JA |
2433 | } |
2434 | } | |
2435 | ||
b63534c4 | 2436 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2437 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2438 | { |
ab454438 | 2439 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2440 | |
ab454438 PB |
2441 | if (!rw) |
2442 | return !io_req_prep_async(req); | |
2443 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2444 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2445 | return true; | |
b63534c4 | 2446 | } |
b63534c4 | 2447 | |
3e6a0d3c | 2448 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2449 | { |
355afaeb | 2450 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2451 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2452 | |
355afaeb JA |
2453 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2454 | return false; | |
3e6a0d3c JA |
2455 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2456 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2457 | return false; |
7c977a58 JA |
2458 | /* |
2459 | * If ref is dying, we might be running poll reap from the exit work. | |
2460 | * Don't attempt to reissue from that path, just let it fail with | |
2461 | * -EAGAIN. | |
2462 | */ | |
3e6a0d3c JA |
2463 | if (percpu_ref_is_dying(&ctx->refs)) |
2464 | return false; | |
2465 | return true; | |
2466 | } | |
e82ad485 | 2467 | #else |
a1ff1e3f | 2468 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2469 | { |
2470 | return false; | |
2471 | } | |
e82ad485 | 2472 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2473 | { |
b63534c4 JA |
2474 | return false; |
2475 | } | |
3e6a0d3c | 2476 | #endif |
b63534c4 | 2477 | |
a1d7c393 | 2478 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2479 | unsigned int issue_flags) |
a1d7c393 | 2480 | { |
2f8e45f1 PB |
2481 | int cflags = 0; |
2482 | ||
b65c128f PB |
2483 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2484 | kiocb_end_write(req); | |
9532b99b PB |
2485 | if (res != req->result) { |
2486 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2487 | io_rw_should_reissue(req)) { | |
2488 | req->flags |= REQ_F_REISSUE; | |
2489 | return; | |
2490 | } | |
93d2bcd2 | 2491 | req_set_fail(req); |
9532b99b | 2492 | } |
2f8e45f1 PB |
2493 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2494 | cflags = io_put_rw_kbuf(req); | |
2495 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2496 | } |
2497 | ||
2498 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2499 | { | |
9adbd45d | 2500 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2501 | |
889fca73 | 2502 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2503 | } |
2504 | ||
def596e9 JA |
2505 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2506 | { | |
9adbd45d | 2507 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2508 | |
491381ce JA |
2509 | if (kiocb->ki_flags & IOCB_WRITE) |
2510 | kiocb_end_write(req); | |
9532b99b | 2511 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2512 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2513 | io_resubmit_prep(req))) { | |
93d2bcd2 | 2514 | req_set_fail(req); |
9532b99b PB |
2515 | req->flags |= REQ_F_DONT_REISSUE; |
2516 | } | |
8c130827 | 2517 | } |
bbde017a XW |
2518 | |
2519 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2520 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2521 | smp_wmb(); |
2522 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2523 | } |
2524 | ||
2525 | /* | |
2526 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2527 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2528 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2529 | * accessing the kiocb cookie. |
2530 | */ | |
cb3d8972 | 2531 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2532 | { |
2533 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2534 | const bool in_async = io_wq_current_is_worker(); |
2535 | ||
2536 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2537 | if (unlikely(in_async)) | |
2538 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2539 | |
2540 | /* | |
2541 | * Track whether we have multiple files in our lists. This will impact | |
2542 | * how we do polling eventually, not spinning if we're on potentially | |
2543 | * different devices. | |
2544 | */ | |
540e32a0 | 2545 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2546 | ctx->poll_multi_file = false; |
2547 | } else if (!ctx->poll_multi_file) { | |
2548 | struct io_kiocb *list_req; | |
2549 | ||
540e32a0 | 2550 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2551 | inflight_entry); |
9adbd45d | 2552 | if (list_req->file != req->file) |
def596e9 JA |
2553 | ctx->poll_multi_file = true; |
2554 | } | |
2555 | ||
2556 | /* | |
2557 | * For fast devices, IO may have already completed. If it has, add | |
2558 | * it to the front so we find it first. | |
2559 | */ | |
65a6543d | 2560 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2561 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2562 | else |
d21ffe7e | 2563 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2564 | |
cb3d8972 PB |
2565 | if (unlikely(in_async)) { |
2566 | /* | |
2567 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2568 | * in sq thread task context or in io worker task context. If | |
2569 | * current task context is sq thread, we don't need to check | |
2570 | * whether should wake up sq thread. | |
2571 | */ | |
2572 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2573 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2574 | wake_up(&ctx->sq_data->wait); | |
2575 | ||
2576 | mutex_unlock(&ctx->uring_lock); | |
2577 | } | |
def596e9 JA |
2578 | } |
2579 | ||
9f13c35b PB |
2580 | static inline void io_state_file_put(struct io_submit_state *state) |
2581 | { | |
02b23a9a PB |
2582 | if (state->file_refs) { |
2583 | fput_many(state->file, state->file_refs); | |
2584 | state->file_refs = 0; | |
2585 | } | |
9a56a232 JA |
2586 | } |
2587 | ||
2588 | /* | |
2589 | * Get as many references to a file as we have IOs left in this submission, | |
2590 | * assuming most submissions are for one file, or at least that each file | |
2591 | * has more than one submission. | |
2592 | */ | |
8da11c19 | 2593 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2594 | { |
2595 | if (!state) | |
2596 | return fget(fd); | |
2597 | ||
6e1271e6 | 2598 | if (state->file_refs) { |
9a56a232 | 2599 | if (state->fd == fd) { |
6e1271e6 | 2600 | state->file_refs--; |
9a56a232 JA |
2601 | return state->file; |
2602 | } | |
02b23a9a | 2603 | io_state_file_put(state); |
9a56a232 JA |
2604 | } |
2605 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2606 | if (unlikely(!state->file)) |
9a56a232 JA |
2607 | return NULL; |
2608 | ||
2609 | state->fd = fd; | |
6e1271e6 | 2610 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2611 | return state->file; |
2612 | } | |
2613 | ||
4503b767 JA |
2614 | static bool io_bdev_nowait(struct block_device *bdev) |
2615 | { | |
9ba0d0c8 | 2616 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2617 | } |
2618 | ||
2b188cc1 JA |
2619 | /* |
2620 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2621 | * any file. For now, just ensure that anything potentially problematic is done | |
2622 | * inline. | |
2623 | */ | |
7b29f92d | 2624 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2625 | { |
2626 | umode_t mode = file_inode(file)->i_mode; | |
2627 | ||
4503b767 | 2628 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2629 | if (IS_ENABLED(CONFIG_BLOCK) && |
2630 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2631 | return true; |
2632 | return false; | |
2633 | } | |
976517f1 | 2634 | if (S_ISSOCK(mode)) |
2b188cc1 | 2635 | return true; |
4503b767 | 2636 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2637 | if (IS_ENABLED(CONFIG_BLOCK) && |
2638 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2639 | file->f_op != &io_uring_fops) |
2640 | return true; | |
2641 | return false; | |
2642 | } | |
2b188cc1 | 2643 | |
c5b85625 JA |
2644 | /* any ->read/write should understand O_NONBLOCK */ |
2645 | if (file->f_flags & O_NONBLOCK) | |
2646 | return true; | |
2647 | ||
af197f50 JA |
2648 | if (!(file->f_mode & FMODE_NOWAIT)) |
2649 | return false; | |
2650 | ||
2651 | if (rw == READ) | |
2652 | return file->f_op->read_iter != NULL; | |
2653 | ||
2654 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2655 | } |
2656 | ||
7b29f92d JA |
2657 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2658 | { | |
2659 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2660 | return true; | |
2661 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2662 | return true; | |
2663 | ||
2664 | return __io_file_supports_async(req->file, rw); | |
2665 | } | |
2666 | ||
a88fc400 | 2667 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2668 | { |
def596e9 | 2669 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2670 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2671 | struct file *file = req->file; |
09bb8394 JA |
2672 | unsigned ioprio; |
2673 | int ret; | |
2b188cc1 | 2674 | |
7b29f92d | 2675 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2676 | req->flags |= REQ_F_ISREG; |
2677 | ||
2b188cc1 | 2678 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2679 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2680 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2681 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2682 | } |
2b188cc1 | 2683 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2684 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2685 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2686 | if (unlikely(ret)) | |
2687 | return ret; | |
2b188cc1 | 2688 | |
75c668cd PB |
2689 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2690 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2691 | req->flags |= REQ_F_NOWAIT; | |
2692 | ||
2b188cc1 JA |
2693 | ioprio = READ_ONCE(sqe->ioprio); |
2694 | if (ioprio) { | |
2695 | ret = ioprio_check_cap(ioprio); | |
2696 | if (ret) | |
09bb8394 | 2697 | return ret; |
2b188cc1 JA |
2698 | |
2699 | kiocb->ki_ioprio = ioprio; | |
2700 | } else | |
2701 | kiocb->ki_ioprio = get_current_ioprio(); | |
2702 | ||
def596e9 | 2703 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2704 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2705 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2706 | return -EOPNOTSUPP; |
2b188cc1 | 2707 | |
def596e9 JA |
2708 | kiocb->ki_flags |= IOCB_HIPRI; |
2709 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2710 | req->iopoll_completed = 0; |
def596e9 | 2711 | } else { |
09bb8394 JA |
2712 | if (kiocb->ki_flags & IOCB_HIPRI) |
2713 | return -EINVAL; | |
def596e9 JA |
2714 | kiocb->ki_complete = io_complete_rw; |
2715 | } | |
9adbd45d | 2716 | |
eae071c9 PB |
2717 | if (req->opcode == IORING_OP_READ_FIXED || |
2718 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2719 | req->imu = NULL; | |
2720 | io_req_set_rsrc_node(req); | |
2721 | } | |
2722 | ||
3529d8c2 JA |
2723 | req->rw.addr = READ_ONCE(sqe->addr); |
2724 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2725 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2726 | return 0; |
2b188cc1 JA |
2727 | } |
2728 | ||
2729 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2730 | { | |
2731 | switch (ret) { | |
2732 | case -EIOCBQUEUED: | |
2733 | break; | |
2734 | case -ERESTARTSYS: | |
2735 | case -ERESTARTNOINTR: | |
2736 | case -ERESTARTNOHAND: | |
2737 | case -ERESTART_RESTARTBLOCK: | |
2738 | /* | |
2739 | * We can't just restart the syscall, since previously | |
2740 | * submitted sqes may already be in progress. Just fail this | |
2741 | * IO with EINTR. | |
2742 | */ | |
2743 | ret = -EINTR; | |
df561f66 | 2744 | fallthrough; |
2b188cc1 JA |
2745 | default: |
2746 | kiocb->ki_complete(kiocb, ret, 0); | |
2747 | } | |
2748 | } | |
2749 | ||
a1d7c393 | 2750 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2751 | unsigned int issue_flags) |
ba816ad6 | 2752 | { |
ba04291e | 2753 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2754 | struct io_async_rw *io = req->async_data; |
97284637 | 2755 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2756 | |
227c0c96 | 2757 | /* add previously done IO, if any */ |
e8c2bc1f | 2758 | if (io && io->bytes_done > 0) { |
227c0c96 | 2759 | if (ret < 0) |
e8c2bc1f | 2760 | ret = io->bytes_done; |
227c0c96 | 2761 | else |
e8c2bc1f | 2762 | ret += io->bytes_done; |
227c0c96 JA |
2763 | } |
2764 | ||
ba04291e JA |
2765 | if (req->flags & REQ_F_CUR_POS) |
2766 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2767 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2768 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2769 | else |
2770 | io_rw_done(kiocb, ret); | |
97284637 | 2771 | |
fe7e3257 | 2772 | if (check_reissue && (req->flags & REQ_F_REISSUE)) { |
97284637 | 2773 | req->flags &= ~REQ_F_REISSUE; |
a7be7c23 | 2774 | if (io_resubmit_prep(req)) { |
8c130827 PB |
2775 | req_ref_get(req); |
2776 | io_queue_async_work(req); | |
2777 | } else { | |
97284637 PB |
2778 | int cflags = 0; |
2779 | ||
93d2bcd2 | 2780 | req_set_fail(req); |
97284637 PB |
2781 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2782 | cflags = io_put_rw_kbuf(req); | |
2783 | __io_req_complete(req, issue_flags, ret, cflags); | |
2784 | } | |
2785 | } | |
ba816ad6 JA |
2786 | } |
2787 | ||
eae071c9 PB |
2788 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2789 | struct io_mapped_ubuf *imu) | |
edafccee | 2790 | { |
9adbd45d | 2791 | size_t len = req->rw.len; |
75769e3f | 2792 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2793 | size_t offset; |
edafccee | 2794 | |
75769e3f | 2795 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2796 | return -EFAULT; |
2797 | /* not inside the mapped region */ | |
4751f53d | 2798 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2799 | return -EFAULT; |
2800 | ||
2801 | /* | |
2802 | * May not be a start of buffer, set size appropriately | |
2803 | * and advance us to the beginning. | |
2804 | */ | |
2805 | offset = buf_addr - imu->ubuf; | |
2806 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2807 | |
2808 | if (offset) { | |
2809 | /* | |
2810 | * Don't use iov_iter_advance() here, as it's really slow for | |
2811 | * using the latter parts of a big fixed buffer - it iterates | |
2812 | * over each segment manually. We can cheat a bit here, because | |
2813 | * we know that: | |
2814 | * | |
2815 | * 1) it's a BVEC iter, we set it up | |
2816 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2817 | * first and last bvec | |
2818 | * | |
2819 | * So just find our index, and adjust the iterator afterwards. | |
2820 | * If the offset is within the first bvec (or the whole first | |
2821 | * bvec, just use iov_iter_advance(). This makes it easier | |
2822 | * since we can just skip the first segment, which may not | |
2823 | * be PAGE_SIZE aligned. | |
2824 | */ | |
2825 | const struct bio_vec *bvec = imu->bvec; | |
2826 | ||
2827 | if (offset <= bvec->bv_len) { | |
2828 | iov_iter_advance(iter, offset); | |
2829 | } else { | |
2830 | unsigned long seg_skip; | |
2831 | ||
2832 | /* skip first vec */ | |
2833 | offset -= bvec->bv_len; | |
2834 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2835 | ||
2836 | iter->bvec = bvec + seg_skip; | |
2837 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2838 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2839 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2840 | } |
2841 | } | |
2842 | ||
847595de | 2843 | return 0; |
edafccee JA |
2844 | } |
2845 | ||
eae071c9 PB |
2846 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
2847 | { | |
2848 | struct io_ring_ctx *ctx = req->ctx; | |
2849 | struct io_mapped_ubuf *imu = req->imu; | |
2850 | u16 index, buf_index = req->buf_index; | |
2851 | ||
2852 | if (likely(!imu)) { | |
2853 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
2854 | return -EFAULT; | |
2855 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2856 | imu = READ_ONCE(ctx->user_bufs[index]); | |
2857 | req->imu = imu; | |
2858 | } | |
2859 | return __io_import_fixed(req, rw, iter, imu); | |
2860 | } | |
2861 | ||
bcda7baa JA |
2862 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2863 | { | |
2864 | if (needs_lock) | |
2865 | mutex_unlock(&ctx->uring_lock); | |
2866 | } | |
2867 | ||
2868 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2869 | { | |
2870 | /* | |
2871 | * "Normal" inline submissions always hold the uring_lock, since we | |
2872 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2873 | * The only exception is when we've detached the request and issue it | |
2874 | * from an async worker thread, grab the lock for that case. | |
2875 | */ | |
2876 | if (needs_lock) | |
2877 | mutex_lock(&ctx->uring_lock); | |
2878 | } | |
2879 | ||
2880 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2881 | int bgid, struct io_buffer *kbuf, | |
2882 | bool needs_lock) | |
2883 | { | |
2884 | struct io_buffer *head; | |
2885 | ||
2886 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2887 | return kbuf; | |
2888 | ||
2889 | io_ring_submit_lock(req->ctx, needs_lock); | |
2890 | ||
2891 | lockdep_assert_held(&req->ctx->uring_lock); | |
2892 | ||
9e15c3a0 | 2893 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2894 | if (head) { |
2895 | if (!list_empty(&head->list)) { | |
2896 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2897 | list); | |
2898 | list_del(&kbuf->list); | |
2899 | } else { | |
2900 | kbuf = head; | |
9e15c3a0 | 2901 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2902 | } |
2903 | if (*len > kbuf->len) | |
2904 | *len = kbuf->len; | |
2905 | } else { | |
2906 | kbuf = ERR_PTR(-ENOBUFS); | |
2907 | } | |
2908 | ||
2909 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2910 | ||
2911 | return kbuf; | |
2912 | } | |
2913 | ||
4d954c25 JA |
2914 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2915 | bool needs_lock) | |
2916 | { | |
2917 | struct io_buffer *kbuf; | |
4f4eeba8 | 2918 | u16 bgid; |
4d954c25 JA |
2919 | |
2920 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2921 | bgid = req->buf_index; |
4d954c25 JA |
2922 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2923 | if (IS_ERR(kbuf)) | |
2924 | return kbuf; | |
2925 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2926 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2927 | return u64_to_user_ptr(kbuf->addr); | |
2928 | } | |
2929 | ||
2930 | #ifdef CONFIG_COMPAT | |
2931 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2932 | bool needs_lock) | |
2933 | { | |
2934 | struct compat_iovec __user *uiov; | |
2935 | compat_ssize_t clen; | |
2936 | void __user *buf; | |
2937 | ssize_t len; | |
2938 | ||
2939 | uiov = u64_to_user_ptr(req->rw.addr); | |
2940 | if (!access_ok(uiov, sizeof(*uiov))) | |
2941 | return -EFAULT; | |
2942 | if (__get_user(clen, &uiov->iov_len)) | |
2943 | return -EFAULT; | |
2944 | if (clen < 0) | |
2945 | return -EINVAL; | |
2946 | ||
2947 | len = clen; | |
2948 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2949 | if (IS_ERR(buf)) | |
2950 | return PTR_ERR(buf); | |
2951 | iov[0].iov_base = buf; | |
2952 | iov[0].iov_len = (compat_size_t) len; | |
2953 | return 0; | |
2954 | } | |
2955 | #endif | |
2956 | ||
2957 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2958 | bool needs_lock) | |
2959 | { | |
2960 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2961 | void __user *buf; | |
2962 | ssize_t len; | |
2963 | ||
2964 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2965 | return -EFAULT; | |
2966 | ||
2967 | len = iov[0].iov_len; | |
2968 | if (len < 0) | |
2969 | return -EINVAL; | |
2970 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2971 | if (IS_ERR(buf)) | |
2972 | return PTR_ERR(buf); | |
2973 | iov[0].iov_base = buf; | |
2974 | iov[0].iov_len = len; | |
2975 | return 0; | |
2976 | } | |
2977 | ||
2978 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2979 | bool needs_lock) | |
2980 | { | |
dddb3e26 JA |
2981 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2982 | struct io_buffer *kbuf; | |
2983 | ||
2984 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2985 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2986 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2987 | return 0; |
dddb3e26 | 2988 | } |
dd201662 | 2989 | if (req->rw.len != 1) |
4d954c25 JA |
2990 | return -EINVAL; |
2991 | ||
2992 | #ifdef CONFIG_COMPAT | |
2993 | if (req->ctx->compat) | |
2994 | return io_compat_import(req, iov, needs_lock); | |
2995 | #endif | |
2996 | ||
2997 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2998 | } | |
2999 | ||
847595de PB |
3000 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3001 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3002 | { |
9adbd45d JA |
3003 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3004 | size_t sqe_len = req->rw.len; | |
847595de | 3005 | u8 opcode = req->opcode; |
4d954c25 | 3006 | ssize_t ret; |
edafccee | 3007 | |
7d009165 | 3008 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3009 | *iovec = NULL; |
9adbd45d | 3010 | return io_import_fixed(req, rw, iter); |
edafccee | 3011 | } |
2b188cc1 | 3012 | |
bcda7baa | 3013 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3014 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3015 | return -EINVAL; |
3016 | ||
3a6820f2 | 3017 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3018 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3019 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3020 | if (IS_ERR(buf)) |
4d954c25 | 3021 | return PTR_ERR(buf); |
3f9d6441 | 3022 | req->rw.len = sqe_len; |
bcda7baa JA |
3023 | } |
3024 | ||
3a6820f2 JA |
3025 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3026 | *iovec = NULL; | |
10fc72e4 | 3027 | return ret; |
3a6820f2 JA |
3028 | } |
3029 | ||
4d954c25 JA |
3030 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3031 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3032 | if (!ret) |
3033 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3034 | *iovec = NULL; |
3035 | return ret; | |
3036 | } | |
3037 | ||
89cd35c5 CH |
3038 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3039 | req->ctx->compat); | |
2b188cc1 JA |
3040 | } |
3041 | ||
0fef9483 JA |
3042 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3043 | { | |
5b09e37e | 3044 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3045 | } |
3046 | ||
31b51510 | 3047 | /* |
32960613 JA |
3048 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3049 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3050 | */ |
4017eb91 | 3051 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3052 | { |
4017eb91 JA |
3053 | struct kiocb *kiocb = &req->rw.kiocb; |
3054 | struct file *file = req->file; | |
32960613 JA |
3055 | ssize_t ret = 0; |
3056 | ||
3057 | /* | |
3058 | * Don't support polled IO through this interface, and we can't | |
3059 | * support non-blocking either. For the latter, this just causes | |
3060 | * the kiocb to be handled from an async context. | |
3061 | */ | |
3062 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3063 | return -EOPNOTSUPP; | |
3064 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3065 | return -EAGAIN; | |
3066 | ||
3067 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3068 | struct iovec iovec; |
32960613 JA |
3069 | ssize_t nr; |
3070 | ||
311ae9e1 PB |
3071 | if (!iov_iter_is_bvec(iter)) { |
3072 | iovec = iov_iter_iovec(iter); | |
3073 | } else { | |
4017eb91 JA |
3074 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3075 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3076 | } |
3077 | ||
32960613 JA |
3078 | if (rw == READ) { |
3079 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3080 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3081 | } else { |
3082 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3083 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3084 | } |
3085 | ||
3086 | if (nr < 0) { | |
3087 | if (!ret) | |
3088 | ret = nr; | |
3089 | break; | |
3090 | } | |
3091 | ret += nr; | |
3092 | if (nr != iovec.iov_len) | |
3093 | break; | |
4017eb91 JA |
3094 | req->rw.len -= nr; |
3095 | req->rw.addr += nr; | |
32960613 JA |
3096 | iov_iter_advance(iter, nr); |
3097 | } | |
3098 | ||
3099 | return ret; | |
3100 | } | |
3101 | ||
ff6165b2 JA |
3102 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3103 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3104 | { |
e8c2bc1f | 3105 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3106 | |
ff6165b2 | 3107 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3108 | rw->free_iovec = iovec; |
227c0c96 | 3109 | rw->bytes_done = 0; |
ff6165b2 | 3110 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3111 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3112 | return; |
b64e3444 | 3113 | if (!iovec) { |
ff6165b2 JA |
3114 | unsigned iov_off = 0; |
3115 | ||
3116 | rw->iter.iov = rw->fast_iov; | |
3117 | if (iter->iov != fast_iov) { | |
3118 | iov_off = iter->iov - fast_iov; | |
3119 | rw->iter.iov += iov_off; | |
3120 | } | |
3121 | if (rw->fast_iov != fast_iov) | |
3122 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3123 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3124 | } else { |
3125 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3126 | } |
3127 | } | |
3128 | ||
6cb78689 | 3129 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3130 | { |
e8c2bc1f JA |
3131 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3132 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3133 | return req->async_data == NULL; | |
3d9932a8 XW |
3134 | } |
3135 | ||
ff6165b2 JA |
3136 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3137 | const struct iovec *fast_iov, | |
227c0c96 | 3138 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3139 | { |
26f0505a | 3140 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3141 | return 0; |
e8c2bc1f | 3142 | if (!req->async_data) { |
6cb78689 | 3143 | if (io_alloc_async_data(req)) { |
6bf985dc | 3144 | kfree(iovec); |
5d204bcf | 3145 | return -ENOMEM; |
6bf985dc | 3146 | } |
b7bb4f7d | 3147 | |
ff6165b2 | 3148 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3149 | } |
b7bb4f7d | 3150 | return 0; |
f67676d1 JA |
3151 | } |
3152 | ||
73debe68 | 3153 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3154 | { |
e8c2bc1f | 3155 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3156 | struct iovec *iov = iorw->fast_iov; |
847595de | 3157 | int ret; |
c3e330a4 | 3158 | |
2846c481 | 3159 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3160 | if (unlikely(ret < 0)) |
3161 | return ret; | |
3162 | ||
ab0b196c PB |
3163 | iorw->bytes_done = 0; |
3164 | iorw->free_iovec = iov; | |
3165 | if (iov) | |
3166 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3167 | return 0; |
3168 | } | |
3169 | ||
73debe68 | 3170 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3171 | { |
3529d8c2 JA |
3172 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3173 | return -EBADF; | |
93642ef8 | 3174 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3175 | } |
3176 | ||
c1dd91d1 JA |
3177 | /* |
3178 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3179 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3180 | * This gets called when the page is unlocked, and we generally expect that to | |
3181 | * happen when the page IO is completed and the page is now uptodate. This will | |
3182 | * queue a task_work based retry of the operation, attempting to copy the data | |
3183 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3184 | * do a thread based blocking retry of the operation. That's the unexpected | |
3185 | * slow path. | |
3186 | */ | |
bcf5a063 JA |
3187 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3188 | int sync, void *arg) | |
3189 | { | |
3190 | struct wait_page_queue *wpq; | |
3191 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3192 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3193 | |
3194 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3195 | ||
cdc8fcb4 LT |
3196 | if (!wake_page_match(wpq, key)) |
3197 | return 0; | |
3198 | ||
c8d317aa | 3199 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3200 | list_del_init(&wait->entry); |
3201 | ||
bcf5a063 | 3202 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3203 | req_ref_get(req); |
921b9054 | 3204 | io_req_task_queue(req); |
bcf5a063 JA |
3205 | return 1; |
3206 | } | |
3207 | ||
c1dd91d1 JA |
3208 | /* |
3209 | * This controls whether a given IO request should be armed for async page | |
3210 | * based retry. If we return false here, the request is handed to the async | |
3211 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3212 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3213 | * will either succeed because the page is now uptodate and unlocked, or it | |
3214 | * will register a callback when the page is unlocked at IO completion. Through | |
3215 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3216 | * That retry will attempt the buffered read again. The retry will generally | |
3217 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3218 | * async worker threads for a blocking retry. | |
3219 | */ | |
227c0c96 | 3220 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3221 | { |
e8c2bc1f JA |
3222 | struct io_async_rw *rw = req->async_data; |
3223 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3224 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3225 | |
bcf5a063 JA |
3226 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3227 | if (req->flags & REQ_F_NOWAIT) | |
3228 | return false; | |
f67676d1 | 3229 | |
227c0c96 | 3230 | /* Only for buffered IO */ |
3b2a4439 | 3231 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3232 | return false; |
3b2a4439 | 3233 | |
bcf5a063 JA |
3234 | /* |
3235 | * just use poll if we can, and don't attempt if the fs doesn't | |
3236 | * support callback based unlocks | |
3237 | */ | |
3238 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3239 | return false; | |
f67676d1 | 3240 | |
3b2a4439 JA |
3241 | wait->wait.func = io_async_buf_func; |
3242 | wait->wait.private = req; | |
3243 | wait->wait.flags = 0; | |
3244 | INIT_LIST_HEAD(&wait->wait.entry); | |
3245 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3246 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3247 | kiocb->ki_waitq = wait; |
3b2a4439 | 3248 | return true; |
bcf5a063 JA |
3249 | } |
3250 | ||
aeab9506 | 3251 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3252 | { |
3253 | if (req->file->f_op->read_iter) | |
3254 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3255 | else if (req->file->f_op->read) |
4017eb91 | 3256 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3257 | else |
3258 | return -EINVAL; | |
f67676d1 JA |
3259 | } |
3260 | ||
889fca73 | 3261 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3262 | { |
3263 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3264 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3265 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3266 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3267 | ssize_t io_size, ret, ret2; |
45d189c6 | 3268 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3269 | |
2846c481 | 3270 | if (rw) { |
e8c2bc1f | 3271 | iter = &rw->iter; |
2846c481 PB |
3272 | iovec = NULL; |
3273 | } else { | |
3274 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3275 | if (ret < 0) | |
3276 | return ret; | |
3277 | } | |
632546c4 | 3278 | io_size = iov_iter_count(iter); |
fa15bafb | 3279 | req->result = io_size; |
2b188cc1 | 3280 | |
fd6c2e4c JA |
3281 | /* Ensure we clear previously set non-block flag */ |
3282 | if (!force_nonblock) | |
29de5f6a | 3283 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3284 | else |
3285 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3286 | ||
24c74678 | 3287 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3288 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3289 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3290 | return ret ?: -EAGAIN; |
6713e7a6 | 3291 | } |
9e645e11 | 3292 | |
632546c4 | 3293 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3294 | if (unlikely(ret)) { |
3295 | kfree(iovec); | |
3296 | return ret; | |
3297 | } | |
2b188cc1 | 3298 | |
227c0c96 | 3299 | ret = io_iter_do_read(req, iter); |
32960613 | 3300 | |
230d50d4 | 3301 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3302 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3303 | /* IOPOLL retry should happen for io-wq threads */ |
3304 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3305 | goto done; |
75c668cd PB |
3306 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3307 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3308 | goto done; |
84216315 | 3309 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3310 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3311 | ret = 0; |
230d50d4 JA |
3312 | } else if (ret == -EIOCBQUEUED) { |
3313 | goto out_free; | |
7335e3bf | 3314 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3315 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3316 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3317 | goto done; |
227c0c96 JA |
3318 | } |
3319 | ||
227c0c96 | 3320 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3321 | if (ret2) |
3322 | return ret2; | |
3323 | ||
fe1cdd55 | 3324 | iovec = NULL; |
e8c2bc1f | 3325 | rw = req->async_data; |
227c0c96 | 3326 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3327 | iter = &rw->iter; |
227c0c96 | 3328 | |
b23df91b PB |
3329 | do { |
3330 | io_size -= ret; | |
3331 | rw->bytes_done += ret; | |
3332 | /* if we can retry, do so with the callbacks armed */ | |
3333 | if (!io_rw_should_retry(req)) { | |
3334 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3335 | return -EAGAIN; | |
3336 | } | |
3337 | ||
3338 | /* | |
3339 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3340 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3341 | * desired page gets unlocked. We can also get a partial read | |
3342 | * here, and if we do, then just retry at the new offset. | |
3343 | */ | |
3344 | ret = io_iter_do_read(req, iter); | |
3345 | if (ret == -EIOCBQUEUED) | |
3346 | return 0; | |
227c0c96 | 3347 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3348 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3349 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3350 | done: |
889fca73 | 3351 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3352 | out_free: |
3353 | /* it's faster to check here then delegate to kfree */ | |
3354 | if (iovec) | |
3355 | kfree(iovec); | |
5ea5dd45 | 3356 | return 0; |
2b188cc1 JA |
3357 | } |
3358 | ||
73debe68 | 3359 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3360 | { |
3529d8c2 JA |
3361 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3362 | return -EBADF; | |
93642ef8 | 3363 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3364 | } |
3365 | ||
889fca73 | 3366 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3367 | { |
3368 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3369 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3370 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3371 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3372 | ssize_t ret, ret2, io_size; |
45d189c6 | 3373 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3374 | |
2846c481 | 3375 | if (rw) { |
e8c2bc1f | 3376 | iter = &rw->iter; |
2846c481 PB |
3377 | iovec = NULL; |
3378 | } else { | |
3379 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3380 | if (ret < 0) | |
3381 | return ret; | |
3382 | } | |
632546c4 | 3383 | io_size = iov_iter_count(iter); |
fa15bafb | 3384 | req->result = io_size; |
2b188cc1 | 3385 | |
fd6c2e4c JA |
3386 | /* Ensure we clear previously set non-block flag */ |
3387 | if (!force_nonblock) | |
a88fc400 PB |
3388 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3389 | else | |
3390 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3391 | |
24c74678 | 3392 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3393 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3394 | goto copy_iov; |
31b51510 | 3395 | |
10d59345 JA |
