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; | |
915b3dde | 437 | bool poll_multi_queue; |
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 */ | |
9011bf9a PB |
468 | struct llist_head fallback_llist; |
469 | struct delayed_work fallback_work; | |
b986af7e PB |
470 | struct work_struct exit_work; |
471 | struct list_head tctx_list; | |
472 | struct completion ref_comp; | |
473 | }; | |
2b188cc1 JA |
474 | }; |
475 | ||
53e043b2 SM |
476 | struct io_uring_task { |
477 | /* submission side */ | |
09899b19 | 478 | int cached_refs; |
53e043b2 SM |
479 | struct xarray xa; |
480 | struct wait_queue_head wait; | |
ee53fb2b SM |
481 | const struct io_ring_ctx *last; |
482 | struct io_wq *io_wq; | |
53e043b2 | 483 | struct percpu_counter inflight; |
b303fe2e | 484 | atomic_t inflight_tracked; |
53e043b2 | 485 | atomic_t in_idle; |
53e043b2 SM |
486 | |
487 | spinlock_t task_lock; | |
488 | struct io_wq_work_list task_list; | |
489 | unsigned long task_state; | |
490 | struct callback_head task_work; | |
491 | }; | |
492 | ||
09bb8394 JA |
493 | /* |
494 | * First field must be the file pointer in all the | |
495 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
496 | */ | |
221c5eb2 JA |
497 | struct io_poll_iocb { |
498 | struct file *file; | |
018043be | 499 | struct wait_queue_head *head; |
221c5eb2 | 500 | __poll_t events; |
8c838788 | 501 | bool done; |
221c5eb2 | 502 | bool canceled; |
392edb45 | 503 | struct wait_queue_entry wait; |
221c5eb2 JA |
504 | }; |
505 | ||
9d805892 | 506 | struct io_poll_update { |
018043be | 507 | struct file *file; |
9d805892 PB |
508 | u64 old_user_data; |
509 | u64 new_user_data; | |
510 | __poll_t events; | |
b69de288 JA |
511 | bool update_events; |
512 | bool update_user_data; | |
018043be PB |
513 | }; |
514 | ||
b5dba59e JA |
515 | struct io_close { |
516 | struct file *file; | |
b5dba59e JA |
517 | int fd; |
518 | }; | |
519 | ||
ad8a48ac JA |
520 | struct io_timeout_data { |
521 | struct io_kiocb *req; | |
522 | struct hrtimer timer; | |
523 | struct timespec64 ts; | |
524 | enum hrtimer_mode mode; | |
525 | }; | |
526 | ||
8ed8d3c3 JA |
527 | struct io_accept { |
528 | struct file *file; | |
529 | struct sockaddr __user *addr; | |
530 | int __user *addr_len; | |
531 | int flags; | |
09952e3e | 532 | unsigned long nofile; |
8ed8d3c3 JA |
533 | }; |
534 | ||
535 | struct io_sync { | |
536 | struct file *file; | |
537 | loff_t len; | |
538 | loff_t off; | |
539 | int flags; | |
d63d1b5e | 540 | int mode; |
8ed8d3c3 JA |
541 | }; |
542 | ||
fbf23849 JA |
543 | struct io_cancel { |
544 | struct file *file; | |
545 | u64 addr; | |
546 | }; | |
547 | ||
b29472ee JA |
548 | struct io_timeout { |
549 | struct file *file; | |
bfe68a22 PB |
550 | u32 off; |
551 | u32 target_seq; | |
135fcde8 | 552 | struct list_head list; |
90cd7e42 PB |
553 | /* head of the link, used by linked timeouts only */ |
554 | struct io_kiocb *head; | |
b29472ee JA |
555 | }; |
556 | ||
0bdf7a2d PB |
557 | struct io_timeout_rem { |
558 | struct file *file; | |
559 | u64 addr; | |
9c8e11b3 PB |
560 | |
561 | /* timeout update */ | |
562 | struct timespec64 ts; | |
563 | u32 flags; | |
0bdf7a2d PB |
564 | }; |
565 | ||
9adbd45d JA |
566 | struct io_rw { |
567 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
568 | struct kiocb kiocb; | |
569 | u64 addr; | |
570 | u64 len; | |
571 | }; | |
572 | ||
3fbb51c1 JA |
573 | struct io_connect { |
574 | struct file *file; | |
575 | struct sockaddr __user *addr; | |
576 | int addr_len; | |
577 | }; | |
578 | ||
e47293fd JA |
579 | struct io_sr_msg { |
580 | struct file *file; | |
fddaface | 581 | union { |
4af3417a PB |
582 | struct compat_msghdr __user *umsg_compat; |
583 | struct user_msghdr __user *umsg; | |
584 | void __user *buf; | |
fddaface | 585 | }; |
e47293fd | 586 | int msg_flags; |
bcda7baa | 587 | int bgid; |
fddaface | 588 | size_t len; |
bcda7baa | 589 | struct io_buffer *kbuf; |
e47293fd JA |
590 | }; |
591 | ||
15b71abe JA |
592 | struct io_open { |
593 | struct file *file; | |
594 | int dfd; | |
15b71abe | 595 | struct filename *filename; |
c12cedf2 | 596 | struct open_how how; |
4022e7af | 597 | unsigned long nofile; |
15b71abe JA |
598 | }; |
599 | ||
269bbe5f | 600 | struct io_rsrc_update { |
05f3fb3c JA |
601 | struct file *file; |
602 | u64 arg; | |
603 | u32 nr_args; | |
604 | u32 offset; | |
605 | }; | |
606 | ||
4840e418 JA |
607 | struct io_fadvise { |
608 | struct file *file; | |
609 | u64 offset; | |
610 | u32 len; | |
611 | u32 advice; | |
612 | }; | |
613 | ||
c1ca757b JA |
614 | struct io_madvise { |
615 | struct file *file; | |
616 | u64 addr; | |
617 | u32 len; | |
618 | u32 advice; | |
619 | }; | |
620 | ||
3e4827b0 JA |
621 | struct io_epoll { |
622 | struct file *file; | |
623 | int epfd; | |
624 | int op; | |
625 | int fd; | |
626 | struct epoll_event event; | |
e47293fd JA |
627 | }; |
628 | ||
7d67af2c PB |
629 | struct io_splice { |
630 | struct file *file_out; | |
631 | struct file *file_in; | |
632 | loff_t off_out; | |
633 | loff_t off_in; | |
634 | u64 len; | |
635 | unsigned int flags; | |
636 | }; | |
637 | ||
ddf0322d JA |
638 | struct io_provide_buf { |
639 | struct file *file; | |
640 | __u64 addr; | |
38134ada | 641 | __u32 len; |
ddf0322d JA |
642 | __u32 bgid; |
643 | __u16 nbufs; | |
644 | __u16 bid; | |
645 | }; | |
646 | ||
1d9e1288 BM |
647 | struct io_statx { |
648 | struct file *file; | |
649 | int dfd; | |
650 | unsigned int mask; | |
651 | unsigned int flags; | |
e62753e4 | 652 | const char __user *filename; |
1d9e1288 BM |
653 | struct statx __user *buffer; |
654 | }; | |
655 | ||
36f4fa68 JA |
656 | struct io_shutdown { |
657 | struct file *file; | |
658 | int how; | |
659 | }; | |
660 | ||
80a261fd JA |
661 | struct io_rename { |
662 | struct file *file; | |
663 | int old_dfd; | |
664 | int new_dfd; | |
665 | struct filename *oldpath; | |
666 | struct filename *newpath; | |
667 | int flags; | |
668 | }; | |
669 | ||
14a1143b JA |
670 | struct io_unlink { |
671 | struct file *file; | |
672 | int dfd; | |
673 | int flags; | |
674 | struct filename *filename; | |
675 | }; | |
676 | ||
3ca405eb PB |
677 | struct io_completion { |
678 | struct file *file; | |
679 | struct list_head list; | |
8c3f9cd1 | 680 | u32 cflags; |
3ca405eb PB |
681 | }; |
682 | ||
f499a021 JA |
683 | struct io_async_connect { |
684 | struct sockaddr_storage address; | |
685 | }; | |
686 | ||
03b1230c JA |
687 | struct io_async_msghdr { |
688 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
689 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
690 | struct iovec *free_iov; | |
03b1230c JA |
691 | struct sockaddr __user *uaddr; |
692 | struct msghdr msg; | |
b537916c | 693 | struct sockaddr_storage addr; |
03b1230c JA |
694 | }; |
695 | ||
f67676d1 JA |
696 | struct io_async_rw { |
697 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
698 | const struct iovec *free_iovec; |
699 | struct iov_iter iter; | |
227c0c96 | 700 | size_t bytes_done; |
bcf5a063 | 701 | struct wait_page_queue wpq; |
f67676d1 JA |
702 | }; |
703 | ||
6b47ee6e PB |
704 | enum { |
705 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
706 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
707 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
708 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
709 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 710 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 711 | |
dddca226 | 712 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 713 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
714 | REQ_F_INFLIGHT_BIT, |
715 | REQ_F_CUR_POS_BIT, | |
716 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 717 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 718 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 719 | REQ_F_POLLED_BIT, |
bcda7baa | 720 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 721 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 722 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 723 | REQ_F_REISSUE_BIT, |
8c130827 | 724 | REQ_F_DONT_REISSUE_BIT, |
b8e64b53 | 725 | REQ_F_CREDS_BIT, |
7b29f92d JA |
726 | /* keep async read/write and isreg together and in order */ |
727 | REQ_F_ASYNC_READ_BIT, | |
728 | REQ_F_ASYNC_WRITE_BIT, | |
729 | REQ_F_ISREG_BIT, | |
84557871 JA |
730 | |
731 | /* not a real bit, just to check we're not overflowing the space */ | |
732 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
733 | }; |
734 | ||
735 | enum { | |
736 | /* ctx owns file */ | |
737 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
738 | /* drain existing IO first */ | |
739 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
740 | /* linked sqes */ | |
741 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
742 | /* doesn't sever on completion < 0 */ | |
743 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
744 | /* IOSQE_ASYNC */ | |
745 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
746 | /* IOSQE_BUFFER_SELECT */ |
747 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 748 | |
6b47ee6e | 749 | /* fail rest of links */ |
93d2bcd2 | 750 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 751 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
752 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
753 | /* read/write uses file position */ | |
754 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
755 | /* must not punt to workers */ | |
756 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 757 | /* has or had linked timeout */ |
6b47ee6e | 758 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
759 | /* needs cleanup */ |
760 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
761 | /* already went through poll handler */ |
762 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
763 | /* buffer already selected */ |
764 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
765 | /* linked timeout is active, i.e. prepared by link's head */ |
766 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
767 | /* completion is deferred through io_comp_state */ |
768 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
769 | /* caller should reissue async */ |
770 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
771 | /* don't attempt request reissue, see io_rw_reissue() */ |
772 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
773 | /* supports async reads */ |
774 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
775 | /* supports async writes */ | |
776 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
777 | /* regular file */ | |
778 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
779 | /* has creds assigned */ |
780 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
d7718a9d JA |
781 | }; |
782 | ||
783 | struct async_poll { | |
784 | struct io_poll_iocb poll; | |
807abcb0 | 785 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
786 | }; |
787 | ||
5b0a6acc PB |
788 | typedef void (*io_req_tw_func_t)(struct io_kiocb *req); |
789 | ||
7cbf1722 | 790 | struct io_task_work { |
5b0a6acc PB |
791 | union { |
792 | struct io_wq_work_node node; | |
793 | struct llist_node fallback_node; | |
794 | }; | |
795 | io_req_tw_func_t func; | |
7cbf1722 JA |
796 | }; |
797 | ||
992da01a PB |
798 | enum { |
799 | IORING_RSRC_FILE = 0, | |
800 | IORING_RSRC_BUFFER = 1, | |
801 | }; | |
802 | ||
09bb8394 JA |
803 | /* |
804 | * NOTE! Each of the iocb union members has the file pointer | |
805 | * as the first entry in their struct definition. So you can | |
806 | * access the file pointer through any of the sub-structs, | |
807 | * or directly as just 'ki_filp' in this struct. | |
808 | */ | |
2b188cc1 | 809 | struct io_kiocb { |
221c5eb2 | 810 | union { |
09bb8394 | 811 | struct file *file; |
9adbd45d | 812 | struct io_rw rw; |
221c5eb2 | 813 | struct io_poll_iocb poll; |
9d805892 | 814 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
815 | struct io_accept accept; |
816 | struct io_sync sync; | |
fbf23849 | 817 | struct io_cancel cancel; |
b29472ee | 818 | struct io_timeout timeout; |
0bdf7a2d | 819 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 820 | struct io_connect connect; |
e47293fd | 821 | struct io_sr_msg sr_msg; |
15b71abe | 822 | struct io_open open; |
b5dba59e | 823 | struct io_close close; |
269bbe5f | 824 | struct io_rsrc_update rsrc_update; |
4840e418 | 825 | struct io_fadvise fadvise; |
c1ca757b | 826 | struct io_madvise madvise; |
3e4827b0 | 827 | struct io_epoll epoll; |
7d67af2c | 828 | struct io_splice splice; |
ddf0322d | 829 | struct io_provide_buf pbuf; |
1d9e1288 | 830 | struct io_statx statx; |
36f4fa68 | 831 | struct io_shutdown shutdown; |
80a261fd | 832 | struct io_rename rename; |
14a1143b | 833 | struct io_unlink unlink; |
3ca405eb PB |
834 | /* use only after cleaning per-op data, see io_clean_op() */ |
835 | struct io_completion compl; | |
221c5eb2 | 836 | }; |
2b188cc1 | 837 | |
e8c2bc1f JA |
838 | /* opcode allocated if it needs to store data for async defer */ |
839 | void *async_data; | |
d625c6ee | 840 | u8 opcode; |
65a6543d XW |
841 | /* polled IO has completed */ |
842 | u8 iopoll_completed; | |
2b188cc1 | 843 | |
4f4eeba8 | 844 | u16 buf_index; |
9cf7c104 | 845 | u32 result; |
4f4eeba8 | 846 | |
010e8e6b PB |
847 | struct io_ring_ctx *ctx; |
848 | unsigned int flags; | |
abc54d63 | 849 | atomic_t refs; |
010e8e6b PB |
850 | struct task_struct *task; |
851 | u64 user_data; | |
d7718a9d | 852 | |
f2f87370 | 853 | struct io_kiocb *link; |
269bbe5f | 854 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 855 | |
b303fe2e | 856 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 857 | struct list_head inflight_entry; |
5b0a6acc | 858 | struct io_task_work io_task_work; |
010e8e6b PB |
859 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
860 | struct hlist_node hash_node; | |
861 | struct async_poll *apoll; | |
862 | struct io_wq_work work; | |
fe7e3257 | 863 | const struct cred *creds; |
c10d1f98 | 864 | |
eae071c9 PB |
865 | /* store used ubuf, so we can prevent reloading */ |
866 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 867 | }; |
05589553 | 868 | |
13bf43f5 PB |
869 | struct io_tctx_node { |
870 | struct list_head ctx_node; | |
871 | struct task_struct *task; | |
13bf43f5 PB |
872 | struct io_ring_ctx *ctx; |
873 | }; | |
874 | ||
27dc8338 PB |
875 | struct io_defer_entry { |
876 | struct list_head list; | |
877 | struct io_kiocb *req; | |
9cf7c104 | 878 | u32 seq; |
2b188cc1 JA |
879 | }; |
880 | ||
d3656344 | 881 | struct io_op_def { |
d3656344 JA |
882 | /* needs req->file assigned */ |
883 | unsigned needs_file : 1; | |
d3656344 JA |
884 | /* hash wq insertion if file is a regular file */ |
885 | unsigned hash_reg_file : 1; | |
886 | /* unbound wq insertion if file is a non-regular file */ | |
887 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
888 | /* opcode is not supported by this kernel */ |
889 | unsigned not_supported : 1; | |
8a72758c JA |
890 | /* set if opcode supports polled "wait" */ |
891 | unsigned pollin : 1; | |
892 | unsigned pollout : 1; | |
bcda7baa JA |
893 | /* op supports buffer selection */ |
894 | unsigned buffer_select : 1; | |
26f0505a PB |
895 | /* do prep async if is going to be punted */ |
896 | unsigned needs_async_setup : 1; | |
27926b68 JA |
897 | /* should block plug */ |
898 | unsigned plug : 1; | |
e8c2bc1f JA |
899 | /* size of async data needed, if any */ |
900 | unsigned short async_size; | |
d3656344 JA |
901 | }; |
902 | ||
0918682b | 903 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
904 | [IORING_OP_NOP] = {}, |
905 | [IORING_OP_READV] = { | |
d3656344 JA |
906 | .needs_file = 1, |
907 | .unbound_nonreg_file = 1, | |
8a72758c | 908 | .pollin = 1, |
4d954c25 | 909 | .buffer_select = 1, |
26f0505a | 910 | .needs_async_setup = 1, |
27926b68 | 911 | .plug = 1, |
e8c2bc1f | 912 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 913 | }, |
0463b6c5 | 914 | [IORING_OP_WRITEV] = { |
d3656344 JA |
915 | .needs_file = 1, |
916 | .hash_reg_file = 1, | |
917 | .unbound_nonreg_file = 1, | |
8a72758c | 918 | .pollout = 1, |
26f0505a | 919 | .needs_async_setup = 1, |
27926b68 | 920 | .plug = 1, |
e8c2bc1f | 921 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 922 | }, |
0463b6c5 | 923 | [IORING_OP_FSYNC] = { |
d3656344 JA |
924 | .needs_file = 1, |
925 | }, | |
0463b6c5 | 926 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
927 | .needs_file = 1, |
928 | .unbound_nonreg_file = 1, | |
8a72758c | 929 | .pollin = 1, |
27926b68 | 930 | .plug = 1, |
e8c2bc1f | 931 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 932 | }, |
0463b6c5 | 933 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
934 | .needs_file = 1, |
935 | .hash_reg_file = 1, | |
936 | .unbound_nonreg_file = 1, | |
8a72758c | 937 | .pollout = 1, |
27926b68 | 938 | .plug = 1, |
e8c2bc1f | 939 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 940 | }, |
0463b6c5 | 941 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
942 | .needs_file = 1, |
943 | .unbound_nonreg_file = 1, | |
944 | }, | |
0463b6c5 PB |
945 | [IORING_OP_POLL_REMOVE] = {}, |
946 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
947 | .needs_file = 1, |
948 | }, | |
0463b6c5 | 949 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
950 | .needs_file = 1, |
951 | .unbound_nonreg_file = 1, | |
8a72758c | 952 | .pollout = 1, |
26f0505a | 953 | .needs_async_setup = 1, |
e8c2bc1f | 954 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 955 | }, |
0463b6c5 | 956 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
957 | .needs_file = 1, |
958 | .unbound_nonreg_file = 1, | |
8a72758c | 959 | .pollin = 1, |
52de1fe1 | 960 | .buffer_select = 1, |
26f0505a | 961 | .needs_async_setup = 1, |
e8c2bc1f | 962 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 963 | }, |
0463b6c5 | 964 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 965 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 966 | }, |
9c8e11b3 PB |
967 | [IORING_OP_TIMEOUT_REMOVE] = { |
968 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 969 | }, |
0463b6c5 | 970 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
971 | .needs_file = 1, |
972 | .unbound_nonreg_file = 1, | |
8a72758c | 973 | .pollin = 1, |
d3656344 | 974 | }, |
0463b6c5 PB |
975 | [IORING_OP_ASYNC_CANCEL] = {}, |
976 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 977 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 978 | }, |
0463b6c5 | 979 | [IORING_OP_CONNECT] = { |
d3656344 JA |
980 | .needs_file = 1, |
981 | .unbound_nonreg_file = 1, | |
8a72758c | 982 | .pollout = 1, |
26f0505a | 983 | .needs_async_setup = 1, |
e8c2bc1f | 984 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 985 | }, |
0463b6c5 | 986 | [IORING_OP_FALLOCATE] = { |
d3656344 | 987 | .needs_file = 1, |
d3656344 | 988 | }, |
44526bed JA |
989 | [IORING_OP_OPENAT] = {}, |
990 | [IORING_OP_CLOSE] = {}, | |
991 | [IORING_OP_FILES_UPDATE] = {}, | |
992 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 993 | [IORING_OP_READ] = { |
3a6820f2 JA |
994 | .needs_file = 1, |
995 | .unbound_nonreg_file = 1, | |
8a72758c | 996 | .pollin = 1, |
bcda7baa | 997 | .buffer_select = 1, |
27926b68 | 998 | .plug = 1, |
e8c2bc1f | 999 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1000 | }, |
0463b6c5 | 1001 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
1002 | .needs_file = 1, |
1003 | .unbound_nonreg_file = 1, | |
8a72758c | 1004 | .pollout = 1, |
27926b68 | 1005 | .plug = 1, |
e8c2bc1f | 1006 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1007 | }, |
0463b6c5 | 1008 | [IORING_OP_FADVISE] = { |
4840e418 | 1009 | .needs_file = 1, |
c1ca757b | 1010 | }, |
44526bed | 1011 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1012 | [IORING_OP_SEND] = { |
fddaface JA |
1013 | .needs_file = 1, |
1014 | .unbound_nonreg_file = 1, | |
8a72758c | 1015 | .pollout = 1, |
fddaface | 1016 | }, |
0463b6c5 | 1017 | [IORING_OP_RECV] = { |
fddaface JA |
1018 | .needs_file = 1, |
1019 | .unbound_nonreg_file = 1, | |
8a72758c | 1020 | .pollin = 1, |
bcda7baa | 1021 | .buffer_select = 1, |
fddaface | 1022 | }, |
0463b6c5 | 1023 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1024 | }, |
3e4827b0 JA |
1025 | [IORING_OP_EPOLL_CTL] = { |
1026 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1027 | }, |
7d67af2c PB |
1028 | [IORING_OP_SPLICE] = { |
1029 | .needs_file = 1, | |
1030 | .hash_reg_file = 1, | |
1031 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1032 | }, |
1033 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1034 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1035 | [IORING_OP_TEE] = { |
1036 | .needs_file = 1, | |
1037 | .hash_reg_file = 1, | |
1038 | .unbound_nonreg_file = 1, | |
1039 | }, | |
36f4fa68 JA |
1040 | [IORING_OP_SHUTDOWN] = { |
1041 | .needs_file = 1, | |
1042 | }, | |
44526bed JA |
1043 | [IORING_OP_RENAMEAT] = {}, |
1044 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1045 | }; |
1046 | ||
7a612350 | 1047 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1048 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1049 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1050 | struct task_struct *task, | |
3dd0c97a | 1051 | bool cancel_all); |
78cc687b | 1052 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
b895c9a6 | 1053 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1054 | |
d4d19c19 PB |
1055 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1056 | long res, unsigned int cflags); | |
ec9c02ad | 1057 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1058 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1059 | static void io_dismantle_req(struct io_kiocb *req); |
1060 | static void io_put_task(struct task_struct *task, int nr); | |
94ae5e77 JA |
1061 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1062 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
fdecb662 | 1063 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1064 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1065 | unsigned nr_args); |
68fb8979 | 1066 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1067 | static struct file *io_file_get(struct io_submit_state *state, |
1068 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1069 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1070 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1071 | |
907d1df3 | 1072 | static void io_req_task_queue(struct io_kiocb *req); |
2a2758f2 | 1073 | static void io_submit_flush_completions(struct io_ring_ctx *ctx); |
5082620f | 1074 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1075 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1076 | |
9011bf9a PB |
1077 | static void io_fallback_req_func(struct work_struct *unused); |
1078 | ||
2b188cc1 JA |
1079 | static struct kmem_cache *req_cachep; |
1080 | ||
0918682b | 1081 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1082 | |
1083 | struct sock *io_uring_get_socket(struct file *file) | |
1084 | { | |
1085 | #if defined(CONFIG_UNIX) | |
1086 | if (file->f_op == &io_uring_fops) { | |
1087 | struct io_ring_ctx *ctx = file->private_data; | |
1088 | ||
1089 | return ctx->ring_sock->sk; | |
1090 | } | |
1091 | #endif | |
1092 | return NULL; | |
1093 | } | |
1094 | EXPORT_SYMBOL(io_uring_get_socket); | |
1095 | ||
f2f87370 PB |
1096 | #define io_for_each_link(pos, head) \ |
1097 | for (pos = (head); pos; pos = pos->link) | |
1098 | ||
b895c9a6 | 1099 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1100 | { |
1101 | struct io_ring_ctx *ctx = req->ctx; | |
1102 | ||
269bbe5f | 1103 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1104 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1105 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1106 | } |
1107 | } | |
1108 | ||
f70865db PB |
1109 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1110 | { | |
1111 | bool got = percpu_ref_tryget(ref); | |
1112 | ||
1113 | /* already at zero, wait for ->release() */ | |
1114 | if (!got) | |
1115 | wait_for_completion(compl); | |
1116 | percpu_ref_resurrect(ref); | |
1117 | if (got) | |
1118 | percpu_ref_put(ref); | |
1119 | } | |
1120 | ||
3dd0c97a PB |
1121 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1122 | bool cancel_all) | |
08d23634 PB |
1123 | { |
1124 | struct io_kiocb *req; | |
1125 | ||
68207680 | 1126 | if (task && head->task != task) |
08d23634 | 1127 | return false; |
3dd0c97a | 1128 | if (cancel_all) |
08d23634 PB |
1129 | return true; |
1130 | ||
1131 | io_for_each_link(req, head) { | |
b05a1bcd | 1132 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1133 | return true; |
08d23634 PB |
1134 | } |
1135 | return false; | |
1136 | } | |
1137 | ||
93d2bcd2 | 1138 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1139 | { |
93d2bcd2 | 1140 | req->flags |= REQ_F_FAIL; |
c40f6379 | 1141 | } |
4a38aed2 | 1142 | |
2b188cc1 JA |
1143 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1144 | { | |
1145 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1146 | ||
0f158b4c | 1147 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1148 | } |
1149 | ||
8eb7e2d0 PB |
1150 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1151 | { | |
1152 | return !req->timeout.off; | |
1153 | } | |
1154 | ||
2b188cc1 JA |
1155 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1156 | { | |
1157 | struct io_ring_ctx *ctx; | |
78076bb6 | 1158 | int hash_bits; |
2b188cc1 JA |
1159 | |
1160 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1161 | if (!ctx) | |
1162 | return NULL; | |
1163 | ||
78076bb6 JA |
1164 | /* |
1165 | * Use 5 bits less than the max cq entries, that should give us around | |
1166 | * 32 entries per hash list if totally full and uniformly spread. | |
1167 | */ | |
1168 | hash_bits = ilog2(p->cq_entries); | |
1169 | hash_bits -= 5; | |
1170 | if (hash_bits <= 0) | |
1171 | hash_bits = 1; | |
1172 | ctx->cancel_hash_bits = hash_bits; | |
1173 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1174 | GFP_KERNEL); | |
1175 | if (!ctx->cancel_hash) | |
1176 | goto err; | |
1177 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1178 | ||
6224843d PB |
1179 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1180 | if (!ctx->dummy_ubuf) | |
1181 | goto err; | |
1182 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1183 | ctx->dummy_ubuf->ubuf = -1UL; | |
1184 | ||
21482896 | 1185 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1186 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1187 | goto err; | |
2b188cc1 JA |
1188 | |
1189 | ctx->flags = p->flags; | |
90554200 | 1190 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1191 | INIT_LIST_HEAD(&ctx->sqd_list); |
311997b3 | 1192 | init_waitqueue_head(&ctx->poll_wait); |
1d7bb1d5 | 1193 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1194 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1195 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1196 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1197 | mutex_init(&ctx->uring_lock); |
311997b3 | 1198 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1199 | spin_lock_init(&ctx->completion_lock); |
540e32a0 | 1200 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1201 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1202 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1203 | spin_lock_init(&ctx->rsrc_ref_lock); |
1204 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1205 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1206 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1207 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1208 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
d0acdee2 | 1209 | INIT_LIST_HEAD(&ctx->locked_free_list); |
9011bf9a | 1210 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
2b188cc1 | 1211 | return ctx; |
206aefde | 1212 | err: |
6224843d | 1213 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1214 | kfree(ctx->cancel_hash); |
206aefde JA |
1215 | kfree(ctx); |
1216 | return NULL; | |
2b188cc1 JA |
1217 | } |
1218 | ||
8f6ed49a PB |
1219 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1220 | { | |
1221 | struct io_rings *r = ctx->rings; | |
1222 | ||
1223 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1224 | ctx->cq_extra--; | |
1225 | } | |
1226 | ||
9cf7c104 | 1227 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1228 | { |
2bc9930e JA |
1229 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1230 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1231 | |
8f6ed49a | 1232 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1233 | } |
de0617e4 | 1234 | |
9d858b21 | 1235 | return false; |
de0617e4 JA |
1236 | } |
1237 | ||
ce3d5aae PB |
1238 | static void io_req_track_inflight(struct io_kiocb *req) |
1239 | { | |
ce3d5aae | 1240 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1241 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1242 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1243 | } |
1244 | } | |
1245 | ||
1e6fa521 JA |
1246 | static void io_prep_async_work(struct io_kiocb *req) |
1247 | { | |
1248 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1249 | struct io_ring_ctx *ctx = req->ctx; |
1250 | ||
b8e64b53 PB |
1251 | if (!(req->flags & REQ_F_CREDS)) { |
1252 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1253 | req->creds = get_current_cred(); |
b8e64b53 | 1254 | } |
003e8dcc | 1255 | |
e1d675df PB |
1256 | req->work.list.next = NULL; |
1257 | req->work.flags = 0; | |
feaadc4f PB |
1258 | if (req->flags & REQ_F_FORCE_ASYNC) |
1259 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1260 | ||
1e6fa521 JA |
1261 | if (req->flags & REQ_F_ISREG) { |
1262 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1263 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1264 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1265 | if (def->unbound_nonreg_file) |
1266 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1267 | } | |
e1d675df PB |
1268 | |
1269 | switch (req->opcode) { | |
1270 | case IORING_OP_SPLICE: | |
1271 | case IORING_OP_TEE: | |
e1d675df PB |
1272 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1273 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1274 | break; | |
1275 | } | |
561fb04a | 1276 | } |
cccf0ee8 | 1277 | |
cbdcb435 | 1278 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1279 | { |
cbdcb435 | 1280 | struct io_kiocb *cur; |
54a91f3b | 1281 | |
f2f87370 PB |
1282 | io_for_each_link(cur, req) |
1283 | io_prep_async_work(cur); | |
561fb04a JA |
1284 | } |
1285 | ||
ebf93667 | 1286 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1287 | { |
a197f664 | 1288 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1289 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1290 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1291 | |
3bfe6106 JA |
1292 | BUG_ON(!tctx); |
1293 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1294 | |
cbdcb435 PB |
1295 | /* init ->work of the whole link before punting */ |
1296 | io_prep_async_link(req); | |
d07f1e8a PB |
1297 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1298 | &req->work, req->flags); | |
ebf93667 | 1299 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1300 | if (link) |
1301 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1302 | } |
1303 | ||
1ee4160c | 1304 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1305 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1306 | { |
e8c2bc1f | 1307 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1308 | |
fd9c7bc5 | 1309 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1310 | atomic_set(&req->ctx->cq_timeouts, |
1311 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1312 | list_del_init(&req->timeout.list); |
d4d19c19 | 1313 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
216578e5 | 1314 | io_put_req_deferred(req, 1); |
5262f567 JA |
1315 | } |
1316 | } | |
1317 | ||
441b8a78 | 1318 | static void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1319 | { |
441b8a78 | 1320 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1321 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1322 | struct io_defer_entry, list); | |
de0617e4 | 1323 | |
9cf7c104 | 1324 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1325 | break; |
27dc8338 | 1326 | list_del_init(&de->list); |
907d1df3 | 1327 | io_req_task_queue(de->req); |
27dc8338 | 1328 | kfree(de); |
441b8a78 | 1329 | } |
04518945 PB |
1330 | } |
1331 | ||
360428f8 | 1332 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1333 | { |
441b8a78 | 1334 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1335 | |
f18ee4cf | 1336 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1337 | u32 events_needed, events_got; |
360428f8 | 1338 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1339 | struct io_kiocb, timeout.list); |
de0617e4 | 1340 | |
8eb7e2d0 | 1341 | if (io_is_timeout_noseq(req)) |
360428f8 | 1342 | break; |
f010505b MDG |
1343 | |
1344 | /* | |
1345 | * Since seq can easily wrap around over time, subtract | |
1346 | * the last seq at which timeouts were flushed before comparing. | |
1347 | * Assuming not more than 2^31-1 events have happened since, | |
1348 | * these subtractions won't have wrapped, so we can check if | |
1349 | * target is in [last_seq, current_seq] by comparing the two. | |
1350 | */ | |
1351 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1352 | events_got = seq - ctx->cq_last_tm_flush; | |
1353 | if (events_got < events_needed) | |
360428f8 | 1354 | break; |
bfe68a22 | 1355 | |
135fcde8 | 1356 | list_del_init(&req->timeout.list); |
1ee4160c | 1357 | io_kill_timeout(req, 0); |
f18ee4cf | 1358 | } |
f010505b | 1359 | ctx->cq_last_tm_flush = seq; |
360428f8 | 1360 | } |
5262f567 | 1361 | |
2335f6f5 | 1362 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1363 | { |
2335f6f5 PB |
1364 | if (ctx->off_timeout_used) |
1365 | io_flush_timeouts(ctx); | |
1366 | if (ctx->drain_active) | |
1367 | io_queue_deferred(ctx); | |
1368 | } | |
1369 | ||
1370 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) | |
1371 | { | |
1372 | if (unlikely(ctx->off_timeout_used || ctx->drain_active)) | |
1373 | __io_commit_cqring_flush(ctx); | |
ec30e04b PB |
1374 | /* order cqe stores with ring update */ |
1375 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 JA |
1376 | } |
1377 | ||
90554200 JA |
1378 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1379 | { | |
1380 | struct io_rings *r = ctx->rings; | |
1381 | ||
a566c556 | 1382 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1383 | } |
1384 | ||
888aae2e PB |
1385 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1386 | { | |
1387 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1388 | } | |
1389 | ||