3396 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3397 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3398 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3399 | goto copy_iov; |
31b51510 | 3400 | |
632546c4 | 3401 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3402 | if (unlikely(ret)) |
3403 | goto out_free; | |
4ed734b0 | 3404 | |
fa15bafb PB |
3405 | /* |
3406 | * Open-code file_start_write here to grab freeze protection, | |
3407 | * which will be released by another thread in | |
3408 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3409 | * released so that it doesn't complain about the held lock when | |
3410 | * we return to userspace. | |
3411 | */ | |
3412 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3413 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3414 | __sb_writers_release(file_inode(req->file)->i_sb, |
3415 | SB_FREEZE_WRITE); | |
3416 | } | |
3417 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3418 | |
fa15bafb | 3419 | if (req->file->f_op->write_iter) |
ff6165b2 | 3420 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3421 | else if (req->file->f_op->write) |
4017eb91 | 3422 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3423 | else |
3424 | ret2 = -EINVAL; | |
4ed734b0 | 3425 | |
6ad7f233 PB |
3426 | if (req->flags & REQ_F_REISSUE) { |
3427 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3428 | ret2 = -EAGAIN; |
6ad7f233 | 3429 | } |
230d50d4 | 3430 | |
fa15bafb PB |
3431 | /* |
3432 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3433 | * retry them without IOCB_NOWAIT. | |
3434 | */ | |
3435 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3436 | ret2 = -EAGAIN; | |
75c668cd PB |
3437 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3438 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3439 | goto done; |
fa15bafb | 3440 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3441 | /* IOPOLL retry should happen for io-wq threads */ |
3442 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3443 | goto copy_iov; | |
355afaeb | 3444 | done: |
889fca73 | 3445 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3446 | } else { |
f67676d1 | 3447 | copy_iov: |
84216315 | 3448 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3449 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3450 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3451 | return ret ?: -EAGAIN; |
2b188cc1 | 3452 | } |
31b51510 | 3453 | out_free: |
f261c168 | 3454 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3455 | if (iovec) |
6f2cc166 | 3456 | kfree(iovec); |
2b188cc1 JA |
3457 | return ret; |
3458 | } | |
3459 | ||
80a261fd JA |
3460 | static int io_renameat_prep(struct io_kiocb *req, |
3461 | const struct io_uring_sqe *sqe) | |
3462 | { | |
3463 | struct io_rename *ren = &req->rename; | |
3464 | const char __user *oldf, *newf; | |
3465 | ||
ed7eb259 JA |
3466 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3467 | return -EINVAL; | |
3468 | if (sqe->ioprio || sqe->buf_index) | |
3469 | return -EINVAL; | |
80a261fd JA |
3470 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3471 | return -EBADF; | |
3472 | ||
3473 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3474 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3475 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3476 | ren->new_dfd = READ_ONCE(sqe->len); | |
3477 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3478 | ||
3479 | ren->oldpath = getname(oldf); | |
3480 | if (IS_ERR(ren->oldpath)) | |
3481 | return PTR_ERR(ren->oldpath); | |
3482 | ||
3483 | ren->newpath = getname(newf); | |
3484 | if (IS_ERR(ren->newpath)) { | |
3485 | putname(ren->oldpath); | |
3486 | return PTR_ERR(ren->newpath); | |
3487 | } | |
3488 | ||
3489 | req->flags |= REQ_F_NEED_CLEANUP; | |
3490 | return 0; | |
3491 | } | |
3492 | ||
45d189c6 | 3493 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3494 | { |
3495 | struct io_rename *ren = &req->rename; | |
3496 | int ret; | |
3497 | ||
45d189c6 | 3498 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3499 | return -EAGAIN; |
3500 | ||
3501 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3502 | ren->newpath, ren->flags); | |
3503 | ||
3504 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3505 | if (ret < 0) | |
93d2bcd2 | 3506 | req_set_fail(req); |
80a261fd JA |
3507 | io_req_complete(req, ret); |
3508 | return 0; | |
3509 | } | |
3510 | ||
14a1143b JA |
3511 | static int io_unlinkat_prep(struct io_kiocb *req, |
3512 | const struct io_uring_sqe *sqe) | |
3513 | { | |
3514 | struct io_unlink *un = &req->unlink; | |
3515 | const char __user *fname; | |
3516 | ||
3517 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3518 | return -EBADF; | |
3519 | ||
3520 | un->dfd = READ_ONCE(sqe->fd); | |
3521 | ||
3522 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3523 | if (un->flags & ~AT_REMOVEDIR) | |
3524 | return -EINVAL; | |
3525 | ||
3526 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3527 | un->filename = getname(fname); | |
3528 | if (IS_ERR(un->filename)) | |
3529 | return PTR_ERR(un->filename); | |
3530 | ||
3531 | req->flags |= REQ_F_NEED_CLEANUP; | |
3532 | return 0; | |
3533 | } | |
3534 | ||
45d189c6 | 3535 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3536 | { |
3537 | struct io_unlink *un = &req->unlink; | |
3538 | int ret; | |
3539 | ||
45d189c6 | 3540 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3541 | return -EAGAIN; |
3542 | ||
3543 | if (un->flags & AT_REMOVEDIR) | |
3544 | ret = do_rmdir(un->dfd, un->filename); | |
3545 | else | |
3546 | ret = do_unlinkat(un->dfd, un->filename); | |
3547 | ||
3548 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3549 | if (ret < 0) | |
93d2bcd2 | 3550 | req_set_fail(req); |
14a1143b JA |
3551 | io_req_complete(req, ret); |
3552 | return 0; | |
3553 | } | |
3554 | ||
36f4fa68 JA |
3555 | static int io_shutdown_prep(struct io_kiocb *req, |
3556 | const struct io_uring_sqe *sqe) | |
3557 | { | |
3558 | #if defined(CONFIG_NET) | |
3559 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3560 | return -EINVAL; | |
3561 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3562 | sqe->buf_index) | |
3563 | return -EINVAL; | |
3564 | ||
3565 | req->shutdown.how = READ_ONCE(sqe->len); | |
3566 | return 0; | |
3567 | #else | |
3568 | return -EOPNOTSUPP; | |
3569 | #endif | |
3570 | } | |
3571 | ||
45d189c6 | 3572 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3573 | { |
3574 | #if defined(CONFIG_NET) | |
3575 | struct socket *sock; | |
3576 | int ret; | |
3577 | ||
45d189c6 | 3578 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3579 | return -EAGAIN; |
3580 | ||
48aba79b | 3581 | sock = sock_from_file(req->file); |
36f4fa68 | 3582 | if (unlikely(!sock)) |
48aba79b | 3583 | return -ENOTSOCK; |
36f4fa68 JA |
3584 | |
3585 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3586 | if (ret < 0) |
93d2bcd2 | 3587 | req_set_fail(req); |
36f4fa68 JA |
3588 | io_req_complete(req, ret); |
3589 | return 0; | |
3590 | #else | |
3591 | return -EOPNOTSUPP; | |
3592 | #endif | |
3593 | } | |
3594 | ||
f2a8d5c7 PB |
3595 | static int __io_splice_prep(struct io_kiocb *req, |
3596 | const struct io_uring_sqe *sqe) | |
7d67af2c | 3597 | { |
fe7e3257 | 3598 | struct io_splice *sp = &req->splice; |
7d67af2c | 3599 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 3600 | |
3232dd02 PB |
3601 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3602 | return -EINVAL; | |
7d67af2c PB |
3603 | |
3604 | sp->file_in = NULL; | |
7d67af2c PB |
3605 | sp->len = READ_ONCE(sqe->len); |
3606 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3607 | ||
3608 | if (unlikely(sp->flags & ~valid_flags)) | |
3609 | return -EINVAL; | |
3610 | ||
8371adf5 PB |
3611 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3612 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3613 | if (!sp->file_in) | |
3614 | return -EBADF; | |
7d67af2c | 3615 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3616 | return 0; |
3617 | } | |
3618 | ||
f2a8d5c7 PB |
3619 | static int io_tee_prep(struct io_kiocb *req, |
3620 | const struct io_uring_sqe *sqe) | |
3621 | { | |
3622 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3623 | return -EINVAL; | |
3624 | return __io_splice_prep(req, sqe); | |
3625 | } | |
3626 | ||
45d189c6 | 3627 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3628 | { |
3629 | struct io_splice *sp = &req->splice; | |
3630 | struct file *in = sp->file_in; | |
3631 | struct file *out = sp->file_out; | |
3632 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3633 | long ret = 0; | |
3634 | ||
45d189c6 | 3635 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3636 | return -EAGAIN; |
3637 | if (sp->len) | |
3638 | ret = do_tee(in, out, sp->len, flags); | |
3639 | ||
e1d767f0 PB |
3640 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3641 | io_put_file(in); | |
f2a8d5c7 PB |
3642 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3643 | ||
f2a8d5c7 | 3644 | if (ret != sp->len) |
93d2bcd2 | 3645 | req_set_fail(req); |
e1e16097 | 3646 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3647 | return 0; |
3648 | } | |
3649 | ||
3650 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3651 | { | |
fe7e3257 | 3652 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
3653 | |
3654 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3655 | sp->off_out = READ_ONCE(sqe->off); | |
3656 | return __io_splice_prep(req, sqe); | |
3657 | } | |
3658 | ||
45d189c6 | 3659 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3660 | { |
3661 | struct io_splice *sp = &req->splice; | |
3662 | struct file *in = sp->file_in; | |
3663 | struct file *out = sp->file_out; | |
3664 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3665 | loff_t *poff_in, *poff_out; | |
c9687426 | 3666 | long ret = 0; |
7d67af2c | 3667 | |
45d189c6 | 3668 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3669 | return -EAGAIN; |
7d67af2c PB |
3670 | |
3671 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3672 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3673 | |
948a7749 | 3674 | if (sp->len) |
c9687426 | 3675 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3676 | |
e1d767f0 PB |
3677 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3678 | io_put_file(in); | |
7d67af2c PB |
3679 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3680 | ||
7d67af2c | 3681 | if (ret != sp->len) |
93d2bcd2 | 3682 | req_set_fail(req); |
e1e16097 | 3683 | io_req_complete(req, ret); |
7d67af2c PB |
3684 | return 0; |
3685 | } | |
3686 | ||
2b188cc1 JA |
3687 | /* |
3688 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3689 | */ | |
889fca73 | 3690 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3691 | { |
3692 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3693 | |
def596e9 JA |
3694 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3695 | return -EINVAL; | |
3696 | ||
889fca73 | 3697 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3698 | return 0; |
3699 | } | |
3700 | ||
1155c76a | 3701 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3702 | { |
6b06314c | 3703 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3704 | |
09bb8394 JA |
3705 | if (!req->file) |
3706 | return -EBADF; | |
c992fe29 | 3707 | |
6b06314c | 3708 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3709 | return -EINVAL; |
edafccee | 3710 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3711 | return -EINVAL; |
3712 | ||
8ed8d3c3 JA |
3713 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3714 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3715 | return -EINVAL; | |
3716 | ||
3717 | req->sync.off = READ_ONCE(sqe->off); | |
3718 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3719 | return 0; |
3720 | } | |
3721 | ||
45d189c6 | 3722 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3723 | { |
8ed8d3c3 | 3724 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3725 | int ret; |
3726 | ||
ac45abc0 | 3727 | /* fsync always requires a blocking context */ |
45d189c6 | 3728 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3729 | return -EAGAIN; |
3730 | ||
9adbd45d | 3731 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3732 | end > 0 ? end : LLONG_MAX, |
3733 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3734 | if (ret < 0) | |
93d2bcd2 | 3735 | req_set_fail(req); |
e1e16097 | 3736 | io_req_complete(req, ret); |
c992fe29 CH |
3737 | return 0; |
3738 | } | |
3739 | ||
d63d1b5e JA |
3740 | static int io_fallocate_prep(struct io_kiocb *req, |
3741 | const struct io_uring_sqe *sqe) | |
3742 | { | |
3743 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3744 | return -EINVAL; | |
3232dd02 PB |
3745 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3746 | return -EINVAL; | |
d63d1b5e JA |
3747 | |
3748 | req->sync.off = READ_ONCE(sqe->off); | |
3749 | req->sync.len = READ_ONCE(sqe->addr); | |
3750 | req->sync.mode = READ_ONCE(sqe->len); | |
3751 | return 0; | |
3752 | } | |
3753 | ||
45d189c6 | 3754 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3755 | { |
ac45abc0 PB |
3756 | int ret; |
3757 | ||
d63d1b5e | 3758 | /* fallocate always requiring blocking context */ |
45d189c6 | 3759 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3760 | return -EAGAIN; |
ac45abc0 PB |
3761 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3762 | req->sync.len); | |
ac45abc0 | 3763 | if (ret < 0) |
93d2bcd2 | 3764 | req_set_fail(req); |
e1e16097 | 3765 | io_req_complete(req, ret); |
5d17b4a4 JA |
3766 | return 0; |
3767 | } | |
3768 | ||
ec65fea5 | 3769 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3770 | { |
f8748881 | 3771 | const char __user *fname; |
15b71abe | 3772 | int ret; |
b7bb4f7d | 3773 | |
ec65fea5 | 3774 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3775 | return -EINVAL; |
ec65fea5 | 3776 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3777 | return -EBADF; |
03b1230c | 3778 | |
ec65fea5 PB |
3779 | /* open.how should be already initialised */ |
3780 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3781 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3782 | |
25e72d10 PB |
3783 | req->open.dfd = READ_ONCE(sqe->fd); |
3784 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3785 | req->open.filename = getname(fname); |
15b71abe JA |
3786 | if (IS_ERR(req->open.filename)) { |
3787 | ret = PTR_ERR(req->open.filename); | |
3788 | req->open.filename = NULL; | |
3789 | return ret; | |
3790 | } | |
4022e7af | 3791 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3792 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3793 | return 0; |
03b1230c JA |
3794 | } |
3795 | ||
ec65fea5 PB |
3796 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3797 | { | |
3798 | u64 flags, mode; | |
3799 | ||
14587a46 | 3800 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3801 | return -EINVAL; |
ec65fea5 PB |
3802 | mode = READ_ONCE(sqe->len); |
3803 | flags = READ_ONCE(sqe->open_flags); | |
3804 | req->open.how = build_open_how(flags, mode); | |
3805 | return __io_openat_prep(req, sqe); | |
3806 | } | |
3807 | ||
cebdb986 | 3808 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3809 | { |
cebdb986 | 3810 | struct open_how __user *how; |
cebdb986 | 3811 | size_t len; |
0fa03c62 JA |
3812 | int ret; |
3813 | ||
14587a46 | 3814 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3815 | return -EINVAL; |
cebdb986 JA |
3816 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3817 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3818 | if (len < OPEN_HOW_SIZE_VER0) |
3819 | return -EINVAL; | |
3529d8c2 | 3820 | |
cebdb986 JA |
3821 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3822 | len); | |
3823 | if (ret) | |
3824 | return ret; | |
3529d8c2 | 3825 | |
ec65fea5 | 3826 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3827 | } |
3828 | ||
45d189c6 | 3829 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3830 | { |
3831 | struct open_flags op; | |
15b71abe | 3832 | struct file *file; |
3a81fd02 JA |
3833 | bool nonblock_set; |
3834 | bool resolve_nonblock; | |
15b71abe JA |
3835 | int ret; |
3836 | ||
cebdb986 | 3837 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3838 | if (ret) |
3839 | goto err; | |
3a81fd02 JA |
3840 | nonblock_set = op.open_flag & O_NONBLOCK; |
3841 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3842 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3843 | /* |
3844 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3845 | * it'll always -EAGAIN | |
3846 | */ | |
3847 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3848 | return -EAGAIN; | |
3849 | op.lookup_flags |= LOOKUP_CACHED; | |
3850 | op.open_flag |= O_NONBLOCK; | |
3851 | } | |
15b71abe | 3852 | |
4022e7af | 3853 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3854 | if (ret < 0) |
3855 | goto err; | |
3856 | ||
3857 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 3858 | if (IS_ERR(file)) { |
944d1444 | 3859 | /* |
12dcb58a PB |
3860 | * We could hang on to this 'fd' on retrying, but seems like |
3861 | * marginal gain for something that is now known to be a slower | |
3862 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 3863 | */ |
3a81fd02 | 3864 | put_unused_fd(ret); |
3a81fd02 | 3865 | |
15b71abe | 3866 | ret = PTR_ERR(file); |
12dcb58a PB |
3867 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
3868 | if (ret == -EAGAIN && | |
3869 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
3870 | return -EAGAIN; | |
3871 | goto err; | |
15b71abe | 3872 | } |
12dcb58a PB |
3873 | |
3874 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
3875 | file->f_flags &= ~O_NONBLOCK; | |
3876 | fsnotify_open(file); | |
3877 | fd_install(ret, file); | |
15b71abe JA |
3878 | err: |
3879 | putname(req->open.filename); | |
8fef80bf | 3880 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 3881 | if (ret < 0) |
93d2bcd2 | 3882 | req_set_fail(req); |
0bdf3398 | 3883 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3884 | return 0; |
3885 | } | |
3886 | ||
45d189c6 | 3887 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3888 | { |
e45cff58 | 3889 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3890 | } |
3891 | ||
067524e9 JA |
3892 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3893 | const struct io_uring_sqe *sqe) | |
3894 | { | |
3895 | struct io_provide_buf *p = &req->pbuf; | |
3896 | u64 tmp; | |
3897 | ||
3898 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3899 | return -EINVAL; | |
3900 | ||
3901 | tmp = READ_ONCE(sqe->fd); | |
3902 | if (!tmp || tmp > USHRT_MAX) | |
3903 | return -EINVAL; | |
3904 | ||
3905 | memset(p, 0, sizeof(*p)); | |
3906 | p->nbufs = tmp; | |
3907 | p->bgid = READ_ONCE(sqe->buf_group); | |
3908 | return 0; | |
3909 | } | |
3910 | ||
3911 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3912 | int bgid, unsigned nbufs) | |
3913 | { | |
3914 | unsigned i = 0; | |
3915 | ||
3916 | /* shouldn't happen */ | |
3917 | if (!nbufs) | |
3918 | return 0; | |
3919 | ||
3920 | /* the head kbuf is the list itself */ | |
3921 | while (!list_empty(&buf->list)) { | |
3922 | struct io_buffer *nxt; | |
3923 | ||
3924 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3925 | list_del(&nxt->list); | |
3926 | kfree(nxt); | |
3927 | if (++i == nbufs) | |
3928 | return i; | |
3929 | } | |
3930 | i++; | |
3931 | kfree(buf); | |
9e15c3a0 | 3932 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3933 | |
3934 | return i; | |
3935 | } | |
3936 | ||
889fca73 | 3937 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3938 | { |
3939 | struct io_provide_buf *p = &req->pbuf; | |
3940 | struct io_ring_ctx *ctx = req->ctx; | |
3941 | struct io_buffer *head; | |
3942 | int ret = 0; | |
45d189c6 | 3943 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3944 | |
3945 | io_ring_submit_lock(ctx, !force_nonblock); | |
3946 | ||
3947 | lockdep_assert_held(&ctx->uring_lock); | |
3948 | ||
3949 | ret = -ENOENT; | |
9e15c3a0 | 3950 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3951 | if (head) |
3952 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 3953 | if (ret < 0) |
93d2bcd2 | 3954 | req_set_fail(req); |
067524e9 | 3955 | |
9fb8cb49 PB |
3956 | /* complete before unlock, IOPOLL may need the lock */ |
3957 | __io_req_complete(req, issue_flags, ret, 0); | |
3958 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3959 | return 0; |
3960 | } | |
3961 | ||
ddf0322d JA |
3962 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3963 | const struct io_uring_sqe *sqe) | |
3964 | { | |
38134ada | 3965 | unsigned long size, tmp_check; |
ddf0322d JA |
3966 | struct io_provide_buf *p = &req->pbuf; |
3967 | u64 tmp; | |
3968 | ||
3969 | if (sqe->ioprio || sqe->rw_flags) | |
3970 | return -EINVAL; | |
3971 | ||
3972 | tmp = READ_ONCE(sqe->fd); | |
3973 | if (!tmp || tmp > USHRT_MAX) | |
3974 | return -E2BIG; | |
3975 | p->nbufs = tmp; | |
3976 | p->addr = READ_ONCE(sqe->addr); | |
3977 | p->len = READ_ONCE(sqe->len); | |
3978 | ||
38134ada PB |
3979 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
3980 | &size)) | |
3981 | return -EOVERFLOW; | |
3982 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
3983 | return -EOVERFLOW; | |
3984 | ||
d81269fe PB |
3985 | size = (unsigned long)p->len * p->nbufs; |
3986 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3987 | return -EFAULT; |
3988 | ||
3989 | p->bgid = READ_ONCE(sqe->buf_group); | |
3990 | tmp = READ_ONCE(sqe->off); | |
3991 | if (tmp > USHRT_MAX) | |
3992 | return -E2BIG; | |
3993 | p->bid = tmp; | |
3994 | return 0; | |
3995 | } | |
3996 | ||
3997 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3998 | { | |
3999 | struct io_buffer *buf; | |
4000 | u64 addr = pbuf->addr; | |
4001 | int i, bid = pbuf->bid; | |
4002 | ||
4003 | for (i = 0; i < pbuf->nbufs; i++) { | |
4004 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4005 | if (!buf) | |
4006 | break; | |
4007 | ||
4008 | buf->addr = addr; | |
d1f82808 | 4009 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
4010 | buf->bid = bid; |
4011 | addr += pbuf->len; | |
4012 | bid++; | |
4013 | if (!*head) { | |
4014 | INIT_LIST_HEAD(&buf->list); | |
4015 | *head = buf; | |
4016 | } else { | |
4017 | list_add_tail(&buf->list, &(*head)->list); | |
4018 | } | |
4019 | } | |
4020 | ||
4021 | return i ? i : -ENOMEM; | |
4022 | } | |
4023 | ||
889fca73 | 4024 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4025 | { |
4026 | struct io_provide_buf *p = &req->pbuf; | |
4027 | struct io_ring_ctx *ctx = req->ctx; | |
4028 | struct io_buffer *head, *list; | |
4029 | int ret = 0; | |
45d189c6 | 4030 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4031 | |
4032 | io_ring_submit_lock(ctx, !force_nonblock); | |
4033 | ||
4034 | lockdep_assert_held(&ctx->uring_lock); | |
4035 | ||
9e15c3a0 | 4036 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4037 | |
4038 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4039 | if (ret >= 0 && !list) { |
4040 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4041 | if (ret < 0) | |
067524e9 | 4042 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4043 | } |
ddf0322d | 4044 | if (ret < 0) |
93d2bcd2 | 4045 | req_set_fail(req); |
9fb8cb49 PB |
4046 | /* complete before unlock, IOPOLL may need the lock */ |
4047 | __io_req_complete(req, issue_flags, ret, 0); | |
4048 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4049 | return 0; |
cebdb986 JA |
4050 | } |
4051 | ||
3e4827b0 JA |
4052 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4053 | const struct io_uring_sqe *sqe) | |
4054 | { | |
4055 | #if defined(CONFIG_EPOLL) | |
4056 | if (sqe->ioprio || sqe->buf_index) | |
4057 | return -EINVAL; | |
2d74d042 | 4058 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4059 | return -EINVAL; |
3e4827b0 JA |
4060 | |
4061 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4062 | req->epoll.op = READ_ONCE(sqe->len); | |
4063 | req->epoll.fd = READ_ONCE(sqe->off); | |
4064 | ||
4065 | if (ep_op_has_event(req->epoll.op)) { | |
4066 | struct epoll_event __user *ev; | |
4067 | ||
4068 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4069 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4070 | return -EFAULT; | |
4071 | } | |
4072 | ||
4073 | return 0; | |
4074 | #else | |
4075 | return -EOPNOTSUPP; | |
4076 | #endif | |
4077 | } | |
4078 | ||
889fca73 | 4079 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4080 | { |
4081 | #if defined(CONFIG_EPOLL) | |
4082 | struct io_epoll *ie = &req->epoll; | |
4083 | int ret; | |
45d189c6 | 4084 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4085 | |
4086 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4087 | if (force_nonblock && ret == -EAGAIN) | |
4088 | return -EAGAIN; | |
4089 | ||
4090 | if (ret < 0) | |
93d2bcd2 | 4091 | req_set_fail(req); |
889fca73 | 4092 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4093 | return 0; |
4094 | #else | |
4095 | return -EOPNOTSUPP; | |
4096 | #endif | |
4097 | } | |
4098 | ||
c1ca757b JA |
4099 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4100 | { | |
4101 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4102 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4103 | return -EINVAL; | |
3232dd02 PB |
4104 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4105 | return -EINVAL; | |
c1ca757b JA |
4106 | |
4107 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4108 | req->madvise.len = READ_ONCE(sqe->len); | |
4109 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4110 | return 0; | |
4111 | #else | |
4112 | return -EOPNOTSUPP; | |
4113 | #endif | |
4114 | } | |
4115 | ||
45d189c6 | 4116 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4117 | { |
4118 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4119 | struct io_madvise *ma = &req->madvise; | |
4120 | int ret; | |
4121 | ||
45d189c6 | 4122 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4123 | return -EAGAIN; |
4124 | ||
0726b01e | 4125 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4126 | if (ret < 0) |
93d2bcd2 | 4127 | req_set_fail(req); |
e1e16097 | 4128 | io_req_complete(req, ret); |
c1ca757b JA |
4129 | return 0; |
4130 | #else | |
4131 | return -EOPNOTSUPP; | |
4132 | #endif | |
4133 | } | |
4134 | ||
4840e418 JA |
4135 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4136 | { | |
4137 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4138 | return -EINVAL; | |
3232dd02 PB |
4139 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4140 | return -EINVAL; | |
4840e418 JA |
4141 | |
4142 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4143 | req->fadvise.len = READ_ONCE(sqe->len); | |
4144 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4145 | return 0; | |
4146 | } | |
4147 | ||
45d189c6 | 4148 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4149 | { |
4150 | struct io_fadvise *fa = &req->fadvise; | |
4151 | int ret; | |
4152 | ||
45d189c6 | 4153 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4154 | switch (fa->advice) { |
4155 | case POSIX_FADV_NORMAL: | |
4156 | case POSIX_FADV_RANDOM: | |
4157 | case POSIX_FADV_SEQUENTIAL: | |
4158 | break; | |
4159 | default: | |
4160 | return -EAGAIN; | |
4161 | } | |
4162 | } | |
4840e418 JA |
4163 | |
4164 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4165 | if (ret < 0) | |
93d2bcd2 | 4166 | req_set_fail(req); |
0bdf3398 | 4167 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4168 | return 0; |
4169 | } | |
4170 | ||
eddc7ef5 JA |
4171 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4172 | { | |
2d74d042 | 4173 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4174 | return -EINVAL; |
eddc7ef5 JA |
4175 | if (sqe->ioprio || sqe->buf_index) |
4176 | return -EINVAL; | |
9c280f90 | 4177 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4178 | return -EBADF; |
eddc7ef5 | 4179 | |
1d9e1288 BM |
4180 | req->statx.dfd = READ_ONCE(sqe->fd); |
4181 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4182 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4183 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4184 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4185 | |
4186 | return 0; | |
4187 | } | |
4188 | ||
45d189c6 | 4189 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4190 | { |
1d9e1288 | 4191 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4192 | int ret; |
4193 | ||
59d70013 | 4194 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4195 | return -EAGAIN; |
4196 | ||
e62753e4 BM |
4197 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4198 | ctx->buffer); | |
eddc7ef5 | 4199 | |
eddc7ef5 | 4200 | if (ret < 0) |
93d2bcd2 | 4201 | req_set_fail(req); |
e1e16097 | 4202 | io_req_complete(req, ret); |
eddc7ef5 JA |
4203 | return 0; |
4204 | } | |
4205 | ||
b5dba59e JA |
4206 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4207 | { | |
14587a46 | 4208 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4209 | return -EINVAL; |
b5dba59e JA |
4210 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4211 | sqe->rw_flags || sqe->buf_index) | |
4212 | return -EINVAL; | |
9c280f90 | 4213 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4214 | return -EBADF; |
b5dba59e JA |
4215 | |
4216 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4217 | return 0; |
b5dba59e JA |
4218 | } |
4219 | ||
889fca73 | 4220 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4221 | { |
9eac1904 | 4222 | struct files_struct *files = current->files; |
3af73b28 | 4223 | struct io_close *close = &req->close; |
9eac1904 | 4224 | struct fdtable *fdt; |
a1fde923 PB |
4225 | struct file *file = NULL; |
4226 | int ret = -EBADF; | |
b5dba59e | 4227 | |
9eac1904 JA |
4228 | spin_lock(&files->file_lock); |
4229 | fdt = files_fdtable(files); | |
4230 | if (close->fd >= fdt->max_fds) { | |
4231 | spin_unlock(&files->file_lock); | |
4232 | goto err; | |
4233 | } | |
4234 | file = fdt->fd[close->fd]; | |
a1fde923 | 4235 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4236 | spin_unlock(&files->file_lock); |
4237 | file = NULL; | |
4238 | goto err; | |
3af73b28 | 4239 | } |
b5dba59e JA |
4240 | |
4241 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4242 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4243 | spin_unlock(&files->file_lock); |
0bf0eefd | 4244 | return -EAGAIN; |
a2100672 | 4245 | } |
b5dba59e | 4246 | |
9eac1904 JA |
4247 | ret = __close_fd_get_file(close->fd, &file); |
4248 | spin_unlock(&files->file_lock); | |
4249 | if (ret < 0) { | |
4250 | if (ret == -ENOENT) | |
4251 | ret = -EBADF; | |
4252 | goto err; | |
4253 | } | |
4254 | ||
3af73b28 | 4255 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4256 | ret = filp_close(file, current->files); |
4257 | err: | |
3af73b28 | 4258 | if (ret < 0) |
93d2bcd2 | 4259 | req_set_fail(req); |
9eac1904 JA |
4260 | if (file) |
4261 | fput(file); | |
889fca73 | 4262 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4263 | return 0; |
b5dba59e JA |
4264 | } |
4265 | ||
1155c76a | 4266 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4267 | { |
4268 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4269 | |
5d17b4a4 JA |
4270 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4271 | return -EINVAL; | |
4272 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4273 | return -EINVAL; | |
4274 | ||
8ed8d3c3 JA |
4275 | req->sync.off = READ_ONCE(sqe->off); |
4276 | req->sync.len = READ_ONCE(sqe->len); | |
4277 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4278 | return 0; |
4279 | } | |
4280 | ||
45d189c6 | 4281 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4282 | { |
8ed8d3c3 JA |
4283 | int ret; |
4284 | ||
ac45abc0 | 4285 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4286 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4287 | return -EAGAIN; |
4288 | ||
9adbd45d | 4289 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4290 | req->sync.flags); |
4291 | if (ret < 0) | |
93d2bcd2 | 4292 | req_set_fail(req); |
e1e16097 | 4293 | io_req_complete(req, ret); |
5d17b4a4 JA |
4294 | return 0; |
4295 | } | |
4296 | ||
469956e8 | 4297 | #if defined(CONFIG_NET) |
02d27d89 PB |
4298 | static int io_setup_async_msg(struct io_kiocb *req, |
4299 | struct io_async_msghdr *kmsg) | |
4300 | { | |
e8c2bc1f JA |
4301 | struct io_async_msghdr *async_msg = req->async_data; |
4302 | ||
4303 | if (async_msg) | |
02d27d89 | 4304 | return -EAGAIN; |
e8c2bc1f | 4305 | if (io_alloc_async_data(req)) { |
257e84a5 | 4306 | kfree(kmsg->free_iov); |
02d27d89 PB |
4307 | return -ENOMEM; |
4308 | } | |
e8c2bc1f | 4309 | async_msg = req->async_data; |
02d27d89 | 4310 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4311 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4312 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4313 | /* if were using fast_iov, set it to the new one */ |
4314 | if (!async_msg->free_iov) | |
4315 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4316 | ||
02d27d89 PB |
4317 | return -EAGAIN; |
4318 | } | |
4319 | ||
2ae523ed PB |
4320 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4321 | struct io_async_msghdr *iomsg) | |
4322 | { | |
2ae523ed | 4323 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4324 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4325 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4326 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4327 | } |
4328 | ||
93642ef8 PB |
4329 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4330 | { | |
4331 | int ret; | |
4332 | ||
93642ef8 PB |
4333 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4334 | if (!ret) | |
4335 | req->flags |= REQ_F_NEED_CLEANUP; | |
4336 | return ret; | |
4337 | } | |
4338 | ||
3529d8c2 | 4339 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4340 | { |
e47293fd | 4341 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4342 | |
d2b6f48b PB |
4343 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4344 | return -EINVAL; | |
4345 | ||
270a5940 | 4346 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4347 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4348 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4349 | if (sr->msg_flags & MSG_DONTWAIT) | |
4350 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4351 | |
d8768362 JA |
4352 | #ifdef CONFIG_COMPAT |
4353 | if (req->ctx->compat) | |
4354 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4355 | #endif | |
93642ef8 | 4356 | return 0; |
03b1230c JA |
4357 | } |
4358 | ||
889fca73 | 4359 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4360 | { |
6b754c8b | 4361 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4362 | struct socket *sock; |
7a7cacba | 4363 | unsigned flags; |
0031275d | 4364 | int min_ret = 0; |
0fa03c62 JA |
4365 | int ret; |
4366 | ||
dba4a925 | 4367 | sock = sock_from_file(req->file); |
7a7cacba | 4368 | if (unlikely(!sock)) |
dba4a925 | 4369 | return -ENOTSOCK; |
3529d8c2 | 4370 | |
257e84a5 PB |
4371 | kmsg = req->async_data; |
4372 | if (!kmsg) { | |
7a7cacba PB |
4373 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4374 | if (ret) | |
4375 | return ret; | |
4376 | kmsg = &iomsg; | |
0fa03c62 | 4377 | } |
0fa03c62 | 4378 | |
04411806 PB |
4379 | flags = req->sr_msg.msg_flags; |
4380 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4381 | flags |= MSG_DONTWAIT; |
0031275d SM |
4382 | if (flags & MSG_WAITALL) |
4383 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4384 | ||
7a7cacba | 4385 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4386 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4387 | return io_setup_async_msg(req, kmsg); |
4388 | if (ret == -ERESTARTSYS) | |
4389 | ret = -EINTR; | |
0fa03c62 | 4390 | |
257e84a5 PB |
4391 | /* fast path, check for non-NULL to avoid function call */ |
4392 | if (kmsg->free_iov) | |
4393 | kfree(kmsg->free_iov); | |
99bc4c38 | 4394 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4395 | if (ret < min_ret) |
93d2bcd2 | 4396 | req_set_fail(req); |
889fca73 | 4397 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4398 | return 0; |
03b1230c | 4399 | } |
aa1fa28f | 4400 | |
889fca73 | 4401 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4402 | { |
7a7cacba PB |
4403 | struct io_sr_msg *sr = &req->sr_msg; |
4404 | struct msghdr msg; | |
4405 | struct iovec iov; | |