d068b506 | 1390 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1391 | { |
75b28aff | 1392 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1393 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1394 | |
115e12e5 SB |
1395 | /* |
1396 | * writes to the cq entry need to come after reading head; the | |
1397 | * control dependency is enough as we're using WRITE_ONCE to | |
1398 | * fill the cq entry | |
1399 | */ | |
a566c556 | 1400 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1401 | return NULL; |
1402 | ||
888aae2e | 1403 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1404 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1405 | } |
1406 | ||
f2842ab5 JA |
1407 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1408 | { | |
44c769de | 1409 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1410 | return false; |
7e55a19c SG |
1411 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1412 | return false; | |
44c769de | 1413 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1414 | } |
1415 | ||
b41e9852 | 1416 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1417 | { |
b1445e59 PB |
1418 | /* see waitqueue_active() comment */ |
1419 | smp_mb(); | |
1420 | ||
311997b3 PB |
1421 | if (waitqueue_active(&ctx->cq_wait)) |
1422 | wake_up(&ctx->cq_wait); | |
534ca6d6 JA |
1423 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1424 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1425 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1426 | eventfd_signal(ctx->cq_ev_fd, 1); |
311997b3 PB |
1427 | if (waitqueue_active(&ctx->poll_wait)) { |
1428 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1429 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1430 | } | |
1d7bb1d5 JA |
1431 | } |
1432 | ||
80c18e4a PB |
1433 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1434 | { | |
b1445e59 PB |
1435 | /* see waitqueue_active() comment */ |
1436 | smp_mb(); | |
1437 | ||
80c18e4a | 1438 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
311997b3 PB |
1439 | if (waitqueue_active(&ctx->cq_wait)) |
1440 | wake_up(&ctx->cq_wait); | |
80c18e4a PB |
1441 | } |
1442 | if (io_should_trigger_evfd(ctx)) | |
1443 | eventfd_signal(ctx->cq_ev_fd, 1); | |
311997b3 PB |
1444 | if (waitqueue_active(&ctx->poll_wait)) { |
1445 | wake_up_interruptible(&ctx->poll_wait); | |
4aa84f2f PB |
1446 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); |
1447 | } | |
80c18e4a PB |
1448 | } |
1449 | ||
c4a2ed72 | 1450 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1451 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1452 | { |
1d7bb1d5 | 1453 | unsigned long flags; |
b18032bb | 1454 | bool all_flushed, posted; |
1d7bb1d5 | 1455 | |
a566c556 | 1456 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1457 | return false; |
1d7bb1d5 | 1458 | |
b18032bb | 1459 | posted = false; |
1d7bb1d5 | 1460 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae | 1461 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1462 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1463 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1464 | |
1d7bb1d5 JA |
1465 | if (!cqe && !force) |
1466 | break; | |
6c2450ae PB |
1467 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1468 | struct io_overflow_cqe, list); | |
1469 | if (cqe) | |
1470 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1471 | else | |
8f6ed49a PB |
1472 | io_account_cq_overflow(ctx); |
1473 | ||
b18032bb | 1474 | posted = true; |
6c2450ae PB |
1475 | list_del(&ocqe->list); |
1476 | kfree(ocqe); | |
1d7bb1d5 JA |
1477 | } |
1478 | ||
09e88404 PB |
1479 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1480 | if (all_flushed) { | |
5ed7a37d | 1481 | clear_bit(0, &ctx->check_cq_overflow); |
09e88404 PB |
1482 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; |
1483 | } | |
46930143 | 1484 | |
b18032bb JA |
1485 | if (posted) |
1486 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1487 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1488 | if (posted) |
1489 | io_cqring_ev_posted(ctx); | |
09e88404 | 1490 | return all_flushed; |
1d7bb1d5 JA |
1491 | } |
1492 | ||
6c2450ae | 1493 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1494 | { |
ca0a2651 JA |
1495 | bool ret = true; |
1496 | ||
5ed7a37d | 1497 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1498 | /* iopoll syncs against uring_lock, not completion_lock */ |
1499 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1500 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1501 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1502 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1503 | mutex_unlock(&ctx->uring_lock); | |
1504 | } | |
ca0a2651 JA |
1505 | |
1506 | return ret; | |
6c503150 PB |
1507 | } |
1508 | ||
abc54d63 JA |
1509 | /* |
1510 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1511 | * see commit f958d7b528b1 for details. | |
1512 | */ | |
1513 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1514 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1515 | ||
de9b4cca JA |
1516 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1517 | { | |
abc54d63 | 1518 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1519 | } |
1520 | ||
1521 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1522 | { | |
abc54d63 JA |
1523 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1524 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1525 | } |
1526 | ||
1527 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1528 | { | |
abc54d63 JA |
1529 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1530 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1531 | } |
1532 | ||
1533 | static inline void req_ref_put(struct io_kiocb *req) | |
1534 | { | |
abc54d63 | 1535 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1536 | } |
1537 | ||
1538 | static inline void req_ref_get(struct io_kiocb *req) | |
1539 | { | |
abc54d63 JA |
1540 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1541 | atomic_inc(&req->refs); | |
de9b4cca JA |
1542 | } |
1543 | ||
d4d19c19 PB |
1544 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1545 | long res, unsigned int cflags) | |
2b188cc1 | 1546 | { |
cce4b8b0 | 1547 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1548 | |
cce4b8b0 PB |
1549 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1550 | if (!ocqe) { | |
1551 | /* | |
1552 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1553 | * or cannot allocate an overflow entry, then we need to drop it | |
1554 | * on the floor. | |
1555 | */ | |
8f6ed49a | 1556 | io_account_cq_overflow(ctx); |
cce4b8b0 | 1557 | return false; |
2b188cc1 | 1558 | } |
cce4b8b0 | 1559 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 1560 | set_bit(0, &ctx->check_cq_overflow); |
cce4b8b0 PB |
1561 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
1562 | } | |
d4d19c19 | 1563 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1564 | ocqe->cqe.res = res; |
1565 | ocqe->cqe.flags = cflags; | |
1566 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1567 | return true; | |
2b188cc1 JA |
1568 | } |
1569 | ||
d4d19c19 PB |
1570 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1571 | long res, unsigned int cflags) | |
2b188cc1 JA |
1572 | { |
1573 | struct io_uring_cqe *cqe; | |
1574 | ||
d4d19c19 | 1575 | trace_io_uring_complete(ctx, user_data, res, cflags); |
51c3ff62 | 1576 | |
2b188cc1 JA |
1577 | /* |
1578 | * If we can't get a cq entry, userspace overflowed the | |
1579 | * submission (by quite a lot). Increment the overflow count in | |
1580 | * the ring. | |
1581 | */ | |
d068b506 | 1582 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 1583 | if (likely(cqe)) { |
d4d19c19 | 1584 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 1585 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1586 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 1587 | return true; |
2b188cc1 | 1588 | } |
d4d19c19 | 1589 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
1590 | } |
1591 | ||
8d13326e | 1592 | /* not as hot to bloat with inlining */ |
d4d19c19 PB |
1593 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1594 | long res, unsigned int cflags) | |
bcda7baa | 1595 | { |
d4d19c19 | 1596 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
bcda7baa JA |
1597 | } |
1598 | ||
7a612350 PB |
1599 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1600 | unsigned int cflags) | |
2b188cc1 | 1601 | { |
78e19bbe | 1602 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1603 | unsigned long flags; |
1604 | ||
1605 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
d4d19c19 | 1606 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1607 | /* |
1608 | * If we're the last reference to this request, add to our locked | |
1609 | * free_list cache. | |
1610 | */ | |
de9b4cca | 1611 | if (req_ref_put_and_test(req)) { |
7a612350 | 1612 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
93d2bcd2 | 1613 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) |
7a612350 PB |
1614 | io_disarm_next(req); |
1615 | if (req->link) { | |
1616 | io_req_task_queue(req->link); | |
1617 | req->link = NULL; | |
1618 | } | |
1619 | } | |
c7dae4ba JA |
1620 | io_dismantle_req(req); |
1621 | io_put_task(req->task, 1); | |
d0acdee2 PB |
1622 | list_add(&req->compl.list, &ctx->locked_free_list); |
1623 | ctx->locked_free_nr++; | |
180f829f PB |
1624 | } else { |
1625 | if (!percpu_ref_tryget(&ctx->refs)) | |
1626 | req = NULL; | |
1627 | } | |
7a612350 | 1628 | io_commit_cqring(ctx); |
2b188cc1 | 1629 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1630 | |
180f829f PB |
1631 | if (req) { |
1632 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1633 | percpu_ref_put(&ctx->refs); |
180f829f | 1634 | } |
229a7b63 JA |
1635 | } |
1636 | ||
4e3d9ff9 JA |
1637 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1638 | { | |
c854357b | 1639 | return req->flags & IO_REQ_CLEAN_FLAGS; |
4e3d9ff9 JA |
1640 | } |
1641 | ||
a38d68db | 1642 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1643 | unsigned int cflags) |
229a7b63 | 1644 | { |
4e3d9ff9 | 1645 | if (io_req_needs_clean(req)) |
68fb8979 | 1646 | io_clean_op(req); |
a38d68db PB |
1647 | req->result = res; |
1648 | req->compl.cflags = cflags; | |
e342c807 | 1649 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1650 | } |
1651 | ||
889fca73 PB |
1652 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1653 | long res, unsigned cflags) | |
bcda7baa | 1654 | { |
889fca73 PB |
1655 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1656 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1657 | else |
c7dae4ba | 1658 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1659 | } |
1660 | ||
a38d68db | 1661 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1662 | { |
889fca73 | 1663 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1664 | } |
1665 | ||
f41db273 PB |
1666 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1667 | { | |
93d2bcd2 | 1668 | req_set_fail(req); |
f41db273 PB |
1669 | io_put_req(req); |
1670 | io_req_complete_post(req, res, 0); | |
1671 | } | |
1672 | ||
dac7a098 PB |
1673 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1674 | struct io_comp_state *cs) | |
1675 | { | |
1676 | spin_lock_irq(&ctx->completion_lock); | |
d0acdee2 PB |
1677 | list_splice_init(&ctx->locked_free_list, &cs->free_list); |
1678 | ctx->locked_free_nr = 0; | |
dac7a098 PB |
1679 | spin_unlock_irq(&ctx->completion_lock); |
1680 | } | |
1681 | ||
dd78f492 | 1682 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1683 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1684 | { |
c7dae4ba JA |
1685 | struct io_submit_state *state = &ctx->submit_state; |
1686 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1687 | int nr; |
0ddf92e8 | 1688 | |
c7dae4ba JA |
1689 | /* |
1690 | * If we have more than a batch's worth of requests in our IRQ side | |
1691 | * locked cache, grab the lock and move them over to our submission | |
1692 | * side cache. | |
1693 | */ | |
d0acdee2 | 1694 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
dac7a098 | 1695 | io_flush_cached_locked_reqs(ctx, cs); |
0ddf92e8 | 1696 | |
dd78f492 | 1697 | nr = state->free_reqs; |
c7dae4ba | 1698 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1699 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1700 | struct io_kiocb, compl.list); | |
1701 | ||
1b4c351f | 1702 | list_del(&req->compl.list); |
dd78f492 PB |
1703 | state->reqs[nr++] = req; |
1704 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1705 | break; |
1b4c351f JA |
1706 | } |
1707 | ||
dd78f492 PB |
1708 | state->free_reqs = nr; |
1709 | return nr != 0; | |
0ddf92e8 JA |
1710 | } |
1711 | ||
e5d1bc0a | 1712 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1713 | { |
e5d1bc0a PB |
1714 | struct io_submit_state *state = &ctx->submit_state; |
1715 | ||
fe7e3257 | 1716 | BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); |
e5d1bc0a | 1717 | |
f6b6c7d6 | 1718 | if (!state->free_reqs) { |
291b2821 | 1719 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
99ebe4ef | 1720 | int ret, i; |
2579f913 | 1721 | |
c7dae4ba | 1722 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1723 | goto got_req; |
1724 | ||
bf019da7 PB |
1725 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1726 | state->reqs); | |
fd6fab2c JA |
1727 | |
1728 | /* | |
1729 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1730 | * retry single alloc to be on the safe side. | |
1731 | */ | |
1732 | if (unlikely(ret <= 0)) { | |
1733 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1734 | if (!state->reqs[0]) | |
3893f39f | 1735 | return NULL; |
fd6fab2c JA |
1736 | ret = 1; |
1737 | } | |
99ebe4ef PB |
1738 | |
1739 | /* | |
1740 | * Don't initialise the fields below on every allocation, but | |
1741 | * do that in advance and keep valid on free. | |
1742 | */ | |
1743 | for (i = 0; i < ret; i++) { | |
1744 | struct io_kiocb *req = state->reqs[i]; | |
1745 | ||
1746 | req->ctx = ctx; | |
1747 | req->link = NULL; | |
1748 | req->async_data = NULL; | |
1749 | /* not necessary, but safer to zero */ | |
1750 | req->result = 0; | |
1751 | } | |
291b2821 | 1752 | state->free_reqs = ret; |
2b188cc1 | 1753 | } |
e5d1bc0a | 1754 | got_req: |
291b2821 PB |
1755 | state->free_reqs--; |
1756 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1757 | } |
1758 | ||
e1d767f0 | 1759 | static inline void io_put_file(struct file *file) |
8da11c19 | 1760 | { |
e1d767f0 | 1761 | if (file) |
8da11c19 PB |
1762 | fput(file); |
1763 | } | |
1764 | ||
4edf20f9 | 1765 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1766 | { |
094bae49 | 1767 | unsigned int flags = req->flags; |
929a3af9 | 1768 | |
3a0a6902 PB |
1769 | if (io_req_needs_clean(req)) |
1770 | io_clean_op(req); | |
e1d767f0 PB |
1771 | if (!(flags & REQ_F_FIXED_FILE)) |
1772 | io_put_file(req->file); | |
269bbe5f BM |
1773 | if (req->fixed_rsrc_refs) |
1774 | percpu_ref_put(req->fixed_rsrc_refs); | |
99ebe4ef | 1775 | if (req->async_data) { |
094bae49 | 1776 | kfree(req->async_data); |
99ebe4ef PB |
1777 | req->async_data = NULL; |
1778 | } | |
e65ef56d JA |
1779 | } |
1780 | ||
b23fcf47 | 1781 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1782 | static inline void io_put_task(struct task_struct *task, int nr) |
1783 | { | |
1784 | struct io_uring_task *tctx = task->io_uring; | |
1785 | ||
1786 | percpu_counter_sub(&tctx->inflight, nr); | |
1787 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1788 | wake_up(&tctx->wait); | |
1789 | put_task_struct_many(task, nr); | |
1790 | } | |
1791 | ||
216578e5 | 1792 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1793 | { |
51a4cc11 | 1794 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1795 | |
216578e5 | 1796 | io_dismantle_req(req); |
7c660731 | 1797 | io_put_task(req->task, 1); |
c6ca97b3 | 1798 | |
3893f39f | 1799 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1800 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1801 | } |
1802 | ||
f2f87370 PB |
1803 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1804 | { | |
1805 | struct io_kiocb *nxt = req->link; | |
1806 | ||
1807 | req->link = nxt->link; | |
1808 | nxt->link = NULL; | |
1809 | } | |
1810 | ||
33cc89a9 PB |
1811 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1812 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1813 | { |
33cc89a9 | 1814 | struct io_kiocb *link = req->link; |
f2f87370 | 1815 | |
900fad45 PB |
1816 | /* |
1817 | * Can happen if a linked timeout fired and link had been like | |
1818 | * req -> link t-out -> link t-out [-> ...] | |
1819 | */ | |
c9abd7ad PB |
1820 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1821 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1822 | |
f2f87370 | 1823 | io_remove_next_linked(req); |
90cd7e42 | 1824 | link->timeout.head = NULL; |
fd9c7bc5 | 1825 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
d4d19c19 PB |
1826 | io_cqring_fill_event(link->ctx, link->user_data, |
1827 | -ECANCELED, 0); | |
33cc89a9 | 1828 | io_put_req_deferred(link, 1); |
d4729fbd | 1829 | return true; |
c9abd7ad PB |
1830 | } |
1831 | } | |
d4729fbd | 1832 | return false; |
7c86ffee PB |
1833 | } |
1834 | ||
d148ca4b | 1835 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1836 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1837 | { |
33cc89a9 | 1838 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1839 | |
f2f87370 | 1840 | req->link = NULL; |
f2f87370 PB |
1841 | while (link) { |
1842 | nxt = link->link; | |
1843 | link->link = NULL; | |
2665abfd | 1844 | |
f2f87370 | 1845 | trace_io_uring_fail_link(req, link); |
d4d19c19 | 1846 | io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0); |
1575f21a | 1847 | io_put_req_deferred(link, 2); |
f2f87370 | 1848 | link = nxt; |
9e645e11 | 1849 | } |
33cc89a9 | 1850 | } |
9e645e11 | 1851 | |
33cc89a9 PB |
1852 | static bool io_disarm_next(struct io_kiocb *req) |
1853 | __must_hold(&req->ctx->completion_lock) | |
1854 | { | |
1855 | bool posted = false; | |
1856 | ||
1857 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1858 | posted = io_kill_linked_timeout(req); | |
93d2bcd2 | 1859 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 1860 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
1861 | posted |= (req->link != NULL); |
1862 | io_fail_links(req); | |
1863 | } | |
1864 | return posted; | |
9e645e11 JA |
1865 | } |
1866 | ||
3fa5e0f3 | 1867 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1868 | { |
33cc89a9 | 1869 | struct io_kiocb *nxt; |
944e58bf | 1870 | |
9e645e11 JA |
1871 | /* |
1872 | * If LINK is set, we have dependent requests in this chain. If we | |
1873 | * didn't fail this request, queue the first one up, moving any other | |
1874 | * dependencies to the next request. In case of failure, fail the rest | |
1875 | * of the chain. | |
1876 | */ | |
93d2bcd2 | 1877 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) { |
33cc89a9 PB |
1878 | struct io_ring_ctx *ctx = req->ctx; |
1879 | unsigned long flags; | |
1880 | bool posted; | |
1881 | ||
1882 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1883 | posted = io_disarm_next(req); | |
1884 | if (posted) | |
1885 | io_commit_cqring(req->ctx); | |
1886 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1887 | if (posted) | |
1888 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1889 | } |
33cc89a9 PB |
1890 | nxt = req->link; |
1891 | req->link = NULL; | |
1892 | return nxt; | |
4d7dd462 | 1893 | } |
9e645e11 | 1894 | |
f2f87370 | 1895 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1896 | { |
cdbff982 | 1897 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1898 | return NULL; |
1899 | return __io_req_find_next(req); | |
1900 | } | |
1901 | ||
2c32395d PB |
1902 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1903 | { | |
1904 | if (!ctx) | |
1905 | return; | |
1906 | if (ctx->submit_state.comp.nr) { | |
1907 | mutex_lock(&ctx->uring_lock); | |
2a2758f2 | 1908 | io_submit_flush_completions(ctx); |
2c32395d PB |
1909 | mutex_unlock(&ctx->uring_lock); |
1910 | } | |
1911 | percpu_ref_put(&ctx->refs); | |
1912 | } | |
1913 | ||
7cbf1722 | 1914 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1915 | { |
ebd0df2e | 1916 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
1917 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
1918 | task_work); | |
c40f6379 | 1919 | |
16f72070 | 1920 | while (1) { |
3f18407d PB |
1921 | struct io_wq_work_node *node; |
1922 | ||
1923 | spin_lock_irq(&tctx->task_lock); | |
c6538be9 | 1924 | node = tctx->task_list.first; |
3f18407d PB |
1925 | INIT_WQ_LIST(&tctx->task_list); |
1926 | spin_unlock_irq(&tctx->task_lock); | |
1927 | ||
3f18407d PB |
1928 | while (node) { |
1929 | struct io_wq_work_node *next = node->next; | |
1930 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
1931 | io_task_work.node); | |
1932 | ||
1933 | if (req->ctx != ctx) { | |
1934 | ctx_flush_and_put(ctx); | |
1935 | ctx = req->ctx; | |
1936 | percpu_ref_get(&ctx->refs); | |
1937 | } | |
5b0a6acc | 1938 | req->io_task_work.func(req); |
3f18407d PB |
1939 | node = next; |
1940 | } | |
7a778f9d PB |
1941 | if (wq_list_empty(&tctx->task_list)) { |
1942 | clear_bit(0, &tctx->task_state); | |
1943 | if (wq_list_empty(&tctx->task_list)) | |
1944 | break; | |
1945 | /* another tctx_task_work() is enqueued, yield */ | |
1946 | if (test_and_set_bit(0, &tctx->task_state)) | |
1947 | break; | |
1948 | } | |
7cbf1722 | 1949 | cond_resched(); |
3f18407d | 1950 | } |
ebd0df2e PB |
1951 | |
1952 | ctx_flush_and_put(ctx); | |
7cbf1722 JA |
1953 | } |
1954 | ||
e09ee510 | 1955 | static void io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1956 | { |
c15b79de | 1957 | struct task_struct *tsk = req->task; |
7cbf1722 | 1958 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1959 | enum task_work_notify_mode notify; |
e09ee510 | 1960 | struct io_wq_work_node *node; |
0b81e80c | 1961 | unsigned long flags; |
7cbf1722 JA |
1962 | |
1963 | WARN_ON_ONCE(!tctx); | |
1964 | ||
0b81e80c | 1965 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1966 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1967 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1968 | |
1969 | /* task_work already pending, we're done */ | |
1970 | if (test_bit(0, &tctx->task_state) || | |
1971 | test_and_set_bit(0, &tctx->task_state)) | |
e09ee510 PB |
1972 | return; |
1973 | if (unlikely(tsk->flags & PF_EXITING)) | |
1974 | goto fail; | |
7cbf1722 | 1975 | |
c15b79de PB |
1976 | /* |
1977 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1978 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1979 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1980 | * will do the job. | |
1981 | */ | |
1982 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
c15b79de PB |
1983 | if (!task_work_add(tsk, &tctx->task_work, notify)) { |
1984 | wake_up_process(tsk); | |
e09ee510 | 1985 | return; |
c15b79de | 1986 | } |
e09ee510 PB |
1987 | fail: |
1988 | clear_bit(0, &tctx->task_state); | |
0b81e80c | 1989 | spin_lock_irqsave(&tctx->task_lock, flags); |
e09ee510 PB |
1990 | node = tctx->task_list.first; |
1991 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1992 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 1993 | |
e09ee510 PB |
1994 | while (node) { |
1995 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
1996 | node = node->next; | |
1997 | if (llist_add(&req->io_task_work.fallback_node, | |
1998 | &req->ctx->fallback_llist)) | |
1999 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
2000 | } | |
eab30c4d PB |
2001 | } |
2002 | ||
5b0a6acc | 2003 | static void io_req_task_cancel(struct io_kiocb *req) |
c40f6379 | 2004 | { |
87ceb6a6 | 2005 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2006 | |
e83acd7d | 2007 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2008 | mutex_lock(&ctx->uring_lock); |
2593553a | 2009 | io_req_complete_failed(req, req->result); |
792bb6eb | 2010 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2011 | } |
2012 | ||
5b0a6acc | 2013 | static void io_req_task_submit(struct io_kiocb *req) |
c40f6379 JA |
2014 | { |
2015 | struct io_ring_ctx *ctx = req->ctx; | |
2016 | ||
04fc6c80 | 2017 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2018 | mutex_lock(&ctx->uring_lock); |
9c688260 | 2019 | if (!(req->task->flags & PF_EXITING) && !req->task->in_execve) |
c5eef2b9 | 2020 | __io_queue_sqe(req); |
81b6d05c | 2021 | else |
2593553a | 2022 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2023 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2024 | } |
2025 | ||
2c4b8eb6 | 2026 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2027 | { |
2c4b8eb6 | 2028 | req->result = ret; |
5b0a6acc | 2029 | req->io_task_work.func = io_req_task_cancel; |
e09ee510 | 2030 | io_req_task_work_add(req); |
c40f6379 JA |
2031 | } |
2032 | ||
2c4b8eb6 | 2033 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2034 | { |
5b0a6acc | 2035 | req->io_task_work.func = io_req_task_submit; |
e09ee510 | 2036 | io_req_task_work_add(req); |
a3df7698 PB |
2037 | } |
2038 | ||
f2f87370 | 2039 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2040 | { |
9b5f7bd9 | 2041 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2042 | |
2043 | if (nxt) | |
906a8c3f | 2044 | io_req_task_queue(nxt); |
c69f8dbe JL |
2045 | } |
2046 | ||
c3524383 | 2047 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2048 | { |
c3524383 PB |
2049 | io_queue_next(req); |
2050 | __io_free_req(req); | |
2051 | } | |
8766dd51 | 2052 | |
2d6500d4 | 2053 | struct req_batch { |
5af1d13e PB |
2054 | struct task_struct *task; |
2055 | int task_refs; | |
1b4c351f | 2056 | int ctx_refs; |
2d6500d4 PB |
2057 | }; |
2058 | ||
5af1d13e PB |
2059 | static inline void io_init_req_batch(struct req_batch *rb) |
2060 | { | |
5af1d13e | 2061 | rb->task_refs = 0; |
9ae72463 | 2062 | rb->ctx_refs = 0; |
5af1d13e PB |
2063 | rb->task = NULL; |
2064 | } | |
2065 | ||
2d6500d4 PB |
2066 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2067 | struct req_batch *rb) | |
2068 | { | |
6e833d53 | 2069 | if (rb->task) |
7c660731 | 2070 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2071 | if (rb->ctx_refs) |
2072 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2073 | } |
2074 | ||
6ff119a6 PB |
2075 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2076 | struct io_submit_state *state) | |
2d6500d4 | 2077 | { |
f2f87370 | 2078 | io_queue_next(req); |
96670657 | 2079 | io_dismantle_req(req); |
2d6500d4 | 2080 | |
e3bc8e9d | 2081 | if (req->task != rb->task) { |
7c660731 PB |
2082 | if (rb->task) |
2083 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2084 | rb->task = req->task; |
2085 | rb->task_refs = 0; | |
5af1d13e | 2086 | } |
e3bc8e9d | 2087 | rb->task_refs++; |
9ae72463 | 2088 | rb->ctx_refs++; |
5af1d13e | 2089 | |
bd759045 | 2090 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2091 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2092 | else |
2093 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2094 | } |
2095 | ||
2a2758f2 | 2096 | static void io_submit_flush_completions(struct io_ring_ctx *ctx) |
905c172f | 2097 | { |
2a2758f2 | 2098 | struct io_comp_state *cs = &ctx->submit_state.comp; |
905c172f | 2099 | int i, nr = cs->nr; |
905c172f PB |
2100 | struct req_batch rb; |
2101 | ||
905c172f PB |
2102 | spin_lock_irq(&ctx->completion_lock); |
2103 | for (i = 0; i < nr; i++) { | |
5182ed2e PB |
2104 | struct io_kiocb *req = cs->reqs[i]; |
2105 | ||
d4d19c19 PB |
2106 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2107 | req->compl.cflags); | |
905c172f PB |
2108 | } |
2109 | io_commit_cqring(ctx); | |
2110 | spin_unlock_irq(&ctx->completion_lock); | |
905c172f | 2111 | io_cqring_ev_posted(ctx); |
5182ed2e PB |
2112 | |
2113 | io_init_req_batch(&rb); | |
905c172f | 2114 | for (i = 0; i < nr; i++) { |
5182ed2e | 2115 | struct io_kiocb *req = cs->reqs[i]; |
905c172f PB |
2116 | |
2117 | /* submission and completion refs */ | |
de9b4cca | 2118 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2119 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2120 | } |
2121 | ||
2122 | io_req_free_batch_finish(ctx, &rb); | |
2123 | cs->nr = 0; | |
7a743e22 PB |
2124 | } |
2125 | ||
ba816ad6 JA |
2126 | /* |
2127 | * Drop reference to request, return next in chain (if there is one) if this | |
2128 | * was the last reference to this request. | |
2129 | */ | |
0d85035a | 2130 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2131 | { |
9b5f7bd9 PB |
2132 | struct io_kiocb *nxt = NULL; |
2133 | ||
de9b4cca | 2134 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2135 | nxt = io_req_find_next(req); |
4d7dd462 | 2136 | __io_free_req(req); |
2a44f467 | 2137 | } |
9b5f7bd9 | 2138 | return nxt; |
2b188cc1 JA |
2139 | } |
2140 | ||
0d85035a | 2141 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2142 | { |
de9b4cca | 2143 | if (req_ref_put_and_test(req)) |
e65ef56d | 2144 | io_free_req(req); |
2b188cc1 JA |
2145 | } |
2146 | ||
216578e5 PB |
2147 | static void io_free_req_deferred(struct io_kiocb *req) |
2148 | { | |
5b0a6acc | 2149 | req->io_task_work.func = io_free_req; |
e09ee510 | 2150 | io_req_task_work_add(req); |
216578e5 PB |
2151 | } |
2152 | ||
2153 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2154 | { | |
de9b4cca | 2155 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2156 | io_free_req_deferred(req); |
2157 | } | |
2158 | ||
6c503150 | 2159 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2160 | { |
2161 | /* See comment at the top of this file */ | |
2162 | smp_rmb(); | |
e23de15f | 2163 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2164 | } |
2165 | ||
fb5ccc98 PB |
2166 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2167 | { | |
2168 | struct io_rings *rings = ctx->rings; | |
2169 | ||
2170 | /* make sure SQ entry isn't read before tail */ | |
2171 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2172 | } | |
2173 | ||
8ff069bf | 2174 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2175 | { |
8ff069bf | 2176 | unsigned int cflags; |
e94f141b | 2177 | |
bcda7baa JA |
2178 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2179 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2180 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2181 | kfree(kbuf); |
2182 | return cflags; | |
e94f141b JA |
2183 | } |
2184 | ||
8ff069bf | 2185 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2186 | { |
4d954c25 | 2187 | struct io_buffer *kbuf; |
bcda7baa | 2188 | |
4d954c25 | 2189 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2190 | return io_put_kbuf(req, kbuf); |
2191 | } | |
2192 | ||
4c6e277c JA |
2193 | static inline bool io_run_task_work(void) |
2194 | { | |
2195 | if (current->task_works) { | |
2196 | __set_current_state(TASK_RUNNING); | |
2197 | task_work_run(); | |
2198 | return true; | |
2199 | } | |
2200 | ||
2201 | return false; | |
bcda7baa JA |
2202 | } |
2203 | ||
def596e9 JA |
2204 | /* |
2205 | * Find and free completed poll iocbs | |
2206 | */ | |
2207 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2208 | struct list_head *done) | |
2209 | { | |
8237e045 | 2210 | struct req_batch rb; |
def596e9 | 2211 | struct io_kiocb *req; |
bbde017a XW |
2212 | |
2213 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2214 | smp_rmb(); | |
def596e9 | 2215 | |
5af1d13e | 2216 | io_init_req_batch(&rb); |
def596e9 | 2217 | while (!list_empty(done)) { |
bcda7baa JA |
2218 | int cflags = 0; |
2219 | ||
d21ffe7e | 2220 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2221 | list_del(&req->inflight_entry); |
2222 | ||
8c130827 PB |
2223 | if (READ_ONCE(req->result) == -EAGAIN && |
2224 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2225 | req->iopoll_completed = 0; |
8c130827 PB |
2226 | req_ref_get(req); |
2227 | io_queue_async_work(req); | |
2228 | continue; | |
bbde017a | 2229 | } |
def596e9 | 2230 | |
bcda7baa | 2231 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2232 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2233 | |
d4d19c19 | 2234 | __io_cqring_fill_event(ctx, req->user_data, req->result, cflags); |
def596e9 JA |
2235 | (*nr_events)++; |
2236 | ||
de9b4cca | 2237 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2238 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2239 | } |
def596e9 | 2240 | |
09bb8394 | 2241 | io_commit_cqring(ctx); |
80c18e4a | 2242 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2243 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2244 | } |
2245 | ||
def596e9 JA |
2246 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2247 | long min) | |
2248 | { | |
2249 | struct io_kiocb *req, *tmp; | |
2250 | LIST_HEAD(done); | |
2251 | bool spin; | |
2252 | int ret; | |
2253 | ||
2254 | /* | |
2255 | * Only spin for completions if we don't have multiple devices hanging | |
2256 | * off our complete list, and we're under the requested amount. | |
2257 | */ | |
915b3dde | 2258 | spin = !ctx->poll_multi_queue && *nr_events < min; |
def596e9 JA |
2259 | |
2260 | ret = 0; | |
d21ffe7e | 2261 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2262 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2263 | |
2264 | /* | |
581f9810 BM |
2265 | * Move completed and retryable entries to our local lists. |
2266 | * If we find a request that requires polling, break out | |
2267 | * and complete those lists first, if we have entries there. | |
def596e9 | 2268 | */ |
65a6543d | 2269 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2270 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2271 | continue; |
2272 | } | |
2273 | if (!list_empty(&done)) | |
2274 | break; | |
2275 | ||
2276 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2277 | if (ret < 0) | |
2278 | break; | |
2279 | ||
3aadc23e PB |
2280 | /* iopoll may have completed current req */ |
2281 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2282 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2283 | |
def596e9 JA |
2284 | if (ret && spin) |
2285 | spin = false; | |
2286 | ret = 0; | |
2287 | } | |
2288 | ||
2289 | if (!list_empty(&done)) | |
2290 | io_iopoll_complete(ctx, nr_events, &done); | |
2291 | ||
2292 | return ret; | |
2293 | } | |
2294 | ||
def596e9 JA |
2295 | /* |
2296 | * We can't just wait for polled events to come to us, we have to actively | |
2297 | * find and complete them. | |
2298 | */ | |
b2edc0a7 | 2299 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2300 | { |
2301 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2302 | return; | |
2303 | ||
2304 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2305 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2306 | unsigned int nr_events = 0; |
2307 | ||
b2edc0a7 | 2308 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2309 | |
b2edc0a7 PB |
2310 | /* let it sleep and repeat later if can't complete a request */ |
2311 | if (nr_events == 0) | |
2312 | break; | |
08f5439f JA |
2313 | /* |
2314 | * Ensure we allow local-to-the-cpu processing to take place, | |
2315 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2316 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2317 | */ |
3fcee5a6 PB |
2318 | if (need_resched()) { |
2319 | mutex_unlock(&ctx->uring_lock); | |
2320 | cond_resched(); | |
2321 | mutex_lock(&ctx->uring_lock); | |
2322 | } | |
def596e9 JA |
2323 | } |
2324 | mutex_unlock(&ctx->uring_lock); | |
2325 | } | |
2326 | ||
7668b92a | 2327 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2328 | { |
7668b92a | 2329 | unsigned int nr_events = 0; |
e9979b36 | 2330 | int ret = 0; |
500f9fba | 2331 | |
c7849be9 XW |
2332 | /* |
2333 | * We disallow the app entering submit/complete with polling, but we | |
2334 | * still need to lock the ring to prevent racing with polled issue | |
2335 | * that got punted to a workqueue. | |
2336 | */ | |
2337 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2338 | /* |
2339 | * Don't enter poll loop if we already have events pending. | |
2340 | * If we do, we can potentially be spinning for commands that | |
2341 | * already triggered a CQE (eg in error). | |
2342 | */ | |
5ed7a37d | 2343 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2344 | __io_cqring_overflow_flush(ctx, false); |
2345 | if (io_cqring_events(ctx)) | |
2346 | goto out; | |
def596e9 | 2347 | do { |
500f9fba JA |
2348 | /* |
2349 | * If a submit got punted to a workqueue, we can have the | |
2350 | * application entering polling for a command before it gets | |
2351 | * issued. That app will hold the uring_lock for the duration | |
2352 | * of the poll right here, so we need to take a breather every | |
2353 | * now and then to ensure that the issue has a chance to add | |
2354 | * the poll to the issued list. Otherwise we can spin here | |
2355 | * forever, while the workqueue is stuck trying to acquire the | |
2356 | * very same mutex. | |
2357 | */ | |
e9979b36 | 2358 | if (list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
2359 | u32 tail = ctx->cached_cq_tail; |
2360 | ||
500f9fba | 2361 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2362 | io_run_task_work(); |
500f9fba | 2363 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2364 | |
8f487ef2 PB |
2365 | /* some requests don't go through iopoll_list */ |
2366 | if (tail != ctx->cached_cq_tail || | |
2367 | list_empty(&ctx->iopoll_list)) | |
e9979b36 | 2368 | break; |
500f9fba | 2369 | } |
f39c8a5b PB |
2370 | ret = io_do_iopoll(ctx, &nr_events, min); |
2371 | } while (!ret && nr_events < min && !need_resched()); | |
2372 | out: | |
500f9fba | 2373 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2374 | return ret; |
2375 | } | |
2376 | ||
491381ce | 2377 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2378 | { |
491381ce JA |
2379 | /* |
2380 | * Tell lockdep we inherited freeze protection from submission | |
2381 | * thread. | |
2382 | */ | |
2383 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2384 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2385 | |
1c98679d PB |
2386 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2387 | sb_end_write(sb); | |
2b188cc1 JA |
2388 | } |
2389 | } | |
2390 | ||
b63534c4 | 2391 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2392 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2393 | { |
ab454438 | 2394 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2395 | |
ab454438 PB |
2396 | if (!rw) |
2397 | return !io_req_prep_async(req); | |
2398 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2399 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2400 | return true; | |
b63534c4 | 2401 | } |
b63534c4 | 2402 | |
3e6a0d3c | 2403 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2404 | { |
355afaeb | 2405 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2406 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2407 | |
355afaeb JA |
2408 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2409 | return false; | |
3e6a0d3c JA |
2410 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2411 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2412 | return false; |
7c977a58 JA |
2413 | /* |
2414 | * If ref is dying, we might be running poll reap from the exit work. | |
2415 | * Don't attempt to reissue from that path, just let it fail with | |
2416 | * -EAGAIN. | |
2417 | */ | |
3e6a0d3c JA |
2418 | if (percpu_ref_is_dying(&ctx->refs)) |
2419 | return false; | |
2420 | return true; | |
2421 | } | |
e82ad485 | 2422 | #else |
a1ff1e3f | 2423 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2424 | { |
2425 | return false; | |
2426 | } | |
e82ad485 | 2427 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2428 | { |
b63534c4 JA |
2429 | return false; |
2430 | } | |
3e6a0d3c | 2431 | #endif |
b63534c4 | 2432 | |
9011bf9a PB |
2433 | static void io_fallback_req_func(struct work_struct *work) |
2434 | { | |
2435 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
2436 | fallback_work.work); | |
2437 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
2438 | struct io_kiocb *req, *tmp; | |
2439 | ||
5b0a6acc PB |
2440 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) |
2441 | req->io_task_work.func(req); | |
9011bf9a PB |
2442 | } |
2443 | ||
a1d7c393 | 2444 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2445 | unsigned int issue_flags) |
a1d7c393 | 2446 | { |
2f8e45f1 PB |
2447 | int cflags = 0; |
2448 | ||
b65c128f PB |
2449 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2450 | kiocb_end_write(req); | |
9532b99b PB |
2451 | if (res != req->result) { |
2452 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2453 | io_rw_should_reissue(req)) { | |
2454 | req->flags |= REQ_F_REISSUE; | |
2455 | return; | |
2456 | } | |
93d2bcd2 | 2457 | req_set_fail(req); |
9532b99b | 2458 | } |
2f8e45f1 PB |
2459 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2460 | cflags = io_put_rw_kbuf(req); | |
2461 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2462 | } |
2463 | ||
2464 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2465 | { | |
9adbd45d | 2466 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2467 | |
889fca73 | 2468 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2469 | } |
2470 | ||
def596e9 JA |
2471 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2472 | { | |
9adbd45d | 2473 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2474 | |
491381ce JA |
2475 | if (kiocb->ki_flags & IOCB_WRITE) |
2476 | kiocb_end_write(req); | |
9532b99b | 2477 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2478 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2479 | io_resubmit_prep(req))) { | |
93d2bcd2 | 2480 | req_set_fail(req); |
9532b99b PB |
2481 | req->flags |= REQ_F_DONT_REISSUE; |
2482 | } | |
8c130827 | 2483 | } |
bbde017a XW |
2484 | |
2485 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2486 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2487 | smp_wmb(); |
2488 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2489 | } |
2490 | ||
2491 | /* | |
2492 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2493 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2494 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2495 | * accessing the kiocb cookie. |
2496 | */ | |
cb3d8972 | 2497 | static void io_iopoll_req_issued(struct io_kiocb *req) |
def596e9 JA |
2498 | { |
2499 | struct io_ring_ctx *ctx = req->ctx; | |
cb3d8972 PB |
2500 | const bool in_async = io_wq_current_is_worker(); |
2501 | ||
2502 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
2503 | if (unlikely(in_async)) | |
2504 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
2505 | |
2506 | /* | |
2507 | * Track whether we have multiple files in our lists. This will impact | |
2508 | * how we do polling eventually, not spinning if we're on potentially | |
2509 | * different devices. | |
2510 | */ | |
540e32a0 | 2511 | if (list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
2512 | ctx->poll_multi_queue = false; |
2513 | } else if (!ctx->poll_multi_queue) { | |
def596e9 | 2514 | struct io_kiocb *list_req; |
915b3dde | 2515 | unsigned int queue_num0, queue_num1; |
def596e9 | 2516 | |
540e32a0 | 2517 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2518 | inflight_entry); |
915b3dde HX |
2519 | |
2520 | if (list_req->file != req->file) { | |
2521 | ctx->poll_multi_queue = true; | |
2522 | } else { | |
2523 | queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie); | |
2524 | queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie); | |
2525 | if (queue_num0 != queue_num1) | |
2526 | ctx->poll_multi_queue = true; | |
2527 | } | |
def596e9 JA |
2528 | } |
2529 | ||
2530 | /* | |
2531 | * For fast devices, IO may have already completed. If it has, add | |
2532 | * it to the front so we find it first. | |
2533 | */ | |
65a6543d | 2534 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2535 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2536 | else |
d21ffe7e | 2537 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2538 | |
cb3d8972 PB |
2539 | if (unlikely(in_async)) { |
2540 | /* | |
2541 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
2542 | * in sq thread task context or in io worker task context. If | |
2543 | * current task context is sq thread, we don't need to check | |
2544 | * whether should wake up sq thread. | |
2545 | */ | |
2546 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2547 | wq_has_sleeper(&ctx->sq_data->wait)) | |
2548 | wake_up(&ctx->sq_data->wait); | |
2549 | ||
2550 | mutex_unlock(&ctx->uring_lock); | |
2551 | } | |
def596e9 JA |
2552 | } |
2553 | ||
9f13c35b PB |
2554 | static inline void io_state_file_put(struct io_submit_state *state) |
2555 | { | |
02b23a9a PB |
2556 | if (state->file_refs) { |
2557 | fput_many(state->file, state->file_refs); | |
2558 | state->file_refs = 0; | |
2559 | } | |
9a56a232 JA |
2560 | } |
2561 | ||
2562 | /* | |
2563 | * Get as many references to a file as we have IOs left in this submission, | |
2564 | * assuming most submissions are for one file, or at least that each file | |
2565 | * has more than one submission. | |
2566 | */ | |
8da11c19 | 2567 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2568 | { |
2569 | if (!state) | |
2570 | return fget(fd); | |
2571 | ||
6e1271e6 | 2572 | if (state->file_refs) { |
9a56a232 | 2573 | if (state->fd == fd) { |
6e1271e6 | 2574 | state->file_refs--; |
9a56a232 JA |
2575 | return state->file; |
2576 | } | |
02b23a9a | 2577 | io_state_file_put(state); |
9a56a232 JA |
2578 | } |
2579 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2580 | if (unlikely(!state->file)) |
9a56a232 JA |
2581 | return NULL; |
2582 | ||
2583 | state->fd = fd; | |
6e1271e6 | 2584 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2585 | return state->file; |
2586 | } | |
2587 | ||
4503b767 JA |
2588 | static bool io_bdev_nowait(struct block_device *bdev) |
2589 | { | |
9ba0d0c8 | 2590 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2591 | } |
2592 | ||
2b188cc1 JA |
2593 | /* |
2594 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2595 | * any file. For now, just ensure that anything potentially problematic is done | |
2596 | * inline. | |
2597 | */ | |
7b29f92d | 2598 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2599 | { |
2600 | umode_t mode = file_inode(file)->i_mode; | |
2601 | ||
4503b767 | 2602 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2603 | if (IS_ENABLED(CONFIG_BLOCK) && |
2604 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2605 | return true; |
2606 | return false; | |
2607 | } | |
976517f1 | 2608 | if (S_ISSOCK(mode)) |
2b188cc1 | 2609 | return true; |
4503b767 | 2610 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2611 | if (IS_ENABLED(CONFIG_BLOCK) && |
2612 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2613 | file->f_op != &io_uring_fops) |
2614 | return true; | |
2615 | return false; | |
2616 | } | |
2b188cc1 | 2617 | |
c5b85625 JA |
2618 | /* any ->read/write should understand O_NONBLOCK */ |
2619 | if (file->f_flags & O_NONBLOCK) | |
2620 | return true; | |
2621 | ||
af197f50 JA |
2622 | if (!(file->f_mode & FMODE_NOWAIT)) |
2623 | return false; | |
2624 | ||
2625 | if (rw == READ) | |
2626 | return file->f_op->read_iter != NULL; | |
2627 | ||
2628 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2629 | } |
2630 | ||
7b29f92d JA |
2631 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2632 | { | |
2633 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2634 | return true; | |
2635 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2636 | return true; | |
2637 | ||
2638 | return __io_file_supports_async(req->file, rw); | |
2639 | } | |
2640 | ||
a88fc400 | 2641 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2642 | { |
def596e9 | 2643 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2644 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2645 | struct file *file = req->file; |
09bb8394 JA |
2646 | unsigned ioprio; |
2647 | int ret; | |
2b188cc1 | 2648 | |
7b29f92d | 2649 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2650 | req->flags |= REQ_F_ISREG; |
2651 | ||
2b188cc1 | 2652 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2653 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2654 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2655 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2656 | } |
2b188cc1 | 2657 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2658 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2659 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2660 | if (unlikely(ret)) | |
2661 | return ret; | |
2b188cc1 | 2662 | |
75c668cd PB |
2663 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2664 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2665 | req->flags |= REQ_F_NOWAIT; | |
2666 | ||
2b188cc1 JA |
2667 | ioprio = READ_ONCE(sqe->ioprio); |
2668 | if (ioprio) { | |
2669 | ret = ioprio_check_cap(ioprio); | |
2670 | if (ret) | |
09bb8394 | 2671 | return ret; |
2b188cc1 JA |
2672 | |
2673 | kiocb->ki_ioprio = ioprio; | |
2674 | } else | |
2675 | kiocb->ki_ioprio = get_current_ioprio(); | |
2676 | ||
def596e9 | 2677 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2678 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2679 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2680 | return -EOPNOTSUPP; |
2b188cc1 | 2681 | |
def596e9 JA |
2682 | kiocb->ki_flags |= IOCB_HIPRI; |
2683 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2684 | req->iopoll_completed = 0; |
def596e9 | 2685 | } else { |
09bb8394 JA |
2686 | if (kiocb->ki_flags & IOCB_HIPRI) |
2687 | return -EINVAL; | |
def596e9 JA |
2688 | kiocb->ki_complete = io_complete_rw; |
2689 | } | |
9adbd45d | 2690 | |
eae071c9 PB |
2691 | if (req->opcode == IORING_OP_READ_FIXED || |
2692 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2693 | req->imu = NULL; | |
2694 | io_req_set_rsrc_node(req); | |
2695 | } | |
2696 | ||
3529d8c2 JA |
2697 | req->rw.addr = READ_ONCE(sqe->addr); |
2698 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2699 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2700 | return 0; |
2b188cc1 JA |
2701 | } |
2702 | ||
2703 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2704 | { | |
2705 | switch (ret) { | |
2706 | case -EIOCBQUEUED: | |
2707 | break; | |
2708 | case -ERESTARTSYS: | |
2709 | case -ERESTARTNOINTR: | |
2710 | case -ERESTARTNOHAND: | |
2711 | case -ERESTART_RESTARTBLOCK: | |
2712 | /* | |
2713 | * We can't just restart the syscall, since previously | |
2714 | * submitted sqes may already be in progress. Just fail this | |
2715 | * IO with EINTR. | |
2716 | */ | |
2717 | ret = -EINTR; | |
df561f66 | 2718 | fallthrough; |
2b188cc1 JA |
2719 | default: |
2720 | kiocb->ki_complete(kiocb, ret, 0); | |
2721 | } | |
2722 | } | |
2723 | ||
a1d7c393 | 2724 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2725 | unsigned int issue_flags) |
ba816ad6 | 2726 | { |
ba04291e | 2727 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2728 | struct io_async_rw *io = req->async_data; |
97284637 | 2729 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2730 | |
227c0c96 | 2731 | /* add previously done IO, if any */ |
e8c2bc1f | 2732 | if (io && io->bytes_done > 0) { |
227c0c96 | 2733 | if (ret < 0) |
e8c2bc1f | 2734 | ret = io->bytes_done; |
227c0c96 | 2735 | else |
e8c2bc1f | 2736 | ret += io->bytes_done; |
227c0c96 JA |
2737 | } |
2738 | ||
ba04291e JA |
2739 | if (req->flags & REQ_F_CUR_POS) |
2740 | req->file->f_pos = kiocb->ki_pos; | |
e149bd74 | 2741 | if (ret >= 0 && check_reissue) |
889fca73 | 2742 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2743 | else |
2744 | io_rw_done(kiocb, ret); | |
97284637 | 2745 | |
fe7e3257 | 2746 | if (check_reissue && (req->flags & REQ_F_REISSUE)) { |
97284637 | 2747 | req->flags &= ~REQ_F_REISSUE; |
a7be7c23 | 2748 | if (io_resubmit_prep(req)) { |
8c130827 PB |
2749 | req_ref_get(req); |
2750 | io_queue_async_work(req); | |
2751 | } else { | |
97284637 PB |
2752 | int cflags = 0; |
2753 | ||
93d2bcd2 | 2754 | req_set_fail(req); |
97284637 PB |
2755 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2756 | cflags = io_put_rw_kbuf(req); | |
2757 | __io_req_complete(req, issue_flags, ret, cflags); | |
2758 | } | |
2759 | } | |
ba816ad6 JA |
2760 | } |
2761 | ||
eae071c9 PB |
2762 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2763 | struct io_mapped_ubuf *imu) | |
edafccee | 2764 | { |
9adbd45d | 2765 | size_t len = req->rw.len; |
75769e3f | 2766 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2767 | size_t offset; |
edafccee | 2768 | |
75769e3f | 2769 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2770 | return -EFAULT; |
2771 | /* not inside the mapped region */ | |
4751f53d | 2772 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2773 | return -EFAULT; |
2774 | ||
2775 | /* | |
2776 | * May not be a start of buffer, set size appropriately | |
2777 | * and advance us to the beginning. | |
2778 | */ | |
2779 | offset = buf_addr - imu->ubuf; | |
2780 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2781 | |
2782 | if (offset) { | |
2783 | /* | |
2784 | * Don't use iov_iter_advance() here, as it's really slow for | |
2785 | * using the latter parts of a big fixed buffer - it iterates | |
2786 | * over each segment manually. We can cheat a bit here, because | |
2787 | * we know that: | |
2788 | * | |
2789 | * 1) it's a BVEC iter, we set it up | |
2790 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2791 | * first and last bvec | |
2792 | * | |
2793 | * So just find our index, and adjust the iterator afterwards. | |
2794 | * If the offset is within the first bvec (or the whole first | |
2795 | * bvec, just use iov_iter_advance(). This makes it easier | |
2796 | * since we can just skip the first segment, which may not | |
2797 | * be PAGE_SIZE aligned. | |
2798 | */ | |
2799 | const struct bio_vec *bvec = imu->bvec; | |
2800 | ||
2801 | if (offset <= bvec->bv_len) { | |
2802 | iov_iter_advance(iter, offset); | |
2803 | } else { | |
2804 | unsigned long seg_skip; | |
2805 | ||
2806 | /* skip first vec */ | |
2807 | offset -= bvec->bv_len; | |
2808 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2809 | ||
2810 | iter->bvec = bvec + seg_skip; | |
2811 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2812 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2813 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2814 | } |
2815 | } | |
2816 | ||
847595de | 2817 | return 0; |
edafccee JA |
2818 | } |
2819 | ||
eae071c9 PB |
2820 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
2821 | { | |
2822 | struct io_ring_ctx *ctx = req->ctx; | |
2823 | struct io_mapped_ubuf *imu = req->imu; | |
2824 | u16 index, buf_index = req->buf_index; | |
2825 | ||
2826 | if (likely(!imu)) { | |
2827 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
2828 | return -EFAULT; | |
2829 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2830 | imu = READ_ONCE(ctx->user_bufs[index]); | |
2831 | req->imu = imu; | |
2832 | } | |
2833 | return __io_import_fixed(req, rw, iter, imu); | |
2834 | } | |
2835 | ||
bcda7baa JA |
2836 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2837 | { | |
2838 | if (needs_lock) | |
2839 | mutex_unlock(&ctx->uring_lock); | |
2840 | } | |
2841 | ||
2842 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2843 | { | |
2844 | /* | |
2845 | * "Normal" inline submissions always hold the uring_lock, since we | |
2846 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2847 | * The only exception is when we've detached the request and issue it | |
2848 | * from an async worker thread, grab the lock for that case. | |
2849 | */ | |
2850 | if (needs_lock) | |
2851 | mutex_lock(&ctx->uring_lock); | |
2852 | } | |
2853 | ||
2854 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2855 | int bgid, struct io_buffer *kbuf, | |
2856 | bool needs_lock) | |
2857 | { | |
2858 | struct io_buffer *head; | |
2859 | ||
2860 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2861 | return kbuf; | |
2862 | ||
2863 | io_ring_submit_lock(req->ctx, needs_lock); | |
2864 | ||
2865 | lockdep_assert_held(&req->ctx->uring_lock); | |
2866 | ||
9e15c3a0 | 2867 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2868 | if (head) { |
2869 | if (!list_empty(&head->list)) { | |
2870 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2871 | list); | |
2872 | list_del(&kbuf->list); | |
2873 | } else { | |
2874 | kbuf = head; | |
9e15c3a0 | 2875 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2876 | } |
2877 | if (*len > kbuf->len) | |
2878 | *len = kbuf->len; | |
2879 | } else { | |
2880 | kbuf = ERR_PTR(-ENOBUFS); | |
2881 | } | |
2882 | ||
2883 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2884 | ||
2885 | return kbuf; | |
2886 | } | |
2887 | ||
4d954c25 JA |
2888 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2889 | bool needs_lock) | |
2890 | { | |
2891 | struct io_buffer *kbuf; | |
4f4eeba8 | 2892 | u16 bgid; |
4d954c25 JA |
2893 | |
2894 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2895 | bgid = req->buf_index; |
4d954c25 JA |
2896 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2897 | if (IS_ERR(kbuf)) | |
2898 | return kbuf; | |
2899 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2900 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2901 | return u64_to_user_ptr(kbuf->addr); | |
2902 | } | |
2903 | ||
2904 | #ifdef CONFIG_COMPAT | |
2905 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2906 | bool needs_lock) | |
2907 | { | |
2908 | struct compat_iovec __user *uiov; | |
2909 | compat_ssize_t clen; | |
2910 | void __user *buf; | |
2911 | ssize_t len; | |
2912 | ||
2913 | uiov = u64_to_user_ptr(req->rw.addr); | |
2914 | if (!access_ok(uiov, sizeof(*uiov))) | |
2915 | return -EFAULT; | |
2916 | if (__get_user(clen, &uiov->iov_len)) | |
2917 | return -EFAULT; | |
2918 | if (clen < 0) | |
2919 | return -EINVAL; | |
2920 | ||
2921 | len = clen; | |
2922 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2923 | if (IS_ERR(buf)) | |
2924 | return PTR_ERR(buf); | |
2925 | iov[0].iov_base = buf; | |
2926 | iov[0].iov_len = (compat_size_t) len; | |
2927 | return 0; | |
2928 | } | |
2929 | #endif | |
2930 | ||
2931 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2932 | bool needs_lock) | |
2933 | { | |
2934 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2935 | void __user *buf; | |
2936 | ssize_t len; | |
2937 | ||
2938 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2939 | return -EFAULT; | |
2940 | ||
2941 | len = iov[0].iov_len; | |
2942 | if (len < 0) | |
2943 | return -EINVAL; | |
2944 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2945 | if (IS_ERR(buf)) | |
2946 | return PTR_ERR(buf); | |
2947 | iov[0].iov_base = buf; | |
2948 | iov[0].iov_len = len; | |
2949 | return 0; | |
2950 | } | |
2951 | ||
2952 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2953 | bool needs_lock) | |
2954 | { | |
dddb3e26 JA |
2955 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2956 | struct io_buffer *kbuf; | |
2957 | ||
2958 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2959 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2960 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2961 | return 0; |
dddb3e26 | 2962 | } |
dd201662 | 2963 | if (req->rw.len != 1) |
4d954c25 JA |
2964 | return -EINVAL; |
2965 | ||
2966 | #ifdef CONFIG_COMPAT | |
2967 | if (req->ctx->compat) | |
2968 | return io_compat_import(req, iov, needs_lock); | |
2969 | #endif | |
2970 | ||
2971 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2972 | } | |
2973 | ||
847595de PB |
2974 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2975 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2976 | { |
9adbd45d JA |
2977 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2978 | size_t sqe_len = req->rw.len; | |
847595de | 2979 | u8 opcode = req->opcode; |
4d954c25 | 2980 | ssize_t ret; |
edafccee | 2981 | |
7d009165 | 2982 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2983 | *iovec = NULL; |
9adbd45d | 2984 | return io_import_fixed(req, rw, iter); |
edafccee | 2985 | } |
2b188cc1 | 2986 | |
bcda7baa | 2987 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2988 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2989 | return -EINVAL; |
2990 | ||
3a6820f2 | 2991 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2992 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2993 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2994 | if (IS_ERR(buf)) |
4d954c25 | 2995 | return PTR_ERR(buf); |
3f9d6441 | 2996 | req->rw.len = sqe_len; |
bcda7baa JA |
2997 | } |
2998 | ||
3a6820f2 JA |
2999 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3000 | *iovec = NULL; | |
10fc72e4 | 3001 | return ret; |
3a6820f2 JA |
3002 | } |
3003 | ||
4d954c25 JA |
3004 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3005 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3006 | if (!ret) |
3007 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3008 | *iovec = NULL; |
3009 | return ret; | |
3010 | } | |
3011 | ||
89cd35c5 CH |
3012 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3013 | req->ctx->compat); | |
2b188cc1 JA |
3014 | } |
3015 | ||
0fef9483 JA |
3016 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3017 | { | |
5b09e37e | 3018 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3019 | } |
3020 | ||
31b51510 | 3021 | /* |
32960613 JA |
3022 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3023 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3024 | */ |
4017eb91 | 3025 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3026 | { |
4017eb91 JA |
3027 | struct kiocb *kiocb = &req->rw.kiocb; |
3028 | struct file *file = req->file; | |
32960613 JA |
3029 | ssize_t ret = 0; |
3030 | ||
3031 | /* | |
3032 | * Don't support polled IO through this interface, and we can't | |
3033 | * support non-blocking either. For the latter, this just causes | |
3034 | * the kiocb to be handled from an async context. | |
3035 | */ | |
3036 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3037 | return -EOPNOTSUPP; | |
3038 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3039 | return -EAGAIN; | |
3040 | ||
3041 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3042 | struct iovec iovec; |
32960613 JA |
3043 | ssize_t nr; |
3044 | ||
311ae9e1 PB |
3045 | if (!iov_iter_is_bvec(iter)) { |
3046 | iovec = iov_iter_iovec(iter); | |
3047 | } else { | |
4017eb91 JA |
3048 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3049 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3050 | } |
3051 | ||
32960613 JA |
3052 | if (rw == READ) { |
3053 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3054 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3055 | } else { |
3056 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3057 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3058 | } |
3059 | ||
3060 | if (nr < 0) { | |
3061 | if (!ret) | |
3062 | ret = nr; | |
3063 | break; | |
3064 | } | |
3065 | ret += nr; | |
3066 | if (nr != iovec.iov_len) | |
3067 | break; | |
4017eb91 JA |
3068 | req->rw.len -= nr; |
3069 | req->rw.addr += nr; | |
32960613 JA |
3070 | iov_iter_advance(iter, nr); |
3071 | } | |
3072 | ||
3073 | return ret; | |
3074 | } | |
3075 | ||
ff6165b2 JA |
3076 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3077 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3078 | { |
e8c2bc1f | 3079 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3080 | |
ff6165b2 | 3081 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3082 | rw->free_iovec = iovec; |
227c0c96 | 3083 | rw->bytes_done = 0; |
ff6165b2 | 3084 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3085 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3086 | return; |
b64e3444 | 3087 | if (!iovec) { |
ff6165b2 JA |
3088 | unsigned iov_off = 0; |
3089 | ||
3090 | rw->iter.iov = rw->fast_iov; | |
3091 | if (iter->iov != fast_iov) { | |
3092 | iov_off = iter->iov - fast_iov; | |
3093 | rw->iter.iov += iov_off; | |
3094 | } | |
3095 | if (rw->fast_iov != fast_iov) | |
3096 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3097 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3098 | } else { |
3099 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3100 | } |
3101 | } | |
3102 | ||
6cb78689 | 3103 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3104 | { |
e8c2bc1f JA |
3105 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3106 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3107 | return req->async_data == NULL; | |
3d9932a8 XW |
3108 | } |
3109 | ||
ff6165b2 JA |
3110 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3111 | const struct iovec *fast_iov, | |
227c0c96 | 3112 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3113 | { |
26f0505a | 3114 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3115 | return 0; |
e8c2bc1f | 3116 | if (!req->async_data) { |
6cb78689 | 3117 | if (io_alloc_async_data(req)) { |
6bf985dc | 3118 | kfree(iovec); |
5d204bcf | 3119 | return -ENOMEM; |
6bf985dc | 3120 | } |
b7bb4f7d | 3121 | |
ff6165b2 | 3122 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3123 | } |
b7bb4f7d | 3124 | return 0; |
f67676d1 JA |
3125 | } |
3126 | ||
73debe68 | 3127 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3128 | { |
e8c2bc1f | 3129 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3130 | struct iovec *iov = iorw->fast_iov; |
847595de | 3131 | int ret; |
c3e330a4 | 3132 | |
2846c481 | 3133 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3134 | if (unlikely(ret < 0)) |
3135 | return ret; | |
3136 | ||
ab0b196c PB |
3137 | iorw->bytes_done = 0; |
3138 | iorw->free_iovec = iov; | |
3139 | if (iov) | |
3140 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3141 | return 0; |
3142 | } | |
3143 | ||
73debe68 | 3144 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3145 | { |
3529d8c2 JA |
3146 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3147 | return -EBADF; | |
93642ef8 | 3148 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3149 | } |
3150 | ||
c1dd91d1 JA |
3151 | /* |
3152 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3153 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3154 | * This gets called when the page is unlocked, and we generally expect that to | |
3155 | * happen when the page IO is completed and the page is now uptodate. This will | |
3156 | * queue a task_work based retry of the operation, attempting to copy the data | |
3157 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3158 | * do a thread based blocking retry of the operation. That's the unexpected | |
3159 | * slow path. | |
3160 | */ | |
bcf5a063 JA |
3161 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3162 | int sync, void *arg) | |
3163 | { | |
3164 | struct wait_page_queue *wpq; | |
3165 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3166 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3167 | |
3168 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3169 | ||
cdc8fcb4 LT |
3170 | if (!wake_page_match(wpq, key)) |
3171 | return 0; | |
3172 | ||
c8d317aa | 3173 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3174 | list_del_init(&wait->entry); |
3175 | ||
bcf5a063 | 3176 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3177 | req_ref_get(req); |
921b9054 | 3178 | io_req_task_queue(req); |
bcf5a063 JA |
3179 | return 1; |
3180 | } | |
3181 | ||
c1dd91d1 JA |
3182 | /* |
3183 | * This controls whether a given IO request should be armed for async page | |
3184 | * based retry. If we return false here, the request is handed to the async | |
3185 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3186 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3187 | * will either succeed because the page is now uptodate and unlocked, or it | |
3188 | * will register a callback when the page is unlocked at IO completion. Through | |
3189 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3190 | * That retry will attempt the buffered read again. The retry will generally | |
3191 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3192 | * async worker threads for a blocking retry. | |
3193 | */ | |
227c0c96 | 3194 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3195 | { |
e8c2bc1f JA |
3196 | struct io_async_rw *rw = req->async_data; |
3197 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3198 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3199 | |
bcf5a063 JA |
3200 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3201 | if (req->flags & REQ_F_NOWAIT) | |
3202 | return false; | |
f67676d1 | 3203 | |
227c0c96 | 3204 | /* Only for buffered IO */ |
3b2a4439 | 3205 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3206 | return false; |
3b2a4439 | 3207 | |
bcf5a063 JA |
3208 | /* |
3209 | * just use poll if we can, and don't attempt if the fs doesn't | |
3210 | * support callback based unlocks | |
3211 | */ | |
3212 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3213 | return false; | |
f67676d1 | 3214 | |
3b2a4439 JA |
3215 | wait->wait.func = io_async_buf_func; |
3216 | wait->wait.private = req; | |
3217 | wait->wait.flags = 0; | |
3218 | INIT_LIST_HEAD(&wait->wait.entry); | |
3219 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3220 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3221 | kiocb->ki_waitq = wait; |
3b2a4439 | 3222 | return true; |