fddaface | 4406 | struct socket *sock; |
7a7cacba | 4407 | unsigned flags; |
0031275d | 4408 | int min_ret = 0; |
fddaface JA |
4409 | int ret; |
4410 | ||
dba4a925 | 4411 | sock = sock_from_file(req->file); |
7a7cacba | 4412 | if (unlikely(!sock)) |
dba4a925 | 4413 | return -ENOTSOCK; |
fddaface | 4414 | |
7a7cacba PB |
4415 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4416 | if (unlikely(ret)) | |
14db8411 | 4417 | return ret; |
fddaface | 4418 | |
7a7cacba PB |
4419 | msg.msg_name = NULL; |
4420 | msg.msg_control = NULL; | |
4421 | msg.msg_controllen = 0; | |
4422 | msg.msg_namelen = 0; | |
fddaface | 4423 | |
04411806 PB |
4424 | flags = req->sr_msg.msg_flags; |
4425 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4426 | flags |= MSG_DONTWAIT; |
0031275d SM |
4427 | if (flags & MSG_WAITALL) |
4428 | min_ret = iov_iter_count(&msg.msg_iter); | |
4429 | ||
7a7cacba PB |
4430 | msg.msg_flags = flags; |
4431 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4432 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4433 | return -EAGAIN; |
4434 | if (ret == -ERESTARTSYS) | |
4435 | ret = -EINTR; | |
fddaface | 4436 | |
0031275d | 4437 | if (ret < min_ret) |
93d2bcd2 | 4438 | req_set_fail(req); |
889fca73 | 4439 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4440 | return 0; |
fddaface JA |
4441 | } |
4442 | ||
1400e697 PB |
4443 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4444 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4445 | { |
4446 | struct io_sr_msg *sr = &req->sr_msg; | |
4447 | struct iovec __user *uiov; | |
4448 | size_t iov_len; | |
4449 | int ret; | |
4450 | ||
1400e697 PB |
4451 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4452 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4453 | if (ret) |
4454 | return ret; | |
4455 | ||
4456 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4457 | if (iov_len > 1) | |
4458 | return -EINVAL; | |
5476dfed | 4459 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4460 | return -EFAULT; |
5476dfed | 4461 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4462 | iomsg->free_iov = NULL; |
52de1fe1 | 4463 | } else { |
257e84a5 | 4464 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4465 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4466 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4467 | false); |
52de1fe1 JA |
4468 | if (ret > 0) |
4469 | ret = 0; | |
4470 | } | |
4471 | ||
4472 | return ret; | |
4473 | } | |
4474 | ||
4475 | #ifdef CONFIG_COMPAT | |
4476 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4477 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4478 | { |
52de1fe1 JA |
4479 | struct io_sr_msg *sr = &req->sr_msg; |
4480 | struct compat_iovec __user *uiov; | |
4481 | compat_uptr_t ptr; | |
4482 | compat_size_t len; | |
4483 | int ret; | |
4484 | ||
4af3417a PB |
4485 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4486 | &ptr, &len); | |
52de1fe1 JA |
4487 | if (ret) |
4488 | return ret; | |
4489 | ||
4490 | uiov = compat_ptr(ptr); | |
4491 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4492 | compat_ssize_t clen; | |
4493 | ||
4494 | if (len > 1) | |
4495 | return -EINVAL; | |
4496 | if (!access_ok(uiov, sizeof(*uiov))) | |
4497 | return -EFAULT; | |
4498 | if (__get_user(clen, &uiov->iov_len)) | |
4499 | return -EFAULT; | |
4500 | if (clen < 0) | |
4501 | return -EINVAL; | |
2d280bc8 | 4502 | sr->len = clen; |
257e84a5 | 4503 | iomsg->free_iov = NULL; |
52de1fe1 | 4504 | } else { |
257e84a5 | 4505 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4506 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4507 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4508 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4509 | if (ret < 0) |
4510 | return ret; | |
4511 | } | |
4512 | ||
4513 | return 0; | |
4514 | } | |
4515 | #endif | |
4516 | ||
1400e697 PB |
4517 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4518 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4519 | { |
1400e697 | 4520 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4521 | |
4522 | #ifdef CONFIG_COMPAT | |
4523 | if (req->ctx->compat) | |
1400e697 | 4524 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4525 | #endif |
52de1fe1 | 4526 | |
1400e697 | 4527 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4528 | } |
4529 | ||
bcda7baa | 4530 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4531 | bool needs_lock) |
bcda7baa JA |
4532 | { |
4533 | struct io_sr_msg *sr = &req->sr_msg; | |
4534 | struct io_buffer *kbuf; | |
4535 | ||
bcda7baa JA |
4536 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4537 | if (IS_ERR(kbuf)) | |
4538 | return kbuf; | |
4539 | ||
4540 | sr->kbuf = kbuf; | |
4541 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4542 | return kbuf; |
fddaface JA |
4543 | } |
4544 | ||
7fbb1b54 PB |
4545 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4546 | { | |
4547 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4548 | } | |
4549 | ||
93642ef8 | 4550 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4551 | { |
99bc4c38 | 4552 | int ret; |
3529d8c2 | 4553 | |
93642ef8 PB |
4554 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4555 | if (!ret) | |
4556 | req->flags |= REQ_F_NEED_CLEANUP; | |
4557 | return ret; | |
4558 | } | |
4559 | ||
4560 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4561 | { | |
4562 | struct io_sr_msg *sr = &req->sr_msg; | |
4563 | ||
d2b6f48b PB |
4564 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4565 | return -EINVAL; | |
4566 | ||
270a5940 | 4567 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4568 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4569 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4570 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4571 | if (sr->msg_flags & MSG_DONTWAIT) | |
4572 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4573 | |
d8768362 JA |
4574 | #ifdef CONFIG_COMPAT |
4575 | if (req->ctx->compat) | |
4576 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4577 | #endif | |
93642ef8 | 4578 | return 0; |
aa1fa28f JA |
4579 | } |
4580 | ||
889fca73 | 4581 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4582 | { |
6b754c8b | 4583 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4584 | struct socket *sock; |
7fbb1b54 | 4585 | struct io_buffer *kbuf; |
7a7cacba | 4586 | unsigned flags; |
0031275d | 4587 | int min_ret = 0; |
52de1fe1 | 4588 | int ret, cflags = 0; |
45d189c6 | 4589 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4590 | |
dba4a925 | 4591 | sock = sock_from_file(req->file); |
7a7cacba | 4592 | if (unlikely(!sock)) |
dba4a925 | 4593 | return -ENOTSOCK; |
3529d8c2 | 4594 | |
257e84a5 PB |
4595 | kmsg = req->async_data; |
4596 | if (!kmsg) { | |
7a7cacba PB |
4597 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4598 | if (ret) | |
681fda8d | 4599 | return ret; |
7a7cacba PB |
4600 | kmsg = &iomsg; |
4601 | } | |
03b1230c | 4602 | |
bc02ef33 | 4603 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4604 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4605 | if (IS_ERR(kbuf)) |
52de1fe1 | 4606 | return PTR_ERR(kbuf); |
7a7cacba | 4607 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4608 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4609 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4610 | 1, req->sr_msg.len); |
4611 | } | |
52de1fe1 | 4612 | |
04411806 PB |
4613 | flags = req->sr_msg.msg_flags; |
4614 | if (force_nonblock) | |
7a7cacba | 4615 | flags |= MSG_DONTWAIT; |
0031275d SM |
4616 | if (flags & MSG_WAITALL) |
4617 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4618 | ||
7a7cacba PB |
4619 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4620 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4621 | if (force_nonblock && ret == -EAGAIN) |
4622 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4623 | if (ret == -ERESTARTSYS) |
4624 | ret = -EINTR; | |
03b1230c | 4625 | |
7fbb1b54 PB |
4626 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4627 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4628 | /* fast path, check for non-NULL to avoid function call */ |
4629 | if (kmsg->free_iov) | |
4630 | kfree(kmsg->free_iov); | |
99bc4c38 | 4631 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4632 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4633 | req_set_fail(req); |
889fca73 | 4634 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4635 | return 0; |
0fa03c62 | 4636 | } |
5d17b4a4 | 4637 | |
889fca73 | 4638 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4639 | { |
6b754c8b | 4640 | struct io_buffer *kbuf; |
7a7cacba PB |
4641 | struct io_sr_msg *sr = &req->sr_msg; |
4642 | struct msghdr msg; | |
4643 | void __user *buf = sr->buf; | |
fddaface | 4644 | struct socket *sock; |
7a7cacba PB |
4645 | struct iovec iov; |
4646 | unsigned flags; | |
0031275d | 4647 | int min_ret = 0; |
bcda7baa | 4648 | int ret, cflags = 0; |
45d189c6 | 4649 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4650 | |
dba4a925 | 4651 | sock = sock_from_file(req->file); |
7a7cacba | 4652 | if (unlikely(!sock)) |
dba4a925 | 4653 | return -ENOTSOCK; |
fddaface | 4654 | |
bc02ef33 | 4655 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4656 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4657 | if (IS_ERR(kbuf)) |
4658 | return PTR_ERR(kbuf); | |
7a7cacba | 4659 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4660 | } |
bcda7baa | 4661 | |
7a7cacba | 4662 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4663 | if (unlikely(ret)) |
4664 | goto out_free; | |
fddaface | 4665 | |
7a7cacba PB |
4666 | msg.msg_name = NULL; |
4667 | msg.msg_control = NULL; | |
4668 | msg.msg_controllen = 0; | |
4669 | msg.msg_namelen = 0; | |
4670 | msg.msg_iocb = NULL; | |
4671 | msg.msg_flags = 0; | |
fddaface | 4672 | |
04411806 PB |
4673 | flags = req->sr_msg.msg_flags; |
4674 | if (force_nonblock) | |
7a7cacba | 4675 | flags |= MSG_DONTWAIT; |
0031275d SM |
4676 | if (flags & MSG_WAITALL) |
4677 | min_ret = iov_iter_count(&msg.msg_iter); | |
4678 | ||
7a7cacba PB |
4679 | ret = sock_recvmsg(sock, &msg, flags); |
4680 | if (force_nonblock && ret == -EAGAIN) | |
4681 | return -EAGAIN; | |
4682 | if (ret == -ERESTARTSYS) | |
4683 | ret = -EINTR; | |
14c32eee | 4684 | out_free: |
7fbb1b54 PB |
4685 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4686 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4687 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4688 | req_set_fail(req); |
889fca73 | 4689 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4690 | return 0; |
fddaface JA |
4691 | } |
4692 | ||
3529d8c2 | 4693 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4694 | { |
8ed8d3c3 JA |
4695 | struct io_accept *accept = &req->accept; |
4696 | ||
14587a46 | 4697 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4698 | return -EINVAL; |
8042d6ce | 4699 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4700 | return -EINVAL; |
4701 | ||
d55e5f5b JA |
4702 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4703 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4704 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4705 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4706 | return 0; |
8ed8d3c3 | 4707 | } |
17f2fe35 | 4708 | |
889fca73 | 4709 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4710 | { |
4711 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4712 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4713 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4714 | int ret; |
4715 | ||
e697deed JX |
4716 | if (req->file->f_flags & O_NONBLOCK) |
4717 | req->flags |= REQ_F_NOWAIT; | |
4718 | ||
8ed8d3c3 | 4719 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4720 | accept->addr_len, accept->flags, |
4721 | accept->nofile); | |
8ed8d3c3 | 4722 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4723 | return -EAGAIN; |
ac45abc0 PB |
4724 | if (ret < 0) { |
4725 | if (ret == -ERESTARTSYS) | |
4726 | ret = -EINTR; | |
93d2bcd2 | 4727 | req_set_fail(req); |
ac45abc0 | 4728 | } |
889fca73 | 4729 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4730 | return 0; |
8ed8d3c3 JA |
4731 | } |
4732 | ||
93642ef8 PB |
4733 | static int io_connect_prep_async(struct io_kiocb *req) |
4734 | { | |
4735 | struct io_async_connect *io = req->async_data; | |
4736 | struct io_connect *conn = &req->connect; | |
4737 | ||
4738 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4739 | } | |
4740 | ||
3529d8c2 | 4741 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4742 | { |
3529d8c2 | 4743 | struct io_connect *conn = &req->connect; |
f499a021 | 4744 | |
14587a46 | 4745 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4746 | return -EINVAL; |
4747 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4748 | return -EINVAL; | |
4749 | ||
3529d8c2 JA |
4750 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4751 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4752 | return 0; |
f499a021 JA |
4753 | } |
4754 | ||
889fca73 | 4755 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4756 | { |
e8c2bc1f | 4757 | struct io_async_connect __io, *io; |
f8e85cf2 | 4758 | unsigned file_flags; |
3fbb51c1 | 4759 | int ret; |
45d189c6 | 4760 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4761 | |
e8c2bc1f JA |
4762 | if (req->async_data) { |
4763 | io = req->async_data; | |
f499a021 | 4764 | } else { |
3529d8c2 JA |
4765 | ret = move_addr_to_kernel(req->connect.addr, |
4766 | req->connect.addr_len, | |
e8c2bc1f | 4767 | &__io.address); |
f499a021 JA |
4768 | if (ret) |
4769 | goto out; | |
4770 | io = &__io; | |
4771 | } | |
4772 | ||
3fbb51c1 JA |
4773 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4774 | ||
e8c2bc1f | 4775 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4776 | req->connect.addr_len, file_flags); |
87f80d62 | 4777 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4778 | if (req->async_data) |
b7bb4f7d | 4779 | return -EAGAIN; |
e8c2bc1f | 4780 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4781 | ret = -ENOMEM; |
4782 | goto out; | |
4783 | } | |
e8c2bc1f | 4784 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4785 | return -EAGAIN; |
f499a021 | 4786 | } |
f8e85cf2 JA |
4787 | if (ret == -ERESTARTSYS) |
4788 | ret = -EINTR; | |
f499a021 | 4789 | out: |
4e88d6e7 | 4790 | if (ret < 0) |
93d2bcd2 | 4791 | req_set_fail(req); |
889fca73 | 4792 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4793 | return 0; |
469956e8 Y |
4794 | } |
4795 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4796 | #define IO_NETOP_FN(op) \ |
4797 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4798 | { \ | |
4799 | return -EOPNOTSUPP; \ | |
4800 | } | |
4801 | ||
4802 | #define IO_NETOP_PREP(op) \ | |
4803 | IO_NETOP_FN(op) \ | |
4804 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4805 | { \ | |
4806 | return -EOPNOTSUPP; \ | |
4807 | } \ | |
4808 | ||
4809 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4810 | IO_NETOP_PREP(op) \ | |
4811 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4812 | { \ | |
4813 | return -EOPNOTSUPP; \ | |
4814 | } | |
4815 | ||
4816 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4817 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4818 | IO_NETOP_PREP_ASYNC(connect); | |
4819 | IO_NETOP_PREP(accept); | |
4820 | IO_NETOP_FN(send); | |
4821 | IO_NETOP_FN(recv); | |
469956e8 | 4822 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4823 | |
d7718a9d JA |
4824 | struct io_poll_table { |
4825 | struct poll_table_struct pt; | |
4826 | struct io_kiocb *req; | |
4827 | int error; | |
4828 | }; | |
ce593a6c | 4829 | |
d7718a9d JA |
4830 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4831 | __poll_t mask, task_work_func_t func) | |
4832 | { | |
aa96bf8a | 4833 | int ret; |
d7718a9d JA |
4834 | |
4835 | /* for instances that support it check for an event match first: */ | |
4836 | if (mask && !(mask & poll->events)) | |
4837 | return 0; | |
4838 | ||
4839 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4840 | ||
4841 | list_del_init(&poll->wait.entry); | |
4842 | ||
d7718a9d | 4843 | req->result = mask; |
7cbf1722 | 4844 | req->task_work.func = func; |
6d816e08 | 4845 | |
d7718a9d | 4846 | /* |
e3aabf95 JA |
4847 | * If this fails, then the task is exiting. When a task exits, the |
4848 | * work gets canceled, so just cancel this request as well instead | |
4849 | * of executing it. We can't safely execute it anyway, as we may not | |
4850 | * have the needed state needed for it anyway. | |
d7718a9d | 4851 | */ |
355fb9e2 | 4852 | ret = io_req_task_work_add(req); |
aa96bf8a | 4853 | if (unlikely(ret)) { |
e3aabf95 | 4854 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4855 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4856 | } |
d7718a9d JA |
4857 | return 1; |
4858 | } | |
4859 | ||
74ce6ce4 JA |
4860 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4861 | __acquires(&req->ctx->completion_lock) | |
4862 | { | |
4863 | struct io_ring_ctx *ctx = req->ctx; | |
4864 | ||
4865 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4866 | struct poll_table_struct pt = { ._key = poll->events }; | |
4867 | ||
4868 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4869 | } | |
4870 | ||
4871 | spin_lock_irq(&ctx->completion_lock); | |
4872 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4873 | add_wait_queue(poll->head, &poll->wait); | |
4874 | return true; | |
4875 | } | |
4876 | ||
4877 | return false; | |
4878 | } | |
4879 | ||
d4e7cd36 | 4880 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4881 | { |
e8c2bc1f | 4882 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4883 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4884 | return req->async_data; |
d4e7cd36 JA |
4885 | return req->apoll->double_poll; |
4886 | } | |
4887 | ||
4888 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4889 | { | |
4890 | if (req->opcode == IORING_OP_POLL_ADD) | |
4891 | return &req->poll; | |
4892 | return &req->apoll->poll; | |
4893 | } | |
4894 | ||
4895 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4896 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4897 | { |
4898 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4899 | |
4900 | lockdep_assert_held(&req->ctx->completion_lock); | |
4901 | ||
4902 | if (poll && poll->head) { | |
4903 | struct wait_queue_head *head = poll->head; | |
4904 | ||
4905 | spin_lock(&head->lock); | |
4906 | list_del_init(&poll->wait.entry); | |
4907 | if (poll->wait.private) | |
de9b4cca | 4908 | req_ref_put(req); |
18bceab1 JA |
4909 | poll->head = NULL; |
4910 | spin_unlock(&head->lock); | |
4911 | } | |
4912 | } | |
4913 | ||
e27414be | 4914 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4915 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4916 | { |
4917 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4918 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4919 | int error; |
18bceab1 | 4920 | |
e27414be | 4921 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4922 | error = -ECANCELED; |
88e41cf9 | 4923 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4924 | } else { |
5082620f | 4925 | error = mangle_poll(mask); |
e27414be | 4926 | } |
b69de288 JA |
4927 | if (req->poll.events & EPOLLONESHOT) |
4928 | flags = 0; | |
d4d19c19 | 4929 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5082620f | 4930 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4931 | req->poll.done = true; |
4932 | flags = 0; | |
4933 | } | |
7b289c38 HX |
4934 | if (flags & IORING_CQE_F_MORE) |
4935 | ctx->cq_extra++; | |
18bceab1 | 4936 | |
18bceab1 | 4937 | io_commit_cqring(ctx); |
88e41cf9 | 4938 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4939 | } |
4940 | ||
dd221f46 | 4941 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4942 | { |
dd221f46 | 4943 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4944 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4945 | struct io_kiocb *nxt; |
18bceab1 JA |
4946 | |
4947 | if (io_poll_rewait(req, &req->poll)) { | |
4948 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4949 | } else { |
f40b964a | 4950 | bool done; |
18bceab1 | 4951 | |
e27414be | 4952 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4953 | if (done) { |
4954 | hash_del(&req->hash_node); | |
f40b964a | 4955 | } else { |
88e41cf9 JA |
4956 | req->result = 0; |
4957 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4958 | } | |
dd221f46 | 4959 | spin_unlock_irq(&ctx->completion_lock); |
dd221f46 | 4960 | io_cqring_ev_posted(ctx); |
18bceab1 | 4961 | |
88e41cf9 JA |
4962 | if (done) { |
4963 | nxt = io_put_req_find_next(req); | |
4964 | if (nxt) | |
4965 | __io_req_task_submit(nxt); | |
4966 | } | |
dd221f46 | 4967 | } |
18bceab1 JA |
4968 | } |
4969 | ||
4970 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4971 | int sync, void *key) | |
4972 | { | |
4973 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4974 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4975 | __poll_t mask = key_to_poll(key); |
4976 | ||
4977 | /* for instances that support it check for an event match first: */ | |
4978 | if (mask && !(mask & poll->events)) | |
4979 | return 0; | |
88e41cf9 JA |
4980 | if (!(poll->events & EPOLLONESHOT)) |
4981 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4982 | |
8706e04e JA |
4983 | list_del_init(&wait->entry); |
4984 | ||
807abcb0 | 4985 | if (poll && poll->head) { |
18bceab1 JA |
4986 | bool done; |
4987 | ||
807abcb0 JA |
4988 | spin_lock(&poll->head->lock); |
4989 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4990 | if (!done) |
807abcb0 | 4991 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4992 | /* make sure double remove sees this as being gone */ |
4993 | wait->private = NULL; | |
807abcb0 | 4994 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4995 | if (!done) { |
4996 | /* use wait func handler, so it matches the rq type */ | |
4997 | poll->wait.func(&poll->wait, mode, sync, key); | |
4998 | } | |
18bceab1 | 4999 | } |
de9b4cca | 5000 | req_ref_put(req); |
18bceab1 JA |
5001 | return 1; |
5002 | } | |
5003 | ||
5004 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5005 | wait_queue_func_t wake_func) | |
5006 | { | |
5007 | poll->head = NULL; | |
5008 | poll->done = false; | |
5009 | poll->canceled = false; | |
464dca61 JA |
5010 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5011 | /* mask in events that we always want/need */ | |
5012 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5013 | INIT_LIST_HEAD(&poll->wait.entry); |
5014 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5015 | } | |
5016 | ||
5017 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5018 | struct wait_queue_head *head, |
5019 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5020 | { |
5021 | struct io_kiocb *req = pt->req; | |
5022 | ||
5023 | /* | |
5024 | * If poll->head is already set, it's because the file being polled | |
5025 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5026 | * for write). Setup a separate io_poll_iocb if this happens. | |
5027 | */ | |
5028 | if (unlikely(poll->head)) { | |
58852d4d PB |
5029 | struct io_poll_iocb *poll_one = poll; |
5030 | ||
18bceab1 | 5031 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5032 | if (*poll_ptr) { |
18bceab1 JA |
5033 | pt->error = -EINVAL; |
5034 | return; | |
5035 | } | |
ea6a693d JA |
5036 | /* |
5037 | * Can't handle multishot for double wait for now, turn it | |
5038 | * into one-shot mode. | |
5039 | */ | |
7a274727 PB |
5040 | if (!(poll_one->events & EPOLLONESHOT)) |
5041 | poll_one->events |= EPOLLONESHOT; | |
1c3b3e65 | 5042 | /* double add on the same waitqueue head, ignore */ |
7a274727 | 5043 | if (poll_one->head == head) |
1c3b3e65 | 5044 | return; |
18bceab1 JA |
5045 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5046 | if (!poll) { | |
5047 | pt->error = -ENOMEM; | |
5048 | return; | |
5049 | } | |
58852d4d | 5050 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5051 | req_ref_get(req); |
18bceab1 | 5052 | poll->wait.private = req; |
807abcb0 | 5053 | *poll_ptr = poll; |
18bceab1 JA |
5054 | } |
5055 | ||
5056 | pt->error = 0; | |
5057 | poll->head = head; | |
a31eb4a2 JX |
5058 | |
5059 | if (poll->events & EPOLLEXCLUSIVE) | |
5060 | add_wait_queue_exclusive(head, &poll->wait); | |
5061 | else | |
5062 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5063 | } |
5064 | ||
5065 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5066 | struct poll_table_struct *p) | |
5067 | { | |
5068 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5069 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5070 | |
807abcb0 | 5071 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5072 | } |
5073 | ||
d7718a9d JA |
5074 | static void io_async_task_func(struct callback_head *cb) |
5075 | { | |
5076 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5077 | struct async_poll *apoll = req->apoll; | |
5078 | struct io_ring_ctx *ctx = req->ctx; | |
5079 | ||
236daeae | 5080 | trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data); |
d7718a9d | 5081 | |
74ce6ce4 | 5082 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5083 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5084 | return; |
d7718a9d JA |
5085 | } |
5086 | ||
0ea13b44 | 5087 | hash_del(&req->hash_node); |
d4e7cd36 | 5088 | io_poll_remove_double(req); |
74ce6ce4 JA |
5089 | spin_unlock_irq(&ctx->completion_lock); |
5090 | ||
0be0b0e3 PB |
5091 | if (!READ_ONCE(apoll->poll.canceled)) |
5092 | __io_req_task_submit(req); | |
5093 | else | |
2593553a | 5094 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5095 | } |
5096 | ||
5097 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5098 | void *key) | |
5099 | { | |
5100 | struct io_kiocb *req = wait->private; | |
5101 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5102 | ||
5103 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5104 | key_to_poll(key)); | |
5105 | ||
5106 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5107 | } | |
5108 | ||
5109 | static void io_poll_req_insert(struct io_kiocb *req) | |
5110 | { | |
5111 | struct io_ring_ctx *ctx = req->ctx; | |
5112 | struct hlist_head *list; | |
5113 | ||
5114 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5115 | hlist_add_head(&req->hash_node, list); | |
5116 | } | |
5117 | ||
5118 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5119 | struct io_poll_iocb *poll, | |
5120 | struct io_poll_table *ipt, __poll_t mask, | |
5121 | wait_queue_func_t wake_func) | |
5122 | __acquires(&ctx->completion_lock) | |
5123 | { | |
5124 | struct io_ring_ctx *ctx = req->ctx; | |
5125 | bool cancel = false; | |
5126 | ||
4d52f338 | 5127 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5128 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5129 | poll->file = req->file; |
18bceab1 | 5130 | poll->wait.private = req; |
d7718a9d JA |
5131 | |
5132 | ipt->pt._key = mask; | |
5133 | ipt->req = req; | |
5134 | ipt->error = -EINVAL; | |
5135 | ||
d7718a9d JA |
5136 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5137 | ||
5138 | spin_lock_irq(&ctx->completion_lock); | |
5139 | if (likely(poll->head)) { | |
5140 | spin_lock(&poll->head->lock); | |
5141 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5142 | if (ipt->error) | |
5143 | cancel = true; | |
5144 | ipt->error = 0; | |
5145 | mask = 0; | |
5146 | } | |
88e41cf9 | 5147 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5148 | list_del_init(&poll->wait.entry); |
5149 | else if (cancel) | |
5150 | WRITE_ONCE(poll->canceled, true); | |
5151 | else if (!poll->done) /* actually waiting for an event */ | |
5152 | io_poll_req_insert(req); | |
5153 | spin_unlock(&poll->head->lock); | |
5154 | } | |
5155 | ||
5156 | return mask; | |
5157 | } | |
5158 | ||
5159 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5160 | { | |
5161 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5162 | struct io_ring_ctx *ctx = req->ctx; | |
5163 | struct async_poll *apoll; | |
5164 | struct io_poll_table ipt; | |
5165 | __poll_t mask, ret; | |
9dab14b8 | 5166 | int rw; |
d7718a9d JA |
5167 | |
5168 | if (!req->file || !file_can_poll(req->file)) | |
5169 | return false; | |
24c74678 | 5170 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5171 | return false; |
9dab14b8 JA |
5172 | if (def->pollin) |
5173 | rw = READ; | |
5174 | else if (def->pollout) | |
5175 | rw = WRITE; | |
5176 | else | |
5177 | return false; | |
5178 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5179 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5180 | return false; |
5181 | ||
5182 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5183 | if (unlikely(!apoll)) | |
5184 | return false; | |
807abcb0 | 5185 | apoll->double_poll = NULL; |
d7718a9d JA |
5186 | |
5187 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5188 | req->apoll = apoll; |
d7718a9d | 5189 | |
88e41cf9 | 5190 | mask = EPOLLONESHOT; |
d7718a9d | 5191 | if (def->pollin) |
8755d97a | 5192 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5193 | if (def->pollout) |
5194 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5195 | |
5196 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5197 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5198 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5199 | mask &= ~POLLIN; | |
5200 | ||
d7718a9d JA |
5201 | mask |= POLLERR | POLLPRI; |
5202 | ||
5203 | ipt.pt._qproc = io_async_queue_proc; | |
5204 | ||
5205 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5206 | io_async_wake); | |
a36da65c | 5207 | if (ret || ipt.error) { |
d4e7cd36 | 5208 | io_poll_remove_double(req); |
d7718a9d | 5209 | spin_unlock_irq(&ctx->completion_lock); |
d7718a9d JA |
5210 | return false; |
5211 | } | |
5212 | spin_unlock_irq(&ctx->completion_lock); | |
236daeae OL |
5213 | trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, |
5214 | mask, apoll->poll.events); | |
d7718a9d JA |
5215 | return true; |
5216 | } | |
5217 | ||
5218 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5219 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5220 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5221 | { |
b41e9852 | 5222 | bool do_complete = false; |
221c5eb2 | 5223 | |
5082620f JA |
5224 | if (!poll->head) |
5225 | return false; | |
221c5eb2 | 5226 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5227 | if (do_cancel) |
5228 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5229 | if (!list_empty(&poll->wait.entry)) { |
5230 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5231 | do_complete = true; |
221c5eb2 JA |
5232 | } |
5233 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5234 | hash_del(&req->hash_node); |
d7718a9d JA |
5235 | return do_complete; |
5236 | } | |
5237 | ||
b2c3f7e1 | 5238 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5239 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5240 | { |
5241 | bool do_complete; | |
5242 | ||
d4e7cd36 | 5243 | io_poll_remove_double(req); |
e31001a3 | 5244 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5245 | |
e31001a3 | 5246 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
d7718a9d | 5247 | /* non-poll requests have submit ref still */ |
e31001a3 | 5248 | req_ref_put(req); |
b1f573bd | 5249 | } |
b2c3f7e1 JA |
5250 | return do_complete; |
5251 | } | |
5252 | ||
5253 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5254 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5255 | { |
5256 | bool do_complete; | |
b1f573bd | 5257 | |
b2c3f7e1 | 5258 | do_complete = io_poll_remove_waitqs(req); |
b41e9852 | 5259 | if (do_complete) { |
d4d19c19 | 5260 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5261 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5262 | req_set_fail(req); |
216578e5 | 5263 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5264 | } |
5265 | ||
5266 | return do_complete; | |
221c5eb2 JA |
5267 | } |
5268 | ||
76e1b642 JA |
5269 | /* |
5270 | * Returns true if we found and killed one or more poll requests | |
5271 | */ | |
6b81928d | 5272 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5273 | bool cancel_all) |
221c5eb2 | 5274 | { |
78076bb6 | 5275 | struct hlist_node *tmp; |
221c5eb2 | 5276 | struct io_kiocb *req; |
8e2e1faf | 5277 | int posted = 0, i; |
221c5eb2 JA |
5278 | |
5279 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5280 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5281 | struct hlist_head *list; | |
5282 | ||
5283 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5284 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5285 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5286 | posted += io_poll_remove_one(req); |
5287 | } | |
221c5eb2 JA |
5288 | } |
5289 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5290 | |
8e2e1faf JA |
5291 | if (posted) |
5292 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5293 | |
5294 | return posted != 0; | |
221c5eb2 JA |
5295 | } |
5296 | ||
9ba5fac8 PB |
5297 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5298 | bool poll_only) | |
e07785b0 | 5299 | __must_hold(&ctx->completion_lock) |
47f46768 | 5300 | { |
78076bb6 | 5301 | struct hlist_head *list; |
47f46768 JA |
5302 | struct io_kiocb *req; |
5303 | ||
78076bb6 JA |
5304 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5305 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5306 | if (sqe_addr != req->user_data) |
5307 | continue; | |
9ba5fac8 PB |
5308 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5309 | continue; | |
b2cb805f | 5310 | return req; |
47f46768 | 5311 | } |
b2cb805f JA |
5312 | return NULL; |
5313 | } | |
5314 | ||
9ba5fac8 PB |
5315 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5316 | bool poll_only) | |
e07785b0 | 5317 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5318 | { |
5319 | struct io_kiocb *req; | |
5320 | ||
9ba5fac8 | 5321 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5322 | if (!req) |
5323 | return -ENOENT; | |
5324 | if (io_poll_remove_one(req)) | |
5325 | return 0; | |
5326 | ||
5327 | return -EALREADY; | |
47f46768 JA |
5328 | } |
5329 | ||
9096af3e PB |
5330 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5331 | unsigned int flags) | |
5332 | { | |
5333 | u32 events; | |
47f46768 | 5334 | |
9096af3e PB |
5335 | events = READ_ONCE(sqe->poll32_events); |
5336 | #ifdef __BIG_ENDIAN | |
5337 | events = swahw32(events); | |
5338 | #endif | |
5339 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5340 | events |= EPOLLONESHOT; | |
5341 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5342 | } |
5343 | ||
c5de0036 | 5344 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5345 | const struct io_uring_sqe *sqe) |
0969e783 | 5346 | { |
c5de0036 PB |
5347 | struct io_poll_update *upd = &req->poll_update; |
5348 | u32 flags; | |
5349 | ||
0969e783 JA |
5350 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5351 | return -EINVAL; | |
c5de0036 PB |
5352 | if (sqe->ioprio || sqe->buf_index) |
5353 | return -EINVAL; | |
5354 | flags = READ_ONCE(sqe->len); | |
5355 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5356 | IORING_POLL_ADD_MULTI)) | |
5357 | return -EINVAL; | |
5358 | /* meaningless without update */ | |
5359 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5360 | return -EINVAL; |
5361 | ||
c5de0036 PB |
5362 | upd->old_user_data = READ_ONCE(sqe->addr); |
5363 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5364 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5365 | |
c5de0036 PB |
5366 | upd->new_user_data = READ_ONCE(sqe->off); |
5367 | if (!upd->update_user_data && upd->new_user_data) | |
5368 | return -EINVAL; | |
5369 | if (upd->update_events) | |
5370 | upd->events = io_poll_parse_events(sqe, flags); | |
5371 | else if (sqe->poll32_events) | |
5372 | return -EINVAL; | |
221c5eb2 | 5373 | |
221c5eb2 JA |
5374 | return 0; |
5375 | } | |
5376 | ||
221c5eb2 JA |
5377 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5378 | void *key) | |
5379 | { | |
c2f2eb7d JA |
5380 | struct io_kiocb *req = wait->private; |
5381 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5382 | |
d7718a9d | 5383 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5384 | } |
5385 | ||
221c5eb2 JA |