bcf5a063 JA |
3223 | } |
3224 | ||
aeab9506 | 3225 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 JA |
3226 | { |
3227 | if (req->file->f_op->read_iter) | |
3228 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3229 | else if (req->file->f_op->read) |
4017eb91 | 3230 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3231 | else |
3232 | return -EINVAL; | |
f67676d1 JA |
3233 | } |
3234 | ||
889fca73 | 3235 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3236 | { |
3237 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3238 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3239 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3240 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3241 | ssize_t io_size, ret, ret2; |
45d189c6 | 3242 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3243 | |
2846c481 | 3244 | if (rw) { |
e8c2bc1f | 3245 | iter = &rw->iter; |
2846c481 PB |
3246 | iovec = NULL; |
3247 | } else { | |
3248 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3249 | if (ret < 0) | |
3250 | return ret; | |
3251 | } | |
632546c4 | 3252 | io_size = iov_iter_count(iter); |
fa15bafb | 3253 | req->result = io_size; |
2b188cc1 | 3254 | |
fd6c2e4c JA |
3255 | /* Ensure we clear previously set non-block flag */ |
3256 | if (!force_nonblock) | |
29de5f6a | 3257 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3258 | else |
3259 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3260 | ||
24c74678 | 3261 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3262 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3263 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3264 | return ret ?: -EAGAIN; |
6713e7a6 | 3265 | } |
9e645e11 | 3266 | |
632546c4 | 3267 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3268 | if (unlikely(ret)) { |
3269 | kfree(iovec); | |
3270 | return ret; | |
3271 | } | |
2b188cc1 | 3272 | |
227c0c96 | 3273 | ret = io_iter_do_read(req, iter); |
32960613 | 3274 | |
230d50d4 | 3275 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3276 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3277 | /* IOPOLL retry should happen for io-wq threads */ |
3278 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3279 | goto done; |
75c668cd PB |
3280 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3281 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3282 | goto done; |
84216315 | 3283 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3284 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3285 | ret = 0; |
230d50d4 JA |
3286 | } else if (ret == -EIOCBQUEUED) { |
3287 | goto out_free; | |
7335e3bf | 3288 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3289 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3290 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3291 | goto done; |
227c0c96 JA |
3292 | } |
3293 | ||
227c0c96 | 3294 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3295 | if (ret2) |
3296 | return ret2; | |
3297 | ||
fe1cdd55 | 3298 | iovec = NULL; |
e8c2bc1f | 3299 | rw = req->async_data; |
227c0c96 | 3300 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3301 | iter = &rw->iter; |
227c0c96 | 3302 | |
b23df91b PB |
3303 | do { |
3304 | io_size -= ret; | |
3305 | rw->bytes_done += ret; | |
3306 | /* if we can retry, do so with the callbacks armed */ | |
3307 | if (!io_rw_should_retry(req)) { | |
3308 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3309 | return -EAGAIN; | |
3310 | } | |
3311 | ||
3312 | /* | |
3313 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3314 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3315 | * desired page gets unlocked. We can also get a partial read | |
3316 | * here, and if we do, then just retry at the new offset. | |
3317 | */ | |
3318 | ret = io_iter_do_read(req, iter); | |
3319 | if (ret == -EIOCBQUEUED) | |
3320 | return 0; | |
227c0c96 | 3321 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3322 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3323 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3324 | done: |
889fca73 | 3325 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3326 | out_free: |
3327 | /* it's faster to check here then delegate to kfree */ | |
3328 | if (iovec) | |
3329 | kfree(iovec); | |
5ea5dd45 | 3330 | return 0; |
2b188cc1 JA |
3331 | } |
3332 | ||
73debe68 | 3333 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3334 | { |
3529d8c2 JA |
3335 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3336 | return -EBADF; | |
93642ef8 | 3337 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3338 | } |
3339 | ||
889fca73 | 3340 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3341 | { |
3342 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3343 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3344 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3345 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3346 | ssize_t ret, ret2, io_size; |
45d189c6 | 3347 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3348 | |
2846c481 | 3349 | if (rw) { |
e8c2bc1f | 3350 | iter = &rw->iter; |
2846c481 PB |
3351 | iovec = NULL; |
3352 | } else { | |
3353 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3354 | if (ret < 0) | |
3355 | return ret; | |
3356 | } | |
632546c4 | 3357 | io_size = iov_iter_count(iter); |
fa15bafb | 3358 | req->result = io_size; |
2b188cc1 | 3359 | |
fd6c2e4c JA |
3360 | /* Ensure we clear previously set non-block flag */ |
3361 | if (!force_nonblock) | |
a88fc400 PB |
3362 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3363 | else | |
3364 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3365 | |
24c74678 | 3366 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3367 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3368 | goto copy_iov; |
31b51510 | 3369 | |
10d59345 JA |
3370 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3371 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3372 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3373 | goto copy_iov; |
31b51510 | 3374 | |
632546c4 | 3375 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3376 | if (unlikely(ret)) |
3377 | goto out_free; | |
4ed734b0 | 3378 | |
fa15bafb PB |
3379 | /* |
3380 | * Open-code file_start_write here to grab freeze protection, | |
3381 | * which will be released by another thread in | |
3382 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3383 | * released so that it doesn't complain about the held lock when | |
3384 | * we return to userspace. | |
3385 | */ | |
3386 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3387 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3388 | __sb_writers_release(file_inode(req->file)->i_sb, |
3389 | SB_FREEZE_WRITE); | |
3390 | } | |
3391 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3392 | |
fa15bafb | 3393 | if (req->file->f_op->write_iter) |
ff6165b2 | 3394 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3395 | else if (req->file->f_op->write) |
4017eb91 | 3396 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3397 | else |
3398 | ret2 = -EINVAL; | |
4ed734b0 | 3399 | |
6ad7f233 PB |
3400 | if (req->flags & REQ_F_REISSUE) { |
3401 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3402 | ret2 = -EAGAIN; |
6ad7f233 | 3403 | } |
230d50d4 | 3404 | |
fa15bafb PB |
3405 | /* |
3406 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3407 | * retry them without IOCB_NOWAIT. | |
3408 | */ | |
3409 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3410 | ret2 = -EAGAIN; | |
75c668cd PB |
3411 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3412 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3413 | goto done; |
fa15bafb | 3414 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3415 | /* IOPOLL retry should happen for io-wq threads */ |
3416 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3417 | goto copy_iov; | |
355afaeb | 3418 | done: |
889fca73 | 3419 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3420 | } else { |
f67676d1 | 3421 | copy_iov: |
84216315 | 3422 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3423 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3424 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3425 | return ret ?: -EAGAIN; |
2b188cc1 | 3426 | } |
31b51510 | 3427 | out_free: |
f261c168 | 3428 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3429 | if (iovec) |
6f2cc166 | 3430 | kfree(iovec); |
2b188cc1 JA |
3431 | return ret; |
3432 | } | |
3433 | ||
80a261fd JA |
3434 | static int io_renameat_prep(struct io_kiocb *req, |
3435 | const struct io_uring_sqe *sqe) | |
3436 | { | |
3437 | struct io_rename *ren = &req->rename; | |
3438 | const char __user *oldf, *newf; | |
3439 | ||
ed7eb259 JA |
3440 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3441 | return -EINVAL; | |
3442 | if (sqe->ioprio || sqe->buf_index) | |
3443 | return -EINVAL; | |
80a261fd JA |
3444 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3445 | return -EBADF; | |
3446 | ||
3447 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3448 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3449 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3450 | ren->new_dfd = READ_ONCE(sqe->len); | |
3451 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3452 | ||
3453 | ren->oldpath = getname(oldf); | |
3454 | if (IS_ERR(ren->oldpath)) | |
3455 | return PTR_ERR(ren->oldpath); | |
3456 | ||
3457 | ren->newpath = getname(newf); | |
3458 | if (IS_ERR(ren->newpath)) { | |
3459 | putname(ren->oldpath); | |
3460 | return PTR_ERR(ren->newpath); | |
3461 | } | |
3462 | ||
3463 | req->flags |= REQ_F_NEED_CLEANUP; | |
3464 | return 0; | |
3465 | } | |
3466 | ||
45d189c6 | 3467 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3468 | { |
3469 | struct io_rename *ren = &req->rename; | |
3470 | int ret; | |
3471 | ||
45d189c6 | 3472 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3473 | return -EAGAIN; |
3474 | ||
3475 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3476 | ren->newpath, ren->flags); | |
3477 | ||
3478 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3479 | if (ret < 0) | |
93d2bcd2 | 3480 | req_set_fail(req); |
80a261fd JA |
3481 | io_req_complete(req, ret); |
3482 | return 0; | |
3483 | } | |
3484 | ||
14a1143b JA |
3485 | static int io_unlinkat_prep(struct io_kiocb *req, |
3486 | const struct io_uring_sqe *sqe) | |
3487 | { | |
3488 | struct io_unlink *un = &req->unlink; | |
3489 | const char __user *fname; | |
3490 | ||
22634bc5 JA |
3491 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3492 | return -EINVAL; | |
3493 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
3494 | return -EINVAL; | |
14a1143b JA |
3495 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
3496 | return -EBADF; | |
3497 | ||
3498 | un->dfd = READ_ONCE(sqe->fd); | |
3499 | ||
3500 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3501 | if (un->flags & ~AT_REMOVEDIR) | |
3502 | return -EINVAL; | |
3503 | ||
3504 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3505 | un->filename = getname(fname); | |
3506 | if (IS_ERR(un->filename)) | |
3507 | return PTR_ERR(un->filename); | |
3508 | ||
3509 | req->flags |= REQ_F_NEED_CLEANUP; | |
3510 | return 0; | |
3511 | } | |
3512 | ||
45d189c6 | 3513 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3514 | { |
3515 | struct io_unlink *un = &req->unlink; | |
3516 | int ret; | |
3517 | ||
45d189c6 | 3518 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3519 | return -EAGAIN; |
3520 | ||
3521 | if (un->flags & AT_REMOVEDIR) | |
3522 | ret = do_rmdir(un->dfd, un->filename); | |
3523 | else | |
3524 | ret = do_unlinkat(un->dfd, un->filename); | |
3525 | ||
3526 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3527 | if (ret < 0) | |
93d2bcd2 | 3528 | req_set_fail(req); |
14a1143b JA |
3529 | io_req_complete(req, ret); |
3530 | return 0; | |
3531 | } | |
3532 | ||
36f4fa68 JA |
3533 | static int io_shutdown_prep(struct io_kiocb *req, |
3534 | const struct io_uring_sqe *sqe) | |
3535 | { | |
3536 | #if defined(CONFIG_NET) | |
3537 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3538 | return -EINVAL; | |
3539 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3540 | sqe->buf_index) | |
3541 | return -EINVAL; | |
3542 | ||
3543 | req->shutdown.how = READ_ONCE(sqe->len); | |
3544 | return 0; | |
3545 | #else | |
3546 | return -EOPNOTSUPP; | |
3547 | #endif | |
3548 | } | |
3549 | ||
45d189c6 | 3550 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3551 | { |
3552 | #if defined(CONFIG_NET) | |
3553 | struct socket *sock; | |
3554 | int ret; | |
3555 | ||
45d189c6 | 3556 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3557 | return -EAGAIN; |
3558 | ||
48aba79b | 3559 | sock = sock_from_file(req->file); |
36f4fa68 | 3560 | if (unlikely(!sock)) |
48aba79b | 3561 | return -ENOTSOCK; |
36f4fa68 JA |
3562 | |
3563 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 3564 | if (ret < 0) |
93d2bcd2 | 3565 | req_set_fail(req); |
36f4fa68 JA |
3566 | io_req_complete(req, ret); |
3567 | return 0; | |
3568 | #else | |
3569 | return -EOPNOTSUPP; | |
3570 | #endif | |
3571 | } | |
3572 | ||
f2a8d5c7 PB |
3573 | static int __io_splice_prep(struct io_kiocb *req, |
3574 | const struct io_uring_sqe *sqe) | |
7d67af2c | 3575 | { |
fe7e3257 | 3576 | struct io_splice *sp = &req->splice; |
7d67af2c | 3577 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 3578 | |
3232dd02 PB |
3579 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3580 | return -EINVAL; | |
7d67af2c PB |
3581 | |
3582 | sp->file_in = NULL; | |
7d67af2c PB |
3583 | sp->len = READ_ONCE(sqe->len); |
3584 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3585 | ||
3586 | if (unlikely(sp->flags & ~valid_flags)) | |
3587 | return -EINVAL; | |
3588 | ||
8371adf5 PB |
3589 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3590 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3591 | if (!sp->file_in) | |
3592 | return -EBADF; | |
7d67af2c | 3593 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3594 | return 0; |
3595 | } | |
3596 | ||
f2a8d5c7 PB |
3597 | static int io_tee_prep(struct io_kiocb *req, |
3598 | const struct io_uring_sqe *sqe) | |
3599 | { | |
3600 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3601 | return -EINVAL; | |
3602 | return __io_splice_prep(req, sqe); | |
3603 | } | |
3604 | ||
45d189c6 | 3605 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3606 | { |
3607 | struct io_splice *sp = &req->splice; | |
3608 | struct file *in = sp->file_in; | |
3609 | struct file *out = sp->file_out; | |
3610 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3611 | long ret = 0; | |
3612 | ||
45d189c6 | 3613 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3614 | return -EAGAIN; |
3615 | if (sp->len) | |
3616 | ret = do_tee(in, out, sp->len, flags); | |
3617 | ||
e1d767f0 PB |
3618 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3619 | io_put_file(in); | |
f2a8d5c7 PB |
3620 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3621 | ||
f2a8d5c7 | 3622 | if (ret != sp->len) |
93d2bcd2 | 3623 | req_set_fail(req); |
e1e16097 | 3624 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3625 | return 0; |
3626 | } | |
3627 | ||
3628 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3629 | { | |
fe7e3257 | 3630 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
3631 | |
3632 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3633 | sp->off_out = READ_ONCE(sqe->off); | |
3634 | return __io_splice_prep(req, sqe); | |
3635 | } | |
3636 | ||
45d189c6 | 3637 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3638 | { |
3639 | struct io_splice *sp = &req->splice; | |
3640 | struct file *in = sp->file_in; | |
3641 | struct file *out = sp->file_out; | |
3642 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3643 | loff_t *poff_in, *poff_out; | |
c9687426 | 3644 | long ret = 0; |
7d67af2c | 3645 | |
45d189c6 | 3646 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3647 | return -EAGAIN; |
7d67af2c PB |
3648 | |
3649 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3650 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3651 | |
948a7749 | 3652 | if (sp->len) |
c9687426 | 3653 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3654 | |
e1d767f0 PB |
3655 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3656 | io_put_file(in); | |
7d67af2c PB |
3657 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3658 | ||
7d67af2c | 3659 | if (ret != sp->len) |
93d2bcd2 | 3660 | req_set_fail(req); |
e1e16097 | 3661 | io_req_complete(req, ret); |
7d67af2c PB |
3662 | return 0; |
3663 | } | |
3664 | ||
2b188cc1 JA |
3665 | /* |
3666 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3667 | */ | |
889fca73 | 3668 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3669 | { |
3670 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3671 | |
def596e9 JA |
3672 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3673 | return -EINVAL; | |
3674 | ||
889fca73 | 3675 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3676 | return 0; |
3677 | } | |
3678 | ||
1155c76a | 3679 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3680 | { |
6b06314c | 3681 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3682 | |
09bb8394 JA |
3683 | if (!req->file) |
3684 | return -EBADF; | |
c992fe29 | 3685 | |
6b06314c | 3686 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3687 | return -EINVAL; |
edafccee | 3688 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3689 | return -EINVAL; |
3690 | ||
8ed8d3c3 JA |
3691 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3692 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3693 | return -EINVAL; | |
3694 | ||
3695 | req->sync.off = READ_ONCE(sqe->off); | |
3696 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3697 | return 0; |
3698 | } | |
3699 | ||
45d189c6 | 3700 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3701 | { |
8ed8d3c3 | 3702 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3703 | int ret; |
3704 | ||
ac45abc0 | 3705 | /* fsync always requires a blocking context */ |
45d189c6 | 3706 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3707 | return -EAGAIN; |
3708 | ||
9adbd45d | 3709 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3710 | end > 0 ? end : LLONG_MAX, |
3711 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3712 | if (ret < 0) | |
93d2bcd2 | 3713 | req_set_fail(req); |
e1e16097 | 3714 | io_req_complete(req, ret); |
c992fe29 CH |
3715 | return 0; |
3716 | } | |
3717 | ||
d63d1b5e JA |
3718 | static int io_fallocate_prep(struct io_kiocb *req, |
3719 | const struct io_uring_sqe *sqe) | |
3720 | { | |
3721 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3722 | return -EINVAL; | |
3232dd02 PB |
3723 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3724 | return -EINVAL; | |
d63d1b5e JA |
3725 | |
3726 | req->sync.off = READ_ONCE(sqe->off); | |
3727 | req->sync.len = READ_ONCE(sqe->addr); | |
3728 | req->sync.mode = READ_ONCE(sqe->len); | |
3729 | return 0; | |
3730 | } | |
3731 | ||
45d189c6 | 3732 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3733 | { |
ac45abc0 PB |
3734 | int ret; |
3735 | ||
d63d1b5e | 3736 | /* fallocate always requiring blocking context */ |
45d189c6 | 3737 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3738 | return -EAGAIN; |
ac45abc0 PB |
3739 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3740 | req->sync.len); | |
ac45abc0 | 3741 | if (ret < 0) |
93d2bcd2 | 3742 | req_set_fail(req); |
e1e16097 | 3743 | io_req_complete(req, ret); |
5d17b4a4 JA |
3744 | return 0; |
3745 | } | |
3746 | ||
ec65fea5 | 3747 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3748 | { |
f8748881 | 3749 | const char __user *fname; |
15b71abe | 3750 | int ret; |
b7bb4f7d | 3751 | |
ec65fea5 | 3752 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3753 | return -EINVAL; |
ec65fea5 | 3754 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3755 | return -EBADF; |
03b1230c | 3756 | |
ec65fea5 PB |
3757 | /* open.how should be already initialised */ |
3758 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3759 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3760 | |
25e72d10 PB |
3761 | req->open.dfd = READ_ONCE(sqe->fd); |
3762 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3763 | req->open.filename = getname(fname); |
15b71abe JA |
3764 | if (IS_ERR(req->open.filename)) { |
3765 | ret = PTR_ERR(req->open.filename); | |
3766 | req->open.filename = NULL; | |
3767 | return ret; | |
3768 | } | |
4022e7af | 3769 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3770 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3771 | return 0; |
03b1230c JA |
3772 | } |
3773 | ||
ec65fea5 PB |
3774 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3775 | { | |
3776 | u64 flags, mode; | |
3777 | ||
14587a46 | 3778 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3779 | return -EINVAL; |
ec65fea5 PB |
3780 | mode = READ_ONCE(sqe->len); |
3781 | flags = READ_ONCE(sqe->open_flags); | |
3782 | req->open.how = build_open_how(flags, mode); | |
3783 | return __io_openat_prep(req, sqe); | |
3784 | } | |
3785 | ||
cebdb986 | 3786 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3787 | { |
cebdb986 | 3788 | struct open_how __user *how; |
cebdb986 | 3789 | size_t len; |
0fa03c62 JA |
3790 | int ret; |
3791 | ||
14587a46 | 3792 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3793 | return -EINVAL; |
cebdb986 JA |
3794 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3795 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3796 | if (len < OPEN_HOW_SIZE_VER0) |
3797 | return -EINVAL; | |
3529d8c2 | 3798 | |
cebdb986 JA |
3799 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3800 | len); | |
3801 | if (ret) | |
3802 | return ret; | |
3529d8c2 | 3803 | |
ec65fea5 | 3804 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3805 | } |
3806 | ||
45d189c6 | 3807 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3808 | { |
3809 | struct open_flags op; | |
15b71abe | 3810 | struct file *file; |
3a81fd02 JA |
3811 | bool nonblock_set; |
3812 | bool resolve_nonblock; | |
15b71abe JA |
3813 | int ret; |
3814 | ||
cebdb986 | 3815 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3816 | if (ret) |
3817 | goto err; | |
3a81fd02 JA |
3818 | nonblock_set = op.open_flag & O_NONBLOCK; |
3819 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3820 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3821 | /* |
3822 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3823 | * it'll always -EAGAIN | |
3824 | */ | |
3825 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3826 | return -EAGAIN; | |
3827 | op.lookup_flags |= LOOKUP_CACHED; | |
3828 | op.open_flag |= O_NONBLOCK; | |
3829 | } | |
15b71abe | 3830 | |
4022e7af | 3831 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3832 | if (ret < 0) |
3833 | goto err; | |
3834 | ||
3835 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 3836 | if (IS_ERR(file)) { |
944d1444 | 3837 | /* |
12dcb58a PB |
3838 | * We could hang on to this 'fd' on retrying, but seems like |
3839 | * marginal gain for something that is now known to be a slower | |
3840 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 3841 | */ |
3a81fd02 | 3842 | put_unused_fd(ret); |
3a81fd02 | 3843 | |
15b71abe | 3844 | ret = PTR_ERR(file); |
12dcb58a PB |
3845 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
3846 | if (ret == -EAGAIN && | |
3847 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
3848 | return -EAGAIN; | |
3849 | goto err; | |
15b71abe | 3850 | } |
12dcb58a PB |
3851 | |
3852 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
3853 | file->f_flags &= ~O_NONBLOCK; | |
3854 | fsnotify_open(file); | |
3855 | fd_install(ret, file); | |
15b71abe JA |
3856 | err: |
3857 | putname(req->open.filename); | |
8fef80bf | 3858 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 3859 | if (ret < 0) |
93d2bcd2 | 3860 | req_set_fail(req); |
0bdf3398 | 3861 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3862 | return 0; |
3863 | } | |
3864 | ||
45d189c6 | 3865 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3866 | { |
e45cff58 | 3867 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3868 | } |
3869 | ||
067524e9 JA |
3870 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3871 | const struct io_uring_sqe *sqe) | |
3872 | { | |
3873 | struct io_provide_buf *p = &req->pbuf; | |
3874 | u64 tmp; | |
3875 | ||
3876 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3877 | return -EINVAL; | |
3878 | ||
3879 | tmp = READ_ONCE(sqe->fd); | |
3880 | if (!tmp || tmp > USHRT_MAX) | |
3881 | return -EINVAL; | |
3882 | ||
3883 | memset(p, 0, sizeof(*p)); | |
3884 | p->nbufs = tmp; | |
3885 | p->bgid = READ_ONCE(sqe->buf_group); | |
3886 | return 0; | |
3887 | } | |
3888 | ||
3889 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3890 | int bgid, unsigned nbufs) | |
3891 | { | |
3892 | unsigned i = 0; | |
3893 | ||
3894 | /* shouldn't happen */ | |
3895 | if (!nbufs) | |
3896 | return 0; | |
3897 | ||
3898 | /* the head kbuf is the list itself */ | |
3899 | while (!list_empty(&buf->list)) { | |
3900 | struct io_buffer *nxt; | |
3901 | ||
3902 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3903 | list_del(&nxt->list); | |
3904 | kfree(nxt); | |
3905 | if (++i == nbufs) | |
3906 | return i; | |
3907 | } | |
3908 | i++; | |
3909 | kfree(buf); | |
9e15c3a0 | 3910 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3911 | |
3912 | return i; | |
3913 | } | |
3914 | ||
889fca73 | 3915 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3916 | { |
3917 | struct io_provide_buf *p = &req->pbuf; | |
3918 | struct io_ring_ctx *ctx = req->ctx; | |
3919 | struct io_buffer *head; | |
3920 | int ret = 0; | |
45d189c6 | 3921 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3922 | |
3923 | io_ring_submit_lock(ctx, !force_nonblock); | |
3924 | ||
3925 | lockdep_assert_held(&ctx->uring_lock); | |
3926 | ||
3927 | ret = -ENOENT; | |
9e15c3a0 | 3928 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3929 | if (head) |
3930 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 | 3931 | if (ret < 0) |
93d2bcd2 | 3932 | req_set_fail(req); |
067524e9 | 3933 | |
9fb8cb49 PB |
3934 | /* complete before unlock, IOPOLL may need the lock */ |
3935 | __io_req_complete(req, issue_flags, ret, 0); | |
3936 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3937 | return 0; |
3938 | } | |
3939 | ||
ddf0322d JA |
3940 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3941 | const struct io_uring_sqe *sqe) | |
3942 | { | |
38134ada | 3943 | unsigned long size, tmp_check; |
ddf0322d JA |
3944 | struct io_provide_buf *p = &req->pbuf; |
3945 | u64 tmp; | |
3946 | ||
3947 | if (sqe->ioprio || sqe->rw_flags) | |
3948 | return -EINVAL; | |
3949 | ||
3950 | tmp = READ_ONCE(sqe->fd); | |
3951 | if (!tmp || tmp > USHRT_MAX) | |
3952 | return -E2BIG; | |
3953 | p->nbufs = tmp; | |
3954 | p->addr = READ_ONCE(sqe->addr); | |
3955 | p->len = READ_ONCE(sqe->len); | |
3956 | ||
38134ada PB |
3957 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
3958 | &size)) | |
3959 | return -EOVERFLOW; | |
3960 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
3961 | return -EOVERFLOW; | |
3962 | ||
d81269fe PB |
3963 | size = (unsigned long)p->len * p->nbufs; |
3964 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3965 | return -EFAULT; |
3966 | ||
3967 | p->bgid = READ_ONCE(sqe->buf_group); | |
3968 | tmp = READ_ONCE(sqe->off); | |
3969 | if (tmp > USHRT_MAX) | |
3970 | return -E2BIG; | |
3971 | p->bid = tmp; | |
3972 | return 0; | |
3973 | } | |
3974 | ||
3975 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3976 | { | |
3977 | struct io_buffer *buf; | |
3978 | u64 addr = pbuf->addr; | |
3979 | int i, bid = pbuf->bid; | |
3980 | ||
3981 | for (i = 0; i < pbuf->nbufs; i++) { | |
3982 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3983 | if (!buf) | |
3984 | break; | |
3985 | ||
3986 | buf->addr = addr; | |
d1f82808 | 3987 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d JA |
3988 | buf->bid = bid; |
3989 | addr += pbuf->len; | |
3990 | bid++; | |
3991 | if (!*head) { | |
3992 | INIT_LIST_HEAD(&buf->list); | |
3993 | *head = buf; | |
3994 | } else { | |
3995 | list_add_tail(&buf->list, &(*head)->list); | |
3996 | } | |
3997 | } | |
3998 | ||
3999 | return i ? i : -ENOMEM; | |
4000 | } | |
4001 | ||
889fca73 | 4002 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4003 | { |
4004 | struct io_provide_buf *p = &req->pbuf; | |
4005 | struct io_ring_ctx *ctx = req->ctx; | |
4006 | struct io_buffer *head, *list; | |
4007 | int ret = 0; | |
45d189c6 | 4008 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4009 | |
4010 | io_ring_submit_lock(ctx, !force_nonblock); | |
4011 | ||
4012 | lockdep_assert_held(&ctx->uring_lock); | |
4013 | ||
9e15c3a0 | 4014 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4015 | |
4016 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4017 | if (ret >= 0 && !list) { |
4018 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4019 | if (ret < 0) | |
067524e9 | 4020 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4021 | } |
ddf0322d | 4022 | if (ret < 0) |
93d2bcd2 | 4023 | req_set_fail(req); |
9fb8cb49 PB |
4024 | /* complete before unlock, IOPOLL may need the lock */ |
4025 | __io_req_complete(req, issue_flags, ret, 0); | |
4026 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4027 | return 0; |
cebdb986 JA |
4028 | } |
4029 | ||
3e4827b0 JA |
4030 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4031 | const struct io_uring_sqe *sqe) | |
4032 | { | |
4033 | #if defined(CONFIG_EPOLL) | |
4034 | if (sqe->ioprio || sqe->buf_index) | |
4035 | return -EINVAL; | |
2d74d042 | 4036 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4037 | return -EINVAL; |
3e4827b0 JA |
4038 | |
4039 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4040 | req->epoll.op = READ_ONCE(sqe->len); | |
4041 | req->epoll.fd = READ_ONCE(sqe->off); | |
4042 | ||
4043 | if (ep_op_has_event(req->epoll.op)) { | |
4044 | struct epoll_event __user *ev; | |
4045 | ||
4046 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4047 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4048 | return -EFAULT; | |
4049 | } | |
4050 | ||
4051 | return 0; | |
4052 | #else | |
4053 | return -EOPNOTSUPP; | |
4054 | #endif | |
4055 | } | |
4056 | ||
889fca73 | 4057 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4058 | { |
4059 | #if defined(CONFIG_EPOLL) | |
4060 | struct io_epoll *ie = &req->epoll; | |
4061 | int ret; | |
45d189c6 | 4062 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4063 | |
4064 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4065 | if (force_nonblock && ret == -EAGAIN) | |
4066 | return -EAGAIN; | |
4067 | ||
4068 | if (ret < 0) | |
93d2bcd2 | 4069 | req_set_fail(req); |
889fca73 | 4070 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4071 | return 0; |
4072 | #else | |
4073 | return -EOPNOTSUPP; | |
4074 | #endif | |
4075 | } | |
4076 | ||
c1ca757b JA |
4077 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4078 | { | |
4079 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4080 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4081 | return -EINVAL; | |
3232dd02 PB |
4082 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4083 | return -EINVAL; | |
c1ca757b JA |
4084 | |
4085 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4086 | req->madvise.len = READ_ONCE(sqe->len); | |
4087 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4088 | return 0; | |
4089 | #else | |
4090 | return -EOPNOTSUPP; | |
4091 | #endif | |
4092 | } | |
4093 | ||
45d189c6 | 4094 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4095 | { |
4096 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4097 | struct io_madvise *ma = &req->madvise; | |
4098 | int ret; | |
4099 | ||
45d189c6 | 4100 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4101 | return -EAGAIN; |
4102 | ||
0726b01e | 4103 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4104 | if (ret < 0) |
93d2bcd2 | 4105 | req_set_fail(req); |
e1e16097 | 4106 | io_req_complete(req, ret); |
c1ca757b JA |
4107 | return 0; |
4108 | #else | |
4109 | return -EOPNOTSUPP; | |
4110 | #endif | |
4111 | } | |
4112 | ||
4840e418 JA |
4113 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4114 | { | |
4115 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4116 | return -EINVAL; | |
3232dd02 PB |
4117 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4118 | return -EINVAL; | |
4840e418 JA |
4119 | |
4120 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4121 | req->fadvise.len = READ_ONCE(sqe->len); | |
4122 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4123 | return 0; | |
4124 | } | |
4125 | ||
45d189c6 | 4126 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4127 | { |
4128 | struct io_fadvise *fa = &req->fadvise; | |
4129 | int ret; | |
4130 | ||
45d189c6 | 4131 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4132 | switch (fa->advice) { |
4133 | case POSIX_FADV_NORMAL: | |
4134 | case POSIX_FADV_RANDOM: | |
4135 | case POSIX_FADV_SEQUENTIAL: | |
4136 | break; | |
4137 | default: | |
4138 | return -EAGAIN; | |
4139 | } | |
4140 | } | |
4840e418 JA |
4141 | |
4142 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4143 | if (ret < 0) | |
93d2bcd2 | 4144 | req_set_fail(req); |
0bdf3398 | 4145 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4146 | return 0; |
4147 | } | |
4148 | ||
eddc7ef5 JA |
4149 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4150 | { | |
2d74d042 | 4151 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4152 | return -EINVAL; |
eddc7ef5 JA |
4153 | if (sqe->ioprio || sqe->buf_index) |
4154 | return -EINVAL; | |
9c280f90 | 4155 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4156 | return -EBADF; |
eddc7ef5 | 4157 | |
1d9e1288 BM |
4158 | req->statx.dfd = READ_ONCE(sqe->fd); |
4159 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4160 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4161 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4162 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4163 | |
4164 | return 0; | |
4165 | } | |
4166 | ||
45d189c6 | 4167 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4168 | { |
1d9e1288 | 4169 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4170 | int ret; |
4171 | ||
59d70013 | 4172 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4173 | return -EAGAIN; |
4174 | ||
e62753e4 BM |
4175 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4176 | ctx->buffer); | |
eddc7ef5 | 4177 | |
eddc7ef5 | 4178 | if (ret < 0) |
93d2bcd2 | 4179 | req_set_fail(req); |
e1e16097 | 4180 | io_req_complete(req, ret); |
eddc7ef5 JA |
4181 | return 0; |
4182 | } | |
4183 | ||
b5dba59e JA |
4184 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4185 | { | |
14587a46 | 4186 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4187 | return -EINVAL; |
b5dba59e JA |
4188 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4189 | sqe->rw_flags || sqe->buf_index) | |
4190 | return -EINVAL; | |
9c280f90 | 4191 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4192 | return -EBADF; |
b5dba59e JA |
4193 | |
4194 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4195 | return 0; |
b5dba59e JA |
4196 | } |
4197 | ||
889fca73 | 4198 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4199 | { |
9eac1904 | 4200 | struct files_struct *files = current->files; |
3af73b28 | 4201 | struct io_close *close = &req->close; |
9eac1904 | 4202 | struct fdtable *fdt; |
a1fde923 PB |
4203 | struct file *file = NULL; |
4204 | int ret = -EBADF; | |
b5dba59e | 4205 | |
9eac1904 JA |
4206 | spin_lock(&files->file_lock); |
4207 | fdt = files_fdtable(files); | |
4208 | if (close->fd >= fdt->max_fds) { | |
4209 | spin_unlock(&files->file_lock); | |
4210 | goto err; | |
4211 | } | |
4212 | file = fdt->fd[close->fd]; | |
a1fde923 | 4213 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4214 | spin_unlock(&files->file_lock); |
4215 | file = NULL; | |
4216 | goto err; | |
3af73b28 | 4217 | } |
b5dba59e JA |
4218 | |
4219 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4220 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4221 | spin_unlock(&files->file_lock); |
0bf0eefd | 4222 | return -EAGAIN; |
a2100672 | 4223 | } |
b5dba59e | 4224 | |
9eac1904 JA |
4225 | ret = __close_fd_get_file(close->fd, &file); |
4226 | spin_unlock(&files->file_lock); | |
4227 | if (ret < 0) { | |
4228 | if (ret == -ENOENT) | |
4229 | ret = -EBADF; | |
4230 | goto err; | |
4231 | } | |
4232 | ||
3af73b28 | 4233 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4234 | ret = filp_close(file, current->files); |
4235 | err: | |
3af73b28 | 4236 | if (ret < 0) |
93d2bcd2 | 4237 | req_set_fail(req); |
9eac1904 JA |
4238 | if (file) |
4239 | fput(file); | |
889fca73 | 4240 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4241 | return 0; |
b5dba59e JA |
4242 | } |
4243 | ||
1155c76a | 4244 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4245 | { |
4246 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4247 | |
5d17b4a4 JA |
4248 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4249 | return -EINVAL; | |
4250 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4251 | return -EINVAL; | |
4252 | ||
8ed8d3c3 JA |
4253 | req->sync.off = READ_ONCE(sqe->off); |
4254 | req->sync.len = READ_ONCE(sqe->len); | |
4255 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4256 | return 0; |
4257 | } | |
4258 | ||
45d189c6 | 4259 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4260 | { |
8ed8d3c3 JA |
4261 | int ret; |
4262 | ||
ac45abc0 | 4263 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4264 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4265 | return -EAGAIN; |
4266 | ||
9adbd45d | 4267 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4268 | req->sync.flags); |
4269 | if (ret < 0) | |
93d2bcd2 | 4270 | req_set_fail(req); |
e1e16097 | 4271 | io_req_complete(req, ret); |
5d17b4a4 JA |
4272 | return 0; |
4273 | } | |
4274 | ||
469956e8 | 4275 | #if defined(CONFIG_NET) |
02d27d89 PB |
4276 | static int io_setup_async_msg(struct io_kiocb *req, |
4277 | struct io_async_msghdr *kmsg) | |
4278 | { | |
e8c2bc1f JA |
4279 | struct io_async_msghdr *async_msg = req->async_data; |
4280 | ||
4281 | if (async_msg) | |
02d27d89 | 4282 | return -EAGAIN; |
e8c2bc1f | 4283 | if (io_alloc_async_data(req)) { |
257e84a5 | 4284 | kfree(kmsg->free_iov); |
02d27d89 PB |
4285 | return -ENOMEM; |
4286 | } | |
e8c2bc1f | 4287 | async_msg = req->async_data; |
02d27d89 | 4288 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4289 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4290 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4291 | /* if were using fast_iov, set it to the new one */ |
4292 | if (!async_msg->free_iov) | |
4293 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4294 | ||
02d27d89 PB |
4295 | return -EAGAIN; |
4296 | } | |
4297 | ||
2ae523ed PB |
4298 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4299 | struct io_async_msghdr *iomsg) | |
4300 | { | |
2ae523ed | 4301 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4302 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4303 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4304 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4305 | } |
4306 | ||
93642ef8 PB |
4307 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4308 | { | |
4309 | int ret; | |
4310 | ||
93642ef8 PB |
4311 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4312 | if (!ret) | |
4313 | req->flags |= REQ_F_NEED_CLEANUP; | |
4314 | return ret; | |
4315 | } | |
4316 | ||
3529d8c2 | 4317 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4318 | { |
e47293fd | 4319 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4320 | |
d2b6f48b PB |
4321 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4322 | return -EINVAL; | |
4323 | ||
270a5940 | 4324 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4325 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4326 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4327 | if (sr->msg_flags & MSG_DONTWAIT) | |
4328 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4329 | |
d8768362 JA |
4330 | #ifdef CONFIG_COMPAT |
4331 | if (req->ctx->compat) | |
4332 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4333 | #endif | |
93642ef8 | 4334 | return 0; |
03b1230c JA |
4335 | } |
4336 | ||
889fca73 | 4337 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4338 | { |
6b754c8b | 4339 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4340 | struct socket *sock; |
7a7cacba | 4341 | unsigned flags; |
0031275d | 4342 | int min_ret = 0; |
0fa03c62 JA |
4343 | int ret; |
4344 | ||
dba4a925 | 4345 | sock = sock_from_file(req->file); |
7a7cacba | 4346 | if (unlikely(!sock)) |
dba4a925 | 4347 | return -ENOTSOCK; |
3529d8c2 | 4348 | |
257e84a5 PB |
4349 | kmsg = req->async_data; |
4350 | if (!kmsg) { | |
7a7cacba PB |
4351 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4352 | if (ret) | |
4353 | return ret; | |
4354 | kmsg = &iomsg; | |
0fa03c62 | 4355 | } |
0fa03c62 | 4356 | |
04411806 PB |
4357 | flags = req->sr_msg.msg_flags; |
4358 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4359 | flags |= MSG_DONTWAIT; |
0031275d SM |
4360 | if (flags & MSG_WAITALL) |
4361 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4362 | ||
7a7cacba | 4363 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4364 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4365 | return io_setup_async_msg(req, kmsg); |
4366 | if (ret == -ERESTARTSYS) | |
4367 | ret = -EINTR; | |
0fa03c62 | 4368 | |
257e84a5 PB |
4369 | /* fast path, check for non-NULL to avoid function call */ |
4370 | if (kmsg->free_iov) | |
4371 | kfree(kmsg->free_iov); | |
99bc4c38 | 4372 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4373 | if (ret < min_ret) |
93d2bcd2 | 4374 | req_set_fail(req); |
889fca73 | 4375 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4376 | return 0; |
03b1230c | 4377 | } |
aa1fa28f | 4378 | |
889fca73 | 4379 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4380 | { |
7a7cacba PB |
4381 | struct io_sr_msg *sr = &req->sr_msg; |
4382 | struct msghdr msg; | |
4383 | struct iovec iov; | |
fddaface | 4384 | struct socket *sock; |
7a7cacba | 4385 | unsigned flags; |
0031275d | 4386 | int min_ret = 0; |
fddaface JA |
4387 | int ret; |
4388 | ||
dba4a925 | 4389 | sock = sock_from_file(req->file); |
7a7cacba | 4390 | if (unlikely(!sock)) |
dba4a925 | 4391 | return -ENOTSOCK; |
fddaface | 4392 | |
7a7cacba PB |
4393 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4394 | if (unlikely(ret)) | |
14db8411 | 4395 | return ret; |
fddaface | 4396 | |
7a7cacba PB |
4397 | msg.msg_name = NULL; |
4398 | msg.msg_control = NULL; | |
4399 | msg.msg_controllen = 0; | |
4400 | msg.msg_namelen = 0; | |
fddaface | 4401 | |
04411806 PB |
4402 | flags = req->sr_msg.msg_flags; |
4403 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4404 | flags |= MSG_DONTWAIT; |
0031275d SM |
4405 | if (flags & MSG_WAITALL) |
4406 | min_ret = iov_iter_count(&msg.msg_iter); | |
4407 | ||
7a7cacba PB |
4408 | msg.msg_flags = flags; |
4409 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4410 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4411 | return -EAGAIN; |
4412 | if (ret == -ERESTARTSYS) | |
4413 | ret = -EINTR; | |
fddaface | 4414 | |
0031275d | 4415 | if (ret < min_ret) |
93d2bcd2 | 4416 | req_set_fail(req); |
889fca73 | 4417 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4418 | return 0; |
fddaface JA |
4419 | } |
4420 | ||
1400e697 PB |
4421 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4422 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4423 | { |
4424 | struct io_sr_msg *sr = &req->sr_msg; | |
4425 | struct iovec __user *uiov; | |
4426 | size_t iov_len; | |
4427 | int ret; | |
4428 | ||
1400e697 PB |
4429 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4430 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4431 | if (ret) |
4432 | return ret; | |
4433 | ||
4434 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4435 | if (iov_len > 1) | |
4436 | return -EINVAL; | |
5476dfed | 4437 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4438 | return -EFAULT; |
5476dfed | 4439 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4440 | iomsg->free_iov = NULL; |
52de1fe1 | 4441 | } else { |
257e84a5 | 4442 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4443 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4444 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4445 | false); |
52de1fe1 JA |
4446 | if (ret > 0) |
4447 | ret = 0; | |
4448 | } | |
4449 | ||
4450 | return ret; | |
4451 | } | |
4452 | ||
4453 | #ifdef CONFIG_COMPAT | |
4454 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4455 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4456 | { |
52de1fe1 JA |
4457 | struct io_sr_msg *sr = &req->sr_msg; |
4458 | struct compat_iovec __user *uiov; | |
4459 | compat_uptr_t ptr; | |
4460 | compat_size_t len; | |
4461 | int ret; | |
4462 | ||
4af3417a PB |
4463 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4464 | &ptr, &len); | |
52de1fe1 JA |
4465 | if (ret) |
4466 | return ret; | |
4467 | ||
4468 | uiov = compat_ptr(ptr); | |
4469 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4470 | compat_ssize_t clen; | |
4471 | ||
4472 | if (len > 1) | |
4473 | return -EINVAL; | |
4474 | if (!access_ok(uiov, sizeof(*uiov))) | |
4475 | return -EFAULT; | |
4476 | if (__get_user(clen, &uiov->iov_len)) | |
4477 | return -EFAULT; | |
4478 | if (clen < 0) | |
4479 | return -EINVAL; | |
2d280bc8 | 4480 | sr->len = clen; |
257e84a5 | 4481 | iomsg->free_iov = NULL; |
52de1fe1 | 4482 | } else { |
257e84a5 | 4483 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4484 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4485 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4486 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4487 | if (ret < 0) |
4488 | return ret; | |
4489 | } | |
4490 | ||
4491 | return 0; | |
4492 | } | |
4493 | #endif | |
4494 | ||
1400e697 PB |
4495 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4496 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4497 | { |
1400e697 | 4498 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4499 | |
4500 | #ifdef CONFIG_COMPAT | |
4501 | if (req->ctx->compat) | |
1400e697 | 4502 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4503 | #endif |
52de1fe1 | 4504 | |
1400e697 | 4505 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4506 | } |
4507 | ||
bcda7baa | 4508 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4509 | bool needs_lock) |
bcda7baa JA |
4510 | { |
4511 | struct io_sr_msg *sr = &req->sr_msg; | |
4512 | struct io_buffer *kbuf; | |
4513 | ||
bcda7baa JA |
4514 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4515 | if (IS_ERR(kbuf)) | |
4516 | return kbuf; | |
4517 | ||
4518 | sr->kbuf = kbuf; | |
4519 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4520 | return kbuf; |
fddaface JA |
4521 | } |
4522 | ||
7fbb1b54 PB |
4523 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4524 | { | |
4525 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4526 | } | |
4527 | ||
93642ef8 | 4528 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4529 | { |
99bc4c38 | 4530 | int ret; |
3529d8c2 | 4531 | |
93642ef8 PB |
4532 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4533 | if (!ret) | |
4534 | req->flags |= REQ_F_NEED_CLEANUP; | |
4535 | return ret; | |
4536 | } | |
4537 | ||
4538 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4539 | { | |
4540 | struct io_sr_msg *sr = &req->sr_msg; | |
4541 | ||
d2b6f48b PB |
4542 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4543 | return -EINVAL; | |
4544 | ||
270a5940 | 4545 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4546 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4547 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4548 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4549 | if (sr->msg_flags & MSG_DONTWAIT) | |
4550 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4551 | |
d8768362 JA |
4552 | #ifdef CONFIG_COMPAT |
4553 | if (req->ctx->compat) | |
4554 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4555 | #endif | |
93642ef8 | 4556 | return 0; |
aa1fa28f JA |
4557 | } |
4558 | ||
889fca73 | 4559 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4560 | { |
6b754c8b | 4561 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4562 | struct socket *sock; |
7fbb1b54 | 4563 | struct io_buffer *kbuf; |
7a7cacba | 4564 | unsigned flags; |
0031275d | 4565 | int min_ret = 0; |
52de1fe1 | 4566 | int ret, cflags = 0; |
45d189c6 | 4567 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4568 | |
dba4a925 | 4569 | sock = sock_from_file(req->file); |
7a7cacba | 4570 | if (unlikely(!sock)) |
dba4a925 | 4571 | return -ENOTSOCK; |
3529d8c2 | 4572 | |
257e84a5 PB |
4573 | kmsg = req->async_data; |
4574 | if (!kmsg) { | |
7a7cacba PB |
4575 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4576 | if (ret) | |
681fda8d | 4577 | return ret; |
7a7cacba PB |
4578 | kmsg = &iomsg; |
4579 | } | |
03b1230c | 4580 | |
bc02ef33 | 4581 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4582 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4583 | if (IS_ERR(kbuf)) |
52de1fe1 | 4584 | return PTR_ERR(kbuf); |
7a7cacba | 4585 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4586 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4587 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4588 | 1, req->sr_msg.len); |
4589 | } | |
52de1fe1 | 4590 | |
04411806 PB |
4591 | flags = req->sr_msg.msg_flags; |
4592 | if (force_nonblock) | |
7a7cacba | 4593 | flags |= MSG_DONTWAIT; |
0031275d SM |
4594 | if (flags & MSG_WAITALL) |
4595 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4596 | ||
7a7cacba PB |
4597 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4598 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4599 | if (force_nonblock && ret == -EAGAIN) |
4600 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4601 | if (ret == -ERESTARTSYS) |
4602 | ret = -EINTR; | |
03b1230c | 4603 | |
7fbb1b54 PB |
4604 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4605 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4606 | /* fast path, check for non-NULL to avoid function call */ |
4607 | if (kmsg->free_iov) | |
4608 | kfree(kmsg->free_iov); | |
99bc4c38 | 4609 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4610 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4611 | req_set_fail(req); |
889fca73 | 4612 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4613 | return 0; |
0fa03c62 | 4614 | } |
5d17b4a4 | 4615 | |
889fca73 | 4616 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4617 | { |
6b754c8b | 4618 | struct io_buffer *kbuf; |
7a7cacba PB |
4619 | struct io_sr_msg *sr = &req->sr_msg; |
4620 | struct msghdr msg; | |
4621 | void __user *buf = sr->buf; | |
fddaface | 4622 | struct socket *sock; |
7a7cacba PB |
4623 | struct iovec iov; |
4624 | unsigned flags; | |
0031275d | 4625 | int min_ret = 0; |
bcda7baa | 4626 | int ret, cflags = 0; |
45d189c6 | 4627 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4628 | |
dba4a925 | 4629 | sock = sock_from_file(req->file); |
7a7cacba | 4630 | if (unlikely(!sock)) |
dba4a925 | 4631 | return -ENOTSOCK; |
fddaface | 4632 | |
bc02ef33 | 4633 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4634 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4635 | if (IS_ERR(kbuf)) |
4636 | return PTR_ERR(kbuf); | |
7a7cacba | 4637 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4638 | } |
bcda7baa | 4639 | |
7a7cacba | 4640 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4641 | if (unlikely(ret)) |
4642 | goto out_free; | |
fddaface | 4643 | |
7a7cacba PB |
4644 | msg.msg_name = NULL; |
4645 | msg.msg_control = NULL; | |
4646 | msg.msg_controllen = 0; | |
4647 | msg.msg_namelen = 0; | |
4648 | msg.msg_iocb = NULL; | |
4649 | msg.msg_flags = 0; | |
fddaface | 4650 | |
04411806 PB |
4651 | flags = req->sr_msg.msg_flags; |
4652 | if (force_nonblock) | |
7a7cacba | 4653 | flags |= MSG_DONTWAIT; |
0031275d SM |
4654 | if (flags & MSG_WAITALL) |
4655 | min_ret = iov_iter_count(&msg.msg_iter); | |
4656 | ||
7a7cacba PB |
4657 | ret = sock_recvmsg(sock, &msg, flags); |
4658 | if (force_nonblock && ret == -EAGAIN) | |
4659 | return -EAGAIN; | |
4660 | if (ret == -ERESTARTSYS) | |
4661 | ret = -EINTR; | |
14c32eee | 4662 | out_free: |
7fbb1b54 PB |
4663 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4664 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4665 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
93d2bcd2 | 4666 | req_set_fail(req); |
889fca73 | 4667 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4668 | return 0; |
fddaface JA |
4669 | } |
4670 | ||
3529d8c2 | 4671 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4672 | { |
8ed8d3c3 JA |
4673 | struct io_accept *accept = &req->accept; |
4674 | ||
14587a46 | 4675 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4676 | return -EINVAL; |
8042d6ce | 4677 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4678 | return -EINVAL; |
4679 | ||
d55e5f5b JA |
4680 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4681 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4682 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4683 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4684 | return 0; |
8ed8d3c3 | 4685 | } |
17f2fe35 | 4686 | |
889fca73 | 4687 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4688 | { |
4689 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4690 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4691 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4692 | int ret; |
4693 | ||
e697deed JX |
4694 | if (req->file->f_flags & O_NONBLOCK) |
4695 | req->flags |= REQ_F_NOWAIT; | |
4696 | ||
8ed8d3c3 | 4697 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4698 | accept->addr_len, accept->flags, |
4699 | accept->nofile); | |
8ed8d3c3 | 4700 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4701 | return -EAGAIN; |
ac45abc0 PB |
4702 | if (ret < 0) { |
4703 | if (ret == -ERESTARTSYS) | |
4704 | ret = -EINTR; | |
93d2bcd2 | 4705 | req_set_fail(req); |
ac45abc0 | 4706 | } |
889fca73 | 4707 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4708 | return 0; |
8ed8d3c3 JA |
4709 | } |
4710 | ||
93642ef8 PB |
4711 | static int io_connect_prep_async(struct io_kiocb *req) |
4712 | { | |
4713 | struct io_async_connect *io = req->async_data; | |
4714 | struct io_connect *conn = &req->connect; | |
4715 | ||
4716 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4717 | } | |
4718 | ||
3529d8c2 | 4719 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4720 | { |
3529d8c2 | 4721 | struct io_connect *conn = &req->connect; |
f499a021 | 4722 | |
14587a46 | 4723 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4724 | return -EINVAL; |
4725 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4726 | return -EINVAL; | |
4727 | ||
3529d8c2 JA |
4728 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4729 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4730 | return 0; |
f499a021 JA |
4731 | } |
4732 | ||
889fca73 | 4733 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4734 | { |
e8c2bc1f | 4735 | struct io_async_connect __io, *io; |
f8e85cf2 | 4736 | unsigned file_flags; |
3fbb51c1 | 4737 | int ret; |
45d189c6 | 4738 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4739 | |
e8c2bc1f JA |
4740 | if (req->async_data) { |
4741 | io = req->async_data; | |
f499a021 | 4742 | } else { |
3529d8c2 JA |
4743 | ret = move_addr_to_kernel(req->connect.addr, |
4744 | req->connect.addr_len, | |
e8c2bc1f | 4745 | &__io.address); |
f499a021 JA |
4746 | if (ret) |
4747 | goto out; | |
4748 | io = &__io; | |
4749 | } | |
4750 | ||
3fbb51c1 JA |
4751 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4752 | ||
e8c2bc1f | 4753 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4754 | req->connect.addr_len, file_flags); |
87f80d62 | 4755 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4756 | if (req->async_data) |
b7bb4f7d | 4757 | return -EAGAIN; |
e8c2bc1f | 4758 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4759 | ret = -ENOMEM; |
4760 | goto out; | |
4761 | } | |
e8c2bc1f | 4762 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4763 | return -EAGAIN; |
f499a021 | 4764 | } |
f8e85cf2 JA |
4765 | if (ret == -ERESTARTSYS) |
4766 | ret = -EINTR; | |
f499a021 | 4767 | out: |
4e88d6e7 | 4768 | if (ret < 0) |
93d2bcd2 | 4769 | req_set_fail(req); |
889fca73 | 4770 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4771 | return 0; |
469956e8 Y |
4772 | } |
4773 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4774 | #define IO_NETOP_FN(op) \ |
4775 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4776 | { \ | |
4777 | return -EOPNOTSUPP; \ | |
4778 | } | |
4779 | ||
4780 | #define IO_NETOP_PREP(op) \ | |
4781 | IO_NETOP_FN(op) \ | |
4782 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4783 | { \ | |
4784 | return -EOPNOTSUPP; \ | |
4785 | } \ | |
4786 | ||
4787 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4788 | IO_NETOP_PREP(op) \ | |
4789 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4790 | { \ | |
4791 | return -EOPNOTSUPP; \ | |
4792 | } | |
4793 | ||
4794 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4795 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4796 | IO_NETOP_PREP_ASYNC(connect); | |
4797 | IO_NETOP_PREP(accept); | |
4798 | IO_NETOP_FN(send); | |
4799 | IO_NETOP_FN(recv); | |
469956e8 | 4800 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4801 | |
d7718a9d JA |
4802 | struct io_poll_table { |
4803 | struct poll_table_struct pt; | |
4804 | struct io_kiocb *req; | |
4805 | int error; | |
4806 | }; | |
ce593a6c | 4807 | |
d7718a9d | 4808 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5b0a6acc | 4809 | __poll_t mask, io_req_tw_func_t func) |
d7718a9d | 4810 | { |
d7718a9d JA |
4811 | /* for instances that support it check for an event match first: */ |
4812 | if (mask && !(mask & poll->events)) | |
4813 | return 0; | |
4814 | ||
4815 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4816 | ||
4817 | list_del_init(&poll->wait.entry); | |
4818 | ||
d7718a9d | 4819 | req->result = mask; |
5b0a6acc | 4820 | req->io_task_work.func = func; |
6d816e08 | 4821 | |
d7718a9d | 4822 | /* |
e3aabf95 JA |
4823 | * If this fails, then the task is exiting. When a task exits, the |
4824 | * work gets canceled, so just cancel this request as well instead | |
4825 | * of executing it. We can't safely execute it anyway, as we may not | |
4826 | * have the needed state needed for it anyway. | |
d7718a9d | 4827 | */ |
e09ee510 | 4828 | io_req_task_work_add(req); |
d7718a9d JA |
4829 | return 1; |
4830 | } | |
4831 | ||
74ce6ce4 JA |
4832 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4833 | __acquires(&req->ctx->completion_lock) | |
4834 | { | |
4835 | struct io_ring_ctx *ctx = req->ctx; | |
4836 | ||
e09ee510 PB |
4837 | if (unlikely(req->task->flags & PF_EXITING)) |
4838 | WRITE_ONCE(poll->canceled, true); | |
4839 | ||
74ce6ce4 JA |
4840 | if (!req->result && !READ_ONCE(poll->canceled)) { |
4841 | struct poll_table_struct pt = { ._key = poll->events }; | |
4842 | ||
4843 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4844 | } | |
4845 | ||
4846 | spin_lock_irq(&ctx->completion_lock); | |
4847 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4848 | add_wait_queue(poll->head, &poll->wait); | |
4849 | return true; | |
4850 | } | |
4851 | ||
4852 | return false; | |
4853 | } | |
4854 | ||
d4e7cd36 | 4855 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4856 | { |
e8c2bc1f | 4857 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4858 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4859 | return req->async_data; |
d4e7cd36 JA |
4860 | return req->apoll->double_poll; |
4861 | } | |
4862 | ||
4863 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4864 | { | |
4865 | if (req->opcode == IORING_OP_POLL_ADD) | |
4866 | return &req->poll; | |
4867 | return &req->apoll->poll; | |
4868 | } | |
4869 | ||
4870 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4871 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4872 | { |
4873 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4874 | |
4875 | lockdep_assert_held(&req->ctx->completion_lock); | |
4876 | ||
4877 | if (poll && poll->head) { | |
4878 | struct wait_queue_head *head = poll->head; | |
4879 | ||
4880 | spin_lock(&head->lock); | |
4881 | list_del_init(&poll->wait.entry); | |
4882 | if (poll->wait.private) | |
de9b4cca | 4883 | req_ref_put(req); |
18bceab1 JA |
4884 | poll->head = NULL; |
4885 | spin_unlock(&head->lock); | |
4886 | } | |
4887 | } | |
4888 | ||
e27414be | 4889 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4890 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4891 | { |
4892 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4893 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4894 | int error; |
18bceab1 | 4895 | |
e27414be | 4896 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4897 | error = -ECANCELED; |
88e41cf9 | 4898 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4899 | } else { |
5082620f | 4900 | error = mangle_poll(mask); |
e27414be | 4901 | } |
b69de288 JA |
4902 | if (req->poll.events & EPOLLONESHOT) |
4903 | flags = 0; | |
d4d19c19 | 4904 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5082620f | 4905 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4906 | req->poll.done = true; |
4907 | flags = 0; | |
4908 | } | |
7b289c38 HX |
4909 | if (flags & IORING_CQE_F_MORE) |
4910 | ctx->cq_extra++; | |
18bceab1 | 4911 | |
18bceab1 | 4912 | io_commit_cqring(ctx); |
88e41cf9 | 4913 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4914 | } |
4915 | ||
5b0a6acc | 4916 | static void io_poll_task_func(struct io_kiocb *req) |
18bceab1 JA |
4917 | { |
4918 | struct io_ring_ctx *ctx = req->ctx; | |
dd221f46 | 4919 | struct io_kiocb *nxt; |
18bceab1 JA |
4920 | |
4921 | if (io_poll_rewait(req, &req->poll)) { | |
4922 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4923 | } else { |
f40b964a | 4924 | bool done; |
18bceab1 | 4925 | |
e27414be | 4926 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4927 | if (done) { |
4928 | hash_del(&req->hash_node); | |
f40b964a | 4929 | } else { |
88e41cf9 JA |
4930 | req->result = 0; |
4931 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4932 | } | |
dd221f46 | 4933 | spin_unlock_irq(&ctx->completion_lock); |
dd221f46 | 4934 | io_cqring_ev_posted(ctx); |
18bceab1 | 4935 | |
88e41cf9 JA |
4936 | if (done) { |
4937 | nxt = io_put_req_find_next(req); | |
4938 | if (nxt) | |
5b0a6acc | 4939 | io_req_task_submit(nxt); |
88e41cf9 | 4940 | } |
dd221f46 | 4941 | } |
18bceab1 JA |
4942 | } |
4943 | ||
4944 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4945 | int sync, void *key) | |
4946 | { | |
4947 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4948 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4949 | __poll_t mask = key_to_poll(key); |
4950 | ||
4951 | /* for instances that support it check for an event match first: */ | |
4952 | if (mask && !(mask & poll->events)) | |
4953 | return 0; | |
88e41cf9 JA |
4954 | if (!(poll->events & EPOLLONESHOT)) |
4955 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4956 | |
8706e04e JA |
4957 | list_del_init(&wait->entry); |
4958 | ||
9ce85ef2 | 4959 | if (poll->head) { |
18bceab1 JA |
4960 | bool done; |
4961 | ||
807abcb0 JA |
4962 | spin_lock(&poll->head->lock); |
4963 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4964 | if (!done) |
807abcb0 | 4965 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4966 | /* make sure double remove sees this as being gone */ |
4967 | wait->private = NULL; | |
807abcb0 | 4968 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4969 | if (!done) { |
4970 | /* use wait func handler, so it matches the rq type */ | |
4971 | poll->wait.func(&poll->wait, mode, sync, key); | |
4972 | } | |
18bceab1 | 4973 | } |
de9b4cca | 4974 | req_ref_put(req); |
18bceab1 JA |
4975 | return 1; |
4976 | } | |
4977 | ||
4978 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4979 | wait_queue_func_t wake_func) | |
4980 | { | |
4981 | poll->head = NULL; | |
4982 | poll->done = false; | |
4983 | poll->canceled = false; | |
464dca61 JA |
4984 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
4985 | /* mask in events that we always want/need */ | |
4986 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
4987 | INIT_LIST_HEAD(&poll->wait.entry); |
4988 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4989 | } | |
4990 | ||
4991 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4992 | struct wait_queue_head *head, |
4993 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4994 | { |
4995 | struct io_kiocb *req = pt->req; | |
4996 | ||
4997 | /* | |
4998 | * If poll->head is already set, it's because the file being polled | |
4999 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5000 | * for write). Setup a separate io_poll_iocb if this happens. | |
5001 | */ | |
5002 | if (unlikely(poll->head)) { | |
58852d4d PB |
5003 | struct io_poll_iocb *poll_one = poll; |
5004 | ||
18bceab1 | 5005 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5006 | if (*poll_ptr) { |
18bceab1 JA |
5007 | pt->error = -EINVAL; |
5008 | return; | |
5009 | } | |
ea6a693d JA |
5010 | /* |
5011 | * Can't handle multishot for double wait for now, turn it | |
5012 | * into one-shot mode. | |
5013 | */ | |
7a274727 PB |
5014 | if (!(poll_one->events & EPOLLONESHOT)) |
5015 | poll_one->events |= EPOLLONESHOT; | |
1c3b3e65 | 5016 | /* double add on the same waitqueue head, ignore */ |
7a274727 | 5017 | if (poll_one->head == head) |
1c3b3e65 | 5018 | return; |
18bceab1 JA |
5019 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5020 | if (!poll) { | |
5021 | pt->error = -ENOMEM; | |
5022 | return; | |
5023 | } | |
58852d4d | 5024 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5025 | req_ref_get(req); |
18bceab1 | 5026 | poll->wait.private = req; |
807abcb0 | 5027 | *poll_ptr = poll; |
18bceab1 JA |
5028 | } |
5029 | ||
5030 | pt->error = 0; | |
5031 | poll->head = head; | |
a31eb4a2 JX |
5032 | |
5033 | if (poll->events & EPOLLEXCLUSIVE) | |
5034 | add_wait_queue_exclusive(head, &poll->wait); | |
5035 | else | |
5036 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5037 | } |
5038 | ||
5039 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5040 | struct poll_table_struct *p) | |
5041 | { | |
5042 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5043 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5044 | |
807abcb0 | 5045 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5046 | } |
5047 | ||
5b0a6acc | 5048 | static void io_async_task_func(struct io_kiocb *req) |
d7718a9d | 5049 | { |
d7718a9d JA |
5050 | struct async_poll *apoll = req->apoll; |
5051 | struct io_ring_ctx *ctx = req->ctx; | |
5052 | ||
236daeae | 5053 | trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data); |
d7718a9d | 5054 | |
74ce6ce4 | 5055 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5056 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5057 | return; |
d7718a9d JA |
5058 | } |
5059 | ||
0ea13b44 | 5060 | hash_del(&req->hash_node); |
d4e7cd36 | 5061 | io_poll_remove_double(req); |
74ce6ce4 JA |
5062 | spin_unlock_irq(&ctx->completion_lock); |
5063 | ||
0be0b0e3 | 5064 | if (!READ_ONCE(apoll->poll.canceled)) |
5b0a6acc | 5065 | io_req_task_submit(req); |
0be0b0e3 | 5066 | else |
2593553a | 5067 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5068 | } |
5069 | ||
5070 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5071 | void *key) | |
5072 | { | |
5073 | struct io_kiocb *req = wait->private; | |
5074 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5075 | ||
5076 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5077 | key_to_poll(key)); | |
5078 | ||
5079 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5080 | } | |
5081 | ||
5082 | static void io_poll_req_insert(struct io_kiocb *req) | |
5083 | { | |
5084 | struct io_ring_ctx *ctx = req->ctx; | |
5085 | struct hlist_head *list; | |
5086 | ||
5087 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5088 | hlist_add_head(&req->hash_node, list); | |
5089 | } | |
5090 | ||
5091 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5092 | struct io_poll_iocb *poll, | |
5093 | struct io_poll_table *ipt, __poll_t mask, | |
5094 | wait_queue_func_t wake_func) | |
5095 | __acquires(&ctx->completion_lock) | |
5096 | { | |
5097 | struct io_ring_ctx *ctx = req->ctx; | |
5098 | bool cancel = false; | |
5099 | ||
4d52f338 | 5100 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5101 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5102 | poll->file = req->file; |
18bceab1 | 5103 | poll->wait.private = req; |
d7718a9d JA |
5104 | |
5105 | ipt->pt._key = mask; | |
5106 | ipt->req = req; | |
5107 | ipt->error = -EINVAL; | |
5108 | ||
d7718a9d JA |
5109 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5110 | ||
5111 | spin_lock_irq(&ctx->completion_lock); | |
5112 | if (likely(poll->head)) { | |