5386 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5387 | struct poll_table_struct *p) | |
5388 | { | |
5389 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5390 | ||
e8c2bc1f | 5391 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5392 | } |
5393 | ||
3529d8c2 | 5394 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5395 | { |
5396 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5397 | u32 flags; |
221c5eb2 JA |
5398 | |
5399 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5400 | return -EINVAL; | |
c5de0036 | 5401 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5402 | return -EINVAL; |
5403 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5404 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5405 | return -EINVAL; |
5406 | ||
c5de0036 | 5407 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5408 | return 0; |
5409 | } | |
5410 | ||
61e98203 | 5411 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5412 | { |
5413 | struct io_poll_iocb *poll = &req->poll; | |
5414 | struct io_ring_ctx *ctx = req->ctx; | |
5415 | struct io_poll_table ipt; | |
0969e783 | 5416 | __poll_t mask; |
0969e783 | 5417 | |
d7718a9d | 5418 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5419 | |
d7718a9d JA |
5420 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5421 | io_poll_wake); | |
221c5eb2 | 5422 | |
8c838788 | 5423 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5424 | ipt.error = 0; |
e27414be | 5425 | io_poll_complete(req, mask); |
221c5eb2 | 5426 | } |
221c5eb2 JA |
5427 | spin_unlock_irq(&ctx->completion_lock); |
5428 | ||
8c838788 JA |
5429 | if (mask) { |
5430 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5431 | if (poll->events & EPOLLONESHOT) |
5432 | io_put_req(req); | |
221c5eb2 | 5433 | } |
8c838788 | 5434 | return ipt.error; |
221c5eb2 JA |
5435 | } |
5436 | ||
c5de0036 | 5437 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5438 | { |
5439 | struct io_ring_ctx *ctx = req->ctx; | |
5440 | struct io_kiocb *preq; | |
cb3b200e | 5441 | bool completing; |
b69de288 JA |
5442 | int ret; |
5443 | ||
5444 | spin_lock_irq(&ctx->completion_lock); | |
9ba5fac8 | 5445 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5446 | if (!preq) { |
5447 | ret = -ENOENT; | |
5448 | goto err; | |
b69de288 | 5449 | } |
cb3b200e | 5450 | |
c5de0036 PB |
5451 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5452 | completing = true; | |
5453 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5454 | goto err; | |
5455 | } | |
5456 | ||
cb3b200e JA |
5457 | /* |
5458 | * Don't allow racy completion with singleshot, as we cannot safely | |
5459 | * update those. For multishot, if we're racing with completion, just | |
5460 | * let completion re-add it. | |
5461 | */ | |
5462 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5463 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5464 | ret = -EALREADY; | |
5465 | goto err; | |
b69de288 JA |
5466 | } |
5467 | /* we now have a detached poll request. reissue. */ | |
5468 | ret = 0; | |
5469 | err: | |
b69de288 | 5470 | if (ret < 0) { |
cb3b200e | 5471 | spin_unlock_irq(&ctx->completion_lock); |
93d2bcd2 | 5472 | req_set_fail(req); |
b69de288 JA |
5473 | io_req_complete(req, ret); |
5474 | return 0; | |
5475 | } | |
5476 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5477 | if (req->poll_update.update_events) { |
b69de288 | 5478 | preq->poll.events &= ~0xffff; |
9d805892 | 5479 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5480 | preq->poll.events |= IO_POLL_UNMASK; |
5481 | } | |
9d805892 PB |
5482 | if (req->poll_update.update_user_data) |
5483 | preq->user_data = req->poll_update.new_user_data; | |
cb3b200e JA |
5484 | spin_unlock_irq(&ctx->completion_lock); |
5485 | ||
b69de288 JA |
5486 | /* complete update request, we're done with it */ |
5487 | io_req_complete(req, ret); | |
5488 | ||
cb3b200e | 5489 | if (!completing) { |
c5de0036 | 5490 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5491 | if (ret < 0) { |
93d2bcd2 | 5492 | req_set_fail(preq); |
cb3b200e JA |
5493 | io_req_complete(preq, ret); |
5494 | } | |
b69de288 JA |
5495 | } |
5496 | return 0; | |
5497 | } | |
5498 | ||
5262f567 JA |
5499 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5500 | { | |
ad8a48ac JA |
5501 | struct io_timeout_data *data = container_of(timer, |
5502 | struct io_timeout_data, timer); | |
5503 | struct io_kiocb *req = data->req; | |
5504 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5505 | unsigned long flags; |
5506 | ||
5262f567 | 5507 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5508 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5509 | atomic_set(&req->ctx->cq_timeouts, |
5510 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5511 | ||
d4d19c19 | 5512 | io_cqring_fill_event(ctx, req->user_data, -ETIME, 0); |
5262f567 JA |
5513 | io_commit_cqring(ctx); |
5514 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5515 | ||
5516 | io_cqring_ev_posted(ctx); | |
93d2bcd2 | 5517 | req_set_fail(req); |
5262f567 JA |
5518 | io_put_req(req); |
5519 | return HRTIMER_NORESTART; | |
5520 | } | |
5521 | ||
fbd15848 PB |
5522 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5523 | __u64 user_data) | |
e07785b0 | 5524 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5525 | { |
fbd15848 | 5526 | struct io_timeout_data *io; |
47f46768 | 5527 | struct io_kiocb *req; |
fd9c7bc5 | 5528 | bool found = false; |
f254ac04 | 5529 | |
135fcde8 | 5530 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5531 | found = user_data == req->user_data; |
5532 | if (found) | |
47f46768 | 5533 | break; |
47f46768 | 5534 | } |
fd9c7bc5 PB |
5535 | if (!found) |
5536 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5537 | |
5538 | io = req->async_data; | |
fd9c7bc5 | 5539 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5540 | return ERR_PTR(-EALREADY); |
a71976f3 | 5541 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5542 | return req; |
5543 | } | |
47f46768 | 5544 | |
fbd15848 | 5545 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5546 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5547 | { |
5548 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5549 | ||
5550 | if (IS_ERR(req)) | |
5551 | return PTR_ERR(req); | |
f254ac04 | 5552 | |
93d2bcd2 | 5553 | req_set_fail(req); |
d4d19c19 | 5554 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
216578e5 | 5555 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5556 | return 0; |
5557 | } | |
5558 | ||
9c8e11b3 PB |
5559 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5560 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5561 | __must_hold(&ctx->completion_lock) |
47f46768 | 5562 | { |
9c8e11b3 PB |
5563 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5564 | struct io_timeout_data *data; | |
47f46768 | 5565 | |
9c8e11b3 PB |
5566 | if (IS_ERR(req)) |
5567 | return PTR_ERR(req); | |
47f46768 | 5568 | |
9c8e11b3 PB |
5569 | req->timeout.off = 0; /* noseq */ |
5570 | data = req->async_data; | |
5571 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5572 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5573 | data->timer.function = io_timeout_fn; | |
5574 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5575 | return 0; | |
47f46768 JA |
5576 | } |
5577 | ||
3529d8c2 JA |
5578 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5579 | const struct io_uring_sqe *sqe) | |
b29472ee | 5580 | { |
9c8e11b3 PB |
5581 | struct io_timeout_rem *tr = &req->timeout_rem; |
5582 | ||
b29472ee JA |
5583 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5584 | return -EINVAL; | |
61710e43 DA |
5585 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5586 | return -EINVAL; | |
9c8e11b3 | 5587 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5588 | return -EINVAL; |
5589 | ||
9c8e11b3 PB |
5590 | tr->addr = READ_ONCE(sqe->addr); |
5591 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5592 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5593 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5594 | return -EINVAL; | |
5595 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5596 | return -EFAULT; | |
5597 | } else if (tr->flags) { | |
5598 | /* timeout removal doesn't support flags */ | |
b29472ee | 5599 | return -EINVAL; |
9c8e11b3 | 5600 | } |
b29472ee | 5601 | |
b29472ee JA |
5602 | return 0; |
5603 | } | |
5604 | ||
8662daec PB |
5605 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5606 | { | |
5607 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5608 | : HRTIMER_MODE_REL; | |
5609 | } | |
5610 | ||
11365043 JA |
5611 | /* |
5612 | * Remove or update an existing timeout command | |
5613 | */ | |
61e98203 | 5614 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5615 | { |
9c8e11b3 | 5616 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5617 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5618 | int ret; |
11365043 | 5619 | |
11365043 | 5620 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5621 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5622 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5623 | else |
5624 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5625 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5626 | |
d4d19c19 | 5627 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
11365043 JA |
5628 | io_commit_cqring(ctx); |
5629 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5630 | io_cqring_ev_posted(ctx); |
4e88d6e7 | 5631 | if (ret < 0) |
93d2bcd2 | 5632 | req_set_fail(req); |
ec9c02ad | 5633 | io_put_req(req); |
11365043 | 5634 | return 0; |
5262f567 JA |
5635 | } |
5636 | ||
3529d8c2 | 5637 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5638 | bool is_timeout_link) |
5262f567 | 5639 | { |
ad8a48ac | 5640 | struct io_timeout_data *data; |
a41525ab | 5641 | unsigned flags; |
56080b02 | 5642 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5643 | |
ad8a48ac | 5644 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5645 | return -EINVAL; |
ad8a48ac | 5646 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5647 | return -EINVAL; |
56080b02 | 5648 | if (off && is_timeout_link) |
2d28390a | 5649 | return -EINVAL; |
a41525ab JA |
5650 | flags = READ_ONCE(sqe->timeout_flags); |
5651 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5652 | return -EINVAL; |
bdf20073 | 5653 | |
bfe68a22 | 5654 | req->timeout.off = off; |
f18ee4cf PB |
5655 | if (unlikely(off && !req->ctx->off_timeout_used)) |
5656 | req->ctx->off_timeout_used = true; | |
26a61679 | 5657 | |
e8c2bc1f | 5658 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5659 | return -ENOMEM; |
5660 | ||
e8c2bc1f | 5661 | data = req->async_data; |
ad8a48ac | 5662 | data->req = req; |
ad8a48ac JA |
5663 | |
5664 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5665 | return -EFAULT; |
5666 | ||
8662daec | 5667 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5668 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5669 | if (is_timeout_link) |
5670 | io_req_track_inflight(req); | |
ad8a48ac JA |
5671 | return 0; |
5672 | } | |
5673 | ||
61e98203 | 5674 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5675 | { |
ad8a48ac | 5676 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5677 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5678 | struct list_head *entry; |
bfe68a22 | 5679 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5680 | |
733f5c95 | 5681 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5682 | |
5262f567 JA |
5683 | /* |
5684 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5685 | * timeout event to be satisfied. If it isn't set, then this is |
5686 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5687 | */ |
8eb7e2d0 | 5688 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5689 | entry = ctx->timeout_list.prev; |
5690 | goto add; | |
5691 | } | |
5262f567 | 5692 | |
bfe68a22 PB |
5693 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5694 | req->timeout.target_seq = tail + off; | |
5262f567 | 5695 | |
f010505b MDG |
5696 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5697 | * This is safe because ->completion_lock is held, and submissions | |
5698 | * and completions are never mixed in the same ->completion_lock section. | |
5699 | */ | |
5700 | ctx->cq_last_tm_flush = tail; | |
5701 | ||
5262f567 JA |
5702 | /* |
5703 | * Insertion sort, ensuring the first entry in the list is always | |
5704 | * the one we need first. | |
5705 | */ | |
5262f567 | 5706 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5707 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5708 | timeout.list); | |
5262f567 | 5709 | |
8eb7e2d0 | 5710 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5711 | continue; |
bfe68a22 PB |
5712 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5713 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5714 | break; |
5715 | } | |
93bd25bb | 5716 | add: |
135fcde8 | 5717 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5718 | data->timer.function = io_timeout_fn; |
5719 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5720 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5721 | return 0; |
5722 | } | |
5262f567 | 5723 | |
f458dd84 PB |
5724 | struct io_cancel_data { |
5725 | struct io_ring_ctx *ctx; | |
5726 | u64 user_data; | |
5727 | }; | |
5728 | ||
62755e35 JA |
5729 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5730 | { | |
5731 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5732 | struct io_cancel_data *cd = data; |
62755e35 | 5733 | |
f458dd84 | 5734 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5735 | } |
5736 | ||
f458dd84 PB |
5737 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5738 | struct io_ring_ctx *ctx) | |
62755e35 | 5739 | { |
f458dd84 | 5740 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5741 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5742 | int ret = 0; |
5743 | ||
f458dd84 | 5744 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5745 | return -ENOENT; |
5746 | ||
f458dd84 | 5747 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5748 | switch (cancel_ret) { |
5749 | case IO_WQ_CANCEL_OK: | |
5750 | ret = 0; | |
5751 | break; | |
5752 | case IO_WQ_CANCEL_RUNNING: | |
5753 | ret = -EALREADY; | |
5754 | break; | |
5755 | case IO_WQ_CANCEL_NOTFOUND: | |
5756 | ret = -ENOENT; | |
5757 | break; | |
5758 | } | |
5759 | ||
e977d6d3 JA |
5760 | return ret; |
5761 | } | |
5762 | ||
47f46768 JA |
5763 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5764 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5765 | int success_ret) |
47f46768 JA |
5766 | { |
5767 | unsigned long flags; | |
5768 | int ret; | |
5769 | ||
f458dd84 | 5770 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5771 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5772 | if (ret != -ENOENT) |
5773 | goto done; | |
47f46768 JA |
5774 | ret = io_timeout_cancel(ctx, sqe_addr); |
5775 | if (ret != -ENOENT) | |
5776 | goto done; | |
9ba5fac8 | 5777 | ret = io_poll_cancel(ctx, sqe_addr, false); |
47f46768 | 5778 | done: |
b0dd8a41 JA |
5779 | if (!ret) |
5780 | ret = success_ret; | |
d4d19c19 | 5781 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
47f46768 JA |
5782 | io_commit_cqring(ctx); |
5783 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5784 | io_cqring_ev_posted(ctx); | |
5785 | ||
4e88d6e7 | 5786 | if (ret < 0) |
93d2bcd2 | 5787 | req_set_fail(req); |
47f46768 JA |
5788 | } |
5789 | ||
3529d8c2 JA |
5790 | static int io_async_cancel_prep(struct io_kiocb *req, |
5791 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5792 | { |
fbf23849 | 5793 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5794 | return -EINVAL; |
61710e43 DA |
5795 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5796 | return -EINVAL; | |
5797 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5798 | return -EINVAL; |
5799 | ||
fbf23849 JA |
5800 | req->cancel.addr = READ_ONCE(sqe->addr); |
5801 | return 0; | |
5802 | } | |
5803 | ||
61e98203 | 5804 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5805 | { |
5806 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5807 | u64 sqe_addr = req->cancel.addr; |
5808 | struct io_tctx_node *node; | |
5809 | int ret; | |
5810 | ||
5811 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5812 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5813 | spin_lock_irq(&ctx->completion_lock); | |
5814 | if (ret != -ENOENT) | |
5815 | goto done; | |
5816 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5817 | if (ret != -ENOENT) | |
5818 | goto done; | |
9ba5fac8 | 5819 | ret = io_poll_cancel(ctx, sqe_addr, false); |
58f99373 PB |
5820 | if (ret != -ENOENT) |
5821 | goto done; | |
5822 | spin_unlock_irq(&ctx->completion_lock); | |
5823 | ||
5824 | /* slow path, try all io-wq's */ | |
5825 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5826 | ret = -ENOENT; | |
5827 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5828 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5829 | |
58f99373 PB |
5830 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5831 | if (ret != -ENOENT) | |
5832 | break; | |
5833 | } | |
5834 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5835 | ||
5836 | spin_lock_irq(&ctx->completion_lock); | |
5837 | done: | |
d4d19c19 | 5838 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
58f99373 PB |
5839 | io_commit_cqring(ctx); |
5840 | spin_unlock_irq(&ctx->completion_lock); | |
5841 | io_cqring_ev_posted(ctx); | |
5842 | ||
5843 | if (ret < 0) | |
93d2bcd2 | 5844 | req_set_fail(req); |
58f99373 | 5845 | io_put_req(req); |
5262f567 JA |
5846 | return 0; |
5847 | } | |
5848 | ||
269bbe5f | 5849 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5850 | const struct io_uring_sqe *sqe) |
5851 | { | |
61710e43 DA |
5852 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5853 | return -EINVAL; | |
5854 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5855 | return -EINVAL; |
5856 | ||
269bbe5f BM |
5857 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5858 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5859 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5860 | return -EINVAL; |
269bbe5f | 5861 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5862 | return 0; |
5863 | } | |
5864 | ||
889fca73 | 5865 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5866 | { |
5867 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 5868 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 5869 | int ret; |
fbf23849 | 5870 | |
45d189c6 | 5871 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5872 | return -EAGAIN; |
05f3fb3c | 5873 | |
269bbe5f BM |
5874 | up.offset = req->rsrc_update.offset; |
5875 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
5876 | up.nr = 0; |
5877 | up.tags = 0; | |
615cee49 | 5878 | up.resv = 0; |
05f3fb3c JA |
5879 | |
5880 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 5881 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 5882 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5883 | mutex_unlock(&ctx->uring_lock); |
5884 | ||
5885 | if (ret < 0) | |
93d2bcd2 | 5886 | req_set_fail(req); |
889fca73 | 5887 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5888 | return 0; |
5889 | } | |
5890 | ||
bfe76559 | 5891 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5892 | { |
d625c6ee | 5893 | switch (req->opcode) { |
e781573e | 5894 | case IORING_OP_NOP: |
bfe76559 | 5895 | return 0; |
f67676d1 JA |
5896 | case IORING_OP_READV: |
5897 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5898 | case IORING_OP_READ: |
bfe76559 | 5899 | return io_read_prep(req, sqe); |
f67676d1 JA |
5900 | case IORING_OP_WRITEV: |
5901 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5902 | case IORING_OP_WRITE: |
bfe76559 | 5903 | return io_write_prep(req, sqe); |
0969e783 | 5904 | case IORING_OP_POLL_ADD: |
bfe76559 | 5905 | return io_poll_add_prep(req, sqe); |
0969e783 | 5906 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5907 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5908 | case IORING_OP_FSYNC: |
1155c76a | 5909 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5910 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5911 | return io_sfr_prep(req, sqe); |
03b1230c | 5912 | case IORING_OP_SENDMSG: |
fddaface | 5913 | case IORING_OP_SEND: |
bfe76559 | 5914 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5915 | case IORING_OP_RECVMSG: |
fddaface | 5916 | case IORING_OP_RECV: |
bfe76559 | 5917 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5918 | case IORING_OP_CONNECT: |
bfe76559 | 5919 | return io_connect_prep(req, sqe); |
2d28390a | 5920 | case IORING_OP_TIMEOUT: |
bfe76559 | 5921 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5922 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5923 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5924 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5925 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5926 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5927 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5928 | case IORING_OP_ACCEPT: |
bfe76559 | 5929 | return io_accept_prep(req, sqe); |
d63d1b5e | 5930 | case IORING_OP_FALLOCATE: |
bfe76559 | 5931 | return io_fallocate_prep(req, sqe); |
15b71abe | 5932 | case IORING_OP_OPENAT: |
bfe76559 | 5933 | return io_openat_prep(req, sqe); |
b5dba59e | 5934 | case IORING_OP_CLOSE: |
bfe76559 | 5935 | return io_close_prep(req, sqe); |
05f3fb3c | 5936 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5937 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5938 | case IORING_OP_STATX: |
bfe76559 | 5939 | return io_statx_prep(req, sqe); |
4840e418 | 5940 | case IORING_OP_FADVISE: |
bfe76559 | 5941 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5942 | case IORING_OP_MADVISE: |
bfe76559 | 5943 | return io_madvise_prep(req, sqe); |
cebdb986 | 5944 | case IORING_OP_OPENAT2: |
bfe76559 | 5945 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5946 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5947 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5948 | case IORING_OP_SPLICE: |
bfe76559 | 5949 | return io_splice_prep(req, sqe); |
ddf0322d | 5950 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5951 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5952 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5953 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5954 | case IORING_OP_TEE: |
bfe76559 | 5955 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5956 | case IORING_OP_SHUTDOWN: |
5957 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5958 | case IORING_OP_RENAMEAT: |
5959 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5960 | case IORING_OP_UNLINKAT: |
5961 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5962 | } |
5963 | ||
bfe76559 PB |
5964 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5965 | req->opcode); | |
bd54b6fe | 5966 | return -EINVAL; |
bfe76559 PB |
5967 | } |
5968 | ||
93642ef8 | 5969 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5970 | { |
b7e298d2 PB |
5971 | if (!io_op_defs[req->opcode].needs_async_setup) |
5972 | return 0; | |
5973 | if (WARN_ON_ONCE(req->async_data)) | |
5974 | return -EFAULT; | |
5975 | if (io_alloc_async_data(req)) | |
5976 | return -EAGAIN; | |
5977 | ||
93642ef8 PB |
5978 | switch (req->opcode) { |
5979 | case IORING_OP_READV: | |
93642ef8 PB |
5980 | return io_rw_prep_async(req, READ); |
5981 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5982 | return io_rw_prep_async(req, WRITE); |
5983 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5984 | return io_sendmsg_prep_async(req); |
5985 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5986 | return io_recvmsg_prep_async(req); |
5987 | case IORING_OP_CONNECT: | |
5988 | return io_connect_prep_async(req); | |
5989 | } | |
b7e298d2 PB |
5990 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5991 | req->opcode); | |
5992 | return -EFAULT; | |
f67676d1 JA |
5993 | } |
5994 | ||
9cf7c104 PB |
5995 | static u32 io_get_sequence(struct io_kiocb *req) |
5996 | { | |
a3dbdf54 | 5997 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 5998 | |
a3dbdf54 PB |
5999 | /* need original cached_sq_head, but it was increased for each req */ |
6000 | io_for_each_link(req, req) | |
6001 | seq--; | |
6002 | return seq; | |
9cf7c104 PB |
6003 | } |
6004 | ||
76cc33d7 | 6005 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 6006 | { |
3c19966d | 6007 | struct io_kiocb *pos; |
a197f664 | 6008 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6009 | struct io_defer_entry *de; |
f67676d1 | 6010 | int ret; |
9cf7c104 | 6011 | u32 seq; |
de0617e4 | 6012 | |
3c19966d PB |
6013 | /* |
6014 | * If we need to drain a request in the middle of a link, drain the | |
6015 | * head request and the next request/link after the current link. | |
6016 | * Considering sequential execution of links, IOSQE_IO_DRAIN will be | |
6017 | * maintained for every request of our link. | |
6018 | */ | |
6019 | if (ctx->drain_next) { | |
6020 | req->flags |= REQ_F_IO_DRAIN; | |
6021 | ctx->drain_next = false; | |
6022 | } | |
6023 | /* not interested in head, start from the first linked */ | |
6024 | io_for_each_link(pos, req->link) { | |
6025 | if (pos->flags & REQ_F_IO_DRAIN) { | |
6026 | ctx->drain_next = true; | |
6027 | req->flags |= REQ_F_IO_DRAIN; | |
6028 | break; | |
6029 | } | |
6030 | } | |
6031 | ||
9d858b21 | 6032 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 | 6033 | if (likely(list_empty_careful(&ctx->defer_list) && |
10c66904 PB |
6034 | !(req->flags & REQ_F_IO_DRAIN))) { |
6035 | ctx->drain_active = false; | |
76cc33d7 | 6036 | return false; |
10c66904 | 6037 | } |
9cf7c104 PB |
6038 | |
6039 | seq = io_get_sequence(req); | |
6040 | /* Still a chance to pass the sequence check */ | |
6041 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6042 | return false; |
de0617e4 | 6043 | |
b7e298d2 | 6044 | ret = io_req_prep_async(req); |
be7053b7 PB |
6045 | if (ret) |
6046 | return ret; | |
cbdcb435 | 6047 | io_prep_async_link(req); |
27dc8338 | 6048 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 PB |
6049 | if (!de) { |
6050 | io_req_complete_failed(req, ret); | |
6051 | return true; | |
6052 | } | |
2d28390a | 6053 | |
de0617e4 | 6054 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6055 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6056 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6057 | kfree(de); |
ae34817b | 6058 | io_queue_async_work(req); |
76cc33d7 | 6059 | return true; |
de0617e4 JA |
6060 | } |
6061 | ||
915967f6 | 6062 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6063 | de->req = req; |
9cf7c104 | 6064 | de->seq = seq; |
27dc8338 | 6065 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 | 6066 | spin_unlock_irq(&ctx->completion_lock); |
76cc33d7 | 6067 | return true; |
de0617e4 JA |
6068 | } |
6069 | ||
68fb8979 | 6070 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6071 | { |
0e1b6fe3 PB |
6072 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6073 | switch (req->opcode) { | |
6074 | case IORING_OP_READV: | |
6075 | case IORING_OP_READ_FIXED: | |
6076 | case IORING_OP_READ: | |
bcda7baa | 6077 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6078 | break; |
6079 | case IORING_OP_RECVMSG: | |
6080 | case IORING_OP_RECV: | |
bcda7baa | 6081 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6082 | break; |
6083 | } | |
99bc4c38 PB |
6084 | } |
6085 | ||
0e1b6fe3 PB |
6086 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6087 | switch (req->opcode) { | |
6088 | case IORING_OP_READV: | |
6089 | case IORING_OP_READ_FIXED: | |
6090 | case IORING_OP_READ: | |
6091 | case IORING_OP_WRITEV: | |
6092 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6093 | case IORING_OP_WRITE: { |
6094 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
6095 | |
6096 | kfree(io->free_iovec); | |
0e1b6fe3 | 6097 | break; |
e8c2bc1f | 6098 | } |
0e1b6fe3 | 6099 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6100 | case IORING_OP_SENDMSG: { |
6101 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6102 | |
6103 | kfree(io->free_iov); | |
0e1b6fe3 | 6104 | break; |
e8c2bc1f | 6105 | } |
0e1b6fe3 PB |
6106 | case IORING_OP_SPLICE: |
6107 | case IORING_OP_TEE: | |
e1d767f0 PB |
6108 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6109 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6110 | break; |
f3cd4850 JA |
6111 | case IORING_OP_OPENAT: |
6112 | case IORING_OP_OPENAT2: | |
6113 | if (req->open.filename) | |
6114 | putname(req->open.filename); | |
6115 | break; | |
80a261fd JA |
6116 | case IORING_OP_RENAMEAT: |
6117 | putname(req->rename.oldpath); | |
6118 | putname(req->rename.newpath); | |
6119 | break; | |
14a1143b JA |
6120 | case IORING_OP_UNLINKAT: |
6121 | putname(req->unlink.filename); | |
6122 | break; | |
0e1b6fe3 | 6123 | } |
99bc4c38 | 6124 | } |
75652a30 JA |
6125 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6126 | kfree(req->apoll->double_poll); | |
6127 | kfree(req->apoll); | |
6128 | req->apoll = NULL; | |
6129 | } | |
3a0a6902 PB |
6130 | if (req->flags & REQ_F_INFLIGHT) { |
6131 | struct io_uring_task *tctx = req->task->io_uring; | |
6132 | ||
6133 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 6134 | } |
c854357b | 6135 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 6136 | put_cred(req->creds); |
c854357b PB |
6137 | |
6138 | req->flags &= ~IO_REQ_CLEAN_FLAGS; | |
99bc4c38 PB |
6139 | } |
6140 | ||
889fca73 | 6141 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6142 | { |
a197f664 | 6143 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6144 | const struct cred *creds = NULL; |
d625c6ee | 6145 | int ret; |
2b188cc1 | 6146 | |
b8e64b53 | 6147 | if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) |
c10d1f98 | 6148 | creds = override_creds(req->creds); |
5730b27e | 6149 | |
d625c6ee | 6150 | switch (req->opcode) { |
2b188cc1 | 6151 | case IORING_OP_NOP: |
889fca73 | 6152 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6153 | break; |
6154 | case IORING_OP_READV: | |
edafccee | 6155 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6156 | case IORING_OP_READ: |
889fca73 | 6157 | ret = io_read(req, issue_flags); |
edafccee | 6158 | break; |
3529d8c2 | 6159 | case IORING_OP_WRITEV: |
edafccee | 6160 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6161 | case IORING_OP_WRITE: |
889fca73 | 6162 | ret = io_write(req, issue_flags); |
2b188cc1 | 6163 | break; |
c992fe29 | 6164 | case IORING_OP_FSYNC: |
45d189c6 | 6165 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6166 | break; |
221c5eb2 | 6167 | case IORING_OP_POLL_ADD: |
61e98203 | 6168 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6169 | break; |
6170 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6171 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6172 | break; |
5d17b4a4 | 6173 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6174 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6175 | break; |
0fa03c62 | 6176 | case IORING_OP_SENDMSG: |
889fca73 | 6177 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6178 | break; |
fddaface | 6179 | case IORING_OP_SEND: |
889fca73 | 6180 | ret = io_send(req, issue_flags); |
0fa03c62 | 6181 | break; |
aa1fa28f | 6182 | case IORING_OP_RECVMSG: |
889fca73 | 6183 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6184 | break; |
fddaface | 6185 | case IORING_OP_RECV: |
889fca73 | 6186 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6187 | break; |
5262f567 | 6188 | case IORING_OP_TIMEOUT: |
61e98203 | 6189 | ret = io_timeout(req, issue_flags); |
5262f567 | 6190 | break; |
11365043 | 6191 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6192 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6193 | break; |
17f2fe35 | 6194 | case IORING_OP_ACCEPT: |
889fca73 | 6195 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6196 | break; |
f8e85cf2 | 6197 | case IORING_OP_CONNECT: |
889fca73 | 6198 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6199 | break; |
62755e35 | 6200 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6201 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6202 | break; |
d63d1b5e | 6203 | case IORING_OP_FALLOCATE: |
45d189c6 | 6204 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6205 | break; |
15b71abe | 6206 | case IORING_OP_OPENAT: |
45d189c6 | 6207 | ret = io_openat(req, issue_flags); |
15b71abe | 6208 | break; |
b5dba59e | 6209 | case IORING_OP_CLOSE: |
889fca73 | 6210 | ret = io_close(req, issue_flags); |
b5dba59e | 6211 | break; |
05f3fb3c | 6212 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6213 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6214 | break; |
eddc7ef5 | 6215 | case IORING_OP_STATX: |
45d189c6 | 6216 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6217 | break; |
4840e418 | 6218 | case IORING_OP_FADVISE: |
45d189c6 | 6219 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6220 | break; |
c1ca757b | 6221 | case IORING_OP_MADVISE: |
45d189c6 | 6222 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6223 | break; |
cebdb986 | 6224 | case IORING_OP_OPENAT2: |
45d189c6 | 6225 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6226 | break; |
3e4827b0 | 6227 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6228 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6229 | break; |
7d67af2c | 6230 | case IORING_OP_SPLICE: |
45d189c6 | 6231 | ret = io_splice(req, issue_flags); |
7d67af2c | 6232 | break; |
ddf0322d | 6233 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6234 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6235 | break; |
067524e9 | 6236 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6237 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6238 | break; |
f2a8d5c7 | 6239 | case IORING_OP_TEE: |
45d189c6 | 6240 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6241 | break; |
36f4fa68 | 6242 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6243 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6244 | break; |
80a261fd | 6245 | case IORING_OP_RENAMEAT: |
45d189c6 | 6246 | ret = io_renameat(req, issue_flags); |
80a261fd | 6247 | break; |
14a1143b | 6248 | case IORING_OP_UNLINKAT: |
45d189c6 | 6249 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6250 | break; |
2b188cc1 JA |
6251 | default: |
6252 | ret = -EINVAL; | |
6253 | break; | |
6254 | } | |
6255 | ||
5730b27e JA |
6256 | if (creds) |
6257 | revert_creds(creds); | |
def596e9 JA |
6258 | if (ret) |
6259 | return ret; | |
b532576e | 6260 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6261 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6262 | io_iopoll_req_issued(req); | |
def596e9 JA |
6263 | |
6264 | return 0; | |
2b188cc1 JA |
6265 | } |
6266 | ||
5280f7e5 | 6267 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6268 | { |
6269 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6270 | struct io_kiocb *timeout; |
561fb04a | 6271 | int ret = 0; |
2b188cc1 | 6272 | |
6df1db6b PB |
6273 | timeout = io_prep_linked_timeout(req); |
6274 | if (timeout) | |
6275 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6276 | |
4014d943 | 6277 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6278 | ret = -ECANCELED; |
31b51510 | 6279 | |
561fb04a | 6280 | if (!ret) { |
561fb04a | 6281 | do { |
889fca73 | 6282 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6283 | /* |
6284 | * We can get EAGAIN for polled IO even though we're | |
6285 | * forcing a sync submission from here, since we can't | |
6286 | * wait for request slots on the block side. | |
6287 | */ | |
6288 | if (ret != -EAGAIN) | |
6289 | break; | |
6290 | cond_resched(); | |
6291 | } while (1); | |
6292 | } | |
31b51510 | 6293 | |
a3df7698 | 6294 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6295 | if (ret) { |
a3df7698 | 6296 | /* io-wq is going to take one down */ |
de9b4cca | 6297 | req_ref_get(req); |
a3df7698 | 6298 | io_req_task_queue_fail(req, ret); |
edafccee | 6299 | } |
2b188cc1 JA |
6300 | } |
6301 | ||
7b29f92d JA |
6302 | #define FFS_ASYNC_READ 0x1UL |
6303 | #define FFS_ASYNC_WRITE 0x2UL | |
6304 | #ifdef CONFIG_64BIT | |
6305 | #define FFS_ISREG 0x4UL | |
6306 | #else | |