5113 | spin_lock(&poll->head->lock); | |
5114 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5115 | if (ipt->error) | |
5116 | cancel = true; | |
5117 | ipt->error = 0; | |
5118 | mask = 0; | |
5119 | } | |
88e41cf9 | 5120 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5121 | list_del_init(&poll->wait.entry); |
5122 | else if (cancel) | |
5123 | WRITE_ONCE(poll->canceled, true); | |
5124 | else if (!poll->done) /* actually waiting for an event */ | |
5125 | io_poll_req_insert(req); | |
5126 | spin_unlock(&poll->head->lock); | |
5127 | } | |
5128 | ||
5129 | return mask; | |
5130 | } | |
5131 | ||
59b735ae OL |
5132 | enum { |
5133 | IO_APOLL_OK, | |
5134 | IO_APOLL_ABORTED, | |
5135 | IO_APOLL_READY | |
5136 | }; | |
5137 | ||
5138 | static int io_arm_poll_handler(struct io_kiocb *req) | |
d7718a9d JA |
5139 | { |
5140 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5141 | struct io_ring_ctx *ctx = req->ctx; | |
5142 | struct async_poll *apoll; | |
5143 | struct io_poll_table ipt; | |
b2d9c3da | 5144 | __poll_t ret, mask = EPOLLONESHOT | POLLERR | POLLPRI; |
9dab14b8 | 5145 | int rw; |
d7718a9d JA |
5146 | |
5147 | if (!req->file || !file_can_poll(req->file)) | |
59b735ae | 5148 | return IO_APOLL_ABORTED; |
24c74678 | 5149 | if (req->flags & REQ_F_POLLED) |
59b735ae | 5150 | return IO_APOLL_ABORTED; |
b2d9c3da PB |
5151 | if (!def->pollin && !def->pollout) |
5152 | return IO_APOLL_ABORTED; | |
5153 | ||
5154 | if (def->pollin) { | |
9dab14b8 | 5155 | rw = READ; |
b2d9c3da PB |
5156 | mask |= POLLIN | POLLRDNORM; |
5157 | ||
5158 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5159 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5160 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5161 | mask &= ~POLLIN; | |
5162 | } else { | |
9dab14b8 | 5163 | rw = WRITE; |
b2d9c3da PB |
5164 | mask |= POLLOUT | POLLWRNORM; |
5165 | } | |
5166 | ||
9dab14b8 | 5167 | /* if we can't nonblock try, then no point in arming a poll handler */ |
7b29f92d | 5168 | if (!io_file_supports_async(req, rw)) |
59b735ae | 5169 | return IO_APOLL_ABORTED; |
d7718a9d JA |
5170 | |
5171 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5172 | if (unlikely(!apoll)) | |
59b735ae | 5173 | return IO_APOLL_ABORTED; |
807abcb0 | 5174 | apoll->double_poll = NULL; |
d7718a9d | 5175 | req->apoll = apoll; |
b2d9c3da | 5176 | req->flags |= REQ_F_POLLED; |
d7718a9d JA |
5177 | ipt.pt._qproc = io_async_queue_proc; |
5178 | ||
5179 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5180 | io_async_wake); | |
a36da65c | 5181 | if (ret || ipt.error) { |
d4e7cd36 | 5182 | io_poll_remove_double(req); |
d7718a9d | 5183 | spin_unlock_irq(&ctx->completion_lock); |
59b735ae OL |
5184 | if (ret) |
5185 | return IO_APOLL_READY; | |
5186 | return IO_APOLL_ABORTED; | |
d7718a9d JA |
5187 | } |
5188 | spin_unlock_irq(&ctx->completion_lock); | |
236daeae OL |
5189 | trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, |
5190 | mask, apoll->poll.events); | |
59b735ae | 5191 | return IO_APOLL_OK; |
d7718a9d JA |
5192 | } |
5193 | ||
5194 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5195 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5196 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5197 | { |
b41e9852 | 5198 | bool do_complete = false; |
221c5eb2 | 5199 | |
5082620f JA |
5200 | if (!poll->head) |
5201 | return false; | |
221c5eb2 | 5202 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5203 | if (do_cancel) |
5204 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5205 | if (!list_empty(&poll->wait.entry)) { |
5206 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5207 | do_complete = true; |
221c5eb2 JA |
5208 | } |
5209 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5210 | hash_del(&req->hash_node); |
d7718a9d JA |
5211 | return do_complete; |
5212 | } | |
5213 | ||
b2c3f7e1 | 5214 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5215 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5216 | { |
5217 | bool do_complete; | |
5218 | ||
d4e7cd36 | 5219 | io_poll_remove_double(req); |
e31001a3 | 5220 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5221 | |
e31001a3 | 5222 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
d7718a9d | 5223 | /* non-poll requests have submit ref still */ |
e31001a3 | 5224 | req_ref_put(req); |
b1f573bd | 5225 | } |
b2c3f7e1 JA |
5226 | return do_complete; |
5227 | } | |
5228 | ||
5229 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5230 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5231 | { |
5232 | bool do_complete; | |
b1f573bd | 5233 | |
b2c3f7e1 | 5234 | do_complete = io_poll_remove_waitqs(req); |
b41e9852 | 5235 | if (do_complete) { |
d4d19c19 | 5236 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5237 | io_commit_cqring(req->ctx); |
93d2bcd2 | 5238 | req_set_fail(req); |
216578e5 | 5239 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5240 | } |
5241 | ||
5242 | return do_complete; | |
221c5eb2 JA |
5243 | } |
5244 | ||
76e1b642 JA |
5245 | /* |
5246 | * Returns true if we found and killed one or more poll requests | |
5247 | */ | |
6b81928d | 5248 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
3dd0c97a | 5249 | bool cancel_all) |
221c5eb2 | 5250 | { |
78076bb6 | 5251 | struct hlist_node *tmp; |
221c5eb2 | 5252 | struct io_kiocb *req; |
8e2e1faf | 5253 | int posted = 0, i; |
221c5eb2 JA |
5254 | |
5255 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5256 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5257 | struct hlist_head *list; | |
5258 | ||
5259 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5260 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
3dd0c97a | 5261 | if (io_match_task(req, tsk, cancel_all)) |
f3606e3a JA |
5262 | posted += io_poll_remove_one(req); |
5263 | } | |
221c5eb2 JA |
5264 | } |
5265 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5266 | |
8e2e1faf JA |
5267 | if (posted) |
5268 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5269 | |
5270 | return posted != 0; | |
221c5eb2 JA |
5271 | } |
5272 | ||
9ba5fac8 PB |
5273 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5274 | bool poll_only) | |
e07785b0 | 5275 | __must_hold(&ctx->completion_lock) |
47f46768 | 5276 | { |
78076bb6 | 5277 | struct hlist_head *list; |
47f46768 JA |
5278 | struct io_kiocb *req; |
5279 | ||
78076bb6 JA |
5280 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5281 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5282 | if (sqe_addr != req->user_data) |
5283 | continue; | |
9ba5fac8 PB |
5284 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5285 | continue; | |
b2cb805f | 5286 | return req; |
47f46768 | 5287 | } |
b2cb805f JA |
5288 | return NULL; |
5289 | } | |
5290 | ||
9ba5fac8 PB |
5291 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5292 | bool poll_only) | |
e07785b0 | 5293 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5294 | { |
5295 | struct io_kiocb *req; | |
5296 | ||
9ba5fac8 | 5297 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5298 | if (!req) |
5299 | return -ENOENT; | |
5300 | if (io_poll_remove_one(req)) | |
5301 | return 0; | |
5302 | ||
5303 | return -EALREADY; | |
47f46768 JA |
5304 | } |
5305 | ||
9096af3e PB |
5306 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5307 | unsigned int flags) | |
5308 | { | |
5309 | u32 events; | |
47f46768 | 5310 | |
9096af3e PB |
5311 | events = READ_ONCE(sqe->poll32_events); |
5312 | #ifdef __BIG_ENDIAN | |
5313 | events = swahw32(events); | |
5314 | #endif | |
5315 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5316 | events |= EPOLLONESHOT; | |
5317 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
5318 | } |
5319 | ||
c5de0036 | 5320 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5321 | const struct io_uring_sqe *sqe) |
0969e783 | 5322 | { |
c5de0036 PB |
5323 | struct io_poll_update *upd = &req->poll_update; |
5324 | u32 flags; | |
5325 | ||
0969e783 JA |
5326 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5327 | return -EINVAL; | |
c5de0036 PB |
5328 | if (sqe->ioprio || sqe->buf_index) |
5329 | return -EINVAL; | |
5330 | flags = READ_ONCE(sqe->len); | |
5331 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5332 | IORING_POLL_ADD_MULTI)) | |
5333 | return -EINVAL; | |
5334 | /* meaningless without update */ | |
5335 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5336 | return -EINVAL; |
5337 | ||
c5de0036 PB |
5338 | upd->old_user_data = READ_ONCE(sqe->addr); |
5339 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5340 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5341 | |
c5de0036 PB |
5342 | upd->new_user_data = READ_ONCE(sqe->off); |
5343 | if (!upd->update_user_data && upd->new_user_data) | |
5344 | return -EINVAL; | |
5345 | if (upd->update_events) | |
5346 | upd->events = io_poll_parse_events(sqe, flags); | |
5347 | else if (sqe->poll32_events) | |
5348 | return -EINVAL; | |
221c5eb2 | 5349 | |
221c5eb2 JA |
5350 | return 0; |
5351 | } | |
5352 | ||
221c5eb2 JA |
5353 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5354 | void *key) | |
5355 | { | |
c2f2eb7d JA |
5356 | struct io_kiocb *req = wait->private; |
5357 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5358 | |
d7718a9d | 5359 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5360 | } |
5361 | ||
221c5eb2 JA |
5362 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5363 | struct poll_table_struct *p) | |
5364 | { | |
5365 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5366 | ||
e8c2bc1f | 5367 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5368 | } |
5369 | ||
3529d8c2 | 5370 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5371 | { |
5372 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 5373 | u32 flags; |
221c5eb2 JA |
5374 | |
5375 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5376 | return -EINVAL; | |
c5de0036 | 5377 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5378 | return -EINVAL; |
5379 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5380 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 JA |
5381 | return -EINVAL; |
5382 | ||
c5de0036 | 5383 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5384 | return 0; |
5385 | } | |
5386 | ||
61e98203 | 5387 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5388 | { |
5389 | struct io_poll_iocb *poll = &req->poll; | |
5390 | struct io_ring_ctx *ctx = req->ctx; | |
5391 | struct io_poll_table ipt; | |
0969e783 | 5392 | __poll_t mask; |
0969e783 | 5393 | |
d7718a9d | 5394 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5395 | |
d7718a9d JA |
5396 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5397 | io_poll_wake); | |
221c5eb2 | 5398 | |
8c838788 | 5399 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5400 | ipt.error = 0; |
e27414be | 5401 | io_poll_complete(req, mask); |
221c5eb2 | 5402 | } |
221c5eb2 JA |
5403 | spin_unlock_irq(&ctx->completion_lock); |
5404 | ||
8c838788 JA |
5405 | if (mask) { |
5406 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5407 | if (poll->events & EPOLLONESHOT) |
5408 | io_put_req(req); | |
221c5eb2 | 5409 | } |
8c838788 | 5410 | return ipt.error; |
221c5eb2 JA |
5411 | } |
5412 | ||
c5de0036 | 5413 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5414 | { |
5415 | struct io_ring_ctx *ctx = req->ctx; | |
5416 | struct io_kiocb *preq; | |
cb3b200e | 5417 | bool completing; |
b69de288 JA |
5418 | int ret; |
5419 | ||
5420 | spin_lock_irq(&ctx->completion_lock); | |
9ba5fac8 | 5421 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5422 | if (!preq) { |
5423 | ret = -ENOENT; | |
5424 | goto err; | |
b69de288 | 5425 | } |
cb3b200e | 5426 | |
c5de0036 PB |
5427 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5428 | completing = true; | |
5429 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5430 | goto err; | |
5431 | } | |
5432 | ||
cb3b200e JA |
5433 | /* |
5434 | * Don't allow racy completion with singleshot, as we cannot safely | |
5435 | * update those. For multishot, if we're racing with completion, just | |
5436 | * let completion re-add it. | |
5437 | */ | |
5438 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5439 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5440 | ret = -EALREADY; | |
5441 | goto err; | |
b69de288 JA |
5442 | } |
5443 | /* we now have a detached poll request. reissue. */ | |
5444 | ret = 0; | |
5445 | err: | |
b69de288 | 5446 | if (ret < 0) { |
cb3b200e | 5447 | spin_unlock_irq(&ctx->completion_lock); |
93d2bcd2 | 5448 | req_set_fail(req); |
b69de288 JA |
5449 | io_req_complete(req, ret); |
5450 | return 0; | |
5451 | } | |
5452 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5453 | if (req->poll_update.update_events) { |
b69de288 | 5454 | preq->poll.events &= ~0xffff; |
9d805892 | 5455 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5456 | preq->poll.events |= IO_POLL_UNMASK; |
5457 | } | |
9d805892 PB |
5458 | if (req->poll_update.update_user_data) |
5459 | preq->user_data = req->poll_update.new_user_data; | |
cb3b200e JA |
5460 | spin_unlock_irq(&ctx->completion_lock); |
5461 | ||
b69de288 JA |
5462 | /* complete update request, we're done with it */ |
5463 | io_req_complete(req, ret); | |
5464 | ||
cb3b200e | 5465 | if (!completing) { |
c5de0036 | 5466 | ret = io_poll_add(preq, issue_flags); |
cb3b200e | 5467 | if (ret < 0) { |
93d2bcd2 | 5468 | req_set_fail(preq); |
cb3b200e JA |
5469 | io_req_complete(preq, ret); |
5470 | } | |
b69de288 JA |
5471 | } |
5472 | return 0; | |
5473 | } | |
5474 | ||
5262f567 JA |
5475 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5476 | { | |
ad8a48ac JA |
5477 | struct io_timeout_data *data = container_of(timer, |
5478 | struct io_timeout_data, timer); | |
5479 | struct io_kiocb *req = data->req; | |
5480 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5481 | unsigned long flags; |
5482 | ||
5262f567 | 5483 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5484 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5485 | atomic_set(&req->ctx->cq_timeouts, |
5486 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5487 | ||
d4d19c19 | 5488 | io_cqring_fill_event(ctx, req->user_data, -ETIME, 0); |
5262f567 JA |
5489 | io_commit_cqring(ctx); |
5490 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5491 | ||
5492 | io_cqring_ev_posted(ctx); | |
93d2bcd2 | 5493 | req_set_fail(req); |
5262f567 JA |
5494 | io_put_req(req); |
5495 | return HRTIMER_NORESTART; | |
5496 | } | |
5497 | ||
fbd15848 PB |
5498 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5499 | __u64 user_data) | |
e07785b0 | 5500 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5501 | { |
fbd15848 | 5502 | struct io_timeout_data *io; |
47f46768 | 5503 | struct io_kiocb *req; |
fd9c7bc5 | 5504 | bool found = false; |
f254ac04 | 5505 | |
135fcde8 | 5506 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5507 | found = user_data == req->user_data; |
5508 | if (found) | |
47f46768 | 5509 | break; |
47f46768 | 5510 | } |
fd9c7bc5 PB |
5511 | if (!found) |
5512 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5513 | |
5514 | io = req->async_data; | |
fd9c7bc5 | 5515 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5516 | return ERR_PTR(-EALREADY); |
a71976f3 | 5517 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5518 | return req; |
5519 | } | |
47f46768 | 5520 | |
fbd15848 | 5521 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5522 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5523 | { |
5524 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5525 | ||
5526 | if (IS_ERR(req)) | |
5527 | return PTR_ERR(req); | |
f254ac04 | 5528 | |
93d2bcd2 | 5529 | req_set_fail(req); |
d4d19c19 | 5530 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
216578e5 | 5531 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5532 | return 0; |
5533 | } | |
5534 | ||
9c8e11b3 PB |
5535 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5536 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5537 | __must_hold(&ctx->completion_lock) |
47f46768 | 5538 | { |
9c8e11b3 PB |
5539 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5540 | struct io_timeout_data *data; | |
47f46768 | 5541 | |
9c8e11b3 PB |
5542 | if (IS_ERR(req)) |
5543 | return PTR_ERR(req); | |
47f46768 | 5544 | |
9c8e11b3 PB |
5545 | req->timeout.off = 0; /* noseq */ |
5546 | data = req->async_data; | |
5547 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5548 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5549 | data->timer.function = io_timeout_fn; | |
5550 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5551 | return 0; | |
47f46768 JA |
5552 | } |
5553 | ||
3529d8c2 JA |
5554 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5555 | const struct io_uring_sqe *sqe) | |
b29472ee | 5556 | { |
9c8e11b3 PB |
5557 | struct io_timeout_rem *tr = &req->timeout_rem; |
5558 | ||
b29472ee JA |
5559 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5560 | return -EINVAL; | |
61710e43 DA |
5561 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5562 | return -EINVAL; | |
9c8e11b3 | 5563 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5564 | return -EINVAL; |
5565 | ||
9c8e11b3 PB |
5566 | tr->addr = READ_ONCE(sqe->addr); |
5567 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5568 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5569 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5570 | return -EINVAL; | |
5571 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5572 | return -EFAULT; | |
5573 | } else if (tr->flags) { | |
5574 | /* timeout removal doesn't support flags */ | |
b29472ee | 5575 | return -EINVAL; |
9c8e11b3 | 5576 | } |
b29472ee | 5577 | |
b29472ee JA |
5578 | return 0; |
5579 | } | |
5580 | ||
8662daec PB |
5581 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5582 | { | |
5583 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5584 | : HRTIMER_MODE_REL; | |
5585 | } | |
5586 | ||
11365043 JA |
5587 | /* |
5588 | * Remove or update an existing timeout command | |
5589 | */ | |
61e98203 | 5590 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5591 | { |
9c8e11b3 | 5592 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5593 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5594 | int ret; |
11365043 | 5595 | |
11365043 | 5596 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5597 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5598 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5599 | else |
5600 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5601 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5602 | |
d4d19c19 | 5603 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
11365043 JA |
5604 | io_commit_cqring(ctx); |
5605 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5606 | io_cqring_ev_posted(ctx); |
4e88d6e7 | 5607 | if (ret < 0) |
93d2bcd2 | 5608 | req_set_fail(req); |
ec9c02ad | 5609 | io_put_req(req); |
11365043 | 5610 | return 0; |
5262f567 JA |
5611 | } |
5612 | ||
3529d8c2 | 5613 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5614 | bool is_timeout_link) |
5262f567 | 5615 | { |
ad8a48ac | 5616 | struct io_timeout_data *data; |
a41525ab | 5617 | unsigned flags; |
56080b02 | 5618 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5619 | |
ad8a48ac | 5620 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5621 | return -EINVAL; |
ad8a48ac | 5622 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5623 | return -EINVAL; |
56080b02 | 5624 | if (off && is_timeout_link) |
2d28390a | 5625 | return -EINVAL; |
a41525ab JA |
5626 | flags = READ_ONCE(sqe->timeout_flags); |
5627 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5628 | return -EINVAL; |
bdf20073 | 5629 | |
bfe68a22 | 5630 | req->timeout.off = off; |
f18ee4cf PB |
5631 | if (unlikely(off && !req->ctx->off_timeout_used)) |
5632 | req->ctx->off_timeout_used = true; | |
26a61679 | 5633 | |
e8c2bc1f | 5634 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5635 | return -ENOMEM; |
5636 | ||
e8c2bc1f | 5637 | data = req->async_data; |
ad8a48ac | 5638 | data->req = req; |
ad8a48ac JA |
5639 | |
5640 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5641 | return -EFAULT; |
5642 | ||
8662daec | 5643 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5644 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5645 | if (is_timeout_link) |
5646 | io_req_track_inflight(req); | |
ad8a48ac JA |
5647 | return 0; |
5648 | } | |
5649 | ||
61e98203 | 5650 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5651 | { |
ad8a48ac | 5652 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5653 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5654 | struct list_head *entry; |
bfe68a22 | 5655 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5656 | |
733f5c95 | 5657 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5658 | |
5262f567 JA |
5659 | /* |
5660 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5661 | * timeout event to be satisfied. If it isn't set, then this is |
5662 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5663 | */ |
8eb7e2d0 | 5664 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5665 | entry = ctx->timeout_list.prev; |
5666 | goto add; | |
5667 | } | |
5262f567 | 5668 | |
bfe68a22 PB |
5669 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5670 | req->timeout.target_seq = tail + off; | |
5262f567 | 5671 | |
f010505b MDG |
5672 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5673 | * This is safe because ->completion_lock is held, and submissions | |
5674 | * and completions are never mixed in the same ->completion_lock section. | |
5675 | */ | |
5676 | ctx->cq_last_tm_flush = tail; | |
5677 | ||
5262f567 JA |
5678 | /* |
5679 | * Insertion sort, ensuring the first entry in the list is always | |
5680 | * the one we need first. | |
5681 | */ | |
5262f567 | 5682 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5683 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5684 | timeout.list); | |
5262f567 | 5685 | |
8eb7e2d0 | 5686 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5687 | continue; |
bfe68a22 PB |
5688 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5689 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5690 | break; |
5691 | } | |
93bd25bb | 5692 | add: |
135fcde8 | 5693 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5694 | data->timer.function = io_timeout_fn; |
5695 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5696 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5697 | return 0; |
5698 | } | |
5262f567 | 5699 | |
f458dd84 PB |
5700 | struct io_cancel_data { |
5701 | struct io_ring_ctx *ctx; | |
5702 | u64 user_data; | |
5703 | }; | |
5704 | ||
62755e35 JA |
5705 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5706 | { | |
5707 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5708 | struct io_cancel_data *cd = data; |
62755e35 | 5709 | |
f458dd84 | 5710 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5711 | } |
5712 | ||
f458dd84 PB |
5713 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5714 | struct io_ring_ctx *ctx) | |
62755e35 | 5715 | { |
f458dd84 | 5716 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5717 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5718 | int ret = 0; |
5719 | ||
f458dd84 | 5720 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5721 | return -ENOENT; |
5722 | ||
f458dd84 | 5723 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5724 | switch (cancel_ret) { |
5725 | case IO_WQ_CANCEL_OK: | |
5726 | ret = 0; | |
5727 | break; | |
5728 | case IO_WQ_CANCEL_RUNNING: | |
5729 | ret = -EALREADY; | |
5730 | break; | |
5731 | case IO_WQ_CANCEL_NOTFOUND: | |
5732 | ret = -ENOENT; | |
5733 | break; | |
5734 | } | |
5735 | ||
e977d6d3 JA |
5736 | return ret; |
5737 | } | |
5738 | ||
47f46768 JA |
5739 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5740 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5741 | int success_ret) |
47f46768 JA |
5742 | { |
5743 | unsigned long flags; | |
5744 | int ret; | |
5745 | ||
f458dd84 | 5746 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5747 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5748 | if (ret != -ENOENT) |
5749 | goto done; | |
47f46768 JA |
5750 | ret = io_timeout_cancel(ctx, sqe_addr); |
5751 | if (ret != -ENOENT) | |
5752 | goto done; | |
9ba5fac8 | 5753 | ret = io_poll_cancel(ctx, sqe_addr, false); |
47f46768 | 5754 | done: |
b0dd8a41 JA |
5755 | if (!ret) |
5756 | ret = success_ret; | |
d4d19c19 | 5757 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
47f46768 JA |
5758 | io_commit_cqring(ctx); |
5759 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5760 | io_cqring_ev_posted(ctx); | |
5761 | ||
4e88d6e7 | 5762 | if (ret < 0) |
93d2bcd2 | 5763 | req_set_fail(req); |
47f46768 JA |
5764 | } |
5765 | ||
3529d8c2 JA |
5766 | static int io_async_cancel_prep(struct io_kiocb *req, |
5767 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5768 | { |
fbf23849 | 5769 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5770 | return -EINVAL; |
61710e43 DA |
5771 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5772 | return -EINVAL; | |
5773 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5774 | return -EINVAL; |
5775 | ||
fbf23849 JA |
5776 | req->cancel.addr = READ_ONCE(sqe->addr); |
5777 | return 0; | |
5778 | } | |
5779 | ||
61e98203 | 5780 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5781 | { |
5782 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5783 | u64 sqe_addr = req->cancel.addr; |
5784 | struct io_tctx_node *node; | |
5785 | int ret; | |
5786 | ||
5787 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5788 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5789 | spin_lock_irq(&ctx->completion_lock); | |
5790 | if (ret != -ENOENT) | |
5791 | goto done; | |
5792 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5793 | if (ret != -ENOENT) | |
5794 | goto done; | |
9ba5fac8 | 5795 | ret = io_poll_cancel(ctx, sqe_addr, false); |
58f99373 PB |
5796 | if (ret != -ENOENT) |
5797 | goto done; | |
5798 | spin_unlock_irq(&ctx->completion_lock); | |
5799 | ||
5800 | /* slow path, try all io-wq's */ | |
5801 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5802 | ret = -ENOENT; | |
5803 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5804 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5805 | |
58f99373 PB |
5806 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5807 | if (ret != -ENOENT) | |
5808 | break; | |
5809 | } | |
5810 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5811 | ||
5812 | spin_lock_irq(&ctx->completion_lock); | |
5813 | done: | |
d4d19c19 | 5814 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
58f99373 PB |
5815 | io_commit_cqring(ctx); |
5816 | spin_unlock_irq(&ctx->completion_lock); | |
5817 | io_cqring_ev_posted(ctx); | |
5818 | ||
5819 | if (ret < 0) | |
93d2bcd2 | 5820 | req_set_fail(req); |
58f99373 | 5821 | io_put_req(req); |
5262f567 JA |
5822 | return 0; |
5823 | } | |
5824 | ||
269bbe5f | 5825 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5826 | const struct io_uring_sqe *sqe) |
5827 | { | |
61710e43 DA |
5828 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5829 | return -EINVAL; | |
5830 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5831 | return -EINVAL; |
5832 | ||
269bbe5f BM |
5833 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5834 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5835 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5836 | return -EINVAL; |
269bbe5f | 5837 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5838 | return 0; |
5839 | } | |
5840 | ||
889fca73 | 5841 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5842 | { |
5843 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 5844 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 5845 | int ret; |
fbf23849 | 5846 | |
45d189c6 | 5847 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5848 | return -EAGAIN; |
05f3fb3c | 5849 | |
269bbe5f BM |
5850 | up.offset = req->rsrc_update.offset; |
5851 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
5852 | up.nr = 0; |
5853 | up.tags = 0; | |
615cee49 | 5854 | up.resv = 0; |
05f3fb3c JA |
5855 | |
5856 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 5857 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 5858 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5859 | mutex_unlock(&ctx->uring_lock); |
5860 | ||
5861 | if (ret < 0) | |
93d2bcd2 | 5862 | req_set_fail(req); |
889fca73 | 5863 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5864 | return 0; |
5865 | } | |
5866 | ||
bfe76559 | 5867 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5868 | { |
d625c6ee | 5869 | switch (req->opcode) { |
e781573e | 5870 | case IORING_OP_NOP: |
bfe76559 | 5871 | return 0; |
f67676d1 JA |
5872 | case IORING_OP_READV: |
5873 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5874 | case IORING_OP_READ: |
bfe76559 | 5875 | return io_read_prep(req, sqe); |
f67676d1 JA |
5876 | case IORING_OP_WRITEV: |
5877 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5878 | case IORING_OP_WRITE: |
bfe76559 | 5879 | return io_write_prep(req, sqe); |
0969e783 | 5880 | case IORING_OP_POLL_ADD: |
bfe76559 | 5881 | return io_poll_add_prep(req, sqe); |
0969e783 | 5882 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5883 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5884 | case IORING_OP_FSYNC: |
1155c76a | 5885 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5886 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5887 | return io_sfr_prep(req, sqe); |
03b1230c | 5888 | case IORING_OP_SENDMSG: |
fddaface | 5889 | case IORING_OP_SEND: |
bfe76559 | 5890 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5891 | case IORING_OP_RECVMSG: |
fddaface | 5892 | case IORING_OP_RECV: |
bfe76559 | 5893 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5894 | case IORING_OP_CONNECT: |
bfe76559 | 5895 | return io_connect_prep(req, sqe); |
2d28390a | 5896 | case IORING_OP_TIMEOUT: |
bfe76559 | 5897 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5898 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5899 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5900 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5901 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5902 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5903 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5904 | case IORING_OP_ACCEPT: |
bfe76559 | 5905 | return io_accept_prep(req, sqe); |
d63d1b5e | 5906 | case IORING_OP_FALLOCATE: |
bfe76559 | 5907 | return io_fallocate_prep(req, sqe); |
15b71abe | 5908 | case IORING_OP_OPENAT: |
bfe76559 | 5909 | return io_openat_prep(req, sqe); |
b5dba59e | 5910 | case IORING_OP_CLOSE: |
bfe76559 | 5911 | return io_close_prep(req, sqe); |
05f3fb3c | 5912 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5913 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5914 | case IORING_OP_STATX: |
bfe76559 | 5915 | return io_statx_prep(req, sqe); |
4840e418 | 5916 | case IORING_OP_FADVISE: |
bfe76559 | 5917 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5918 | case IORING_OP_MADVISE: |
bfe76559 | 5919 | return io_madvise_prep(req, sqe); |
cebdb986 | 5920 | case IORING_OP_OPENAT2: |
bfe76559 | 5921 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5922 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5923 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5924 | case IORING_OP_SPLICE: |
bfe76559 | 5925 | return io_splice_prep(req, sqe); |
ddf0322d | 5926 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5927 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5928 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5929 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5930 | case IORING_OP_TEE: |
bfe76559 | 5931 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5932 | case IORING_OP_SHUTDOWN: |
5933 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5934 | case IORING_OP_RENAMEAT: |
5935 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5936 | case IORING_OP_UNLINKAT: |
5937 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5938 | } |
5939 | ||
bfe76559 PB |
5940 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5941 | req->opcode); | |
bd54b6fe | 5942 | return -EINVAL; |
bfe76559 PB |
5943 | } |
5944 | ||
93642ef8 | 5945 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5946 | { |
b7e298d2 PB |
5947 | if (!io_op_defs[req->opcode].needs_async_setup) |
5948 | return 0; | |
5949 | if (WARN_ON_ONCE(req->async_data)) | |
5950 | return -EFAULT; | |
5951 | if (io_alloc_async_data(req)) | |
5952 | return -EAGAIN; | |
5953 | ||
93642ef8 PB |
5954 | switch (req->opcode) { |
5955 | case IORING_OP_READV: | |
93642ef8 PB |
5956 | return io_rw_prep_async(req, READ); |
5957 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5958 | return io_rw_prep_async(req, WRITE); |
5959 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5960 | return io_sendmsg_prep_async(req); |
5961 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5962 | return io_recvmsg_prep_async(req); |
5963 | case IORING_OP_CONNECT: | |
5964 | return io_connect_prep_async(req); | |
5965 | } | |
b7e298d2 PB |
5966 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5967 | req->opcode); | |
5968 | return -EFAULT; | |
f67676d1 JA |
5969 | } |
5970 | ||
9cf7c104 PB |
5971 | static u32 io_get_sequence(struct io_kiocb *req) |
5972 | { | |
a3dbdf54 | 5973 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 5974 | |
a3dbdf54 PB |
5975 | /* need original cached_sq_head, but it was increased for each req */ |
5976 | io_for_each_link(req, req) | |
5977 | seq--; | |
5978 | return seq; | |
9cf7c104 PB |
5979 | } |
5980 | ||
76cc33d7 | 5981 | static bool io_drain_req(struct io_kiocb *req) |
de0617e4 | 5982 | { |
3c19966d | 5983 | struct io_kiocb *pos; |
a197f664 | 5984 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5985 | struct io_defer_entry *de; |
f67676d1 | 5986 | int ret; |
9cf7c104 | 5987 | u32 seq; |
de0617e4 | 5988 | |
3c19966d PB |
5989 | /* |
5990 | * If we need to drain a request in the middle of a link, drain the | |
5991 | * head request and the next request/link after the current link. | |
5992 | * Considering sequential execution of links, IOSQE_IO_DRAIN will be | |
5993 | * maintained for every request of our link. | |
5994 | */ | |
5995 | if (ctx->drain_next) { | |
5996 | req->flags |= REQ_F_IO_DRAIN; | |
5997 | ctx->drain_next = false; | |
5998 | } | |
5999 | /* not interested in head, start from the first linked */ | |
6000 | io_for_each_link(pos, req->link) { | |
6001 | if (pos->flags & REQ_F_IO_DRAIN) { | |
6002 | ctx->drain_next = true; | |
6003 | req->flags |= REQ_F_IO_DRAIN; | |
6004 | break; | |
6005 | } | |
6006 | } | |
6007 | ||
9d858b21 | 6008 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 | 6009 | if (likely(list_empty_careful(&ctx->defer_list) && |
10c66904 PB |
6010 | !(req->flags & REQ_F_IO_DRAIN))) { |
6011 | ctx->drain_active = false; | |
76cc33d7 | 6012 | return false; |
10c66904 | 6013 | } |
9cf7c104 PB |
6014 | |
6015 | seq = io_get_sequence(req); | |
6016 | /* Still a chance to pass the sequence check */ | |
6017 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
76cc33d7 | 6018 | return false; |
de0617e4 | 6019 | |
b7e298d2 | 6020 | ret = io_req_prep_async(req); |
be7053b7 PB |
6021 | if (ret) |
6022 | return ret; | |
cbdcb435 | 6023 | io_prep_async_link(req); |
27dc8338 | 6024 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 6025 | if (!de) { |
c32aace0 | 6026 | io_req_complete_failed(req, -ENOMEM); |
76cc33d7 PB |
6027 | return true; |
6028 | } | |
2d28390a | 6029 | |
de0617e4 | 6030 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6031 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6032 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6033 | kfree(de); |
ae34817b | 6034 | io_queue_async_work(req); |
76cc33d7 | 6035 | return true; |
de0617e4 JA |
6036 | } |
6037 | ||
915967f6 | 6038 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6039 | de->req = req; |
9cf7c104 | 6040 | de->seq = seq; |
27dc8338 | 6041 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 | 6042 | spin_unlock_irq(&ctx->completion_lock); |
76cc33d7 | 6043 | return true; |
de0617e4 JA |
6044 | } |
6045 | ||
68fb8979 | 6046 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6047 | { |
0e1b6fe3 PB |
6048 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6049 | switch (req->opcode) { | |
6050 | case IORING_OP_READV: | |
6051 | case IORING_OP_READ_FIXED: | |
6052 | case IORING_OP_READ: | |
bcda7baa | 6053 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6054 | break; |
6055 | case IORING_OP_RECVMSG: | |
6056 | case IORING_OP_RECV: | |
bcda7baa | 6057 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6058 | break; |
6059 | } | |
99bc4c38 PB |
6060 | } |
6061 | ||
0e1b6fe3 PB |
6062 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6063 | switch (req->opcode) { | |
6064 | case IORING_OP_READV: | |
6065 | case IORING_OP_READ_FIXED: | |
6066 | case IORING_OP_READ: | |
6067 | case IORING_OP_WRITEV: | |
6068 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6069 | case IORING_OP_WRITE: { |
6070 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
6071 | |
6072 | kfree(io->free_iovec); | |
0e1b6fe3 | 6073 | break; |
e8c2bc1f | 6074 | } |
0e1b6fe3 | 6075 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6076 | case IORING_OP_SENDMSG: { |
6077 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6078 | |
6079 | kfree(io->free_iov); | |
0e1b6fe3 | 6080 | break; |
e8c2bc1f | 6081 | } |
0e1b6fe3 PB |
6082 | case IORING_OP_SPLICE: |
6083 | case IORING_OP_TEE: | |
e1d767f0 PB |
6084 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6085 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6086 | break; |
f3cd4850 JA |
6087 | case IORING_OP_OPENAT: |
6088 | case IORING_OP_OPENAT2: | |
6089 | if (req->open.filename) | |
6090 | putname(req->open.filename); | |
6091 | break; | |
80a261fd JA |
6092 | case IORING_OP_RENAMEAT: |
6093 | putname(req->rename.oldpath); | |
6094 | putname(req->rename.newpath); | |
6095 | break; | |
14a1143b JA |
6096 | case IORING_OP_UNLINKAT: |
6097 | putname(req->unlink.filename); | |
6098 | break; | |
0e1b6fe3 | 6099 | } |
99bc4c38 | 6100 | } |
75652a30 JA |
6101 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6102 | kfree(req->apoll->double_poll); | |
6103 | kfree(req->apoll); | |
6104 | req->apoll = NULL; | |
6105 | } | |
3a0a6902 PB |
6106 | if (req->flags & REQ_F_INFLIGHT) { |
6107 | struct io_uring_task *tctx = req->task->io_uring; | |
6108 | ||
6109 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 6110 | } |
c854357b | 6111 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 6112 | put_cred(req->creds); |
c854357b PB |
6113 | |
6114 | req->flags &= ~IO_REQ_CLEAN_FLAGS; | |
99bc4c38 PB |
6115 | } |
6116 | ||
889fca73 | 6117 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6118 | { |
a197f664 | 6119 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6120 | const struct cred *creds = NULL; |
d625c6ee | 6121 | int ret; |
2b188cc1 | 6122 | |
b8e64b53 | 6123 | if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) |
c10d1f98 | 6124 | creds = override_creds(req->creds); |
5730b27e | 6125 | |
d625c6ee | 6126 | switch (req->opcode) { |
2b188cc1 | 6127 | case IORING_OP_NOP: |
889fca73 | 6128 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6129 | break; |
6130 | case IORING_OP_READV: | |
edafccee | 6131 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6132 | case IORING_OP_READ: |
889fca73 | 6133 | ret = io_read(req, issue_flags); |
edafccee | 6134 | break; |
3529d8c2 | 6135 | case IORING_OP_WRITEV: |
edafccee | 6136 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6137 | case IORING_OP_WRITE: |
889fca73 | 6138 | ret = io_write(req, issue_flags); |
2b188cc1 | 6139 | break; |
c992fe29 | 6140 | case IORING_OP_FSYNC: |
45d189c6 | 6141 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6142 | break; |
221c5eb2 | 6143 | case IORING_OP_POLL_ADD: |
61e98203 | 6144 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6145 | break; |
6146 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6147 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6148 | break; |
5d17b4a4 | 6149 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6150 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6151 | break; |
0fa03c62 | 6152 | case IORING_OP_SENDMSG: |
889fca73 | 6153 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6154 | break; |
fddaface | 6155 | case IORING_OP_SEND: |
889fca73 | 6156 | ret = io_send(req, issue_flags); |
0fa03c62 | 6157 | break; |
aa1fa28f | 6158 | case IORING_OP_RECVMSG: |
889fca73 | 6159 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6160 | break; |
fddaface | 6161 | case IORING_OP_RECV: |
889fca73 | 6162 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6163 | break; |
5262f567 | 6164 | case IORING_OP_TIMEOUT: |
61e98203 | 6165 | ret = io_timeout(req, issue_flags); |
5262f567 | 6166 | break; |
11365043 | 6167 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6168 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6169 | break; |
17f2fe35 | 6170 | case IORING_OP_ACCEPT: |
889fca73 | 6171 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6172 | break; |
f8e85cf2 | 6173 | case IORING_OP_CONNECT: |
889fca73 | 6174 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6175 | break; |
62755e35 | 6176 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6177 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6178 | break; |
d63d1b5e | 6179 | case IORING_OP_FALLOCATE: |
45d189c6 | 6180 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6181 | break; |
15b71abe | 6182 | case IORING_OP_OPENAT: |
45d189c6 | 6183 | ret = io_openat(req, issue_flags); |
15b71abe | 6184 | break; |
b5dba59e | 6185 | case IORING_OP_CLOSE: |
889fca73 | 6186 | ret = io_close(req, issue_flags); |
b5dba59e | 6187 | break; |
05f3fb3c | 6188 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6189 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6190 | break; |
eddc7ef5 | 6191 | case IORING_OP_STATX: |
45d189c6 | 6192 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6193 | break; |
4840e418 | 6194 | case IORING_OP_FADVISE: |
45d189c6 | 6195 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6196 | break; |
c1ca757b | 6197 | case IORING_OP_MADVISE: |
45d189c6 | 6198 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6199 | break; |
cebdb986 | 6200 | case IORING_OP_OPENAT2: |
45d189c6 | 6201 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6202 | break; |
3e4827b0 | 6203 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6204 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6205 | break; |
7d67af2c | 6206 | case IORING_OP_SPLICE: |
45d189c6 | 6207 | ret = io_splice(req, issue_flags); |
7d67af2c | 6208 | break; |
ddf0322d | 6209 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6210 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6211 | break; |
067524e9 | 6212 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6213 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6214 | break; |
f2a8d5c7 | 6215 | case IORING_OP_TEE: |
45d189c6 | 6216 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6217 | break; |
36f4fa68 | 6218 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6219 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6220 | break; |
80a261fd | 6221 | case IORING_OP_RENAMEAT: |
45d189c6 | 6222 | ret = io_renameat(req, issue_flags); |
80a261fd | 6223 | break; |
14a1143b | 6224 | case IORING_OP_UNLINKAT: |
45d189c6 | 6225 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6226 | break; |
2b188cc1 JA |
6227 | default: |
6228 | ret = -EINVAL; | |
6229 | break; | |
6230 | } | |
6231 | ||
5730b27e JA |
6232 | if (creds) |
6233 | revert_creds(creds); | |
def596e9 JA |
6234 | if (ret) |
6235 | return ret; | |
b532576e | 6236 | /* If the op doesn't have a file, we're not polling for it */ |
cb3d8972 PB |
6237 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
6238 | io_iopoll_req_issued(req); | |
def596e9 JA |
6239 | |
6240 | return 0; | |
2b188cc1 JA |
6241 | } |
6242 | ||
5280f7e5 | 6243 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6244 | { |
6245 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6246 | struct io_kiocb *timeout; |
561fb04a | 6247 | int ret = 0; |
2b188cc1 | 6248 | |
6df1db6b PB |
6249 | timeout = io_prep_linked_timeout(req); |
6250 | if (timeout) | |
6251 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6252 | |
4014d943 | 6253 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6254 | ret = -ECANCELED; |
31b51510 | 6255 | |
561fb04a | 6256 | if (!ret) { |
561fb04a | 6257 | do { |
889fca73 | 6258 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6259 | /* |
6260 | * We can get EAGAIN for polled IO even though we're | |
6261 | * forcing a sync submission from here, since we can't | |
6262 | * wait for request slots on the block side. | |
6263 | */ | |
6264 | if (ret != -EAGAIN) | |
6265 | break; | |
6266 | cond_resched(); | |
6267 | } while (1); | |
6268 | } | |
31b51510 | 6269 | |
a3df7698 | 6270 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6271 | if (ret) { |
a3df7698 | 6272 | /* io-wq is going to take one down */ |
de9b4cca | 6273 | req_ref_get(req); |
a3df7698 | 6274 | io_req_task_queue_fail(req, ret); |
edafccee | 6275 | } |
2b188cc1 JA |
6276 | } |
6277 | ||
7b29f92d JA |
6278 | #define FFS_ASYNC_READ 0x1UL |
6279 | #define FFS_ASYNC_WRITE 0x2UL | |
6280 | #ifdef CONFIG_64BIT | |
6281 | #define FFS_ISREG 0x4UL | |
6282 | #else | |
6283 | #define FFS_ISREG 0x0UL | |
6284 | #endif | |
6285 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6286 | ||
aeca241b | 6287 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6288 | unsigned i) |
65e19f54 | 6289 | { |
aeca241b | 6290 | struct io_fixed_file *table_l2; |
65e19f54 | 6291 | |
aeca241b PB |
6292 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6293 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6294 | } |
6295 | ||
65e19f54 JA |
6296 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6297 | int index) | |
6298 | { | |
aeca241b | 6299 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 6300 | |
a04b0ac0 | 6301 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6302 | } |
6303 | ||
a04b0ac0 | 6304 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6305 | { |
6306 | unsigned long file_ptr = (unsigned long) file; | |
6307 | ||
6308 | if (__io_file_supports_async(file, READ)) | |
6309 | file_ptr |= FFS_ASYNC_READ; | |
6310 | if (__io_file_supports_async(file, WRITE)) | |
6311 | file_ptr |= FFS_ASYNC_WRITE; | |
6312 | if (S_ISREG(file_inode(file)->i_mode)) | |
6313 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6314 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
6315 | } |
6316 | ||
8371adf5 PB |
6317 | static struct file *io_file_get(struct io_submit_state *state, |
6318 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6319 | { |
a197f664 | 6320 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6321 | struct file *file; |
09bb8394 | 6322 | |
8da11c19 | 6323 | if (fixed) { |
7b29f92d JA |
6324 | unsigned long file_ptr; |
6325 | ||
479f517b | 6326 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6327 | return NULL; |
b7620121 | 6328 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6329 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6330 | file = (struct file *) (file_ptr & FFS_MASK); |
6331 | file_ptr &= ~FFS_MASK; | |
6332 | /* mask in overlapping REQ_F and FFS bits */ | |
6333 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6334 | io_req_set_rsrc_node(req); |
09bb8394 | 6335 | } else { |
c826bd7a | 6336 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6337 | file = __io_file_get(state, fd); |
d44f554e JA |
6338 | |
6339 | /* we don't allow fixed io_uring files */ | |
6340 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6341 | io_req_track_inflight(req); | |
09bb8394 JA |
6342 | } |
6343 | ||
8371adf5 | 6344 | return file; |
09bb8394 JA |
6345 | } |
6346 | ||
2665abfd | 6347 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6348 | { |
ad8a48ac JA |
6349 | struct io_timeout_data *data = container_of(timer, |
6350 | struct io_timeout_data, timer); | |
90cd7e42 | 6351 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6352 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6353 | unsigned long flags; |
2665abfd JA |
6354 | |
6355 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6356 | prev = req->timeout.head; |
6357 | req->timeout.head = NULL; | |
2665abfd JA |
6358 | |
6359 | /* | |
6360 | * We don't expect the list to be empty, that will only happen if we | |
6361 | * race with the completion of the linked work. | |
6362 | */ | |
447c19f3 | 6363 | if (prev) { |
f2f87370 | 6364 | io_remove_next_linked(prev); |
447c19f3 PB |
6365 | if (!req_ref_inc_not_zero(prev)) |
6366 | prev = NULL; | |
6367 | } | |
2665abfd JA |
6368 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6369 | ||
6370 | if (prev) { | |
014db007 | 6371 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6372 | io_put_req_deferred(prev, 1); |
a298232e | 6373 | io_put_req_deferred(req, 1); |
47f46768 | 6374 | } else { |
9ae1f8dd | 6375 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6376 | } |
2665abfd JA |
6377 | return HRTIMER_NORESTART; |
6378 | } | |
6379 | ||
de968c18 | 6380 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6381 | { |
de968c18 PB |
6382 | struct io_ring_ctx *ctx = req->ctx; |
6383 | ||
6384 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6385 | /* |
f2f87370 PB |
6386 | * If the back reference is NULL, then our linked request finished |
6387 | * before we got a chance to setup the timer | |
76a46e06 | 6388 | */ |
90cd7e42 | 6389 | if (req->timeout.head) { |
e8c2bc1f | 6390 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6391 | |
ad8a48ac JA |
6392 | data->timer.function = io_link_timeout_fn; |
6393 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6394 | data->mode); | |
2665abfd | 6395 | } |
76a46e06 | 6396 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6397 | /* drop submission reference */ |
76a46e06 JA |
6398 | io_put_req(req); |
6399 | } | |
2665abfd | 6400 | |
ad8a48ac | 6401 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6402 | { |
f2f87370 | 6403 | struct io_kiocb *nxt = req->link; |
2665abfd | 6404 | |
f2f87370 PB |
6405 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6406 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6407 | return NULL; |
2665abfd | 6408 | |
90cd7e42 | 6409 | nxt->timeout.head = req; |
900fad45 | 6410 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6411 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6412 | return nxt; |
2665abfd JA |
6413 | } |
6414 | ||
c5eef2b9 | 6415 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6416 | { |
d3d7298d | 6417 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6418 | int ret; |
2b188cc1 | 6419 | |
59b735ae | 6420 | issue_sqe: |
c5eef2b9 | 6421 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6422 | |
491381ce JA |
6423 | /* |
6424 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6425 | * doesn't support non-blocking read/write attempts | |
6426 | */ | |
1840038e | 6427 | if (likely(!ret)) { |
0d63c148 | 6428 | /* drop submission reference */ |
e342c807 | 6429 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6430 | struct io_ring_ctx *ctx = req->ctx; |
6431 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6432 | |
6dd0be1e | 6433 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6434 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
2a2758f2 | 6435 | io_submit_flush_completions(ctx); |
9affd664 | 6436 | } else { |
d3d7298d | 6437 | io_put_req(req); |
0d63c148 | 6438 | } |
1840038e | 6439 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
59b735ae OL |
6440 | switch (io_arm_poll_handler(req)) { |
6441 | case IO_APOLL_READY: | |
6442 | goto issue_sqe; | |
6443 | case IO_APOLL_ABORTED: | |
1840038e PB |
6444 | /* |
6445 | * Queued up for async execution, worker will release | |
6446 | * submit reference when the iocb is actually submitted. | |
6447 | */ | |
6448 | io_queue_async_work(req); | |
59b735ae | 6449 | break; |
1840038e | 6450 | } |
0d63c148 | 6451 | } else { |
f41db273 | 6452 | io_req_complete_failed(req, ret); |
9e645e11 | 6453 | } |
d3d7298d PB |
6454 | if (linked_timeout) |
6455 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6456 | } |
6457 | ||
441b8a78 | 6458 | static inline void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 | 6459 | { |
10c66904 | 6460 | if (unlikely(req->ctx->drain_active) && io_drain_req(req)) |
76cc33d7 | 6461 | return; |
4fe2c963 | 6462 | |
76cc33d7 | 6463 | if (likely(!(req->flags & REQ_F_FORCE_ASYNC))) { |
c5eef2b9 | 6464 | __io_queue_sqe(req); |
76cc33d7 PB |
6465 | } else { |
6466 | int ret = io_req_prep_async(req); | |
6467 | ||
6468 | if (unlikely(ret)) | |
6469 | io_req_complete_failed(req, ret); | |
6470 | else | |
6471 | io_queue_async_work(req); | |
ce35a47a | 6472 | } |
4fe2c963 JL |
6473 | } |
6474 | ||
b16fed66 PB |
6475 | /* |
6476 | * Check SQE restrictions (opcode and flags). | |
6477 | * | |
6478 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6479 | */ | |
6480 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6481 | struct io_kiocb *req, | |
6482 | unsigned int sqe_flags) | |
4fe2c963 | 6483 | { |
4cfb25bf | 6484 | if (likely(!ctx->restricted)) |
b16fed66 PB |
6485 | return true; |
6486 | ||
6487 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6488 | return false; | |
6489 | ||
6490 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6491 | ctx->restrictions.sqe_flags_required) | |
6492 | return false; | |
6493 | ||
6494 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6495 | ctx->restrictions.sqe_flags_required)) | |
6496 | return false; | |
6497 | ||
6498 | return true; | |
4fe2c963 JL |
6499 | } |
6500 | ||
b16fed66 PB |
6501 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6502 | const struct io_uring_sqe *sqe) | |
6503 | { | |
6504 | struct io_submit_state *state; | |
6505 | unsigned int sqe_flags; | |
003e8dcc | 6506 | int personality, ret = 0; |
b16fed66 PB |
6507 | |
6508 | req->opcode = READ_ONCE(sqe->opcode); | |
6509 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6510 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6511 | req->user_data = READ_ONCE(sqe->user_data); | |
b16fed66 | 6512 | req->file = NULL; |
b16fed66 PB |
6513 | req->fixed_rsrc_refs = NULL; |
6514 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6515 | atomic_set(&req->refs, 2); |
b16fed66 | 6516 | req->task = current; |
b16fed66 PB |
6517 | |
6518 | /* enforce forwards compatibility on users */ | |
dddca226 | 6519 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) |
b16fed66 | 6520 | return -EINVAL; |
b16fed66 PB |
6521 | if (unlikely(req->opcode >= IORING_OP_LAST)) |
6522 | return -EINVAL; | |
4cfb25bf | 6523 | if (!io_check_restriction(ctx, req, sqe_flags)) |
b16fed66 PB |
6524 | return -EACCES; |
6525 | ||
6526 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6527 | !io_op_defs[req->opcode].buffer_select) | |
6528 | return -EOPNOTSUPP; | |
3c19966d PB |
6529 | if (unlikely(sqe_flags & IOSQE_IO_DRAIN)) |
6530 | ctx->drain_active = true; | |
863e0560 | 6531 | |
003e8dcc JA |
6532 | personality = READ_ONCE(sqe->personality); |
6533 | if (personality) { | |
c10d1f98 PB |
6534 | req->creds = xa_load(&ctx->personalities, personality); |
6535 | if (!req->creds) | |
003e8dcc | 6536 | return -EINVAL; |
c10d1f98 | 6537 | get_cred(req->creds); |
b8e64b53 | 6538 | req->flags |= REQ_F_CREDS; |
003e8dcc | 6539 | } |
b16fed66 PB |
6540 | state = &ctx->submit_state; |
6541 | ||
6542 | /* | |
6543 | * Plug now if we have more than 1 IO left after this, and the target | |
6544 | * is potentially a read/write to block based storage. | |
6545 | */ | |
6546 | if (!state->plug_started && state->ios_left > 1 && | |
6547 | io_op_defs[req->opcode].plug) { | |
6548 | blk_start_plug(&state->plug); | |
6549 | state->plug_started = true; | |
6550 | } | |
6551 | ||
6552 | if (io_op_defs[req->opcode].needs_file) { | |
6553 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6554 | ||
6555 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6556 | if (unlikely(!req->file)) | |
6557 | ret = -EBADF; | |
6558 | } | |
6559 | ||
6560 | state->ios_left--; | |
6561 | return ret; | |
6562 | } | |
6563 | ||
a6b8cadc | 6564 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6565 | const struct io_uring_sqe *sqe) |
9e645e11 | 6566 | { |
a1ab7b35 | 6567 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6568 | int ret; |
9e645e11 | 6569 | |
a6b8cadc PB |
6570 | ret = io_init_req(ctx, req, sqe); |
6571 | if (unlikely(ret)) { | |
6572 | fail_req: | |
de59bc10 PB |
6573 | if (link->head) { |
6574 | /* fail even hard links since we don't submit */ | |
93d2bcd2 | 6575 | req_set_fail(link->head); |
f41db273 | 6576 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6577 | link->head = NULL; |
6578 | } | |
f41db273 | 6579 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6580 | return ret; |
6581 | } | |
441b8a78 | 6582 | |
be7053b7 PB |
6583 | ret = io_req_prep(req, sqe); |
6584 | if (unlikely(ret)) | |
6585 | goto fail_req; | |
a6b8cadc | 6586 | |
be7053b7 | 6587 | /* don't need @sqe from now on */ |
236daeae OL |
6588 | trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, |
6589 | req->flags, true, | |
6590 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 6591 | |
9e645e11 JA |
6592 | /* |
6593 | * If we already have a head request, queue this one for async | |
6594 | * submittal once the head completes. If we don't have a head but | |
6595 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6596 | * submitted sync once the chain is complete. If none of those | |
6597 | * conditions are true (normal request), then just queue it. | |
6598 | */ | |
863e0560 PB |
6599 | if (link->head) { |
6600 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6601 | |
b7e298d2 | 6602 | ret = io_req_prep_async(req); |
cf109604 | 6603 | if (unlikely(ret)) |
a6b8cadc | 6604 | goto fail_req; |
9d76377f | 6605 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6606 | link->last->link = req; |
863e0560 | 6607 | link->last = req; |
32fe525b PB |
6608 | |
6609 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6610 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
863e0560 | 6611 | link->head = NULL; |
5e159204 | 6612 | io_queue_sqe(head); |
32fe525b | 6613 | } |
9e645e11 | 6614 | } else { |
ef4ff581 | 6615 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6616 | link->head = req; |
6617 | link->last = req; | |
711be031 | 6618 | } else { |
be7053b7 | 6619 | io_queue_sqe(req); |
711be031 | 6620 | } |
9e645e11 | 6621 | } |
2e6e1fde | 6622 | |
1d4240cc | 6623 | return 0; |
9e645e11 JA |
6624 | } |
6625 | ||
9a56a232 JA |
6626 | /* |
6627 | * Batched submission is done, ensure local IO is flushed out. | |
6628 | */ | |
ba88ff11 PB |
6629 | static void io_submit_state_end(struct io_submit_state *state, |
6630 | struct io_ring_ctx *ctx) | |
9a56a232 | 6631 | { |
a1ab7b35 | 6632 | if (state->link.head) |
de59bc10 | 6633 | io_queue_sqe(state->link.head); |
6dd0be1e | 6634 | if (state->comp.nr) |
2a2758f2 | 6635 | io_submit_flush_completions(ctx); |
27926b68 JA |
6636 | if (state->plug_started) |
6637 | blk_finish_plug(&state->plug); | |
9f13c35b | 6638 | io_state_file_put(state); |
9a56a232 JA |
6639 | } |
6640 | ||
6641 | /* | |
6642 | * Start submission side cache. | |
6643 | */ | |
6644 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6645 | unsigned int max_ios) |
9a56a232 | 6646 | { |
27926b68 | 6647 | state->plug_started = false; |
9a56a232 | 6648 | state->ios_left = max_ios; |
a1ab7b35 PB |
6649 | /* set only head, no need to init link_last in advance */ |
6650 | state->link.head = NULL; | |
9a56a232 JA |
6651 | } |
6652 | ||
2b188cc1 JA |
6653 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6654 | { | |
75b28aff | 6655 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6656 | |
caf582c6 PB |
6657 | /* |
6658 | * Ensure any loads from the SQEs are done at this point, | |
6659 | * since once we write the new head, the application could | |
6660 | * write new data to them. | |
6661 | */ | |
6662 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6663 | } |
6664 | ||
2b188cc1 | 6665 | /* |
dd9ae8a0 | 6666 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
6667 | * that is mapped by userspace. This means that care needs to be taken to |
6668 | * ensure that reads are stable, as we cannot rely on userspace always | |
6669 | * being a good citizen. If members of the sqe are validated and then later | |
6670 | * used, it's important that those reads are done through READ_ONCE() to | |
6671 | * prevent a re-load down the line. | |
6672 | */ | |
709b302f | 6673 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6674 | { |
ea5ab3b5 | 6675 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 6676 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
6677 | |
6678 | /* | |
6679 | * The cached sq head (or cq tail) serves two purposes: | |
6680 | * | |
6681 | * 1) allows us to batch the cost of updating the user visible | |
6682 | * head updates. | |
6683 | * 2) allows the kernel side to track the head on its own, even | |
6684 | * though the application is the one updating it. | |
6685 | */ | |
17d3aeb3 | 6686 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
6687 | if (likely(head < ctx->sq_entries)) |
6688 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6689 | |
6690 | /* drop invalid entries */ | |
15641e42 PB |
6691 | ctx->cq_extra--; |
6692 | WRITE_ONCE(ctx->rings->sq_dropped, | |
6693 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
6694 | return NULL; |
6695 | } | |
6696 | ||
0f212204 | 6697 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6698 | { |
09899b19 | 6699 | struct io_uring_task *tctx; |
46c4e16a | 6700 | int submitted = 0; |
6c271ce2 | 6701 | |
ee7d46d9 PB |
6702 | /* make sure SQ entry isn't read before tail */ |
6703 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
2b85edfc PB |
6704 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6705 | return -EAGAIN; | |
6c271ce2 | 6706 | |
09899b19 PB |
6707 | tctx = current->io_uring; |
6708 | tctx->cached_refs -= nr; | |
6709 | if (unlikely(tctx->cached_refs < 0)) { | |
6710 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
6711 | ||
6712 | percpu_counter_add(&tctx->inflight, refill); | |
6713 | refcount_add(refill, ¤t->usage); | |
6714 | tctx->cached_refs += refill; | |
6715 | } | |
ba88ff11 | 6716 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6717 | |
46c4e16a | 6718 | while (submitted < nr) { |
3529d8c2 | 6719 | const struct io_uring_sqe *sqe; |
196be95c | 6720 | struct io_kiocb *req; |
fb5ccc98 | 6721 | |
258b29a9 | 6722 | req = io_alloc_req(ctx); |
196be95c PB |
6723 | if (unlikely(!req)) { |
6724 | if (!submitted) | |
6725 | submitted = -EAGAIN; | |
fb5ccc98 | 6726 | break; |
196be95c | 6727 | } |
4fccfcbb PB |
6728 | sqe = io_get_sqe(ctx); |
6729 | if (unlikely(!sqe)) { | |
6730 | kmem_cache_free(req_cachep, req); | |
6731 | break; | |
6732 | } | |
d3656344 JA |
6733 | /* will complete beyond this point, count as submitted */ |
6734 | submitted++; | |
a1ab7b35 | 6735 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6736 | break; |
6c271ce2 JA |
6737 | } |
6738 | ||
9466f437 PB |
6739 | if (unlikely(submitted != nr)) { |
6740 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 6741 | int unused = nr - ref_used; |
9466f437 | 6742 | |
09899b19 | 6743 | current->io_uring->cached_refs += unused; |
d8a6df10 | 6744 | percpu_ref_put_many(&ctx->refs, unused); |
9466f437 | 6745 | } |
6c271ce2 | 6746 | |
a1ab7b35 | 6747 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6748 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6749 | io_commit_sqring(ctx); | |
6750 | ||
6c271ce2 JA |
6751 | return submitted; |
6752 | } | |
6753 | ||
e4b6d902 PB |
6754 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
6755 | { | |
6756 | return READ_ONCE(sqd->state); | |
6757 | } | |
6758 | ||
23b3628e XW |
6759 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6760 | { | |
6761 | /* Tell userspace we may need a wakeup call */ | |
6762 | spin_lock_irq(&ctx->completion_lock); | |
6763 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6764 | spin_unlock_irq(&ctx->completion_lock); | |
6765 | } | |
6766 | ||
6767 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6768 | { | |
6769 | spin_lock_irq(&ctx->completion_lock); | |
6770 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6771 | spin_unlock_irq(&ctx->completion_lock); | |
6772 | } | |
6773 | ||
08369246 | 6774 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6775 | { |
c8d1ba58 | 6776 | unsigned int to_submit; |
bdcd3eab | 6777 | int ret = 0; |
6c271ce2 | 6778 | |
c8d1ba58 | 6779 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 6780 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
6781 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
6782 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 6783 | |
906a3c6f | 6784 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6785 | unsigned nr_events = 0; |
948e1947 PB |
6786 | const struct cred *creds = NULL; |
6787 | ||
6788 | if (ctx->sq_creds != current_cred()) | |
6789 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 6790 | |
c8d1ba58 | 6791 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6792 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6793 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6794 | |
3b763ba1 PB |
6795 | /* |
6796 | * Don't submit if refs are dying, good for io_uring_register(), | |
6797 | * but also it is relied upon by io_ring_exit_work() | |
6798 | */ | |
0298ef96 PB |
6799 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6800 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6801 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 6802 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 | 6803 | |
acfb381d PB |
6804 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6805 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
6806 | if (creds) |
6807 | revert_creds(creds); | |
acfb381d | 6808 | } |
6c271ce2 | 6809 | |
08369246 XW |
6810 | return ret; |
6811 | } | |
6c271ce2 | 6812 | |
08369246 XW |
6813 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6814 | { | |
6815 | struct io_ring_ctx *ctx; | |
6816 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6817 | |
c9dca27d PB |
6818 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6819 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6820 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6821 | } |
6c271ce2 | 6822 | |
e4b6d902 PB |
6823 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
6824 | { | |
6825 | bool did_sig = false; | |
6826 | struct ksignal ksig; | |
6827 | ||
6828 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
6829 | signal_pending(current)) { | |
6830 | mutex_unlock(&sqd->lock); | |
6831 | if (signal_pending(current)) | |
6832 | did_sig = get_signal(&ksig); | |
6833 | cond_resched(); | |
6834 | mutex_lock(&sqd->lock); | |
6835 | } | |
e4b6d902 PB |
6836 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
6837 | } | |
6838 | ||
c8d1ba58 JA |
6839 | static int io_sq_thread(void *data) |
6840 | { | |
69fb2131 JA |
6841 | struct io_sq_data *sqd = data; |
6842 | struct io_ring_ctx *ctx; | |
a0d9205f | 6843 | unsigned long timeout = 0; |
37d1e2e3 | 6844 | char buf[TASK_COMM_LEN]; |
08369246 | 6845 | DEFINE_WAIT(wait); |
6c271ce2 | 6846 | |
696ee88a | 6847 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6848 | set_task_comm(current, buf); |
37d1e2e3 JA |
6849 | |
6850 | if (sqd->sq_cpu != -1) | |
6851 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6852 | else | |
6853 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6854 | current->flags |= PF_NO_SETAFFINITY; | |
6855 | ||
09a6f4ef | 6856 | mutex_lock(&sqd->lock); |
e4b6d902 | 6857 | while (1) { |
1a924a80 | 6858 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 6859 | |
e4b6d902 PB |
6860 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
6861 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 6862 | break; |
08369246 XW |
6863 | timeout = jiffies + sqd->sq_thread_idle; |
6864 | } | |
e4b6d902 | 6865 | |
e95eee2d | 6866 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6867 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 6868 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 6869 | |
08369246 XW |
6870 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6871 | sqt_spin = true; | |
69fb2131 | 6872 | } |
dd432ea5 PB |
6873 | if (io_run_task_work()) |
6874 | sqt_spin = true; | |
6c271ce2 | 6875 | |
08369246 | 6876 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 6877 | cond_resched(); |
08369246 XW |
6878 | if (sqt_spin) |
6879 | timeout = jiffies + sqd->sq_thread_idle; | |
6880 | continue; | |
6881 | } | |
6882 | ||
08369246 | 6883 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
dd432ea5 | 6884 | if (!io_sqd_events_pending(sqd) && !current->task_works) { |
1a924a80 PB |
6885 | bool needs_sched = true; |
6886 | ||