6307 | #define FFS_ISREG 0x0UL | |
6308 | #endif | |
6309 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6310 | ||
aeca241b | 6311 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6312 | unsigned i) |
65e19f54 | 6313 | { |
aeca241b | 6314 | struct io_fixed_file *table_l2; |
65e19f54 | 6315 | |
aeca241b PB |
6316 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6317 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6318 | } |
6319 | ||
65e19f54 JA |
6320 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6321 | int index) | |
6322 | { | |
aeca241b | 6323 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6324 | |
a04b0ac0 | 6325 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6326 | } |
6327 | ||
a04b0ac0 | 6328 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6329 | { |
6330 | unsigned long file_ptr = (unsigned long) file; | |
6331 | ||
6332 | if (__io_file_supports_async(file, READ)) | |
6333 | file_ptr |= FFS_ASYNC_READ; | |
6334 | if (__io_file_supports_async(file, WRITE)) | |
6335 | file_ptr |= FFS_ASYNC_WRITE; | |
6336 | if (S_ISREG(file_inode(file)->i_mode)) | |
6337 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6338 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6339 | } |
6340 | ||
8371adf5 PB |
6341 | static struct file *io_file_get(struct io_submit_state *state, |
6342 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6343 | { |
a197f664 | 6344 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6345 | struct file *file; |
09bb8394 | 6346 | |
8da11c19 | 6347 | if (fixed) { |
7b29f92d JA |
6348 | unsigned long file_ptr; |
6349 | ||
479f517b | 6350 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6351 | return NULL; |
b7620121 | 6352 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6353 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6354 | file = (struct file *) (file_ptr & FFS_MASK); |
6355 | file_ptr &= ~FFS_MASK; | |
6356 | /* mask in overlapping REQ_F and FFS bits */ | |
6357 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6358 | io_req_set_rsrc_node(req); |
09bb8394 | 6359 | } else { |
c826bd7a | 6360 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6361 | file = __io_file_get(state, fd); |
d44f554e JA |
6362 | |
6363 | /* we don't allow fixed io_uring files */ | |
6364 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6365 | io_req_track_inflight(req); | |
09bb8394 JA |
6366 | } |
6367 | ||
8371adf5 | 6368 | return file; |
09bb8394 JA |
6369 | } |
6370 | ||
2665abfd | 6371 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6372 | { |
ad8a48ac JA |
6373 | struct io_timeout_data *data = container_of(timer, |
6374 | struct io_timeout_data, timer); | |
90cd7e42 | 6375 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6376 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6377 | unsigned long flags; |
2665abfd JA |
6378 | |
6379 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6380 | prev = req->timeout.head; |
6381 | req->timeout.head = NULL; | |
2665abfd JA |
6382 | |
6383 | /* | |
6384 | * We don't expect the list to be empty, that will only happen if we | |
6385 | * race with the completion of the linked work. | |
6386 | */ | |
447c19f3 | 6387 | if (prev) { |
f2f87370 | 6388 | io_remove_next_linked(prev); |
447c19f3 PB |
6389 | if (!req_ref_inc_not_zero(prev)) |
6390 | prev = NULL; | |
6391 | } | |
2665abfd JA |
6392 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6393 | ||
6394 | if (prev) { | |
014db007 | 6395 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6396 | io_put_req_deferred(prev, 1); |
a298232e | 6397 | io_put_req_deferred(req, 1); |
47f46768 | 6398 | } else { |
9ae1f8dd | 6399 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6400 | } |
2665abfd JA |
6401 | return HRTIMER_NORESTART; |
6402 | } | |
6403 | ||
de968c18 | 6404 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6405 | { |
de968c18 PB |
6406 | struct io_ring_ctx *ctx = req->ctx; |
6407 | ||
6408 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6409 | /* |
f2f87370 PB |
6410 | * If the back reference is NULL, then our linked request finished |
6411 | * before we got a chance to setup the timer | |
76a46e06 | 6412 | */ |
90cd7e42 | 6413 | if (req->timeout.head) { |
e8c2bc1f | 6414 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6415 | |
ad8a48ac JA |
6416 | data->timer.function = io_link_timeout_fn; |
6417 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6418 | data->mode); | |
2665abfd | 6419 | } |
76a46e06 | 6420 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6421 | /* drop submission reference */ |
76a46e06 JA |
6422 | io_put_req(req); |
6423 | } | |
2665abfd | 6424 | |
ad8a48ac | 6425 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6426 | { |
f2f87370 | 6427 | struct io_kiocb *nxt = req->link; |
2665abfd | 6428 | |
f2f87370 PB |
6429 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6430 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6431 | return NULL; |
2665abfd | 6432 | |
90cd7e42 | 6433 | nxt->timeout.head = req; |
900fad45 | 6434 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6435 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6436 | return nxt; |
2665abfd JA |
6437 | } |
6438 | ||
c5eef2b9 | 6439 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6440 | { |
d3d7298d | 6441 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6442 | int ret; |
2b188cc1 | 6443 | |
c5eef2b9 | 6444 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6445 | |
491381ce JA |
6446 | /* |
6447 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6448 | * doesn't support non-blocking read/write attempts | |
6449 | */ | |
1840038e | 6450 | if (likely(!ret)) { |
0d63c148 | 6451 | /* drop submission reference */ |
e342c807 | 6452 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6453 | struct io_ring_ctx *ctx = req->ctx; |
6454 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6455 | |
6dd0be1e | 6456 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6457 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
2a2758f2 | 6458 | io_submit_flush_completions(ctx); |
9affd664 | 6459 | } else { |
d3d7298d | 6460 | io_put_req(req); |
0d63c148 | 6461 | } |
1840038e PB |
6462 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6463 | if (!io_arm_poll_handler(req)) { | |
6464 | /* | |
6465 | * Queued up for async execution, worker will release | |
6466 | * submit reference when the iocb is actually submitted. | |
6467 | */ | |
6468 | io_queue_async_work(req); | |
6469 | } | |
0d63c148 | 6470 | } else { |
f41db273 | 6471 | io_req_complete_failed(req, ret); |
9e645e11 | 6472 | } |
d3d7298d PB |
6473 | if (linked_timeout) |
6474 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6475 | } |
6476 | ||
441b8a78 | 6477 | static inline void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 | 6478 | { |
10c66904 | 6479 | if (unlikely(req->ctx->drain_active) && io_drain_req(req)) |
76cc33d7 | 6480 | return; |
4fe2c963 | 6481 | |
76cc33d7 | 6482 | if (likely(!(req->flags & REQ_F_FORCE_ASYNC))) { |
c5eef2b9 | 6483 | __io_queue_sqe(req); |
76cc33d7 PB |
6484 | } else { |
6485 | int ret = io_req_prep_async(req); | |
6486 | ||
6487 | if (unlikely(ret)) | |
6488 | io_req_complete_failed(req, ret); | |
6489 | else | |
6490 | io_queue_async_work(req); | |
ce35a47a | 6491 | } |
4fe2c963 JL |
6492 | } |
6493 | ||
b16fed66 PB |
6494 | /* |
6495 | * Check SQE restrictions (opcode and flags). | |
6496 | * | |
6497 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6498 | */ | |
6499 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6500 | struct io_kiocb *req, | |
6501 | unsigned int sqe_flags) | |
4fe2c963 | 6502 | { |
b16fed66 PB |
6503 | if (!ctx->restricted) |
6504 | return true; | |
6505 | ||
6506 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6507 | return false; | |
6508 | ||
6509 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6510 | ctx->restrictions.sqe_flags_required) | |
6511 | return false; | |
6512 | ||
6513 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6514 | ctx->restrictions.sqe_flags_required)) | |
6515 | return false; | |
6516 | ||
6517 | return true; | |
4fe2c963 JL |
6518 | } |
6519 | ||
b16fed66 PB |
6520 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6521 | const struct io_uring_sqe *sqe) | |
6522 | { | |
6523 | struct io_submit_state *state; | |
6524 | unsigned int sqe_flags; | |
003e8dcc | 6525 | int personality, ret = 0; |
b16fed66 PB |
6526 | |
6527 | req->opcode = READ_ONCE(sqe->opcode); | |
6528 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6529 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6530 | req->user_data = READ_ONCE(sqe->user_data); | |
6531 | req->async_data = NULL; | |
6532 | req->file = NULL; | |
6533 | req->ctx = ctx; | |
6534 | req->link = NULL; | |
6535 | req->fixed_rsrc_refs = NULL; | |
6536 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6537 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6538 | req->task = current; |
6539 | req->result = 0; | |
6540 | ||
6541 | /* enforce forwards compatibility on users */ | |
dddca226 | 6542 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 6543 | return -EINVAL; |
b16fed66 PB |
6544 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
6545 | return -EINVAL; | |
b16fed66 PB |
6546 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6547 | return -EACCES; | |
6548 | ||
6549 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6550 | !io_op_defs[req->opcode].buffer_select) | |
6551 | return -EOPNOTSUPP; | |
3c19966d PB |
6552 | if (unlikely(sqe_flags & IOSQE_IO_DRAIN)) |
6553 | ctx->drain_active = true; | |
863e0560 | 6554 | |
003e8dcc JA |
6555 | personality = READ_ONCE(sqe->personality); |
6556 | if (personality) { | |
c10d1f98 PB |
6557 | req->creds = xa_load(&ctx->personalities, personality); |
6558 | if (!req->creds) | |
003e8dcc | 6559 | return -EINVAL; |
c10d1f98 | 6560 | get_cred(req->creds); |
b8e64b53 | 6561 | req->flags |= REQ_F_CREDS; |
003e8dcc | 6562 | } |
b16fed66 PB |
6563 | state = &ctx->submit_state; |
6564 | ||
6565 | /* | |
6566 | * Plug now if we have more than 1 IO left after this, and the target | |
6567 | * is potentially a read/write to block based storage. | |
6568 | */ | |
6569 | if (!state->plug_started && state->ios_left > 1 && | |
6570 | io_op_defs[req->opcode].plug) { | |
6571 | blk_start_plug(&state->plug); | |
6572 | state->plug_started = true; | |
6573 | } | |
6574 | ||
6575 | if (io_op_defs[req->opcode].needs_file) { | |
6576 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6577 | ||
6578 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6579 | if (unlikely(!req->file)) | |
6580 | ret = -EBADF; | |
6581 | } | |
6582 | ||
6583 | state->ios_left--; | |
6584 | return ret; | |
6585 | } | |
6586 | ||
a6b8cadc | 6587 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6588 | const struct io_uring_sqe *sqe) |
9e645e11 | 6589 | { |
a1ab7b35 | 6590 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6591 | int ret; |
9e645e11 | 6592 | |
a6b8cadc PB |
6593 | ret = io_init_req(ctx, req, sqe); |
6594 | if (unlikely(ret)) { | |
6595 | fail_req: | |
de59bc10 PB |
6596 | if (link->head) { |
6597 | /* fail even hard links since we don't submit */ | |
93d2bcd2 | 6598 | req_set_fail(link->head); |
f41db273 | 6599 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6600 | link->head = NULL; |
6601 | } | |
f41db273 | 6602 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6603 | return ret; |
6604 | } | |
441b8a78 | 6605 | |
be7053b7 PB |
6606 | ret = io_req_prep(req, sqe); |
6607 | if (unlikely(ret)) | |
6608 | goto fail_req; | |
a6b8cadc | 6609 | |
be7053b7 | 6610 | /* don't need @sqe from now on */ |
236daeae OL |
6611 | trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, |
6612 | req->flags, true, | |
6613 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 6614 | |
9e645e11 JA |
6615 | /* |
6616 | * If we already have a head request, queue this one for async | |
6617 | * submittal once the head completes. If we don't have a head but | |
6618 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6619 | * submitted sync once the chain is complete. If none of those | |
6620 | * conditions are true (normal request), then just queue it. | |
6621 | */ | |
863e0560 PB |
6622 | if (link->head) { |
6623 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6624 | |
b7e298d2 | 6625 | ret = io_req_prep_async(req); |
cf109604 | 6626 | if (unlikely(ret)) |
a6b8cadc | 6627 | goto fail_req; |
9d76377f | 6628 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6629 | link->last->link = req; |
863e0560 | 6630 | link->last = req; |
32fe525b PB |
6631 | |
6632 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6633 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 6634 | link->head = NULL; |
5e159204 | 6635 | io_queue_sqe(head); |
32fe525b | 6636 | } |
9e645e11 | 6637 | } else { |
ef4ff581 | 6638 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6639 | link->head = req; |
6640 | link->last = req; | |
711be031 | 6641 | } else { |
be7053b7 | 6642 | io_queue_sqe(req); |
711be031 | 6643 | } |
9e645e11 | 6644 | } |
2e6e1fde | 6645 | |
1d4240cc | 6646 | return 0; |
9e645e11 JA |
6647 | } |
6648 | ||
9a56a232 JA |
6649 | /* |
6650 | * Batched submission is done, ensure local IO is flushed out. | |
6651 | */ | |
ba88ff11 PB |
6652 | static void io_submit_state_end(struct io_submit_state *state, |
6653 | struct io_ring_ctx *ctx) | |
9a56a232 | 6654 | { |
a1ab7b35 | 6655 | if (state->link.head) |
de59bc10 | 6656 | io_queue_sqe(state->link.head); |
6dd0be1e | 6657 | if (state->comp.nr) |
2a2758f2 | 6658 | io_submit_flush_completions(ctx); |
27926b68 JA |
6659 | if (state->plug_started) |
6660 | blk_finish_plug(&state->plug); | |
9f13c35b | 6661 | io_state_file_put(state); |
9a56a232 JA |
6662 | } |
6663 | ||
6664 | /* | |
6665 | * Start submission side cache. | |
6666 | */ | |
6667 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6668 | unsigned int max_ios) |
9a56a232 | 6669 | { |
27926b68 | 6670 | state->plug_started = false; |
9a56a232 | 6671 | state->ios_left = max_ios; |
a1ab7b35 PB |
6672 | /* set only head, no need to init link_last in advance */ |
6673 | state->link.head = NULL; | |
9a56a232 JA |
6674 | } |
6675 | ||
2b188cc1 JA |
6676 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6677 | { | |
75b28aff | 6678 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6679 | |
caf582c6 PB |
6680 | /* |
6681 | * Ensure any loads from the SQEs are done at this point, | |
6682 | * since once we write the new head, the application could | |
6683 | * write new data to them. | |
6684 | */ | |
6685 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6686 | } |
6687 | ||
2b188cc1 | 6688 | /* |
dd9ae8a0 | 6689 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
6690 | * that is mapped by userspace. This means that care needs to be taken to |
6691 | * ensure that reads are stable, as we cannot rely on userspace always | |
6692 | * being a good citizen. If members of the sqe are validated and then later | |
6693 | * used, it's important that those reads are done through READ_ONCE() to | |
6694 | * prevent a re-load down the line. | |
6695 | */ | |
709b302f | 6696 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6697 | { |
ea5ab3b5 | 6698 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 6699 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
6700 | |
6701 | /* | |
6702 | * The cached sq head (or cq tail) serves two purposes: | |
6703 | * | |
6704 | * 1) allows us to batch the cost of updating the user visible | |
6705 | * head updates. | |
6706 | * 2) allows the kernel side to track the head on its own, even | |
6707 | * though the application is the one updating it. | |
6708 | */ | |
17d3aeb3 | 6709 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
6710 | if (likely(head < ctx->sq_entries)) |
6711 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6712 | |
6713 | /* drop invalid entries */ | |
15641e42 PB |
6714 | ctx->cq_extra--; |
6715 | WRITE_ONCE(ctx->rings->sq_dropped, | |
6716 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
6717 | return NULL; |
6718 | } | |
6719 | ||
0f212204 | 6720 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6721 | { |
09899b19 | 6722 | struct io_uring_task *tctx; |
46c4e16a | 6723 | int submitted = 0; |
6c271ce2 | 6724 | |
ee7d46d9 PB |
6725 | /* make sure SQ entry isn't read before tail */ |
6726 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
6727 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6728 | return -EAGAIN; | |
6c271ce2 | 6729 | |
09899b19 PB |
6730 | tctx = current->io_uring; |
6731 | tctx->cached_refs -= nr; | |
6732 | if (unlikely(tctx->cached_refs < 0)) { | |
6733 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
6734 | ||
6735 | percpu_counter_add(&tctx->inflight, refill); | |
6736 | refcount_add(refill, ¤t->usage); | |
6737 | tctx->cached_refs += refill; | |
6738 | } | |
ba88ff11 | 6739 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6740 | |
46c4e16a | 6741 | while (submitted < nr) { |
3529d8c2 | 6742 | const struct io_uring_sqe *sqe; |
196be95c | 6743 | struct io_kiocb *req; |
fb5ccc98 | 6744 | |
258b29a9 | 6745 | req = io_alloc_req(ctx); |
196be95c PB |
6746 | if (unlikely(!req)) { |
6747 | if (!submitted) | |
6748 | submitted = -EAGAIN; | |
fb5ccc98 | 6749 | break; |
196be95c | 6750 | } |
4fccfcbb PB |
6751 | sqe = io_get_sqe(ctx); |
6752 | if (unlikely(!sqe)) { | |
6753 | kmem_cache_free(req_cachep, req); | |
6754 | break; | |
6755 | } | |
d3656344 JA |
6756 | /* will complete beyond this point, count as submitted */ |
6757 | submitted++; | |
a1ab7b35 | 6758 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6759 | break; |
6c271ce2 JA |
6760 | } |
6761 | ||
9466f437 PB |
6762 | if (unlikely(submitted != nr)) { |
6763 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 6764 | int unused = nr - ref_used; |
9466f437 | 6765 | |
09899b19 | 6766 | current->io_uring->cached_refs += unused; |
d8a6df10 | 6767 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 6768 | } |
6c271ce2 | 6769 | |
a1ab7b35 | 6770 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6771 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6772 | io_commit_sqring(ctx); | |
6773 | ||
6c271ce2 JA |
6774 | return submitted; |
6775 | } | |
6776 | ||
e4b6d902 PB |
6777 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
6778 | { | |
6779 | return READ_ONCE(sqd->state); | |
6780 | } | |
6781 | ||
23b3628e XW |
6782 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6783 | { | |
6784 | /* Tell userspace we may need a wakeup call */ | |
6785 | spin_lock_irq(&ctx->completion_lock); | |
6786 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6787 | spin_unlock_irq(&ctx->completion_lock); | |
6788 | } | |
6789 | ||
6790 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6791 | { | |
6792 | spin_lock_irq(&ctx->completion_lock); | |
6793 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6794 | spin_unlock_irq(&ctx->completion_lock); | |
6795 | } | |
6796 | ||
08369246 | 6797 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6798 | { |
c8d1ba58 | 6799 | unsigned int to_submit; |
bdcd3eab | 6800 | int ret = 0; |
6c271ce2 | 6801 | |
c8d1ba58 | 6802 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 6803 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
6804 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
6805 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 6806 | |
906a3c6f | 6807 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6808 | unsigned nr_events = 0; |
948e1947 PB |
6809 | const struct cred *creds = NULL; |
6810 | ||
6811 | if (ctx->sq_creds != current_cred()) | |
6812 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 6813 | |
c8d1ba58 | 6814 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6815 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6816 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6817 | |
3b763ba1 PB |
6818 | /* |
6819 | * Don't submit if refs are dying, good for io_uring_register(), | |
6820 | * but also it is relied upon by io_ring_exit_work() | |
6821 | */ | |
0298ef96 PB |
6822 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6823 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6824 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 6825 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 6826 | |
acfb381d PB |
6827 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6828 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
6829 | if (creds) |
6830 | revert_creds(creds); | |
acfb381d | 6831 | } |
6c271ce2 | 6832 | |
08369246 XW |
6833 | return ret; |
6834 | } | |
6c271ce2 | 6835 | |
08369246 XW |
6836 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6837 | { | |
6838 | struct io_ring_ctx *ctx; | |
6839 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6840 | |
c9dca27d PB |
6841 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6842 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6843 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6844 | } |
6c271ce2 | 6845 | |
e4b6d902 PB |
6846 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
6847 | { | |
6848 | bool did_sig = false; | |
6849 | struct ksignal ksig; | |
6850 | ||
6851 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
6852 | signal_pending(current)) { | |
6853 | mutex_unlock(&sqd->lock); | |
6854 | if (signal_pending(current)) | |
6855 | did_sig = get_signal(&ksig); | |
6856 | cond_resched(); | |
6857 | mutex_lock(&sqd->lock); | |
6858 | } | |
6859 | io_run_task_work(); | |
e4b6d902 PB |
6860 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
6861 | } | |
6862 | ||
c8d1ba58 JA |
6863 | static int io_sq_thread(void *data) |
6864 | { | |
69fb2131 JA |
6865 | struct io_sq_data *sqd = data; |
6866 | struct io_ring_ctx *ctx; | |
a0d9205f | 6867 | unsigned long timeout = 0; |
37d1e2e3 | 6868 | char buf[TASK_COMM_LEN]; |
08369246 | 6869 | DEFINE_WAIT(wait); |
6c271ce2 | 6870 | |
696ee88a | 6871 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6872 | set_task_comm(current, buf); |
37d1e2e3 JA |
6873 | |
6874 | if (sqd->sq_cpu != -1) | |
6875 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6876 | else | |
6877 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6878 | current->flags |= PF_NO_SETAFFINITY; | |
6879 | ||
09a6f4ef | 6880 | mutex_lock(&sqd->lock); |
e4b6d902 | 6881 | while (1) { |
1a924a80 | 6882 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 6883 | |
e4b6d902 PB |
6884 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
6885 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 6886 | break; |
08369246 | 6887 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6888 | continue; |
08369246 | 6889 | } |
e4b6d902 | 6890 | |
e95eee2d | 6891 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6892 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 6893 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 6894 | |
08369246 XW |
6895 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6896 | sqt_spin = true; | |
69fb2131 | 6897 | } |
6c271ce2 | 6898 | |
08369246 | 6899 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6900 | io_run_task_work(); |
6901 | cond_resched(); | |
08369246 XW |
6902 | if (sqt_spin) |
6903 | timeout = jiffies + sqd->sq_thread_idle; | |
6904 | continue; | |
6905 | } | |
6906 | ||
08369246 | 6907 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
99713501 | 6908 | if (!io_sqd_events_pending(sqd) && !io_run_task_work()) { |
1a924a80 PB |
6909 | bool needs_sched = true; |
6910 | ||
724cb4f9 | 6911 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
6912 | io_ring_set_wakeup_flag(ctx); |
6913 | ||
724cb4f9 HX |
6914 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
6915 | !list_empty_careful(&ctx->iopoll_list)) { | |
6916 | needs_sched = false; | |
6917 | break; | |
6918 | } | |
6919 | if (io_sqring_entries(ctx)) { | |
6920 | needs_sched = false; | |
6921 | break; | |
6922 | } | |
6923 | } | |
6924 | ||
6925 | if (needs_sched) { | |
6926 | mutex_unlock(&sqd->lock); | |
6927 | schedule(); | |
6928 | mutex_lock(&sqd->lock); | |
6929 | } | |
69fb2131 JA |
6930 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6931 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6932 | } |
08369246 XW |
6933 | |
6934 | finish_wait(&sqd->wait, &wait); | |
6935 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6936 | } |
28cea78a | 6937 | |
78cc687b | 6938 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 6939 | sqd->thread = NULL; |
05962f95 | 6940 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6941 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 6942 | io_run_task_work(); |
734551df PB |
6943 | mutex_unlock(&sqd->lock); |
6944 | ||
37d1e2e3 JA |
6945 | complete(&sqd->exited); |
6946 | do_exit(0); | |
6c271ce2 JA |
6947 | } |
6948 | ||
bda52162 JA |
6949 | struct io_wait_queue { |
6950 | struct wait_queue_entry wq; | |
6951 | struct io_ring_ctx *ctx; | |
6952 | unsigned to_wait; | |
6953 | unsigned nr_timeouts; | |
6954 | }; | |
6955 | ||
6c503150 | 6956 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6957 | { |
6958 | struct io_ring_ctx *ctx = iowq->ctx; | |
6959 | ||
6960 | /* | |
d195a66e | 6961 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6962 | * started waiting. For timeouts, we always want to return to userspace, |
6963 | * regardless of event count. | |
6964 | */ | |
6c503150 | 6965 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6966 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6967 | } | |
6968 | ||
6969 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6970 | int wake_flags, void *key) | |
6971 | { | |
6972 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6973 | wq); | |
6974 | ||
6c503150 PB |
6975 | /* |
6976 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6977 | * the task, and the next invocation will do it. | |
6978 | */ | |
5ed7a37d | 6979 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
6980 | return autoremove_wake_function(curr, mode, wake_flags, key); |
6981 | return -1; | |
bda52162 JA |
6982 | } |
6983 | ||
af9c1a44 JA |
6984 | static int io_run_task_work_sig(void) |
6985 | { | |
6986 | if (io_run_task_work()) | |
6987 | return 1; | |
6988 | if (!signal_pending(current)) | |
6989 | return 0; | |
0b8cfa97 | 6990 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6991 | return -ERESTARTSYS; |
af9c1a44 JA |
6992 | return -EINTR; |
6993 | } | |
6994 | ||
eeb60b9a PB |
6995 | /* when returns >0, the caller should retry */ |
6996 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6997 | struct io_wait_queue *iowq, | |
6998 | signed long *timeout) | |
6999 | { | |
7000 | int ret; | |
7001 | ||
7002 | /* make sure we run task_work before checking for signals */ | |
7003 | ret = io_run_task_work_sig(); | |
7004 | if (ret || io_should_wake(iowq)) | |
7005 | return ret; | |
7006 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 7007 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
7008 | return 1; |
7009 | ||
7010 | *timeout = schedule_timeout(*timeout); | |
7011 | return !*timeout ? -ETIME : 1; | |
7012 | } | |
7013 | ||
2b188cc1 JA |
7014 | /* |
7015 | * Wait until events become available, if we don't already have some. The | |
7016 | * application must reap them itself, as they reside on the shared cq ring. | |
7017 | */ | |
7018 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7019 | const sigset_t __user *sig, size_t sigsz, |
7020 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7021 | { |
bda52162 JA |
7022 | struct io_wait_queue iowq = { |
7023 | .wq = { | |
7024 | .private = current, | |
7025 | .func = io_wake_function, | |
7026 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7027 | }, | |
7028 | .ctx = ctx, | |
7029 | .to_wait = min_events, | |
7030 | }; | |
75b28aff | 7031 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7032 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7033 | int ret; | |
2b188cc1 | 7034 | |
b41e9852 | 7035 | do { |
6c2450ae | 7036 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 7037 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7038 | return 0; |
4c6e277c | 7039 | if (!io_run_task_work()) |
b41e9852 | 7040 | break; |
b41e9852 | 7041 | } while (1); |
2b188cc1 JA |
7042 | |
7043 | if (sig) { | |
9e75ad5d AB |
7044 | #ifdef CONFIG_COMPAT |
7045 | if (in_compat_syscall()) | |
7046 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7047 | sigsz); |
9e75ad5d AB |
7048 | else |
7049 | #endif | |
b772434b | 7050 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7051 | |
2b188cc1 JA |
7052 | if (ret) |
7053 | return ret; | |
7054 | } | |
7055 | ||
c73ebb68 | 7056 | if (uts) { |
c1d5a224 PB |
7057 | struct timespec64 ts; |
7058 | ||
c73ebb68 HX |
7059 | if (get_timespec64(&ts, uts)) |
7060 | return -EFAULT; | |
7061 | timeout = timespec64_to_jiffies(&ts); | |
7062 | } | |
7063 | ||
bda52162 | 7064 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7065 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7066 | do { |
ca0a2651 | 7067 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7068 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7069 | ret = -EBUSY; |
7070 | break; | |
7071 | } | |
311997b3 | 7072 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7073 | TASK_INTERRUPTIBLE); |
eeb60b9a | 7074 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
311997b3 | 7075 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7076 | cond_resched(); |
eeb60b9a | 7077 | } while (ret > 0); |
bda52162 | 7078 | |
b7db41c9 | 7079 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7080 | |
75b28aff | 7081 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7082 | } |
7083 | ||
9123c8ff | 7084 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7085 | { |
9123c8ff | 7086 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7087 | |
846a4ef2 | 7088 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7089 | kfree(table[i]); |
7090 | kfree(table); | |
7091 | } | |
7092 | ||
7093 | static void **io_alloc_page_table(size_t size) | |
7094 | { | |
7095 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7096 | size_t init_size = size; | |
7097 | void **table; | |
7098 | ||
7099 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL); | |
7100 | if (!table) | |
7101 | return NULL; | |
7102 | ||
7103 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 7104 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff PB |
7105 | |
7106 | table[i] = kzalloc(this_size, GFP_KERNEL); | |
7107 | if (!table[i]) { | |
7108 | io_free_page_table(table, init_size); | |
7109 | return NULL; | |
7110 | } | |
7111 | size -= this_size; | |
7112 | } | |
7113 | return table; | |
05f3fb3c JA |
7114 | } |
7115 | ||
2a63b2d9 | 7116 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7117 | { |
2a63b2d9 | 7118 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7119 | } |
7120 | ||
2a63b2d9 | 7121 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7122 | { |
2a63b2d9 BM |
7123 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7124 | } | |
65e19f54 | 7125 | |
28a9fe25 | 7126 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7127 | { |
28a9fe25 PB |
7128 | percpu_ref_exit(&ref_node->refs); |
7129 | kfree(ref_node); | |
1642b445 PB |
7130 | } |
7131 | ||
a7f0ed5a PB |
7132 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7133 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7134 | { |
a7f0ed5a PB |
7135 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7136 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7137 | |
a7f0ed5a PB |
7138 | if (data_to_kill) { |
7139 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7140 | |
a7f0ed5a PB |
7141 | rsrc_node->rsrc_data = data_to_kill; |
7142 | io_rsrc_ref_lock(ctx); | |
7143 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7144 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7145 | |
3e942498 | 7146 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7147 | percpu_ref_kill(&rsrc_node->refs); |
7148 | ctx->rsrc_node = NULL; | |
7149 | } | |
6b06314c | 7150 | |
a7f0ed5a PB |
7151 | if (!ctx->rsrc_node) { |
7152 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7153 | ctx->rsrc_backup_node = NULL; | |
7154 | } | |
8bad28d8 HX |
7155 | } |
7156 | ||
a7f0ed5a | 7157 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7158 | { |
7159 | if (ctx->rsrc_backup_node) | |
7160 | return 0; | |
b895c9a6 | 7161 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7162 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7163 | } |
7164 | ||
40ae0ff7 | 7165 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7166 | { |
7167 | int ret; | |
05589553 | 7168 | |
215c3902 | 7169 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7170 | if (data->quiesce) |
7171 | return -ENXIO; | |
05589553 | 7172 | |
8bad28d8 | 7173 | data->quiesce = true; |
1ffc5422 | 7174 | do { |
a7f0ed5a | 7175 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7176 | if (ret) |
f2303b1f | 7177 | break; |
a7f0ed5a | 7178 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7179 | |
3e942498 PB |
7180 | /* kill initial ref, already quiesced if zero */ |
7181 | if (atomic_dec_and_test(&data->refs)) | |
7182 | break; | |
8bad28d8 | 7183 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7184 | ret = wait_for_completion_interruptible(&data->done); |
7185 | if (!ret) | |
7186 | break; | |
8bad28d8 | 7187 | |
3e942498 PB |
7188 | atomic_inc(&data->refs); |
7189 | /* wait for all works potentially completing data->done */ | |
7190 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7191 | reinit_completion(&data->done); |
8dd03afe | 7192 | |
8bad28d8 | 7193 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7194 | ret = io_run_task_work_sig(); |
8bad28d8 | 7195 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7196 | } while (ret >= 0); |
8bad28d8 | 7197 | data->quiesce = false; |
05f3fb3c | 7198 | |
8bad28d8 | 7199 | return ret; |
d7954b2b BM |
7200 | } |
7201 | ||
2d091d62 PB |
7202 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7203 | { | |
7204 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7205 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7206 | ||
7207 | return &data->tags[table_idx][off]; | |
7208 | } | |
7209 | ||
44b31f2f | 7210 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7211 | { |
2d091d62 PB |
7212 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7213 | ||
7214 | if (data->tags) | |
7215 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7216 | kfree(data); |
7217 | } | |
7218 | ||
d878c816 PB |
7219 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7220 | u64 __user *utags, unsigned nr, | |
7221 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7222 | { |
b895c9a6 | 7223 | struct io_rsrc_data *data; |
2d091d62 | 7224 | int ret = -ENOMEM; |
d878c816 | 7225 | unsigned i; |
1ad555c6 BM |
7226 | |
7227 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7228 | if (!data) | |
d878c816 | 7229 | return -ENOMEM; |
2d091d62 | 7230 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7231 | if (!data->tags) { |
1ad555c6 | 7232 | kfree(data); |
d878c816 PB |
7233 | return -ENOMEM; |
7234 | } | |
2d091d62 PB |
7235 | |
7236 | data->nr = nr; | |
7237 | data->ctx = ctx; | |
7238 | data->do_put = do_put; | |
d878c816 | 7239 | if (utags) { |
2d091d62 | 7240 | ret = -EFAULT; |
d878c816 | 7241 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7242 | u64 *tag_slot = io_get_tag_slot(data, i); |
7243 | ||
7244 | if (copy_from_user(tag_slot, &utags[i], | |
7245 | sizeof(*tag_slot))) | |
2d091d62 | 7246 | goto fail; |
d878c816 | 7247 | } |
1ad555c6 | 7248 | } |
b60c8dce | 7249 | |
3e942498 | 7250 | atomic_set(&data->refs, 1); |
1ad555c6 | 7251 | init_completion(&data->done); |
d878c816 PB |
7252 | *pdata = data; |
7253 | return 0; | |
2d091d62 PB |
7254 | fail: |
7255 | io_rsrc_data_free(data); | |
7256 | return ret; | |
1ad555c6 BM |
7257 | } |
7258 | ||
9123c8ff PB |
7259 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7260 | { | |
7261 | size_t size = nr_files * sizeof(struct io_fixed_file); | |
7262 | ||
7263 | table->files = (struct io_fixed_file **)io_alloc_page_table(size); | |
7264 | return !!table->files; | |
7265 | } | |
7266 | ||
7267 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) | |
7268 | { | |
7269 | size_t size = nr_files * sizeof(struct io_fixed_file); | |
7270 | ||
7271 | io_free_page_table((void **)table->files, size); | |
7272 | table->files = NULL; | |
7273 | } | |
7274 | ||
fff4db76 | 7275 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7276 | { |
fff4db76 PB |
7277 | #if defined(CONFIG_UNIX) |
7278 | if (ctx->ring_sock) { | |
7279 | struct sock *sock = ctx->ring_sock->sk; | |
7280 | struct sk_buff *skb; | |
7281 | ||
7282 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7283 | kfree_skb(skb); | |
7284 | } | |
7285 | #else | |
7286 | int i; | |
7287 | ||
7288 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7289 | struct file *file; | |
7290 | ||
7291 | file = io_file_from_index(ctx, i); | |
7292 | if (file) | |
7293 | fput(file); | |
7294 | } | |
7295 | #endif | |
7296 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); | |
44b31f2f | 7297 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7298 | ctx->file_data = NULL; |
7299 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7300 | } |
7301 | ||
d7954b2b BM |
7302 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7303 | { | |
d7954b2b BM |
7304 | int ret; |
7305 | ||
08480400 | 7306 | if (!ctx->file_data) |
d7954b2b | 7307 | return -ENXIO; |
08480400 PB |
7308 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7309 | if (!ret) | |
7310 | __io_sqe_files_unregister(ctx); | |
7311 | return ret; | |
6b06314c JA |
7312 | } |
7313 | ||
37d1e2e3 | 7314 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7315 | __releases(&sqd->lock) |
37d1e2e3 | 7316 | { |
521d6a73 PB |
7317 | WARN_ON_ONCE(sqd->thread == current); |
7318 | ||
9e138a48 PB |
7319 | /* |
7320 | * Do the dance but not conditional clear_bit() because it'd race with | |
7321 | * other threads incrementing park_pending and setting the bit. | |
7322 | */ | |
37d1e2e3 | 7323 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7324 | if (atomic_dec_return(&sqd->park_pending)) |
7325 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7326 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7327 | } |
7328 | ||
86e0d676 | 7329 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7330 | __acquires(&sqd->lock) |
37d1e2e3 | 7331 | { |
521d6a73 PB |
7332 | WARN_ON_ONCE(sqd->thread == current); |
7333 | ||
9e138a48 | 7334 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7335 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7336 | mutex_lock(&sqd->lock); |
05962f95 | 7337 | if (sqd->thread) |
86e0d676 | 7338 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7339 | } |
7340 | ||
7341 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7342 | { | |
521d6a73 | 7343 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7344 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7345 | |
05962f95 | 7346 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7347 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7348 | if (sqd->thread) |
7349 | wake_up_process(sqd->thread); | |
09a6f4ef | 7350 | mutex_unlock(&sqd->lock); |
05962f95 | 7351 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7352 | } |
7353 | ||
534ca6d6 | 7354 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7355 | { |
534ca6d6 | 7356 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7357 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7358 | ||
37d1e2e3 JA |
7359 | io_sq_thread_stop(sqd); |
7360 | kfree(sqd); | |
7361 | } | |
7362 | } | |
7363 | ||
7364 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7365 | { | |
7366 | struct io_sq_data *sqd = ctx->sq_data; | |
7367 | ||
7368 | if (sqd) { | |
05962f95 | 7369 | io_sq_thread_park(sqd); |
521d6a73 | 7370 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7371 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7372 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7373 | |
7374 | io_put_sq_data(sqd); | |
7375 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7376 | } |
7377 | } | |
7378 | ||
aa06165d JA |
7379 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7380 | { | |
7381 | struct io_ring_ctx *ctx_attach; | |
7382 | struct io_sq_data *sqd; | |
7383 | struct fd f; | |
7384 | ||
7385 | f = fdget(p->wq_fd); | |
7386 | if (!f.file) | |
7387 | return ERR_PTR(-ENXIO); | |
7388 | if (f.file->f_op != &io_uring_fops) { | |
7389 | fdput(f); | |
7390 | return ERR_PTR(-EINVAL); | |
7391 | } | |
7392 | ||
7393 | ctx_attach = f.file->private_data; | |
7394 | sqd = ctx_attach->sq_data; | |
7395 | if (!sqd) { | |
7396 | fdput(f); | |
7397 | return ERR_PTR(-EINVAL); | |
7398 | } | |
5c2469e0 JA |
7399 | if (sqd->task_tgid != current->tgid) { |
7400 | fdput(f); | |
7401 | return ERR_PTR(-EPERM); | |
7402 | } | |
aa06165d JA |
7403 | |
7404 | refcount_inc(&sqd->refs); | |
7405 | fdput(f); | |
7406 | return sqd; | |
7407 | } | |
7408 | ||
26984fbf PB |
7409 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7410 | bool *attached) | |
534ca6d6 JA |
7411 | { |
7412 | struct io_sq_data *sqd; | |
7413 | ||
26984fbf | 7414 | *attached = false; |
5c2469e0 JA |
7415 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7416 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7417 | if (!IS_ERR(sqd)) { |
7418 | *attached = true; | |
5c2469e0 | 7419 | return sqd; |
26984fbf | 7420 | } |
5c2469e0 JA |
7421 | /* fall through for EPERM case, setup new sqd/task */ |
7422 | if (PTR_ERR(sqd) != -EPERM) | |
7423 | return sqd; | |
7424 | } | |
aa06165d | 7425 | |
534ca6d6 JA |
7426 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7427 | if (!sqd) | |
7428 | return ERR_PTR(-ENOMEM); | |
7429 | ||
9e138a48 | 7430 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7431 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7432 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7433 | mutex_init(&sqd->lock); |
534ca6d6 | 7434 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7435 | init_completion(&sqd->exited); |
534ca6d6 JA |
7436 | return sqd; |
7437 | } | |
7438 | ||
6b06314c | 7439 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7440 | /* |
7441 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7442 | * the io_uring can be safely unregistered on process exit, even if we have | |
7443 | * loops in the file referencing. | |
7444 | */ | |
7445 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7446 | { | |
7447 | struct sock *sk = ctx->ring_sock->sk; | |
7448 | struct scm_fp_list *fpl; | |
7449 | struct sk_buff *skb; | |
08a45173 | 7450 | int i, nr_files; |
6b06314c | 7451 | |
6b06314c JA |
7452 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7453 | if (!fpl) | |
7454 | return -ENOMEM; | |
7455 | ||
7456 | skb = alloc_skb(0, GFP_KERNEL); | |
7457 | if (!skb) { | |
7458 | kfree(fpl); | |
7459 | return -ENOMEM; | |
7460 | } | |
7461 | ||
7462 | skb->sk = sk; | |
6b06314c | 7463 | |
08a45173 | 7464 | nr_files = 0; |
62e398be | 7465 | fpl->user = get_uid(current_user()); |
6b06314c | 7466 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7467 | struct file *file = io_file_from_index(ctx, i + offset); |
7468 | ||
7469 | if (!file) | |
08a45173 | 7470 | continue; |
65e19f54 | 7471 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7472 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7473 | nr_files++; | |
6b06314c JA |
7474 | } |
7475 | ||
08a45173 JA |
7476 | if (nr_files) { |
7477 | fpl->max = SCM_MAX_FD; | |
7478 | fpl->count = nr_files; | |
7479 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7480 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7481 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7482 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7483 | |
08a45173 JA |
7484 | for (i = 0; i < nr_files; i++) |
7485 | fput(fpl->fp[i]); | |
7486 | } else { | |
7487 | kfree_skb(skb); | |
7488 | kfree(fpl); | |
7489 | } | |
6b06314c JA |
7490 | |
7491 | return 0; | |
7492 | } | |
7493 | ||
7494 | /* | |
7495 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7496 | * causes regular reference counting to break down. We rely on the UNIX | |
7497 | * garbage collection to take care of this problem for us. | |
7498 | */ | |
7499 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7500 | { | |
7501 | unsigned left, total; | |
7502 | int ret = 0; | |
7503 | ||
7504 | total = 0; | |
7505 | left = ctx->nr_user_files; | |
7506 | while (left) { | |
7507 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7508 | |
7509 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7510 | if (ret) | |
7511 | break; | |
7512 | left -= this_files; | |
7513 | total += this_files; | |
7514 | } | |
7515 | ||
7516 | if (!ret) | |
7517 | return 0; | |
7518 | ||
7519 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7520 | struct file *file = io_file_from_index(ctx, total); |
7521 | ||
7522 | if (file) | |
7523 | fput(file); | |
6b06314c JA |
7524 | total++; |
7525 | } | |
7526 | ||
7527 | return ret; | |
7528 | } | |
7529 | #else | |
7530 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7531 | { | |
7532 | return 0; | |
7533 | } | |
7534 | #endif | |
7535 | ||
47e90392 | 7536 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7537 | { |
50238531 | 7538 | struct file *file = prsrc->file; |
05f3fb3c JA |
7539 | #if defined(CONFIG_UNIX) |
7540 | struct sock *sock = ctx->ring_sock->sk; | |
7541 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7542 | struct sk_buff *skb; | |
7543 | int i; | |
7544 | ||
7545 | __skb_queue_head_init(&list); | |
7546 | ||
7547 | /* | |
7548 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7549 | * remove this entry and rearrange the file array. | |
7550 | */ | |
7551 | skb = skb_dequeue(head); | |
7552 | while (skb) { | |
7553 | struct scm_fp_list *fp; | |
7554 | ||
7555 | fp = UNIXCB(skb).fp; | |
7556 | for (i = 0; i < fp->count; i++) { | |
7557 | int left; | |
7558 | ||
7559 | if (fp->fp[i] != file) | |
7560 | continue; | |
7561 | ||
7562 | unix_notinflight(fp->user, fp->fp[i]); | |
7563 | left = fp->count - 1 - i; | |
7564 | if (left) { | |
7565 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7566 | left * sizeof(struct file *)); | |
7567 | } | |
7568 | fp->count--; | |
7569 | if (!fp->count) { | |
7570 | kfree_skb(skb); | |
7571 | skb = NULL; | |
7572 | } else { | |
7573 | __skb_queue_tail(&list, skb); | |
7574 | } | |
7575 | fput(file); | |
7576 | file = NULL; | |
7577 | break; | |
7578 | } | |
7579 | ||
7580 | if (!file) | |
7581 | break; | |
7582 | ||
7583 | __skb_queue_tail(&list, skb); | |
7584 | ||
7585 | skb = skb_dequeue(head); | |
7586 | } | |
7587 | ||
7588 | if (skb_peek(&list)) { | |
7589 | spin_lock_irq(&head->lock); | |
7590 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7591 | __skb_queue_tail(head, skb); | |
7592 | spin_unlock_irq(&head->lock); | |
7593 | } | |
7594 | #else | |
7595 | fput(file); | |
7596 | #endif | |
7597 | } | |
7598 | ||
b895c9a6 | 7599 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7600 | { |
b895c9a6 | 7601 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7602 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7603 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7604 | |
269bbe5f BM |
7605 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7606 | list_del(&prsrc->list); | |
b60c8dce PB |
7607 | |
7608 | if (prsrc->tag) { | |
7609 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
7610 | |
7611 | io_ring_submit_lock(ctx, lock_ring); | |
157d257f | 7612 | spin_lock_irq(&ctx->completion_lock); |
b60c8dce | 7613 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 7614 | ctx->cq_extra++; |
b60c8dce | 7615 | io_commit_cqring(ctx); |
157d257f | 7616 | spin_unlock_irq(&ctx->completion_lock); |
b60c8dce PB |
7617 | io_cqring_ev_posted(ctx); |
7618 | io_ring_submit_unlock(ctx, lock_ring); | |
7619 | } | |
7620 | ||
40ae0ff7 | 7621 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7622 | kfree(prsrc); |
65e19f54 | 7623 | } |
05589553 | 7624 | |
28a9fe25 | 7625 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7626 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7627 | complete(&rsrc_data->done); | |
2faf852d | 7628 | } |
65e19f54 | 7629 | |
269bbe5f | 7630 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7631 | { |
7632 | struct io_ring_ctx *ctx; | |
7633 | struct llist_node *node; | |
7634 | ||
269bbe5f BM |
7635 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7636 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7637 | |
7638 | while (node) { | |
b895c9a6 | 7639 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7640 | struct llist_node *next = node->next; |
7641 | ||
b895c9a6 | 7642 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7643 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7644 | node = next; |
7645 | } | |
7646 | } | |
7647 | ||
00835dce | 7648 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7649 | { |
b895c9a6 | 7650 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7651 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7652 | bool first_add = false; |
e297822b | 7653 | |
2a63b2d9 | 7654 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7655 | node->done = true; |
e297822b | 7656 | |
d67d2263 | 7657 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7658 | node = list_first_entry(&ctx->rsrc_ref_list, |
7659 | struct io_rsrc_node, node); | |
e297822b | 7660 | /* recycle ref nodes in order */ |
b895c9a6 | 7661 | if (!node->done) |
e297822b | 7662 | break; |
b895c9a6 PB |
7663 | list_del(&node->node); |
7664 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7665 | } |
2a63b2d9 | 7666 | io_rsrc_ref_unlock(ctx); |
05589553 | 7667 | |
3e942498 PB |
7668 | if (first_add) |
7669 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7670 | } |
65e19f54 | 7671 | |
b895c9a6 | 7672 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7673 | { |
b895c9a6 | 7674 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7675 | |
05589553 XW |
7676 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7677 | if (!ref_node) | |
3e2224c5 | 7678 | return NULL; |
05f3fb3c | 7679 | |
00835dce | 7680 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7681 | 0, GFP_KERNEL)) { |
7682 | kfree(ref_node); | |
3e2224c5 | 7683 | return NULL; |
05589553 XW |
7684 | } |
7685 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7686 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7687 | ref_node->done = false; |
05589553 | 7688 | return ref_node; |
05589553 XW |
7689 | } |
7690 | ||
6b06314c | 7691 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 7692 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
7693 | { |
7694 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7695 | struct file *file; |
f3baed39 | 7696 | int fd, ret; |
846a4ef2 | 7697 | unsigned i; |
6b06314c | 7698 | |
05f3fb3c | 7699 | if (ctx->file_data) |
6b06314c JA |
7700 | return -EBUSY; |
7701 | if (!nr_args) | |
7702 | return -EINVAL; | |
7703 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7704 | return -EMFILE; | |
a7f0ed5a | 7705 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7706 | if (ret) |
7707 | return ret; | |
d878c816 PB |
7708 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
7709 | &ctx->file_data); | |
7710 | if (ret) | |
7711 | return ret; | |
6b06314c | 7712 | |
f3baed39 | 7713 | ret = -ENOMEM; |
aeca241b | 7714 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7715 | goto out_free; |
65e19f54 | 7716 | |
08a45173 | 7717 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 7718 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
7719 | ret = -EFAULT; |
7720 | goto out_fput; | |
7721 | } | |
08a45173 | 7722 | /* allow sparse sets */ |
792e3582 PB |
7723 | if (fd == -1) { |
7724 | ret = -EINVAL; | |
2d091d62 | 7725 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 7726 | goto out_fput; |
08a45173 | 7727 | continue; |
792e3582 | 7728 | } |
6b06314c | 7729 | |
05f3fb3c | 7730 | file = fget(fd); |
6b06314c | 7731 | ret = -EBADF; |
792e3582 | 7732 | if (unlikely(!file)) |
600cf3f8 | 7733 | goto out_fput; |
05f3fb3c | 7734 | |
6b06314c JA |
7735 | /* |
7736 | * Don't allow io_uring instances to be registered. If UNIX | |
7737 | * isn't enabled, then this causes a reference cycle and this | |
7738 | * instance can never get freed. If UNIX is enabled we'll | |
7739 | * handle it just fine, but there's still no point in allowing | |
7740 | * a ring fd as it doesn't support regular read/write anyway. | |
7741 | */ | |
05f3fb3c JA |
7742 | if (file->f_op == &io_uring_fops) { |
7743 | fput(file); | |
600cf3f8 | 7744 | goto out_fput; |
6b06314c | 7745 | } |
aeca241b | 7746 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7747 | } |
7748 | ||
6b06314c | 7749 | ret = io_sqe_files_scm(ctx); |
05589553 | 7750 | if (ret) { |
08480400 | 7751 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7752 | return ret; |
7753 | } | |
6b06314c | 7754 | |
a7f0ed5a | 7755 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7756 | return ret; |
600cf3f8 PB |
7757 | out_fput: |
7758 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7759 | file = io_file_from_index(ctx, i); | |
7760 | if (file) | |
7761 | fput(file); | |
7762 | } | |
aeca241b | 7763 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7764 | ctx->nr_user_files = 0; |
600cf3f8 | 7765 | out_free: |
44b31f2f | 7766 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7767 | ctx->file_data = NULL; |
6b06314c JA |
7768 | return ret; |
7769 | } | |
7770 | ||
c3a31e60 JA |
7771 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7772 | int index) | |
7773 | { | |
7774 | #if defined(CONFIG_UNIX) | |
7775 | struct sock *sock = ctx->ring_sock->sk; | |
7776 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7777 | struct sk_buff *skb; | |
7778 | ||
7779 | /* | |
7780 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7781 | * file set. If there's no room, fall back to allocating a new skb | |
7782 | * and filling it in. | |
7783 | */ | |
7784 | spin_lock_irq(&head->lock); | |
7785 | skb = skb_peek(head); | |
7786 | if (skb) { | |
7787 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7788 | ||
7789 | if (fpl->count < SCM_MAX_FD) { | |
7790 | __skb_unlink(skb, head); | |
7791 | spin_unlock_irq(&head->lock); | |
7792 | fpl->fp[fpl->count] = get_file(file); | |
7793 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7794 | fpl->count++; | |
7795 | spin_lock_irq(&head->lock); | |
7796 | __skb_queue_head(head, skb); | |
7797 | } else { | |
7798 | skb = NULL; | |
7799 | } | |
7800 | } | |
7801 | spin_unlock_irq(&head->lock); | |
7802 | ||
7803 | if (skb) { | |
7804 | fput(file); | |
7805 | return 0; | |
7806 | } | |
7807 | ||
7808 | return __io_sqe_files_scm(ctx, 1, index); | |
7809 | #else | |
7810 | return 0; | |
7811 | #endif | |
7812 | } | |
7813 | ||
b60c8dce | 7814 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 7815 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 7816 | { |
269bbe5f | 7817 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7818 | |
269bbe5f BM |
7819 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7820 | if (!prsrc) | |
a5318d3c | 7821 | return -ENOMEM; |
05f3fb3c | 7822 | |
2d091d62 | 7823 | prsrc->tag = *io_get_tag_slot(data, idx); |
50238531 | 7824 | prsrc->rsrc = rsrc; |
e7c78371 | 7825 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7826 | return 0; |
05f3fb3c JA |
7827 | } |
7828 | ||
7829 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 7830 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
7831 | unsigned nr_args) |
7832 | { | |
c3bdad02 | 7833 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 7834 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 7835 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7836 | struct io_fixed_file *file_slot; |
7837 | struct file *file; | |
98f0b3b4 PB |
7838 | int fd, i, err = 0; |
7839 | unsigned int done; | |
05589553 | 7840 | bool needs_switch = false; |
c3a31e60 | 7841 | |
98f0b3b4 PB |
7842 | if (!ctx->file_data) |
7843 | return -ENXIO; | |
7844 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
7845 | return -EINVAL; |
7846 | ||
67973b93 | 7847 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
7848 | u64 tag = 0; |
7849 | ||
7850 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
7851 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
7852 | err = -EFAULT; |
7853 | break; | |
7854 | } | |
c3bdad02 PB |
7855 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
7856 | err = -EINVAL; | |
7857 | break; | |
7858 | } | |
4e0377a1 | 7859 | if (fd == IORING_REGISTER_FILES_SKIP) |
7860 | continue; | |
7861 | ||
67973b93 | 7862 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7863 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7864 | |
a04b0ac0 PB |
7865 | if (file_slot->file_ptr) { |
7866 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
7867 | err = io_queue_rsrc_removal(data, up->offset + done, |
7868 | ctx->rsrc_node, file); | |
a5318d3c HD |
7869 | if (err) |
7870 | break; | |
a04b0ac0 | 7871 | file_slot->file_ptr = 0; |
05589553 | 7872 | needs_switch = true; |
c3a31e60 JA |
7873 | } |
7874 | if (fd != -1) { | |
c3a31e60 JA |
7875 | file = fget(fd); |
7876 | if (!file) { | |
7877 | err = -EBADF; | |
7878 | break; | |
7879 | } | |
7880 | /* | |
7881 | * Don't allow io_uring instances to be registered. If | |
7882 | * UNIX isn't enabled, then this causes a reference | |
7883 | * cycle and this instance can never get freed. If UNIX | |
7884 | * is enabled we'll handle it just fine, but there's | |
7885 | * still no point in allowing a ring fd as it doesn't | |
7886 | * support regular read/write anyway. | |
7887 | */ | |
7888 | if (file->f_op == &io_uring_fops) { | |
7889 | fput(file); | |
7890 | err = -EBADF; | |
7891 | break; | |
7892 | } | |
2d091d62 | 7893 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 7894 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7895 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7896 | if (err) { |
a04b0ac0 | 7897 | file_slot->file_ptr = 0; |
f3bd9dae | 7898 | fput(file); |
c3a31e60 | 7899 | break; |
f3bd9dae | 7900 | } |
c3a31e60 | 7901 | } |
05f3fb3c JA |
7902 | } |
7903 | ||
a7f0ed5a PB |
7904 | if (needs_switch) |
7905 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7906 | return done ? done : err; |
7907 | } | |
05589553 | 7908 | |
5280f7e5 | 7909 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7910 | { |
7911 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7912 | ||
5280f7e5 PB |
7913 | req = io_put_req_find_next(req); |
7914 | return req ? &req->work : NULL; | |
7d723065 JA |
7915 | } |
7916 | ||
685fe7fe JA |
7917 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7918 | struct task_struct *task) | |
24369c2e | 7919 | { |
e941894e | 7920 | struct io_wq_hash *hash; |
24369c2e | 7921 | struct io_wq_data data; |
24369c2e | 7922 | unsigned int concurrency; |
24369c2e | 7923 | |
e941894e JA |
7924 | hash = ctx->hash_map; |
7925 | if (!hash) { | |
7926 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7927 | if (!hash) | |
7928 | return ERR_PTR(-ENOMEM); | |
7929 | refcount_set(&hash->refs, 1); | |
7930 | init_waitqueue_head(&hash->wait); | |
7931 | ctx->hash_map = hash; | |
24369c2e PB |
7932 | } |
7933 | ||
e941894e | 7934 | data.hash = hash; |
685fe7fe | 7935 | data.task = task; |
e9fd9396 | 7936 | data.free_work = io_free_work; |
f5fa38c5 | 7937 | data.do_work = io_wq_submit_work; |
24369c2e | 7938 | |
d25e3a3d JA |
7939 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7940 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7941 | |
5aa75ed5 | 7942 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7943 | } |
7944 | ||
5aa75ed5 JA |
7945 | static int io_uring_alloc_task_context(struct task_struct *task, |
7946 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7947 | { |
7948 | struct io_uring_task *tctx; | |
d8a6df10 | 7949 | int ret; |
0f212204 | 7950 | |
09899b19 | 7951 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
7952 | if (unlikely(!tctx)) |
7953 | return -ENOMEM; | |
7954 | ||
d8a6df10 JA |
7955 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7956 | if (unlikely(ret)) { | |
7957 | kfree(tctx); | |
7958 | return ret; | |
7959 | } | |
7960 | ||
685fe7fe | 7961 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7962 | if (IS_ERR(tctx->io_wq)) { |
7963 | ret = PTR_ERR(tctx->io_wq); | |
7964 | percpu_counter_destroy(&tctx->inflight); | |
7965 | kfree(tctx); | |
7966 | return ret; | |
7967 | } | |
7968 | ||
0f212204 JA |
7969 | xa_init(&tctx->xa); |
7970 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 7971 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7972 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7973 | task->io_uring = tctx; |
7cbf1722 JA |
7974 | spin_lock_init(&tctx->task_lock); |
7975 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 7976 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
7977 | return 0; |
7978 | } | |
7979 | ||
7980 | void __io_uring_free(struct task_struct *tsk) | |
7981 | { | |
7982 | struct io_uring_task *tctx = tsk->io_uring; | |
7983 | ||
7984 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 7985 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 7986 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 7987 | |
d8a6df10 | 7988 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7989 | kfree(tctx); |
7990 | tsk->io_uring = NULL; | |
7991 | } | |
7992 | ||
7e84e1c7 SG |
7993 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7994 | struct io_uring_params *p) | |
2b188cc1 JA |
7995 | { |
7996 | int ret; | |
7997 | ||
d25e3a3d JA |
7998 | /* Retain compatibility with failing for an invalid attach attempt */ |
7999 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
8000 | IORING_SETUP_ATTACH_WQ) { | |
8001 | struct fd f; | |
8002 | ||
8003 | f = fdget(p->wq_fd); | |
8004 | if (!f.file) | |
8005 | return -ENXIO; | |
d25e3a3d | 8006 | fdput(f); |
f2a48dd0 PB |
8007 | if (f.file->f_op != &io_uring_fops) |
8008 | return -EINVAL; | |
d25e3a3d | 8009 | } |
6c271ce2 | 8010 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 8011 | struct task_struct *tsk; |
534ca6d6 | 8012 | struct io_sq_data *sqd; |
26984fbf | 8013 | bool attached; |
534ca6d6 | 8014 | |
26984fbf | 8015 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8016 | if (IS_ERR(sqd)) { |
8017 | ret = PTR_ERR(sqd); | |
8018 | goto err; | |
8019 | } | |
69fb2131 | 8020 | |
7c30f36a | 8021 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8022 | ctx->sq_data = sqd; |
917257da JA |
8023 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8024 | if (!ctx->sq_thread_idle) | |
8025 | ctx->sq_thread_idle = HZ; | |
8026 | ||
78d7f6ba | 8027 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8028 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8029 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8030 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8031 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8032 | io_sq_thread_unpark(sqd); |
8033 | ||
de75a3d3 PB |
8034 | if (ret < 0) |
8035 | goto err; | |
8036 | if (attached) | |
5aa75ed5 | 8037 | return 0; |
aa06165d | 8038 | |
6c271ce2 | 8039 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8040 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8041 | |
917257da | 8042 | ret = -EINVAL; |
f2a48dd0 | 8043 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8044 | goto err_sqpoll; |
37d1e2e3 | 8045 | sqd->sq_cpu = cpu; |
6c271ce2 | 8046 | } else { |
37d1e2e3 | 8047 | sqd->sq_cpu = -1; |
6c271ce2 | 8048 | } |
37d1e2e3 JA |
8049 | |
8050 | sqd->task_pid = current->pid; | |
5c2469e0 | 8051 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8052 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8053 | if (IS_ERR(tsk)) { | |
8054 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8055 | goto err_sqpoll; |
6c271ce2 | 8056 | } |
97a73a0f | 8057 | |
46fe18b1 | 8058 | sqd->thread = tsk; |
97a73a0f | 8059 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8060 | wake_up_new_task(tsk); |
0f212204 JA |
8061 | if (ret) |
8062 | goto err; | |
6c271ce2 JA |
8063 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8064 | /* Can't have SQ_AFF without SQPOLL */ | |
8065 | ret = -EINVAL; | |
8066 | goto err; | |
8067 | } | |
8068 | ||
2b188cc1 | 8069 | return 0; |
f2a48dd0 PB |
8070 | err_sqpoll: |
8071 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8072 | err: |
37d1e2e3 | 8073 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8074 | return ret; |
8075 | } | |
8076 | ||
a087e2b5 BM |
8077 | static inline void __io_unaccount_mem(struct user_struct *user, |
8078 | unsigned long nr_pages) | |
2b188cc1 JA |
8079 | { |
8080 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8081 | } | |
8082 | ||
a087e2b5 BM |
8083 | static inline int __io_account_mem(struct user_struct *user, |
8084 | unsigned long nr_pages) | |
2b188cc1 JA |
8085 | { |
8086 | unsigned long page_limit, cur_pages, new_pages; | |
8087 | ||
8088 | /* Don't allow more pages than we can safely lock */ | |
8089 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8090 | ||
8091 | do { | |
8092 | cur_pages = atomic_long_read(&user->locked_vm); | |
8093 | new_pages = cur_pages + nr_pages; | |
8094 | if (new_pages > page_limit) | |
8095 | return -ENOMEM; | |
8096 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8097 | new_pages) != cur_pages); | |
8098 | ||
8099 | return 0; | |
8100 | } | |
8101 | ||
26bfa89e | 8102 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8103 | { |
62e398be | 8104 | if (ctx->user) |
a087e2b5 | 8105 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8106 | |
26bfa89e JA |
8107 | if (ctx->mm_account) |
8108 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8109 | } |
8110 | ||
26bfa89e | 8111 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8112 | { |
30975825 BM |
8113 | int ret; |
8114 | ||
62e398be | 8115 | if (ctx->user) { |
30975825 BM |
8116 | ret = __io_account_mem(ctx->user, nr_pages); |
8117 | if (ret) | |
8118 | return ret; | |
8119 | } | |
8120 | ||
26bfa89e JA |
8121 | if (ctx->mm_account) |
8122 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8123 | |
8124 | return 0; | |
8125 | } | |
8126 | ||
2b188cc1 JA |
8127 | static void io_mem_free(void *ptr) |
8128 | { | |
52e04ef4 MR |
8129 | struct page *page; |
8130 | ||
8131 | if (!ptr) | |
8132 | return; | |
2b188cc1 | 8133 | |
52e04ef4 | 8134 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8135 | if (put_page_testzero(page)) |
8136 | free_compound_page(page); | |
8137 | } | |
8138 | ||
8139 | static void *io_mem_alloc(size_t size) | |
8140 | { | |
8141 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8142 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8143 | |
8144 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8145 | } | |
8146 | ||
75b28aff HV |
8147 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8148 | size_t *sq_offset) | |
8149 | { | |
8150 | struct io_rings *rings; | |
8151 | size_t off, sq_array_size; | |
8152 | ||
8153 | off = struct_size(rings, cqes, cq_entries); | |
8154 | if (off == SIZE_MAX) | |
8155 | return SIZE_MAX; | |
8156 | ||
8157 | #ifdef CONFIG_SMP | |
8158 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8159 | if (off == 0) | |
8160 | return SIZE_MAX; | |
8161 | #endif | |
8162 | ||
b36200f5 DV |
8163 | if (sq_offset) |
8164 | *sq_offset = off; | |
8165 | ||
75b28aff HV |
8166 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8167 | if (sq_array_size == SIZE_MAX) | |
8168 | return SIZE_MAX; | |
8169 | ||
8170 | if (check_add_overflow(off, sq_array_size, &off)) | |
8171 | return SIZE_MAX; | |
8172 | ||
75b28aff HV |
8173 | return off; |
8174 | } | |
8175 | ||
41edf1a5 | 8176 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8177 | { |
41edf1a5 | 8178 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8179 | unsigned int i; |
8180 | ||
6224843d PB |
8181 | if (imu != ctx->dummy_ubuf) { |
8182 | for (i = 0; i < imu->nr_bvecs; i++) | |
8183 | unpin_user_page(imu->bvec[i].bv_page); | |
8184 | if (imu->acct_pages) | |
8185 | io_unaccount_mem(ctx, imu->acct_pages); | |
8186 | kvfree(imu); | |
8187 | } | |
41edf1a5 | 8188 | *slot = NULL; |
7f61a1e9 PB |
8189 | } |
8190 | ||
bd54b6fe | 8191 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8192 | { |
634d00df PB |
8193 | io_buffer_unmap(ctx, &prsrc->buf); |
8194 | prsrc->buf = NULL; | |
bd54b6fe | 8195 | } |
edafccee | 8196 | |
bd54b6fe BM |
8197 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8198 | { | |
8199 | unsigned int i; | |
edafccee | 8200 | |
7f61a1e9 PB |
8201 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8202 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8203 | kfree(ctx->user_bufs); |
bb6659cc | 8204 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8205 | ctx->user_bufs = NULL; |
bd54b6fe | 8206 | ctx->buf_data = NULL; |
edafccee | 8207 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8208 | } |
8209 | ||
0a96bbe4 | 8210 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8211 | { |
bd54b6fe | 8212 | int ret; |
edafccee | 8213 | |
bd54b6fe | 8214 | if (!ctx->buf_data) |
edafccee JA |
8215 | return -ENXIO; |
8216 | ||
bd54b6fe BM |
8217 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8218 | if (!ret) | |
8219 | __io_sqe_buffers_unregister(ctx); | |
8220 | return ret; | |
edafccee JA |
8221 | } |
8222 | ||
8223 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8224 | void __user *arg, unsigned index) | |
8225 | { | |
8226 | struct iovec __user *src; | |
8227 | ||
8228 | #ifdef CONFIG_COMPAT | |
8229 | if (ctx->compat) { | |
8230 | struct compat_iovec __user *ciovs; | |
8231 | struct compat_iovec ciov; | |
8232 | ||
8233 | ciovs = (struct compat_iovec __user *) arg; | |
8234 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8235 | return -EFAULT; | |
8236 | ||
d55e5f5b | 8237 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8238 | dst->iov_len = ciov.iov_len; |
8239 | return 0; | |
8240 | } | |
8241 | #endif | |
8242 | src = (struct iovec __user *) arg; | |
8243 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8244 | return -EFAULT; | |
8245 | return 0; | |
8246 | } | |
8247 | ||
de293938 JA |
8248 | /* |
8249 | * Not super efficient, but this is just a registration time. And we do cache | |
8250 | * the last compound head, so generally we'll only do a full search if we don't | |
8251 | * match that one. | |
8252 | * | |
8253 | * We check if the given compound head page has already been accounted, to | |
8254 | * avoid double accounting it. This allows us to account the full size of the | |
8255 | * page, not just the constituent pages of a huge page. | |
8256 | */ | |
8257 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8258 | int nr_pages, struct page *hpage) | |
8259 | { | |
8260 | int i, j; | |
8261 | ||
8262 | /* check current page array */ | |
8263 | for (i = 0; i < nr_pages; i++) { | |
8264 | if (!PageCompound(pages[i])) | |
8265 | continue; | |
8266 | if (compound_head(pages[i]) == hpage) | |
8267 | return true; | |
8268 | } | |
8269 | ||
8270 | /* check previously registered pages */ | |
8271 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8272 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8273 | |
8274 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8275 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8276 | continue; | |
8277 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8278 | return true; | |
8279 | } | |
8280 | } | |
8281 | ||
8282 | return false; | |
8283 | } | |
8284 | ||
8285 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8286 | int nr_pages, struct io_mapped_ubuf *imu, | |
8287 | struct page **last_hpage) | |
8288 | { | |
8289 | int i, ret; | |
8290 | ||
216e5835 | 8291 | imu->acct_pages = 0; |
de293938 JA |
8292 | for (i = 0; i < nr_pages; i++) { |
8293 | if (!PageCompound(pages[i])) { | |
8294 | imu->acct_pages++; | |
8295 | } else { | |
8296 | struct page *hpage; | |
8297 | ||
8298 | hpage = compound_head(pages[i]); | |
8299 | if (hpage == *last_hpage) | |
8300 | continue; | |
8301 | *last_hpage = hpage; | |
8302 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8303 | continue; | |
8304 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8305 | } | |
8306 | } | |
8307 | ||
8308 | if (!imu->acct_pages) | |
8309 | return 0; | |
8310 | ||
26bfa89e | 8311 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8312 | if (ret) |
8313 | imu->acct_pages = 0; | |
8314 | return ret; | |
8315 | } | |
8316 | ||
0a96bbe4 | 8317 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8318 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8319 | struct page **last_hpage) |
edafccee | 8320 | { |
41edf1a5 | 8321 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8322 | struct vm_area_struct **vmas = NULL; |
8323 | struct page **pages = NULL; | |
0a96bbe4 BM |
8324 | unsigned long off, start, end, ubuf; |
8325 | size_t size; | |
8326 | int ret, pret, nr_pages, i; | |
8327 | ||
6224843d PB |
8328 | if (!iov->iov_base) { |
8329 | *pimu = ctx->dummy_ubuf; | |
8330 | return 0; | |
8331 | } | |
8332 | ||
0a96bbe4 BM |
8333 | ubuf = (unsigned long) iov->iov_base; |
8334 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8335 | start = ubuf >> PAGE_SHIFT; | |
8336 | nr_pages = end - start; | |
8337 | ||
41edf1a5 | 8338 | *pimu = NULL; |
0a96bbe4 BM |
8339 | ret = -ENOMEM; |
8340 | ||
8341 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8342 | if (!pages) | |
8343 | goto done; | |
8344 | ||
8345 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8346 | GFP_KERNEL); | |
8347 | if (!vmas) | |
8348 | goto done; | |
edafccee | 8349 | |
41edf1a5 | 8350 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8351 | if (!imu) |
0a96bbe4 BM |
8352 | goto done; |
8353 | ||
8354 | ret = 0; | |
8355 | mmap_read_lock(current->mm); | |
8356 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8357 | pages, vmas); | |
8358 | if (pret == nr_pages) { | |
8359 | /* don't support file backed memory */ | |
8360 | for (i = 0; i < nr_pages; i++) { | |
8361 | struct vm_area_struct *vma = vmas[i]; | |
8362 | ||
40dad765 PB |
8363 | if (vma_is_shmem(vma)) |
8364 | continue; | |
0a96bbe4 BM |
8365 | if (vma->vm_file && |
8366 | !is_file_hugepages(vma->vm_file)) { | |
8367 | ret = -EOPNOTSUPP; | |
8368 | break; | |
8369 | } | |
8370 | } | |
8371 | } else { | |
8372 | ret = pret < 0 ? pret : -EFAULT; | |
8373 | } | |
8374 | mmap_read_unlock(current->mm); | |
8375 | if (ret) { | |
8376 | /* | |
8377 | * if we did partial map, or found file backed vmas, | |
8378 | * release any pages we did get | |
8379 | */ | |
8380 | if (pret > 0) | |
8381 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8382 | goto done; |
8383 | } | |
8384 | ||
8385 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8386 | if (ret) { | |
8387 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8388 | goto done; |
8389 | } | |
8390 | ||
8391 | off = ubuf & ~PAGE_MASK; | |
8392 | size = iov->iov_len; | |
8393 | for (i = 0; i < nr_pages; i++) { | |
8394 | size_t vec_len; | |
8395 | ||
8396 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8397 | imu->bvec[i].bv_page = pages[i]; | |
8398 | imu->bvec[i].bv_len = vec_len; | |
8399 | imu->bvec[i].bv_offset = off; | |
8400 | off = 0; | |
8401 | size -= vec_len; | |
8402 | } | |
8403 | /* store original address for later verification */ | |
8404 | imu->ubuf = ubuf; | |
4751f53d | 8405 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8406 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8407 | *pimu = imu; |
0a96bbe4 BM |
8408 | ret = 0; |
8409 | done: | |
41edf1a5 PB |
8410 | if (ret) |
8411 | kvfree(imu); | |
0a96bbe4 BM |
8412 | kvfree(pages); |
8413 | kvfree(vmas); | |
8414 | return ret; | |
8415 | } | |
8416 | ||
2b358604 | 8417 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8418 | { |
87094465 PB |
8419 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8420 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8421 | } |
edafccee | 8422 | |
2b358604 BM |
8423 | static int io_buffer_validate(struct iovec *iov) |
8424 | { | |
50e96989 PB |
8425 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8426 | ||
2b358604 BM |
8427 | /* |
8428 | * Don't impose further limits on the size and buffer | |
8429 | * constraints here, we'll -EINVAL later when IO is | |
8430 | * submitted if they are wrong. | |
8431 | */ | |
6224843d PB |
8432 | if (!iov->iov_base) |
8433 | return iov->iov_len ? -EFAULT : 0; | |
8434 | if (!iov->iov_len) | |
2b358604 | 8435 | return -EFAULT; |
edafccee | 8436 | |
2b358604 BM |
8437 | /* arbitrary limit, but we need something */ |
8438 | if (iov->iov_len > SZ_1G) | |
8439 | return -EFAULT; | |
edafccee | 8440 | |
50e96989 PB |
8441 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8442 | return -EOVERFLOW; | |
8443 | ||
2b358604 BM |
8444 | return 0; |
8445 | } | |
edafccee | 8446 | |
2b358604 | 8447 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 8448 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 8449 | { |
bd54b6fe BM |
8450 | struct page *last_hpage = NULL; |
8451 | struct io_rsrc_data *data; | |
2b358604 BM |
8452 | int i, ret; |
8453 | struct iovec iov; | |
edafccee | 8454 | |
87094465 PB |
8455 | if (ctx->user_bufs) |
8456 | return -EBUSY; | |
489809e2 | 8457 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 8458 | return -EINVAL; |
bd54b6fe | 8459 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8460 | if (ret) |
8461 | return ret; | |
d878c816 PB |
8462 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
8463 | if (ret) | |
8464 | return ret; | |
bd54b6fe BM |
8465 | ret = io_buffers_map_alloc(ctx, nr_args); |
8466 | if (ret) { | |
bb6659cc | 8467 | io_rsrc_data_free(data); |
bd54b6fe BM |
8468 | return ret; |
8469 | } | |
edafccee | 8470 | |
87094465 | 8471 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
8472 | ret = io_copy_iov(ctx, &iov, arg, i); |
8473 | if (ret) | |
0a96bbe4 | 8474 | break; |
2b358604 BM |
8475 | ret = io_buffer_validate(&iov); |
8476 | if (ret) | |
0a96bbe4 | 8477 | break; |
2d091d62 | 8478 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
8479 | ret = -EINVAL; |
8480 | break; | |
8481 | } | |
edafccee | 8482 | |
41edf1a5 PB |
8483 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
8484 | &last_hpage); | |
0a96bbe4 BM |
8485 | if (ret) |
8486 | break; | |
edafccee | 8487 | } |
0a96bbe4 | 8488 | |
bd54b6fe | 8489 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8490 | |
bd54b6fe BM |
8491 | ctx->buf_data = data; |
8492 | if (ret) | |
8493 | __io_sqe_buffers_unregister(ctx); | |
8494 | else | |
8495 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
8496 | return ret; |
8497 | } | |
8498 | ||
634d00df PB |
8499 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
8500 | struct io_uring_rsrc_update2 *up, | |
8501 | unsigned int nr_args) | |
8502 | { | |
8503 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
8504 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
8505 | struct page *last_hpage = NULL; |
8506 | bool needs_switch = false; | |
8507 | __u32 done; | |
8508 | int i, err; | |
8509 | ||
8510 | if (!ctx->buf_data) | |
8511 | return -ENXIO; | |
8512 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
8513 | return -EINVAL; | |
8514 | ||
8515 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
8516 | struct io_mapped_ubuf *imu; |
8517 | int offset = up->offset + done; | |
634d00df PB |
8518 | u64 tag = 0; |
8519 | ||
8520 | err = io_copy_iov(ctx, &iov, iovs, done); | |
8521 | if (err) | |
8522 | break; | |
8523 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
8524 | err = -EFAULT; | |
8525 | break; | |
8526 | } | |
0b8c0e7c PB |
8527 | err = io_buffer_validate(&iov); |
8528 | if (err) | |
8529 | break; | |
cf3770e7 CIK |
8530 | if (!iov.iov_base && tag) { |
8531 | err = -EINVAL; | |
8532 | break; | |
8533 | } | |
0b8c0e7c PB |
8534 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
8535 | if (err) | |
8536 | break; | |
634d00df | 8537 | |
0b8c0e7c | 8538 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 8539 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
8540 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
8541 | ctx->rsrc_node, ctx->user_bufs[i]); | |
8542 | if (unlikely(err)) { | |
8543 | io_buffer_unmap(ctx, &imu); | |
634d00df | 8544 | break; |
0b8c0e7c | 8545 | } |
634d00df PB |
8546 | ctx->user_bufs[i] = NULL; |
8547 | needs_switch = true; | |
8548 | } | |
8549 | ||
0b8c0e7c | 8550 | ctx->user_bufs[i] = imu; |
2d091d62 | 8551 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
8552 | } |
8553 | ||
8554 | if (needs_switch) | |
8555 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
8556 | return done ? done : err; | |
8557 | } | |
8558 | ||
9b402849 JA |
8559 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8560 | { | |
8561 | __s32 __user *fds = arg; | |
8562 | int fd; | |
8563 | ||
8564 | if (ctx->cq_ev_fd) | |
8565 | return -EBUSY; | |
8566 | ||
8567 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8568 | return -EFAULT; | |
8569 | ||
8570 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8571 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8572 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
fe7e3257 | 8573 | |
9b402849 JA |
8574 | ctx->cq_ev_fd = NULL; |
8575 | return ret; | |
8576 | } | |
8577 | ||
8578 | return 0; | |
8579 | } | |
8580 | ||
8581 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8582 | { | |
8583 | if (ctx->cq_ev_fd) { | |
8584 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8585 | ctx->cq_ev_fd = NULL; | |
8586 | return 0; | |
8587 | } | |
8588 | ||
8589 | return -ENXIO; | |
8590 | } | |
8591 | ||
5a2e745d JA |
8592 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8593 | { | |
9e15c3a0 JA |
8594 | struct io_buffer *buf; |
8595 | unsigned long index; | |
8596 | ||
8597 | xa_for_each(&ctx->io_buffers, index, buf) | |
8598 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8599 | } |
8600 | ||
68e68ee6 | 8601 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8602 | { |
68e68ee6 | 8603 | struct io_kiocb *req, *nxt; |
1b4c351f | 8604 | |
68e68ee6 JA |
8605 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8606 | if (tsk && req->task != tsk) | |
8607 | continue; | |
1b4c351f JA |
8608 | list_del(&req->compl.list); |
8609 | kmem_cache_free(req_cachep, req); | |
8610 | } | |
8611 | } | |
8612 | ||
4010fec4 | 8613 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8614 | { |
bf019da7 | 8615 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8616 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8617 | |
9a4fdbd8 JA |
8618 | mutex_lock(&ctx->uring_lock); |
8619 | ||
8e5c66c4 | 8620 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8621 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8622 | submit_state->reqs); | |
8e5c66c4 PB |
8623 | submit_state->free_reqs = 0; |
8624 | } | |
9a4fdbd8 | 8625 | |
dac7a098 | 8626 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8627 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8628 | mutex_unlock(&ctx->uring_lock); |
8629 | } | |
8630 | ||
bd54b6fe | 8631 | static bool io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 8632 | { |
bd54b6fe BM |
8633 | if (!data) |
8634 | return false; | |
8635 | if (!atomic_dec_and_test(&data->refs)) | |
8636 | wait_for_completion(&data->done); | |
8637 | return true; | |
8638 | } | |
04fc6c80 | 8639 | |
2b188cc1 JA |
8640 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8641 | { | |
37d1e2e3 | 8642 | io_sq_thread_finish(ctx); |
2aede0e4 | 8643 | |
37d1e2e3 | 8644 | if (ctx->mm_account) { |
2aede0e4 JA |
8645 | mmdrop(ctx->mm_account); |
8646 | ctx->mm_account = NULL; | |
30975825 | 8647 | } |
def596e9 | 8648 | |
8bad28d8 | 8649 | mutex_lock(&ctx->uring_lock); |
bd54b6fe BM |
8650 | if (io_wait_rsrc_data(ctx->buf_data)) |
8651 | __io_sqe_buffers_unregister(ctx); | |
8652 | if (io_wait_rsrc_data(ctx->file_data)) | |
08480400 | 8653 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
8654 | if (ctx->rings) |
8655 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8656 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8657 | io_eventfd_unregister(ctx); |
5a2e745d | 8658 | io_destroy_buffers(ctx); |
07db298a PB |
8659 | if (ctx->sq_creds) |
8660 | put_cred(ctx->sq_creds); | |
def596e9 | 8661 | |
a7f0ed5a PB |
8662 | /* there are no registered resources left, nobody uses it */ |
8663 | if (ctx->rsrc_node) | |
8664 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8665 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8666 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8667 | flush_delayed_work(&ctx->rsrc_put_work); |
8668 | ||
8669 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8670 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 8671 | |
2b188cc1 | 8672 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8673 | if (ctx->ring_sock) { |
8674 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8675 | sock_release(ctx->ring_sock); |
355e8d26 | 8676 | } |
2b188cc1 JA |
8677 | #endif |
8678 | ||
75b28aff | 8679 | io_mem_free(ctx->rings); |
2b188cc1 | 8680 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8681 | |
8682 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8683 | free_uid(ctx->user); |
4010fec4 | 8684 | io_req_caches_free(ctx); |
e941894e JA |
8685 | if (ctx->hash_map) |
8686 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8687 | kfree(ctx->cancel_hash); |
6224843d | 8688 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
8689 | kfree(ctx); |
8690 | } | |
8691 | ||
8692 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8693 | { | |
8694 | struct io_ring_ctx *ctx = file->private_data; | |
8695 | __poll_t mask = 0; | |
8696 | ||
311997b3 | 8697 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
8698 | /* |
8699 | * synchronizes with barrier from wq_has_sleeper call in | |
8700 | * io_commit_cqring | |
8701 | */ | |
2b188cc1 | 8702 | smp_rmb(); |
90554200 | 8703 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8704 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8705 | |
8706 | /* | |
8707 | * Don't flush cqring overflow list here, just do a simple check. | |
8708 | * Otherwise there could possible be ABBA deadlock: | |
8709 | * CPU0 CPU1 | |
8710 | * ---- ---- | |
8711 | * lock(&ctx->uring_lock); | |
8712 | * lock(&ep->mtx); | |
8713 | * lock(&ctx->uring_lock); | |
8714 | * lock(&ep->mtx); | |
8715 | * | |
8716 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8717 | * pushs them to do the flush. | |
8718 | */ | |
5ed7a37d | 8719 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
8720 | mask |= EPOLLIN | EPOLLRDNORM; |
8721 | ||
8722 | return mask; | |
8723 | } | |
8724 | ||
8725 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8726 | { | |
8727 | struct io_ring_ctx *ctx = file->private_data; | |
8728 | ||
8729 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8730 | } | |
8731 | ||
0bead8cd | 8732 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8733 | { |
4379bf8b | 8734 | const struct cred *creds; |
071698e1 | 8735 | |
61cf9370 | 8736 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8737 | if (creds) { |
8738 | put_cred(creds); | |
0bead8cd | 8739 | return 0; |
1e6fa521 | 8740 | } |
0bead8cd YD |
8741 | |
8742 | return -EINVAL; | |
8743 | } | |
8744 | ||
9b465711 | 8745 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8746 | { |
9b465711 | 8747 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8748 | } |
8749 | ||
d56d938b PB |
8750 | struct io_tctx_exit { |
8751 | struct callback_head task_work; | |
8752 | struct completion completion; | |
baf186c4 | 8753 | struct io_ring_ctx *ctx; |
d56d938b PB |
8754 | }; |
8755 | ||
8756 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8757 | { | |
8758 | struct io_uring_task *tctx = current->io_uring; | |
8759 | struct io_tctx_exit *work; | |
8760 | ||
8761 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8762 | /* | |
8763 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8764 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8765 | */ | |
8766 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 8767 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
8768 | complete(&work->completion); |
8769 | } | |
8770 | ||
28090c13 PB |
8771 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8772 | { | |
8773 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8774 | ||
8775 | return req->ctx == data; | |
8776 | } | |
8777 | ||
85faa7b8 JA |
8778 | static void io_ring_exit_work(struct work_struct *work) |
8779 | { | |
d56d938b | 8780 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8781 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8782 | struct io_tctx_exit exit; |
8783 | struct io_tctx_node *node; | |
8784 | int ret; | |
85faa7b8 | 8785 | |
56952e91 JA |
8786 | /* |
8787 | * If we're doing polled IO and end up having requests being | |
8788 | * submitted async (out-of-line), then completions can come in while | |
8789 | * we're waiting for refs to drop. We need to reap these manually, | |
8790 | * as nobody else will be looking for them. | |
8791 | */ | |
b2edc0a7 | 8792 | do { |
3dd0c97a | 8793 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
8794 | if (ctx->sq_data) { |
8795 | struct io_sq_data *sqd = ctx->sq_data; | |
8796 | struct task_struct *tsk; | |
8797 | ||
8798 | io_sq_thread_park(sqd); | |
8799 | tsk = sqd->thread; | |
8800 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
8801 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
8802 | io_cancel_ctx_cb, ctx, true); | |
8803 | io_sq_thread_unpark(sqd); | |
8804 | } | |
b5bb3a24 PB |
8805 | |
8806 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8807 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8808 | |
7f00651a PB |
8809 | init_completion(&exit.completion); |
8810 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8811 | exit.ctx = ctx; | |
89b5066e PB |
8812 | /* |
8813 | * Some may use context even when all refs and requests have been put, | |
8814 | * and they are free to do so while still holding uring_lock or | |
8815 | * completion_lock, see __io_req_task_submit(). Apart from other work, | |
8816 | * this lock/unlock section also waits them to finish. | |
8817 | */ | |
d56d938b PB |
8818 | mutex_lock(&ctx->uring_lock); |
8819 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8820 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8821 | ||
d56d938b PB |
8822 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8823 | ctx_node); | |
7f00651a PB |
8824 | /* don't spin on a single task if cancellation failed */ |
8825 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8826 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8827 | if (WARN_ON_ONCE(ret)) | |
8828 | continue; | |
8829 | wake_up_process(node->task); | |
8830 | ||
8831 | mutex_unlock(&ctx->uring_lock); | |
8832 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8833 | mutex_lock(&ctx->uring_lock); |
8834 | } | |
8835 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8836 | spin_lock_irq(&ctx->completion_lock); |
8837 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8838 | |
85faa7b8 JA |
8839 | io_ring_ctx_free(ctx); |
8840 | } | |
8841 | ||
80c4cbdb PB |
8842 | /* Returns true if we found and killed one or more timeouts */ |
8843 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 8844 | bool cancel_all) |
80c4cbdb PB |
8845 | { |
8846 | struct io_kiocb *req, *tmp; | |
8847 | int canceled = 0; | |
8848 | ||
8849 | spin_lock_irq(&ctx->completion_lock); | |
8850 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
3dd0c97a | 8851 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
8852 | io_kill_timeout(req, -ECANCELED); |
8853 | canceled++; | |
8854 | } | |
8855 | } | |
51520426 PB |
8856 | if (canceled != 0) |
8857 | io_commit_cqring(ctx); | |
80c4cbdb | 8858 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8859 | if (canceled != 0) |
8860 | io_cqring_ev_posted(ctx); | |
8861 | return canceled != 0; | |
8862 | } | |
8863 | ||
2b188cc1 JA |
8864 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8865 | { | |
61cf9370 MWO |
8866 | unsigned long index; |
8867 | struct creds *creds; | |
8868 | ||
2b188cc1 JA |
8869 | mutex_lock(&ctx->uring_lock); |
8870 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8871 | if (ctx->rings) |
6c2450ae | 8872 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8873 | xa_for_each(&ctx->personalities, index, creds) |
8874 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8875 | mutex_unlock(&ctx->uring_lock); |
8876 | ||
3dd0c97a PB |
8877 | io_kill_timeouts(ctx, NULL, true); |
8878 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 8879 | |
15dff286 | 8880 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8881 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8882 | |
85faa7b8 | 8883 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8884 | /* |
8885 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8886 | * if we're exiting a ton of rings at the same time. It just adds | |
8887 | * noise and overhead, there's no discernable change in runtime | |
8888 | * over using system_wq. | |
8889 | */ | |
8890 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8891 | } |
8892 | ||
8893 | static int io_uring_release(struct inode *inode, struct file *file) | |
8894 | { | |
8895 | struct io_ring_ctx *ctx = file->private_data; | |
8896 | ||
8897 | file->private_data = NULL; | |
8898 | io_ring_ctx_wait_and_kill(ctx); | |
8899 | return 0; | |
8900 | } | |
8901 | ||
f6edbabb PB |
8902 | struct io_task_cancel { |
8903 | struct task_struct *task; | |
3dd0c97a | 8904 | bool all; |
f6edbabb | 8905 | }; |
f254ac04 | 8906 | |
f6edbabb | 8907 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8908 | { |
9a472ef7 | 8909 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8910 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8911 | bool ret; |
8912 | ||
3dd0c97a | 8913 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8914 | unsigned long flags; |
8915 | struct io_ring_ctx *ctx = req->ctx; | |
8916 | ||
8917 | /* protect against races with linked timeouts */ | |
8918 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
3dd0c97a | 8919 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8920 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8921 | } else { | |
3dd0c97a | 8922 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8923 | } |
8924 | return ret; | |
b711d4ea JA |
8925 | } |
8926 | ||
e1915f76 | 8927 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 8928 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 8929 | { |
e1915f76 | 8930 | struct io_defer_entry *de; |
b7ddce3c PB |
8931 | LIST_HEAD(list); |
8932 | ||
8933 | spin_lock_irq(&ctx->completion_lock); | |
8934 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
3dd0c97a | 8935 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
8936 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8937 | break; | |
8938 | } | |
8939 | } | |
8940 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8941 | if (list_empty(&list)) |
8942 | return false; | |
b7ddce3c PB |
8943 | |
8944 | while (!list_empty(&list)) { | |
8945 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8946 | list_del_init(&de->list); | |
f41db273 | 8947 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8948 | kfree(de); |
8949 | } | |
e1915f76 | 8950 | return true; |
b7ddce3c PB |
8951 | } |
8952 | ||
1b00764f PB |
8953 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
8954 | { | |
8955 | struct io_tctx_node *node; | |
8956 | enum io_wq_cancel cret; | |
8957 | bool ret = false; | |
8958 | ||
8959 | mutex_lock(&ctx->uring_lock); | |
8960 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8961 | struct io_uring_task *tctx = node->task->io_uring; | |
8962 | ||
8963 | /* | |
8964 | * io_wq will stay alive while we hold uring_lock, because it's | |
8965 | * killed after ctx nodes, which requires to take the lock. | |
8966 | */ | |
8967 | if (!tctx || !tctx->io_wq) | |
8968 | continue; | |
8969 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8970 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8971 | } | |
8972 | mutex_unlock(&ctx->uring_lock); | |
8973 | ||
8974 | return ret; | |
8975 | } | |
8976 | ||
9936c7c2 PB |
8977 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8978 | struct task_struct *task, | |
3dd0c97a | 8979 | bool cancel_all) |
9936c7c2 | 8980 | { |
3dd0c97a | 8981 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 8982 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8983 | |
8984 | while (1) { | |
8985 | enum io_wq_cancel cret; | |
8986 | bool ret = false; | |
8987 | ||
1b00764f PB |
8988 | if (!task) { |
8989 | ret |= io_uring_try_cancel_iowq(ctx); | |
8990 | } else if (tctx && tctx->io_wq) { | |
8991 | /* | |
8992 | * Cancels requests of all rings, not only @ctx, but | |
8993 | * it's fine as the task is in exit/exec. | |
8994 | */ | |
5aa75ed5 | 8995 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8996 | &cancel, true); |
8997 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8998 | } | |
8999 | ||
9000 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 9001 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 9002 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
9003 | while (!list_empty_careful(&ctx->iopoll_list)) { |
9004 | io_iopoll_try_reap_events(ctx); | |
9005 | ret = true; | |
9006 | } | |
9007 | } | |
9008 | ||
3dd0c97a PB |
9009 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
9010 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
9011 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
9936c7c2 | 9012 | ret |= io_run_task_work(); |
ba50a036 | 9013 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
9014 | if (!ret) |
9015 | break; | |
9016 | cond_resched(); | |
9017 | } | |
9018 | } | |
9019 | ||
eef51daa | 9020 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 9021 | { |
236434c3 | 9022 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 9023 | struct io_tctx_node *node; |
a528b04e | 9024 | int ret; |
236434c3 MWO |
9025 | |
9026 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9027 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9028 | if (unlikely(ret)) |
9029 | return ret; | |
236434c3 | 9030 | tctx = current->io_uring; |
0f212204 | 9031 | } |
cf27f3b1 PB |
9032 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9033 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9034 | if (!node) | |
9035 | return -ENOMEM; | |
9036 | node->ctx = ctx; | |
9037 | node->task = current; | |
13bf43f5 | 9038 | |
cf27f3b1 PB |
9039 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9040 | node, GFP_KERNEL)); | |
9041 | if (ret) { | |
9042 | kfree(node); | |
9043 | return ret; | |
0f212204 | 9044 | } |
cf27f3b1 PB |
9045 | |
9046 | mutex_lock(&ctx->uring_lock); | |
9047 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9048 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9049 | } |
cf27f3b1 | 9050 | tctx->last = ctx; |
0f212204 JA |
9051 | return 0; |
9052 | } | |
9053 | ||
cf27f3b1 PB |
9054 | /* |
9055 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9056 | */ | |
eef51daa | 9057 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9058 | { |
9059 | struct io_uring_task *tctx = current->io_uring; | |
9060 | ||
9061 | if (likely(tctx && tctx->last == ctx)) | |
9062 | return 0; | |
eef51daa | 9063 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9064 | } |
9065 | ||
0f212204 JA |
9066 | /* |
9067 | * Remove this io_uring_file -> task mapping. | |
9068 | */ | |
eef51daa | 9069 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9070 | { |
9071 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9072 | struct io_tctx_node *node; |
2941267b | 9073 | |
eebd2e37 PB |
9074 | if (!tctx) |
9075 | return; | |
13bf43f5 PB |
9076 | node = xa_erase(&tctx->xa, index); |
9077 | if (!node) | |
2941267b | 9078 | return; |
0f212204 | 9079 | |
13bf43f5 PB |
9080 | WARN_ON_ONCE(current != node->task); |
9081 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9082 | ||
9083 | mutex_lock(&node->ctx->uring_lock); | |
9084 | list_del(&node->ctx_node); | |
9085 | mutex_unlock(&node->ctx->uring_lock); | |
9086 | ||
baf186c4 | 9087 | if (tctx->last == node->ctx) |
0f212204 | 9088 | tctx->last = NULL; |
13bf43f5 | 9089 | kfree(node); |
0f212204 JA |
9090 | } |
9091 | ||
8452d4a6 | 9092 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9093 | { |
ba5ef6dc | 9094 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9095 | struct io_tctx_node *node; |
de7f1d9e PB |
9096 | unsigned long index; |
9097 | ||
13bf43f5 | 9098 | xa_for_each(&tctx->xa, index, node) |
eef51daa | 9099 | io_uring_del_tctx_node(index); |
b16ef427 ME |
9100 | if (wq) { |
9101 | /* | |
9102 | * Must be after io_uring_del_task_file() (removes nodes under | |
9103 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9104 | */ | |
9105 | tctx->io_wq = NULL; | |
ba5ef6dc | 9106 | io_wq_put_and_exit(wq); |
b16ef427 | 9107 | } |
de7f1d9e PB |
9108 | } |
9109 | ||
3f48cf18 | 9110 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9111 | { |
3f48cf18 PB |
9112 | if (tracked) |
9113 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9114 | return percpu_counter_sum(&tctx->inflight); |
9115 | } | |
9116 | ||
09899b19 PB |
9117 | static void io_uring_drop_tctx_refs(struct task_struct *task) |
9118 | { | |
9119 | struct io_uring_task *tctx = task->io_uring; | |
9120 | unsigned int refs = tctx->cached_refs; | |
9121 | ||
9122 | tctx->cached_refs = 0; | |
9123 | percpu_counter_sub(&tctx->inflight, refs); | |
9124 | put_task_struct_many(task, refs); | |
9125 | } | |
9126 | ||
78cc687b PB |
9127 | /* |
9128 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
9129 | * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation. | |
9130 | */ | |
9131 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9132 | { |
521d6a73 | 9133 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9134 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9135 | s64 inflight; |
9136 | DEFINE_WAIT(wait); | |
fdaf083c | 9137 | |
78cc687b PB |
9138 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9139 | ||
6d042ffb PO |
9140 | if (!current->io_uring) |
9141 | return; | |
17a91051 PB |
9142 | if (tctx->io_wq) |
9143 | io_wq_exit_start(tctx->io_wq); | |
9144 | ||
09899b19 | 9145 | io_uring_drop_tctx_refs(current); |
0e9ddb39 PB |
9146 | atomic_inc(&tctx->in_idle); |
9147 | do { | |
9148 | /* read completions before cancelations */ | |
78cc687b | 9149 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9150 | if (!inflight) |
9151 | break; | |
fdaf083c | 9152 | |
78cc687b PB |
9153 | if (!sqd) { |
9154 | struct io_tctx_node *node; | |
9155 | unsigned long index; | |
0f212204 | 9156 | |
78cc687b PB |
9157 | xa_for_each(&tctx->xa, index, node) { |
9158 | /* sqpoll task will cancel all its requests */ | |
9159 | if (node->ctx->sq_data) | |
9160 | continue; | |
9161 | io_uring_try_cancel_requests(node->ctx, current, | |
9162 | cancel_all); | |
9163 | } | |
9164 | } else { | |
9165 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9166 | io_uring_try_cancel_requests(ctx, current, | |
9167 | cancel_all); | |
9168 | } | |
17a91051 | 9169 | |
0f212204 | 9170 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
0f212204 | 9171 | /* |
a1bb3cd5 PB |
9172 | * If we've seen completions, retry without waiting. This |
9173 | * avoids a race where a completion comes in before we did | |
9174 | * prepare_to_wait(). | |
0f212204 | 9175 | */ |
3dd0c97a | 9176 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9177 | schedule(); |
f57555ed | 9178 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9179 | } while (1); |
fdaf083c | 9180 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9181 | |
8452d4a6 | 9182 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9183 | if (cancel_all) { |
3f48cf18 PB |
9184 | /* for exec all current's requests should be gone, kill tctx */ |
9185 | __io_uring_free(current); | |
9186 | } | |
44e728b8 PB |
9187 | } |
9188 | ||
78cc687b PB |
9189 | void __io_uring_cancel(struct files_struct *files) |
9190 | { | |
9191 | io_uring_cancel_generic(!files, NULL); | |
9192 | } | |
9193 | ||
6c5c240e RP |
9194 | static void *io_uring_validate_mmap_request(struct file *file, |
9195 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9196 | { |
2b188cc1 | 9197 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9198 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9199 | struct page *page; |
9200 | void *ptr; | |
9201 | ||
9202 | switch (offset) { | |
9203 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9204 | case IORING_OFF_CQ_RING: |
9205 | ptr = ctx->rings; | |
2b188cc1 JA |
9206 | break; |
9207 | case IORING_OFF_SQES: | |
9208 | ptr = ctx->sq_sqes; | |
9209 | break; | |
2b188cc1 | 9210 | default: |
6c5c240e | 9211 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9212 | } |
9213 | ||
9214 | page = virt_to_head_page(ptr); | |
a50b854e | 9215 | if (sz > page_size(page)) |
6c5c240e RP |
9216 | return ERR_PTR(-EINVAL); |
9217 | ||
9218 | return ptr; | |
9219 | } | |
9220 | ||
9221 | #ifdef CONFIG_MMU | |
9222 | ||
9223 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9224 | { | |
9225 | size_t sz = vma->vm_end - vma->vm_start; | |
9226 | unsigned long pfn; | |
9227 | void *ptr; | |
9228 | ||
9229 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9230 | if (IS_ERR(ptr)) | |
9231 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9232 | |
9233 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9234 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9235 | } | |
9236 | ||
6c5c240e RP |
9237 | #else /* !CONFIG_MMU */ |
9238 | ||
9239 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9240 | { | |
9241 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9242 | } | |
9243 | ||
9244 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9245 | { | |
9246 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9247 | } | |
9248 | ||
9249 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9250 | unsigned long addr, unsigned long len, | |
9251 | unsigned long pgoff, unsigned long flags) | |
9252 | { | |
9253 | void *ptr; | |
9254 | ||
9255 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9256 | if (IS_ERR(ptr)) | |
9257 | return PTR_ERR(ptr); | |
9258 | ||
9259 | return (unsigned long) ptr; | |
9260 | } | |
9261 | ||
9262 | #endif /* !CONFIG_MMU */ | |
9263 | ||
d9d05217 | 9264 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9265 | { |
9266 | DEFINE_WAIT(wait); | |
9267 | ||
9268 | do { | |
9269 | if (!io_sqring_full(ctx)) | |
9270 | break; | |
90554200 JA |
9271 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9272 | ||
9273 | if (!io_sqring_full(ctx)) | |
9274 | break; | |
90554200 JA |
9275 | schedule(); |
9276 | } while (!signal_pending(current)); | |
9277 | ||
9278 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9279 | return 0; |
90554200 JA |
9280 | } |
9281 | ||
c73ebb68 HX |
9282 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9283 | struct __kernel_timespec __user **ts, | |
9284 | const sigset_t __user **sig) | |
9285 | { | |
9286 | struct io_uring_getevents_arg arg; | |
9287 | ||
9288 | /* | |
9289 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9290 | * is just a pointer to the sigset_t. | |
9291 | */ | |
9292 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9293 | *sig = (const sigset_t __user *) argp; | |
9294 | *ts = NULL; | |
9295 | return 0; | |
9296 | } | |
9297 | ||
9298 | /* | |
9299 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9300 | * timespec and sigset_t pointers if good. | |
9301 | */ | |
9302 | if (*argsz != sizeof(arg)) | |
9303 | return -EINVAL; | |
9304 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9305 | return -EFAULT; | |
9306 | *sig = u64_to_user_ptr(arg.sigmask); | |
9307 | *argsz = arg.sigmask_sz; | |
9308 | *ts = u64_to_user_ptr(arg.ts); | |
9309 | return 0; | |
9310 | } | |
9311 | ||
2b188cc1 | 9312 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9313 | u32, min_complete, u32, flags, const void __user *, argp, |
9314 | size_t, argsz) | |
2b188cc1 JA |
9315 | { |
9316 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9317 | int submitted = 0; |
9318 | struct fd f; | |
33f993da | 9319 | long ret; |
2b188cc1 | 9320 | |
4c6e277c | 9321 | io_run_task_work(); |
b41e9852 | 9322 | |
33f993da PB |
9323 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9324 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9325 | return -EINVAL; |
9326 | ||
9327 | f = fdget(fd); | |
33f993da | 9328 | if (unlikely(!f.file)) |
2b188cc1 JA |
9329 | return -EBADF; |
9330 | ||
9331 | ret = -EOPNOTSUPP; | |
33f993da | 9332 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9333 | goto out_fput; |
9334 | ||
9335 | ret = -ENXIO; | |
9336 | ctx = f.file->private_data; | |
33f993da | 9337 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9338 | goto out_fput; |
9339 | ||
7e84e1c7 | 9340 | ret = -EBADFD; |
33f993da | 9341 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9342 | goto out; |
9343 | ||
6c271ce2 JA |
9344 | /* |
9345 | * For SQ polling, the thread will do all submissions and completions. | |
9346 | * Just return the requested submit count, and wake the thread if | |
9347 | * we were asked to. | |
9348 | */ | |
b2a9eada | 9349 | ret = 0; |
6c271ce2 | 9350 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9351 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9352 | |
d9d05217 | 9353 | ret = -EOWNERDEAD; |
fe7e3257 | 9354 | if (unlikely(ctx->sq_data->thread == NULL)) |
04147488 | 9355 | goto out; |
6c271ce2 | 9356 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9357 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9358 | if (flags & IORING_ENTER_SQ_WAIT) { |
9359 | ret = io_sqpoll_wait_sq(ctx); | |
9360 | if (ret) | |
9361 | goto out; | |
9362 | } | |
6c271ce2 | 9363 | submitted = to_submit; |
b2a9eada | 9364 | } else if (to_submit) { |
eef51daa | 9365 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
9366 | if (unlikely(ret)) |
9367 | goto out; | |
2b188cc1 | 9368 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9369 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9370 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9371 | |
9372 | if (submitted != to_submit) | |
9373 | goto out; | |
2b188cc1 JA |
9374 | } |
9375 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9376 | const sigset_t __user *sig; |
9377 | struct __kernel_timespec __user *ts; | |
9378 | ||
9379 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9380 | if (unlikely(ret)) | |
9381 | goto out; | |
9382 | ||
2b188cc1 JA |
9383 | min_complete = min(min_complete, ctx->cq_entries); |
9384 | ||
32b2244a XW |
9385 | /* |
9386 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9387 | * space applications don't need to do io completion events | |
9388 | * polling again, they can rely on io_sq_thread to do polling | |
9389 | * work, which can reduce cpu usage and uring_lock contention. | |
9390 | */ | |
9391 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9392 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9393 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9394 | } else { |
c73ebb68 | 9395 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9396 | } |
2b188cc1 JA |
9397 | } |
9398 | ||
7c504e65 | 9399 | out: |
6805b32e | 9400 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9401 | out_fput: |
9402 | fdput(f); | |
9403 | return submitted ? submitted : ret; | |
9404 | } | |
9405 | ||
bebdb65e | 9406 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9407 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9408 | const struct cred *cred) | |
87ce955b | 9409 | { |
87ce955b JA |
9410 | struct user_namespace *uns = seq_user_ns(m); |
9411 | struct group_info *gi; | |
9412 | kernel_cap_t cap; | |
9413 | unsigned __capi; | |
9414 | int g; | |
9415 | ||
9416 | seq_printf(m, "%5d\n", id); | |
9417 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9418 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9419 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9420 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9421 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9422 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9423 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9424 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9425 | seq_puts(m, "\n\tGroups:\t"); | |
9426 | gi = cred->group_info; | |
9427 | for (g = 0; g < gi->ngroups; g++) { | |
9428 | seq_put_decimal_ull(m, g ? " " : "", | |
9429 | from_kgid_munged(uns, gi->gid[g])); | |
9430 | } | |
9431 | seq_puts(m, "\n\tCapEff:\t"); | |
9432 | cap = cred->cap_effective; | |
9433 | CAP_FOR_EACH_U32(__capi) | |
9434 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9435 | seq_putc(m, '\n'); | |
9436 | return 0; | |
9437 | } | |
9438 | ||
9439 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9440 | { | |
dbbe9c64 | 9441 | struct io_sq_data *sq = NULL; |
fad8e0de | 9442 | bool has_lock; |
87ce955b JA |
9443 | int i; |
9444 | ||
fad8e0de JA |
9445 | /* |
9446 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9447 | * since fdinfo case grabs it in the opposite direction of normal use | |
9448 | * cases. If we fail to get the lock, we just don't iterate any | |
9449 | * structures that could be going away outside the io_uring mutex. | |
9450 | */ | |
9451 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9452 | ||
5f3f26f9 | 9453 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9454 | sq = ctx->sq_data; |
5f3f26f9 JA |
9455 | if (!sq->thread) |
9456 | sq = NULL; | |
9457 | } | |
dbbe9c64 JQ |
9458 | |
9459 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9460 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9461 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9462 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9463 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9464 | |
87ce955b JA |
9465 | if (f) |
9466 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9467 | else | |
9468 | seq_printf(m, "%5u: <none>\n", i); | |
9469 | } | |
9470 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9471 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9472 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9473 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9474 | |
4751f53d | 9475 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9476 | } |
61cf9370 MWO |
9477 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9478 | unsigned long index; | |
9479 | const struct cred *cred; | |
9480 | ||
87ce955b | 9481 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9482 | xa_for_each(&ctx->personalities, index, cred) |
9483 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9484 | } |
d7718a9d JA |
9485 | seq_printf(m, "PollList:\n"); |
9486 | spin_lock_irq(&ctx->completion_lock); | |
9487 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9488 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9489 | struct io_kiocb *req; | |
9490 | ||
9491 | hlist_for_each_entry(req, list, hash_node) | |
9492 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9493 | req->task->task_works != NULL); | |
9494 | } | |
9495 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9496 | if (has_lock) |
9497 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9498 | } |
9499 | ||
9500 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9501 | { | |
9502 | struct io_ring_ctx *ctx = f->private_data; | |
9503 | ||
9504 | if (percpu_ref_tryget(&ctx->refs)) { | |
9505 | __io_uring_show_fdinfo(ctx, m); | |
9506 | percpu_ref_put(&ctx->refs); | |
9507 | } | |
9508 | } | |
bebdb65e | 9509 | #endif |
87ce955b | 9510 | |
2b188cc1 JA |
9511 | static const struct file_operations io_uring_fops = { |
9512 | .release = io_uring_release, | |
9513 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9514 | #ifndef CONFIG_MMU |
9515 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9516 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9517 | #endif | |
2b188cc1 JA |
9518 | .poll = io_uring_poll, |
9519 | .fasync = io_uring_fasync, | |
bebdb65e | 9520 | #ifdef CONFIG_PROC_FS |
87ce955b | 9521 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9522 | #endif |
2b188cc1 JA |
9523 | }; |
9524 | ||
9525 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9526 | struct io_uring_params *p) | |
9527 | { | |
75b28aff HV |
9528 | struct io_rings *rings; |
9529 | size_t size, sq_array_offset; | |
2b188cc1 | 9530 | |
bd740481 JA |
9531 | /* make sure these are sane, as we already accounted them */ |
9532 | ctx->sq_entries = p->sq_entries; | |
9533 | ctx->cq_entries = p->cq_entries; | |
9534 | ||
75b28aff HV |
9535 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9536 | if (size == SIZE_MAX) | |
9537 | return -EOVERFLOW; | |
9538 | ||
9539 | rings = io_mem_alloc(size); | |
9540 | if (!rings) | |
2b188cc1 JA |
9541 | return -ENOMEM; |
9542 | ||
75b28aff HV |
9543 | ctx->rings = rings; |
9544 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9545 | rings->sq_ring_mask = p->sq_entries - 1; | |
9546 | rings->cq_ring_mask = p->cq_entries - 1; | |
9547 | rings->sq_ring_entries = p->sq_entries; | |
9548 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
9549 | |
9550 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9551 | if (size == SIZE_MAX) { |
9552 | io_mem_free(ctx->rings); | |
9553 | ctx->rings = NULL; | |
2b188cc1 | 9554 | return -EOVERFLOW; |
eb065d30 | 9555 | } |
2b188cc1 JA |
9556 | |
9557 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9558 | if (!ctx->sq_sqes) { |
9559 | io_mem_free(ctx->rings); | |
9560 | ctx->rings = NULL; | |
2b188cc1 | 9561 | return -ENOMEM; |
eb065d30 | 9562 | } |
2b188cc1 | 9563 | |
2b188cc1 JA |
9564 | return 0; |
9565 | } | |
9566 | ||
9faadcc8 PB |
9567 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9568 | { | |
9569 | int ret, fd; | |
9570 | ||
9571 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9572 | if (fd < 0) | |
9573 | return fd; | |
9574 | ||
eef51daa | 9575 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
9576 | if (ret) { |
9577 | put_unused_fd(fd); | |
9578 | return ret; | |
9579 | } | |
9580 | fd_install(fd, file); | |
9581 | return fd; | |
9582 | } | |
9583 | ||
2b188cc1 JA |
9584 | /* |
9585 | * Allocate an anonymous fd, this is what constitutes the application | |
9586 | * visible backing of an io_uring instance. The application mmaps this | |
9587 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9588 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9589 | */ | |
9faadcc8 | 9590 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9591 | { |
9592 | struct file *file; | |
9faadcc8 | 9593 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9594 | int ret; |
9595 | ||
2b188cc1 JA |
9596 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9597 | &ctx->ring_sock); | |
9598 | if (ret) | |
9faadcc8 | 9599 | return ERR_PTR(ret); |
2b188cc1 JA |
9600 | #endif |
9601 | ||
2b188cc1 JA |
9602 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9603 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9604 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9605 | if (IS_ERR(file)) { |
9606 | sock_release(ctx->ring_sock); | |
9607 | ctx->ring_sock = NULL; | |
9608 | } else { | |
9609 | ctx->ring_sock->file = file; | |
0f212204 | 9610 | } |
2b188cc1 | 9611 | #endif |
9faadcc8 | 9612 | return file; |
2b188cc1 JA |
9613 | } |
9614 | ||
7f13657d XW |
9615 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9616 | struct io_uring_params __user *params) | |
2b188cc1 | 9617 | { |
2b188cc1 | 9618 | struct io_ring_ctx *ctx; |
9faadcc8 | 9619 | struct file *file; |
2b188cc1 JA |
9620 | int ret; |
9621 | ||
8110c1a6 | 9622 | if (!entries) |
2b188cc1 | 9623 | return -EINVAL; |
8110c1a6 JA |
9624 | if (entries > IORING_MAX_ENTRIES) { |
9625 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9626 | return -EINVAL; | |
9627 | entries = IORING_MAX_ENTRIES; | |
9628 | } | |
2b188cc1 JA |
9629 | |
9630 | /* | |
9631 | * Use twice as many entries for the CQ ring. It's possible for the | |
9632 | * application to drive a higher depth than the size of the SQ ring, | |
9633 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9634 | * some flexibility in overcommitting a bit. If the application has |
9635 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9636 | * of CQ ring entries manually. | |
2b188cc1 JA |
9637 | */ |
9638 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9639 | if (p->flags & IORING_SETUP_CQSIZE) { |
9640 | /* | |
9641 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9642 | * to a power-of-two, if it isn't already. We do NOT impose | |
9643 | * any cq vs sq ring sizing. | |
9644 | */ | |
eb2667b3 | 9645 | if (!p->cq_entries) |
33a107f0 | 9646 | return -EINVAL; |
8110c1a6 JA |
9647 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9648 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9649 | return -EINVAL; | |
9650 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9651 | } | |
eb2667b3 JQ |
9652 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9653 | if (p->cq_entries < p->sq_entries) | |
9654 | return -EINVAL; | |
33a107f0 JA |
9655 | } else { |
9656 | p->cq_entries = 2 * p->sq_entries; | |
9657 | } | |
2b188cc1 | 9658 | |
2b188cc1 | 9659 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9660 | if (!ctx) |
2b188cc1 | 9661 | return -ENOMEM; |
2b188cc1 | 9662 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9663 | if (!capable(CAP_IPC_LOCK)) |
9664 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9665 | |
9666 | /* | |
9667 | * This is just grabbed for accounting purposes. When a process exits, | |
9668 | * the mm is exited and dropped before the files, hence we need to hang | |
9669 | * on to this mm purely for the purposes of being able to unaccount | |
9670 | * memory (locked/pinned vm). It's not used for anything else. | |
9671 | */ | |
6b7898eb | 9672 | mmgrab(current->mm); |
2aede0e4 | 9673 | ctx->mm_account = current->mm; |
6b7898eb | 9674 | |
2b188cc1 JA |
9675 | ret = io_allocate_scq_urings(ctx, p); |
9676 | if (ret) | |
9677 | goto err; | |
9678 | ||
7e84e1c7 | 9679 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9680 | if (ret) |
9681 | goto err; | |
eae071c9 | 9682 | /* always set a rsrc node */ |
47b228ce PB |
9683 | ret = io_rsrc_node_switch_start(ctx); |
9684 | if (ret) | |
9685 | goto err; | |
eae071c9 | 9686 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 9687 | |
2b188cc1 | 9688 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9689 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9690 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9691 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9692 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9693 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9694 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9695 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9696 | |
9697 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9698 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9699 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9700 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9701 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9702 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9703 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9704 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9705 | |
7f13657d XW |
9706 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9707 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9708 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9709 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
9710 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
9711 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
9712 | |
9713 | if (copy_to_user(params, p, sizeof(*p))) { | |
9714 | ret = -EFAULT; | |
9715 | goto err; | |
9716 | } | |
d1719f70 | 9717 | |
9faadcc8 PB |
9718 | file = io_uring_get_file(ctx); |
9719 | if (IS_ERR(file)) { | |
9720 | ret = PTR_ERR(file); | |
9721 | goto err; | |
9722 | } | |
9723 | ||
044c1ab3 JA |
9724 | /* |
9725 | * Install ring fd as the very last thing, so we don't risk someone | |
9726 | * having closed it before we finish setup | |
9727 | */ | |
9faadcc8 PB |
9728 | ret = io_uring_install_fd(ctx, file); |
9729 | if (ret < 0) { | |
9730 | /* fput will clean it up */ | |
9731 | fput(file); | |
9732 | return ret; | |
9733 | } | |
044c1ab3 | 9734 | |
c826bd7a | 9735 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9736 | return ret; |
9737 | err: | |
9738 | io_ring_ctx_wait_and_kill(ctx); | |
9739 | return ret; | |
9740 | } | |
9741 | ||
9742 | /* | |
9743 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9744 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9745 | * params structure passed in. | |
9746 | */ | |
9747 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9748 | { | |
9749 | struct io_uring_params p; | |
2b188cc1 JA |
9750 | int i; |
9751 | ||
9752 | if (copy_from_user(&p, params, sizeof(p))) | |
9753 | return -EFAULT; | |
9754 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9755 | if (p.resv[i]) | |
9756 | return -EINVAL; | |
9757 | } | |
9758 | ||
6c271ce2 | 9759 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9760 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9761 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9762 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9763 | return -EINVAL; |
9764 | ||
7f13657d | 9765 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9766 | } |
9767 | ||
9768 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9769 | struct io_uring_params __user *, params) | |
9770 | { | |
9771 | return io_uring_setup(entries, params); | |
9772 | } | |
9773 | ||
66f4af93 JA |
9774 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9775 | { | |
9776 | struct io_uring_probe *p; | |
9777 | size_t size; | |
9778 | int i, ret; | |
9779 | ||
9780 | size = struct_size(p, ops, nr_args); | |
9781 | if (size == SIZE_MAX) | |
9782 | return -EOVERFLOW; | |
9783 | p = kzalloc(size, GFP_KERNEL); | |
9784 | if (!p) | |
9785 | return -ENOMEM; | |
9786 | ||
9787 | ret = -EFAULT; | |
9788 | if (copy_from_user(p, arg, size)) | |
9789 | goto out; | |
9790 | ret = -EINVAL; | |
9791 | if (memchr_inv(p, 0, size)) | |
9792 | goto out; | |
9793 | ||
9794 | p->last_op = IORING_OP_LAST - 1; | |
9795 | if (nr_args > IORING_OP_LAST) | |
9796 | nr_args = IORING_OP_LAST; | |
9797 | ||
9798 | for (i = 0; i < nr_args; i++) { | |
9799 | p->ops[i].op = i; | |
9800 | if (!io_op_defs[i].not_supported) | |
9801 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9802 | } | |
9803 | p->ops_len = i; | |
9804 | ||
9805 | ret = 0; | |
9806 | if (copy_to_user(arg, p, size)) | |
9807 | ret = -EFAULT; | |
9808 | out: | |
9809 | kfree(p); | |
9810 | return ret; | |
9811 | } | |
9812 | ||
071698e1 JA |
9813 | static int io_register_personality(struct io_ring_ctx *ctx) |
9814 | { | |
4379bf8b | 9815 | const struct cred *creds; |
61cf9370 | 9816 | u32 id; |
1e6fa521 | 9817 | int ret; |
071698e1 | 9818 | |
4379bf8b | 9819 | creds = get_current_cred(); |
1e6fa521 | 9820 | |
61cf9370 MWO |
9821 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9822 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9823 | if (!ret) | |
9824 | return id; | |
9825 | put_cred(creds); | |
1e6fa521 | 9826 | return ret; |
071698e1 JA |
9827 | } |
9828 | ||
21b55dbc SG |
9829 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9830 | unsigned int nr_args) | |
9831 | { | |
9832 | struct io_uring_restriction *res; | |
9833 | size_t size; | |
9834 | int i, ret; | |
9835 | ||
7e84e1c7 SG |
9836 | /* Restrictions allowed only if rings started disabled */ |
9837 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9838 | return -EBADFD; | |
9839 | ||
21b55dbc | 9840 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9841 | if (ctx->restrictions.registered) |
21b55dbc SG |
9842 | return -EBUSY; |
9843 | ||
9844 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9845 | return -EINVAL; | |
9846 | ||
9847 | size = array_size(nr_args, sizeof(*res)); | |
9848 | if (size == SIZE_MAX) | |
9849 | return -EOVERFLOW; | |
9850 | ||
9851 | res = memdup_user(arg, size); | |
9852 | if (IS_ERR(res)) | |
9853 | return PTR_ERR(res); | |
9854 | ||
9855 | ret = 0; | |
9856 | ||
9857 | for (i = 0; i < nr_args; i++) { | |
9858 | switch (res[i].opcode) { | |
9859 | case IORING_RESTRICTION_REGISTER_OP: | |
9860 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9861 | ret = -EINVAL; | |
9862 | goto out; | |
9863 | } | |
9864 | ||
9865 | __set_bit(res[i].register_op, | |
9866 | ctx->restrictions.register_op); | |
9867 | break; | |
9868 | case IORING_RESTRICTION_SQE_OP: | |
9869 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9870 | ret = -EINVAL; | |
9871 | goto out; | |
9872 | } | |
9873 | ||
9874 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9875 | break; | |
9876 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9877 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9878 | break; | |
9879 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9880 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9881 | break; | |
9882 | default: | |
9883 | ret = -EINVAL; | |
9884 | goto out; | |
9885 | } | |
9886 | } | |
9887 | ||
9888 | out: | |
9889 | /* Reset all restrictions if an error happened */ | |
9890 | if (ret != 0) | |
9891 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9892 | else | |
7e84e1c7 | 9893 | ctx->restrictions.registered = true; |
21b55dbc SG |
9894 | |
9895 | kfree(res); | |
9896 | return ret; | |
9897 | } | |
9898 | ||
7e84e1c7 SG |
9899 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9900 | { | |
9901 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9902 | return -EBADFD; | |
9903 | ||
9904 | if (ctx->restrictions.registered) | |
9905 | ctx->restricted = 1; | |
9906 | ||
0298ef96 PB |
9907 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9908 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9909 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9910 | return 0; |
9911 | } | |
9912 | ||
fdecb662 | 9913 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 9914 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
9915 | unsigned nr_args) |
9916 | { | |
9917 | __u32 tmp; | |
9918 | int err; | |
9919 | ||
c3bdad02 PB |
9920 | if (up->resv) |
9921 | return -EINVAL; | |
98f0b3b4 PB |
9922 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
9923 | return -EOVERFLOW; | |
9924 | err = io_rsrc_node_switch_start(ctx); | |
9925 | if (err) | |
9926 | return err; | |
9927 | ||
fdecb662 PB |
9928 | switch (type) { |
9929 | case IORING_RSRC_FILE: | |
98f0b3b4 | 9930 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
9931 | case IORING_RSRC_BUFFER: |
9932 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
9933 | } |
9934 | return -EINVAL; | |
9935 | } | |
9936 | ||
c3bdad02 PB |
9937 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
9938 | unsigned nr_args) | |
98f0b3b4 | 9939 | { |
c3bdad02 | 9940 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
9941 | |
9942 | if (!nr_args) | |
9943 | return -EINVAL; | |
c3bdad02 PB |
9944 | memset(&up, 0, sizeof(up)); |
9945 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
9946 | return -EFAULT; | |
9947 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
9948 | } | |
9949 | ||
9950 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 9951 | unsigned size, unsigned type) |
c3bdad02 PB |
9952 | { |
9953 | struct io_uring_rsrc_update2 up; | |
9954 | ||
9955 | if (size != sizeof(up)) | |
9956 | return -EINVAL; | |
98f0b3b4 PB |
9957 | if (copy_from_user(&up, arg, sizeof(up))) |
9958 | return -EFAULT; | |
992da01a | 9959 | if (!up.nr || up.resv) |
98f0b3b4 | 9960 | return -EINVAL; |
992da01a | 9961 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
9962 | } |
9963 | ||
792e3582 | 9964 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 9965 | unsigned int size, unsigned int type) |
792e3582 PB |
9966 | { |
9967 | struct io_uring_rsrc_register rr; | |
9968 | ||
9969 | /* keep it extendible */ | |
9970 | if (size != sizeof(rr)) | |
9971 | return -EINVAL; | |
9972 | ||
9973 | memset(&rr, 0, sizeof(rr)); | |
9974 | if (copy_from_user(&rr, arg, size)) | |
9975 | return -EFAULT; | |
992da01a | 9976 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
9977 | return -EINVAL; |
9978 | ||
992da01a | 9979 | switch (type) { |
792e3582 PB |
9980 | case IORING_RSRC_FILE: |
9981 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
9982 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
9983 | case IORING_RSRC_BUFFER: |
9984 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
9985 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
9986 | } |
9987 | return -EINVAL; | |
9988 | } | |
9989 | ||
fe76421d JA |
9990 | static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg, |
9991 | unsigned len) | |
9992 | { | |
9993 | struct io_uring_task *tctx = current->io_uring; | |
9994 | cpumask_var_t new_mask; | |
9995 | int ret; | |
9996 | ||
9997 | if (!tctx || !tctx->io_wq) | |
9998 | return -EINVAL; | |
9999 | ||
10000 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
10001 | return -ENOMEM; | |
10002 | ||
10003 | cpumask_clear(new_mask); | |
10004 | if (len > cpumask_size()) | |
10005 | len = cpumask_size(); | |
10006 | ||
10007 | if (copy_from_user(new_mask, arg, len)) { | |
10008 | free_cpumask_var(new_mask); | |
10009 | return -EFAULT; | |
10010 | } | |
10011 | ||
10012 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
10013 | free_cpumask_var(new_mask); | |
10014 | return ret; | |
10015 | } | |
10016 | ||
10017 | static int io_unregister_iowq_aff(struct io_ring_ctx *ctx) | |
10018 | { | |
10019 | struct io_uring_task *tctx = current->io_uring; | |
10020 | ||
10021 | if (!tctx || !tctx->io_wq) | |
10022 | return -EINVAL; | |
10023 | ||
10024 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
10025 | } | |
10026 | ||
071698e1 JA |
10027 | static bool io_register_op_must_quiesce(int op) |
10028 | { | |
10029 | switch (op) { | |
bd54b6fe BM |
10030 | case IORING_REGISTER_BUFFERS: |
10031 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 10032 | case IORING_REGISTER_FILES: |
071698e1 JA |
10033 | case IORING_UNREGISTER_FILES: |
10034 | case IORING_REGISTER_FILES_UPDATE: | |
10035 | case IORING_REGISTER_PROBE: | |
10036 | case IORING_REGISTER_PERSONALITY: | |
10037 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
10038 | case IORING_REGISTER_FILES2: |
10039 | case IORING_REGISTER_FILES_UPDATE2: | |
10040 | case IORING_REGISTER_BUFFERS2: | |
10041 | case IORING_REGISTER_BUFFERS_UPDATE: | |
fe76421d JA |
10042 | case IORING_REGISTER_IOWQ_AFF: |
10043 | case IORING_UNREGISTER_IOWQ_AFF: | |
071698e1 JA |
10044 | return false; |
10045 | default: | |
10046 | return true; | |
10047 | } | |
10048 | } | |
10049 | ||
edafccee JA |
10050 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10051 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10052 | __releases(ctx->uring_lock) |
10053 | __acquires(ctx->uring_lock) | |
edafccee JA |
10054 | { |
10055 | int ret; | |
10056 | ||
35fa71a0 JA |
10057 | /* |
10058 | * We're inside the ring mutex, if the ref is already dying, then | |
10059 | * someone else killed the ctx or is already going through | |
10060 | * io_uring_register(). | |
10061 | */ | |
10062 | if (percpu_ref_is_dying(&ctx->refs)) | |
10063 | return -ENXIO; | |
10064 | ||
75c4021a PB |
10065 | if (ctx->restricted) { |
10066 | if (opcode >= IORING_REGISTER_LAST) | |
10067 | return -EINVAL; | |
10068 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10069 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10070 | return -EACCES; | |
10071 | } | |
10072 | ||
071698e1 | 10073 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10074 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10075 | |
05f3fb3c JA |
10076 | /* |
10077 | * Drop uring mutex before waiting for references to exit. If | |
10078 | * another thread is currently inside io_uring_enter() it might | |
10079 | * need to grab the uring_lock to make progress. If we hold it | |
10080 | * here across the drain wait, then we can deadlock. It's safe | |
10081 | * to drop the mutex here, since no new references will come in | |
10082 | * after we've killed the percpu ref. | |
10083 | */ | |
10084 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10085 | do { |
10086 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10087 | if (!ret) | |
10088 | break; | |
ed6930c9 JA |
10089 | ret = io_run_task_work_sig(); |
10090 | if (ret < 0) | |
10091 | break; | |
af9c1a44 | 10092 | } while (1); |
05f3fb3c | 10093 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10094 | |
c150368b | 10095 | if (ret) { |
f70865db PB |
10096 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
10097 | return ret; | |
c150368b | 10098 | } |
05f3fb3c | 10099 | } |
edafccee JA |
10100 | |
10101 | switch (opcode) { | |
10102 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10103 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10104 | break; |
10105 | case IORING_UNREGISTER_BUFFERS: | |
10106 | ret = -EINVAL; | |
10107 | if (arg || nr_args) | |
10108 | break; | |
0a96bbe4 | 10109 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10110 | break; |
6b06314c | 10111 | case IORING_REGISTER_FILES: |
792e3582 | 10112 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10113 | break; |
10114 | case IORING_UNREGISTER_FILES: | |
10115 | ret = -EINVAL; | |
10116 | if (arg || nr_args) | |
10117 | break; | |
10118 | ret = io_sqe_files_unregister(ctx); | |
10119 | break; | |
c3a31e60 | 10120 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10121 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10122 | break; |
9b402849 | 10123 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10124 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10125 | ret = -EINVAL; |
10126 | if (nr_args != 1) | |
10127 | break; | |
10128 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10129 | if (ret) |
10130 | break; | |
10131 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10132 | ctx->eventfd_async = 1; | |
10133 | else | |
10134 | ctx->eventfd_async = 0; | |
9b402849 JA |
10135 | break; |
10136 | case IORING_UNREGISTER_EVENTFD: | |
10137 | ret = -EINVAL; | |
10138 | if (arg || nr_args) | |
10139 | break; | |
10140 | ret = io_eventfd_unregister(ctx); | |
10141 | break; | |
66f4af93 JA |
10142 | case IORING_REGISTER_PROBE: |
10143 | ret = -EINVAL; | |
10144 | if (!arg || nr_args > 256) | |
10145 | break; | |
10146 | ret = io_probe(ctx, arg, nr_args); | |
10147 | break; | |
071698e1 JA |
10148 | case IORING_REGISTER_PERSONALITY: |
10149 | ret = -EINVAL; | |
10150 | if (arg || nr_args) | |
10151 | break; | |
10152 | ret = io_register_personality(ctx); | |
10153 | break; | |
10154 | case IORING_UNREGISTER_PERSONALITY: | |
10155 | ret = -EINVAL; | |
10156 | if (arg) | |
10157 | break; | |
10158 | ret = io_unregister_personality(ctx, nr_args); | |
10159 | break; | |
7e84e1c7 SG |
10160 | case IORING_REGISTER_ENABLE_RINGS: |
10161 | ret = -EINVAL; | |
10162 | if (arg || nr_args) | |
10163 | break; | |
10164 | ret = io_register_enable_rings(ctx); | |
10165 | break; | |
21b55dbc SG |
10166 | case IORING_REGISTER_RESTRICTIONS: |
10167 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10168 | break; | |
992da01a PB |
10169 | case IORING_REGISTER_FILES2: |
10170 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10171 | break; | |
10172 | case IORING_REGISTER_FILES_UPDATE2: | |
10173 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10174 | IORING_RSRC_FILE); | |
10175 | break; | |
10176 | case IORING_REGISTER_BUFFERS2: | |
10177 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10178 | break; |
992da01a PB |
10179 | case IORING_REGISTER_BUFFERS_UPDATE: |
10180 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10181 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10182 | break; |
fe76421d JA |
10183 | case IORING_REGISTER_IOWQ_AFF: |
10184 | ret = -EINVAL; | |
10185 | if (!arg || !nr_args) | |
10186 | break; | |
10187 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
10188 | break; | |
10189 | case IORING_UNREGISTER_IOWQ_AFF: | |
10190 | ret = -EINVAL; | |
10191 | if (arg || nr_args) | |
10192 | break; | |
10193 | ret = io_unregister_iowq_aff(ctx); | |
10194 | break; | |
edafccee JA |
10195 | default: |
10196 | ret = -EINVAL; | |
10197 | break; | |
10198 | } | |
10199 | ||
071698e1 | 10200 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10201 | /* bring the ctx back to life */ |
05f3fb3c | 10202 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10203 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10204 | } |
edafccee JA |
10205 | return ret; |
10206 | } | |
10207 | ||
10208 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10209 | void __user *, arg, unsigned int, nr_args) | |
10210 | { | |
10211 | struct io_ring_ctx *ctx; | |
10212 | long ret = -EBADF; | |
10213 | struct fd f; | |
10214 | ||
10215 | f = fdget(fd); | |
10216 | if (!f.file) | |
10217 | return -EBADF; | |
10218 | ||
10219 | ret = -EOPNOTSUPP; | |
10220 | if (f.file->f_op != &io_uring_fops) | |
10221 | goto out_fput; | |
10222 | ||
10223 | ctx = f.file->private_data; | |
10224 | ||
b6c23dd5 PB |
10225 | io_run_task_work(); |
10226 | ||
edafccee JA |
10227 | mutex_lock(&ctx->uring_lock); |
10228 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10229 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10230 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10231 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10232 | out_fput: |
10233 | fdput(f); | |
10234 | return ret; | |
10235 | } | |
10236 | ||
2b188cc1 JA |
10237 | static int __init io_uring_init(void) |
10238 | { | |
d7f62e82 SM |
10239 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10240 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10241 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10242 | } while (0) | |
10243 | ||
10244 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10245 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10246 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10247 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10248 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10249 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10250 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10251 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10252 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10253 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10254 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10255 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10256 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10257 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10258 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10259 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10260 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10261 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10262 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10263 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10264 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10265 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10266 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10267 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10268 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10269 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10270 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10271 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10272 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 10273 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 10274 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 10275 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10276 | |
b0d658ec PB |
10277 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10278 | sizeof(struct io_uring_rsrc_update)); | |
10279 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10280 | sizeof(struct io_uring_rsrc_update2)); | |
10281 | /* should fit into one byte */ | |
10282 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10283 | ||
d3656344 | 10284 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10285 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
16340eab | 10286 | |
91f245d5 JA |
10287 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10288 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10289 | return 0; |
10290 | }; | |
10291 | __initcall(io_uring_init); |