724cb4f9 | 6887 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
6888 | io_ring_set_wakeup_flag(ctx); |
6889 | ||
724cb4f9 HX |
6890 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
6891 | !list_empty_careful(&ctx->iopoll_list)) { | |
6892 | needs_sched = false; | |
6893 | break; | |
6894 | } | |
6895 | if (io_sqring_entries(ctx)) { | |
6896 | needs_sched = false; | |
6897 | break; | |
6898 | } | |
6899 | } | |
6900 | ||
6901 | if (needs_sched) { | |
6902 | mutex_unlock(&sqd->lock); | |
6903 | schedule(); | |
6904 | mutex_lock(&sqd->lock); | |
6905 | } | |
69fb2131 JA |
6906 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6907 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6908 | } |
08369246 XW |
6909 | |
6910 | finish_wait(&sqd->wait, &wait); | |
6911 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6912 | } |
28cea78a | 6913 | |
78cc687b | 6914 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 6915 | sqd->thread = NULL; |
05962f95 | 6916 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6917 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 6918 | io_run_task_work(); |
734551df PB |
6919 | mutex_unlock(&sqd->lock); |
6920 | ||
37d1e2e3 JA |
6921 | complete(&sqd->exited); |
6922 | do_exit(0); | |
6c271ce2 JA |
6923 | } |
6924 | ||
bda52162 JA |
6925 | struct io_wait_queue { |
6926 | struct wait_queue_entry wq; | |
6927 | struct io_ring_ctx *ctx; | |
6928 | unsigned to_wait; | |
6929 | unsigned nr_timeouts; | |
6930 | }; | |
6931 | ||
6c503150 | 6932 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6933 | { |
6934 | struct io_ring_ctx *ctx = iowq->ctx; | |
6935 | ||
6936 | /* | |
d195a66e | 6937 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6938 | * started waiting. For timeouts, we always want to return to userspace, |
6939 | * regardless of event count. | |
6940 | */ | |
6c503150 | 6941 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6942 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6943 | } | |
6944 | ||
6945 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6946 | int wake_flags, void *key) | |
6947 | { | |
6948 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6949 | wq); | |
6950 | ||
6c503150 PB |
6951 | /* |
6952 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6953 | * the task, and the next invocation will do it. | |
6954 | */ | |
5ed7a37d | 6955 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
6956 | return autoremove_wake_function(curr, mode, wake_flags, key); |
6957 | return -1; | |
bda52162 JA |
6958 | } |
6959 | ||
af9c1a44 JA |
6960 | static int io_run_task_work_sig(void) |
6961 | { | |
6962 | if (io_run_task_work()) | |
6963 | return 1; | |
6964 | if (!signal_pending(current)) | |
6965 | return 0; | |
0b8cfa97 | 6966 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6967 | return -ERESTARTSYS; |
af9c1a44 JA |
6968 | return -EINTR; |
6969 | } | |
6970 | ||
eeb60b9a PB |
6971 | /* when returns >0, the caller should retry */ |
6972 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6973 | struct io_wait_queue *iowq, | |
6974 | signed long *timeout) | |
6975 | { | |
6976 | int ret; | |
6977 | ||
6978 | /* make sure we run task_work before checking for signals */ | |
6979 | ret = io_run_task_work_sig(); | |
6980 | if (ret || io_should_wake(iowq)) | |
6981 | return ret; | |
6982 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 6983 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
6984 | return 1; |
6985 | ||
6986 | *timeout = schedule_timeout(*timeout); | |
6987 | return !*timeout ? -ETIME : 1; | |
6988 | } | |
6989 | ||
2b188cc1 JA |
6990 | /* |
6991 | * Wait until events become available, if we don't already have some. The | |
6992 | * application must reap them itself, as they reside on the shared cq ring. | |
6993 | */ | |
6994 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6995 | const sigset_t __user *sig, size_t sigsz, |
6996 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6997 | { |
bda52162 JA |
6998 | struct io_wait_queue iowq = { |
6999 | .wq = { | |
7000 | .private = current, | |
7001 | .func = io_wake_function, | |
7002 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7003 | }, | |
7004 | .ctx = ctx, | |
7005 | .to_wait = min_events, | |
7006 | }; | |
75b28aff | 7007 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7008 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7009 | int ret; | |
2b188cc1 | 7010 | |
b41e9852 | 7011 | do { |
6c2450ae | 7012 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 7013 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7014 | return 0; |
4c6e277c | 7015 | if (!io_run_task_work()) |
b41e9852 | 7016 | break; |
b41e9852 | 7017 | } while (1); |
2b188cc1 JA |
7018 | |
7019 | if (sig) { | |
9e75ad5d AB |
7020 | #ifdef CONFIG_COMPAT |
7021 | if (in_compat_syscall()) | |
7022 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7023 | sigsz); |
9e75ad5d AB |
7024 | else |
7025 | #endif | |
b772434b | 7026 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7027 | |
2b188cc1 JA |
7028 | if (ret) |
7029 | return ret; | |
7030 | } | |
7031 | ||
c73ebb68 | 7032 | if (uts) { |
c1d5a224 PB |
7033 | struct timespec64 ts; |
7034 | ||
c73ebb68 HX |
7035 | if (get_timespec64(&ts, uts)) |
7036 | return -EFAULT; | |
7037 | timeout = timespec64_to_jiffies(&ts); | |
7038 | } | |
7039 | ||
bda52162 | 7040 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7041 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7042 | do { |
ca0a2651 | 7043 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7044 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7045 | ret = -EBUSY; |
7046 | break; | |
7047 | } | |
311997b3 | 7048 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 7049 | TASK_INTERRUPTIBLE); |
eeb60b9a | 7050 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
311997b3 | 7051 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 7052 | cond_resched(); |
eeb60b9a | 7053 | } while (ret > 0); |
bda52162 | 7054 | |
b7db41c9 | 7055 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7056 | |
75b28aff | 7057 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7058 | } |
7059 | ||
9123c8ff | 7060 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 7061 | { |
9123c8ff | 7062 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 7063 | |
846a4ef2 | 7064 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
7065 | kfree(table[i]); |
7066 | kfree(table); | |
7067 | } | |
7068 | ||
7069 | static void **io_alloc_page_table(size_t size) | |
7070 | { | |
7071 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
7072 | size_t init_size = size; | |
7073 | void **table; | |
7074 | ||
7075 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL); | |
7076 | if (!table) | |
7077 | return NULL; | |
7078 | ||
7079 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 7080 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff PB |
7081 | |
7082 | table[i] = kzalloc(this_size, GFP_KERNEL); | |
7083 | if (!table[i]) { | |
7084 | io_free_page_table(table, init_size); | |
7085 | return NULL; | |
7086 | } | |
7087 | size -= this_size; | |
7088 | } | |
7089 | return table; | |
05f3fb3c JA |
7090 | } |
7091 | ||
2a63b2d9 | 7092 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7093 | { |
2a63b2d9 | 7094 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7095 | } |
7096 | ||
2a63b2d9 | 7097 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7098 | { |
2a63b2d9 BM |
7099 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7100 | } | |
65e19f54 | 7101 | |
28a9fe25 | 7102 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 7103 | { |
28a9fe25 PB |
7104 | percpu_ref_exit(&ref_node->refs); |
7105 | kfree(ref_node); | |
1642b445 PB |
7106 | } |
7107 | ||
a7f0ed5a PB |
7108 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7109 | struct io_rsrc_data *data_to_kill) | |
6b06314c | 7110 | { |
a7f0ed5a PB |
7111 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7112 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 7113 | |
a7f0ed5a PB |
7114 | if (data_to_kill) { |
7115 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7116 | |
a7f0ed5a PB |
7117 | rsrc_node->rsrc_data = data_to_kill; |
7118 | io_rsrc_ref_lock(ctx); | |
7119 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7120 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7121 | |
3e942498 | 7122 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7123 | percpu_ref_kill(&rsrc_node->refs); |
7124 | ctx->rsrc_node = NULL; | |
7125 | } | |
6b06314c | 7126 | |
a7f0ed5a PB |
7127 | if (!ctx->rsrc_node) { |
7128 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7129 | ctx->rsrc_backup_node = NULL; | |
7130 | } | |
8bad28d8 HX |
7131 | } |
7132 | ||
a7f0ed5a | 7133 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7134 | { |
7135 | if (ctx->rsrc_backup_node) | |
7136 | return 0; | |
b895c9a6 | 7137 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe | 7138 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
7139 | } |
7140 | ||
40ae0ff7 | 7141 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7142 | { |
7143 | int ret; | |
05589553 | 7144 | |
215c3902 | 7145 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7146 | if (data->quiesce) |
7147 | return -ENXIO; | |
05589553 | 7148 | |
8bad28d8 | 7149 | data->quiesce = true; |
1ffc5422 | 7150 | do { |
a7f0ed5a | 7151 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7152 | if (ret) |
f2303b1f | 7153 | break; |
a7f0ed5a | 7154 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 7155 | |
3e942498 PB |
7156 | /* kill initial ref, already quiesced if zero */ |
7157 | if (atomic_dec_and_test(&data->refs)) | |
7158 | break; | |
8bad28d8 | 7159 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7160 | ret = wait_for_completion_interruptible(&data->done); |
7161 | if (!ret) | |
7162 | break; | |
8bad28d8 | 7163 | |
3e942498 PB |
7164 | atomic_inc(&data->refs); |
7165 | /* wait for all works potentially completing data->done */ | |
7166 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7167 | reinit_completion(&data->done); |
8dd03afe | 7168 | |
8bad28d8 | 7169 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7170 | ret = io_run_task_work_sig(); |
8bad28d8 | 7171 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7172 | } while (ret >= 0); |
8bad28d8 | 7173 | data->quiesce = false; |
05f3fb3c | 7174 | |
8bad28d8 | 7175 | return ret; |
d7954b2b BM |
7176 | } |
7177 | ||
2d091d62 PB |
7178 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
7179 | { | |
7180 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
7181 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
7182 | ||
7183 | return &data->tags[table_idx][off]; | |
7184 | } | |
7185 | ||
44b31f2f | 7186 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 7187 | { |
2d091d62 PB |
7188 | size_t size = data->nr * sizeof(data->tags[0][0]); |
7189 | ||
7190 | if (data->tags) | |
7191 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
7192 | kfree(data); |
7193 | } | |
7194 | ||
d878c816 PB |
7195 | static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
7196 | u64 __user *utags, unsigned nr, | |
7197 | struct io_rsrc_data **pdata) | |
1ad555c6 | 7198 | { |
b895c9a6 | 7199 | struct io_rsrc_data *data; |
2d091d62 | 7200 | int ret = -ENOMEM; |
d878c816 | 7201 | unsigned i; |
1ad555c6 BM |
7202 | |
7203 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7204 | if (!data) | |
d878c816 | 7205 | return -ENOMEM; |
2d091d62 | 7206 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 7207 | if (!data->tags) { |
1ad555c6 | 7208 | kfree(data); |
d878c816 PB |
7209 | return -ENOMEM; |
7210 | } | |
2d091d62 PB |
7211 | |
7212 | data->nr = nr; | |
7213 | data->ctx = ctx; | |
7214 | data->do_put = do_put; | |
d878c816 | 7215 | if (utags) { |
2d091d62 | 7216 | ret = -EFAULT; |
d878c816 | 7217 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
7218 | u64 *tag_slot = io_get_tag_slot(data, i); |
7219 | ||
7220 | if (copy_from_user(tag_slot, &utags[i], | |
7221 | sizeof(*tag_slot))) | |
2d091d62 | 7222 | goto fail; |
d878c816 | 7223 | } |
1ad555c6 | 7224 | } |
b60c8dce | 7225 | |
3e942498 | 7226 | atomic_set(&data->refs, 1); |
1ad555c6 | 7227 | init_completion(&data->done); |
d878c816 PB |
7228 | *pdata = data; |
7229 | return 0; | |
2d091d62 PB |
7230 | fail: |
7231 | io_rsrc_data_free(data); | |
7232 | return ret; | |
1ad555c6 BM |
7233 | } |
7234 | ||
9123c8ff PB |
7235 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
7236 | { | |
7237 | size_t size = nr_files * sizeof(struct io_fixed_file); | |
7238 | ||
7239 | table->files = (struct io_fixed_file **)io_alloc_page_table(size); | |
7240 | return !!table->files; | |
7241 | } | |
7242 | ||
7243 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) | |
7244 | { | |
7245 | size_t size = nr_files * sizeof(struct io_fixed_file); | |
7246 | ||
7247 | io_free_page_table((void **)table->files, size); | |
7248 | table->files = NULL; | |
7249 | } | |
7250 | ||
fff4db76 | 7251 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 7252 | { |
fff4db76 PB |
7253 | #if defined(CONFIG_UNIX) |
7254 | if (ctx->ring_sock) { | |
7255 | struct sock *sock = ctx->ring_sock->sk; | |
7256 | struct sk_buff *skb; | |
7257 | ||
7258 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7259 | kfree_skb(skb); | |
7260 | } | |
7261 | #else | |
7262 | int i; | |
7263 | ||
7264 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7265 | struct file *file; | |
7266 | ||
7267 | file = io_file_from_index(ctx, i); | |
7268 | if (file) | |
7269 | fput(file); | |
7270 | } | |
7271 | #endif | |
7272 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); | |
44b31f2f | 7273 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7274 | ctx->file_data = NULL; |
7275 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
7276 | } |
7277 | ||
d7954b2b BM |
7278 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7279 | { | |
d7954b2b BM |
7280 | int ret; |
7281 | ||
08480400 | 7282 | if (!ctx->file_data) |
d7954b2b | 7283 | return -ENXIO; |
08480400 PB |
7284 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7285 | if (!ret) | |
7286 | __io_sqe_files_unregister(ctx); | |
7287 | return ret; | |
6b06314c JA |
7288 | } |
7289 | ||
37d1e2e3 | 7290 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7291 | __releases(&sqd->lock) |
37d1e2e3 | 7292 | { |
521d6a73 PB |
7293 | WARN_ON_ONCE(sqd->thread == current); |
7294 | ||
9e138a48 PB |
7295 | /* |
7296 | * Do the dance but not conditional clear_bit() because it'd race with | |
7297 | * other threads incrementing park_pending and setting the bit. | |
7298 | */ | |
37d1e2e3 | 7299 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7300 | if (atomic_dec_return(&sqd->park_pending)) |
7301 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7302 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7303 | } |
7304 | ||
86e0d676 | 7305 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7306 | __acquires(&sqd->lock) |
37d1e2e3 | 7307 | { |
521d6a73 PB |
7308 | WARN_ON_ONCE(sqd->thread == current); |
7309 | ||
9e138a48 | 7310 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7311 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7312 | mutex_lock(&sqd->lock); |
05962f95 | 7313 | if (sqd->thread) |
86e0d676 | 7314 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7315 | } |
7316 | ||
7317 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7318 | { | |
521d6a73 | 7319 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7320 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7321 | |
05962f95 | 7322 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7323 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7324 | if (sqd->thread) |
7325 | wake_up_process(sqd->thread); | |
09a6f4ef | 7326 | mutex_unlock(&sqd->lock); |
05962f95 | 7327 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7328 | } |
7329 | ||
534ca6d6 | 7330 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7331 | { |
534ca6d6 | 7332 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7333 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7334 | ||
37d1e2e3 JA |
7335 | io_sq_thread_stop(sqd); |
7336 | kfree(sqd); | |
7337 | } | |
7338 | } | |
7339 | ||
7340 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7341 | { | |
7342 | struct io_sq_data *sqd = ctx->sq_data; | |
7343 | ||
7344 | if (sqd) { | |
05962f95 | 7345 | io_sq_thread_park(sqd); |
521d6a73 | 7346 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7347 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7348 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7349 | |
7350 | io_put_sq_data(sqd); | |
7351 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7352 | } |
7353 | } | |
7354 | ||
aa06165d JA |
7355 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7356 | { | |
7357 | struct io_ring_ctx *ctx_attach; | |
7358 | struct io_sq_data *sqd; | |
7359 | struct fd f; | |
7360 | ||
7361 | f = fdget(p->wq_fd); | |
7362 | if (!f.file) | |
7363 | return ERR_PTR(-ENXIO); | |
7364 | if (f.file->f_op != &io_uring_fops) { | |
7365 | fdput(f); | |
7366 | return ERR_PTR(-EINVAL); | |
7367 | } | |
7368 | ||
7369 | ctx_attach = f.file->private_data; | |
7370 | sqd = ctx_attach->sq_data; | |
7371 | if (!sqd) { | |
7372 | fdput(f); | |
7373 | return ERR_PTR(-EINVAL); | |
7374 | } | |
5c2469e0 JA |
7375 | if (sqd->task_tgid != current->tgid) { |
7376 | fdput(f); | |
7377 | return ERR_PTR(-EPERM); | |
7378 | } | |
aa06165d JA |
7379 | |
7380 | refcount_inc(&sqd->refs); | |
7381 | fdput(f); | |
7382 | return sqd; | |
7383 | } | |
7384 | ||
26984fbf PB |
7385 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7386 | bool *attached) | |
534ca6d6 JA |
7387 | { |
7388 | struct io_sq_data *sqd; | |
7389 | ||
26984fbf | 7390 | *attached = false; |
5c2469e0 JA |
7391 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7392 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7393 | if (!IS_ERR(sqd)) { |
7394 | *attached = true; | |
5c2469e0 | 7395 | return sqd; |
26984fbf | 7396 | } |
5c2469e0 JA |
7397 | /* fall through for EPERM case, setup new sqd/task */ |
7398 | if (PTR_ERR(sqd) != -EPERM) | |
7399 | return sqd; | |
7400 | } | |
aa06165d | 7401 | |
534ca6d6 JA |
7402 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7403 | if (!sqd) | |
7404 | return ERR_PTR(-ENOMEM); | |
7405 | ||
9e138a48 | 7406 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7407 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7408 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7409 | mutex_init(&sqd->lock); |
534ca6d6 | 7410 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7411 | init_completion(&sqd->exited); |
534ca6d6 JA |
7412 | return sqd; |
7413 | } | |
7414 | ||
6b06314c | 7415 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7416 | /* |
7417 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7418 | * the io_uring can be safely unregistered on process exit, even if we have | |
7419 | * loops in the file referencing. | |
7420 | */ | |
7421 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7422 | { | |
7423 | struct sock *sk = ctx->ring_sock->sk; | |
7424 | struct scm_fp_list *fpl; | |
7425 | struct sk_buff *skb; | |
08a45173 | 7426 | int i, nr_files; |
6b06314c | 7427 | |
6b06314c JA |
7428 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7429 | if (!fpl) | |
7430 | return -ENOMEM; | |
7431 | ||
7432 | skb = alloc_skb(0, GFP_KERNEL); | |
7433 | if (!skb) { | |
7434 | kfree(fpl); | |
7435 | return -ENOMEM; | |
7436 | } | |
7437 | ||
7438 | skb->sk = sk; | |
6b06314c | 7439 | |
08a45173 | 7440 | nr_files = 0; |
62e398be | 7441 | fpl->user = get_uid(current_user()); |
6b06314c | 7442 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7443 | struct file *file = io_file_from_index(ctx, i + offset); |
7444 | ||
7445 | if (!file) | |
08a45173 | 7446 | continue; |
65e19f54 | 7447 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7448 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7449 | nr_files++; | |
6b06314c JA |
7450 | } |
7451 | ||
08a45173 JA |
7452 | if (nr_files) { |
7453 | fpl->max = SCM_MAX_FD; | |
7454 | fpl->count = nr_files; | |
7455 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7456 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7457 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7458 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7459 | |
08a45173 JA |
7460 | for (i = 0; i < nr_files; i++) |
7461 | fput(fpl->fp[i]); | |
7462 | } else { | |
7463 | kfree_skb(skb); | |
7464 | kfree(fpl); | |
7465 | } | |
6b06314c JA |
7466 | |
7467 | return 0; | |
7468 | } | |
7469 | ||
7470 | /* | |
7471 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7472 | * causes regular reference counting to break down. We rely on the UNIX | |
7473 | * garbage collection to take care of this problem for us. | |
7474 | */ | |
7475 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7476 | { | |
7477 | unsigned left, total; | |
7478 | int ret = 0; | |
7479 | ||
7480 | total = 0; | |
7481 | left = ctx->nr_user_files; | |
7482 | while (left) { | |
7483 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7484 | |
7485 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7486 | if (ret) | |
7487 | break; | |
7488 | left -= this_files; | |
7489 | total += this_files; | |
7490 | } | |
7491 | ||
7492 | if (!ret) | |
7493 | return 0; | |
7494 | ||
7495 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7496 | struct file *file = io_file_from_index(ctx, total); |
7497 | ||
7498 | if (file) | |
7499 | fput(file); | |
6b06314c JA |
7500 | total++; |
7501 | } | |
7502 | ||
7503 | return ret; | |
7504 | } | |
7505 | #else | |
7506 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7507 | { | |
7508 | return 0; | |
7509 | } | |
7510 | #endif | |
7511 | ||
47e90392 | 7512 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7513 | { |
50238531 | 7514 | struct file *file = prsrc->file; |
05f3fb3c JA |
7515 | #if defined(CONFIG_UNIX) |
7516 | struct sock *sock = ctx->ring_sock->sk; | |
7517 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7518 | struct sk_buff *skb; | |
7519 | int i; | |
7520 | ||
7521 | __skb_queue_head_init(&list); | |
7522 | ||
7523 | /* | |
7524 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7525 | * remove this entry and rearrange the file array. | |
7526 | */ | |
7527 | skb = skb_dequeue(head); | |
7528 | while (skb) { | |
7529 | struct scm_fp_list *fp; | |
7530 | ||
7531 | fp = UNIXCB(skb).fp; | |
7532 | for (i = 0; i < fp->count; i++) { | |
7533 | int left; | |
7534 | ||
7535 | if (fp->fp[i] != file) | |
7536 | continue; | |
7537 | ||
7538 | unix_notinflight(fp->user, fp->fp[i]); | |
7539 | left = fp->count - 1 - i; | |
7540 | if (left) { | |
7541 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7542 | left * sizeof(struct file *)); | |
7543 | } | |
7544 | fp->count--; | |
7545 | if (!fp->count) { | |
7546 | kfree_skb(skb); | |
7547 | skb = NULL; | |
7548 | } else { | |
7549 | __skb_queue_tail(&list, skb); | |
7550 | } | |
7551 | fput(file); | |
7552 | file = NULL; | |
7553 | break; | |
7554 | } | |
7555 | ||
7556 | if (!file) | |
7557 | break; | |
7558 | ||
7559 | __skb_queue_tail(&list, skb); | |
7560 | ||
7561 | skb = skb_dequeue(head); | |
7562 | } | |
7563 | ||
7564 | if (skb_peek(&list)) { | |
7565 | spin_lock_irq(&head->lock); | |
7566 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7567 | __skb_queue_tail(head, skb); | |
7568 | spin_unlock_irq(&head->lock); | |
7569 | } | |
7570 | #else | |
7571 | fput(file); | |
7572 | #endif | |
7573 | } | |
7574 | ||
b895c9a6 | 7575 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7576 | { |
b895c9a6 | 7577 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7578 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7579 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7580 | |
269bbe5f BM |
7581 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7582 | list_del(&prsrc->list); | |
b60c8dce PB |
7583 | |
7584 | if (prsrc->tag) { | |
7585 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
7586 | |
7587 | io_ring_submit_lock(ctx, lock_ring); | |
157d257f | 7588 | spin_lock_irq(&ctx->completion_lock); |
b60c8dce | 7589 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); |
2840f710 | 7590 | ctx->cq_extra++; |
b60c8dce | 7591 | io_commit_cqring(ctx); |
157d257f | 7592 | spin_unlock_irq(&ctx->completion_lock); |
b60c8dce PB |
7593 | io_cqring_ev_posted(ctx); |
7594 | io_ring_submit_unlock(ctx, lock_ring); | |
7595 | } | |
7596 | ||
40ae0ff7 | 7597 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7598 | kfree(prsrc); |
65e19f54 | 7599 | } |
05589553 | 7600 | |
28a9fe25 | 7601 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7602 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7603 | complete(&rsrc_data->done); | |
2faf852d | 7604 | } |
65e19f54 | 7605 | |
269bbe5f | 7606 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7607 | { |
7608 | struct io_ring_ctx *ctx; | |
7609 | struct llist_node *node; | |
7610 | ||
269bbe5f BM |
7611 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7612 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7613 | |
7614 | while (node) { | |
b895c9a6 | 7615 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7616 | struct llist_node *next = node->next; |
7617 | ||
b895c9a6 | 7618 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7619 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7620 | node = next; |
7621 | } | |
7622 | } | |
7623 | ||
00835dce | 7624 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7625 | { |
b895c9a6 | 7626 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7627 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7628 | bool first_add = false; |
e297822b | 7629 | |
2a63b2d9 | 7630 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7631 | node->done = true; |
e297822b | 7632 | |
d67d2263 | 7633 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7634 | node = list_first_entry(&ctx->rsrc_ref_list, |
7635 | struct io_rsrc_node, node); | |
e297822b | 7636 | /* recycle ref nodes in order */ |
b895c9a6 | 7637 | if (!node->done) |
e297822b | 7638 | break; |
b895c9a6 PB |
7639 | list_del(&node->node); |
7640 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7641 | } |
2a63b2d9 | 7642 | io_rsrc_ref_unlock(ctx); |
05589553 | 7643 | |
3e942498 PB |
7644 | if (first_add) |
7645 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7646 | } |
65e19f54 | 7647 | |
b895c9a6 | 7648 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7649 | { |
b895c9a6 | 7650 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7651 | |
05589553 XW |
7652 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7653 | if (!ref_node) | |
3e2224c5 | 7654 | return NULL; |
05f3fb3c | 7655 | |
00835dce | 7656 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7657 | 0, GFP_KERNEL)) { |
7658 | kfree(ref_node); | |
3e2224c5 | 7659 | return NULL; |
05589553 XW |
7660 | } |
7661 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7662 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7663 | ref_node->done = false; |
05589553 | 7664 | return ref_node; |
05589553 XW |
7665 | } |
7666 | ||
6b06314c | 7667 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 7668 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
7669 | { |
7670 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7671 | struct file *file; |
f3baed39 | 7672 | int fd, ret; |
846a4ef2 | 7673 | unsigned i; |
6b06314c | 7674 | |
05f3fb3c | 7675 | if (ctx->file_data) |
6b06314c JA |
7676 | return -EBUSY; |
7677 | if (!nr_args) | |
7678 | return -EINVAL; | |
7679 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7680 | return -EMFILE; | |
a7f0ed5a | 7681 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7682 | if (ret) |
7683 | return ret; | |
d878c816 PB |
7684 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
7685 | &ctx->file_data); | |
7686 | if (ret) | |
7687 | return ret; | |
6b06314c | 7688 | |
f3baed39 | 7689 | ret = -ENOMEM; |
aeca241b | 7690 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7691 | goto out_free; |
65e19f54 | 7692 | |
08a45173 | 7693 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 7694 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
7695 | ret = -EFAULT; |
7696 | goto out_fput; | |
7697 | } | |
08a45173 | 7698 | /* allow sparse sets */ |
792e3582 PB |
7699 | if (fd == -1) { |
7700 | ret = -EINVAL; | |
2d091d62 | 7701 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 7702 | goto out_fput; |
08a45173 | 7703 | continue; |
792e3582 | 7704 | } |
6b06314c | 7705 | |
05f3fb3c | 7706 | file = fget(fd); |
6b06314c | 7707 | ret = -EBADF; |
792e3582 | 7708 | if (unlikely(!file)) |
600cf3f8 | 7709 | goto out_fput; |
05f3fb3c | 7710 | |
6b06314c JA |
7711 | /* |
7712 | * Don't allow io_uring instances to be registered. If UNIX | |
7713 | * isn't enabled, then this causes a reference cycle and this | |
7714 | * instance can never get freed. If UNIX is enabled we'll | |
7715 | * handle it just fine, but there's still no point in allowing | |
7716 | * a ring fd as it doesn't support regular read/write anyway. | |
7717 | */ | |
05f3fb3c JA |
7718 | if (file->f_op == &io_uring_fops) { |
7719 | fput(file); | |
600cf3f8 | 7720 | goto out_fput; |
6b06314c | 7721 | } |
aeca241b | 7722 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7723 | } |
7724 | ||
6b06314c | 7725 | ret = io_sqe_files_scm(ctx); |
05589553 | 7726 | if (ret) { |
08480400 | 7727 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7728 | return ret; |
7729 | } | |
6b06314c | 7730 | |
a7f0ed5a | 7731 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7732 | return ret; |
600cf3f8 PB |
7733 | out_fput: |
7734 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7735 | file = io_file_from_index(ctx, i); | |
7736 | if (file) | |
7737 | fput(file); | |
7738 | } | |
aeca241b | 7739 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7740 | ctx->nr_user_files = 0; |
600cf3f8 | 7741 | out_free: |
44b31f2f | 7742 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7743 | ctx->file_data = NULL; |
6b06314c JA |
7744 | return ret; |
7745 | } | |
7746 | ||
c3a31e60 JA |
7747 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7748 | int index) | |
7749 | { | |
7750 | #if defined(CONFIG_UNIX) | |
7751 | struct sock *sock = ctx->ring_sock->sk; | |
7752 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7753 | struct sk_buff *skb; | |
7754 | ||
7755 | /* | |
7756 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7757 | * file set. If there's no room, fall back to allocating a new skb | |
7758 | * and filling it in. | |
7759 | */ | |
7760 | spin_lock_irq(&head->lock); | |
7761 | skb = skb_peek(head); | |
7762 | if (skb) { | |
7763 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7764 | ||
7765 | if (fpl->count < SCM_MAX_FD) { | |
7766 | __skb_unlink(skb, head); | |
7767 | spin_unlock_irq(&head->lock); | |
7768 | fpl->fp[fpl->count] = get_file(file); | |
7769 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7770 | fpl->count++; | |
7771 | spin_lock_irq(&head->lock); | |
7772 | __skb_queue_head(head, skb); | |
7773 | } else { | |
7774 | skb = NULL; | |
7775 | } | |
7776 | } | |
7777 | spin_unlock_irq(&head->lock); | |
7778 | ||
7779 | if (skb) { | |
7780 | fput(file); | |
7781 | return 0; | |
7782 | } | |
7783 | ||
7784 | return __io_sqe_files_scm(ctx, 1, index); | |
7785 | #else | |
7786 | return 0; | |
7787 | #endif | |
7788 | } | |
7789 | ||
b60c8dce | 7790 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 7791 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 7792 | { |
269bbe5f | 7793 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7794 | |
269bbe5f BM |
7795 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7796 | if (!prsrc) | |
a5318d3c | 7797 | return -ENOMEM; |
05f3fb3c | 7798 | |
2d091d62 | 7799 | prsrc->tag = *io_get_tag_slot(data, idx); |
50238531 | 7800 | prsrc->rsrc = rsrc; |
e7c78371 | 7801 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7802 | return 0; |
05f3fb3c JA |
7803 | } |
7804 | ||
7805 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 7806 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
7807 | unsigned nr_args) |
7808 | { | |
c3bdad02 | 7809 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 7810 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 7811 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7812 | struct io_fixed_file *file_slot; |
7813 | struct file *file; | |
98f0b3b4 PB |
7814 | int fd, i, err = 0; |
7815 | unsigned int done; | |
05589553 | 7816 | bool needs_switch = false; |
c3a31e60 | 7817 | |
98f0b3b4 PB |
7818 | if (!ctx->file_data) |
7819 | return -ENXIO; | |
7820 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
7821 | return -EINVAL; |
7822 | ||
67973b93 | 7823 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
7824 | u64 tag = 0; |
7825 | ||
7826 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
7827 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
7828 | err = -EFAULT; |
7829 | break; | |
7830 | } | |
c3bdad02 PB |
7831 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
7832 | err = -EINVAL; | |
7833 | break; | |
7834 | } | |
4e0377a1 | 7835 | if (fd == IORING_REGISTER_FILES_SKIP) |
7836 | continue; | |
7837 | ||
67973b93 | 7838 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7839 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7840 | |
a04b0ac0 PB |
7841 | if (file_slot->file_ptr) { |
7842 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
7843 | err = io_queue_rsrc_removal(data, up->offset + done, |
7844 | ctx->rsrc_node, file); | |
a5318d3c HD |
7845 | if (err) |
7846 | break; | |
a04b0ac0 | 7847 | file_slot->file_ptr = 0; |
05589553 | 7848 | needs_switch = true; |
c3a31e60 JA |
7849 | } |
7850 | if (fd != -1) { | |
c3a31e60 JA |
7851 | file = fget(fd); |
7852 | if (!file) { | |
7853 | err = -EBADF; | |
7854 | break; | |
7855 | } | |
7856 | /* | |
7857 | * Don't allow io_uring instances to be registered. If | |
7858 | * UNIX isn't enabled, then this causes a reference | |
7859 | * cycle and this instance can never get freed. If UNIX | |
7860 | * is enabled we'll handle it just fine, but there's | |
7861 | * still no point in allowing a ring fd as it doesn't | |
7862 | * support regular read/write anyway. | |
7863 | */ | |
7864 | if (file->f_op == &io_uring_fops) { | |
7865 | fput(file); | |
7866 | err = -EBADF; | |
7867 | break; | |
7868 | } | |
2d091d62 | 7869 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 7870 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7871 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7872 | if (err) { |
a04b0ac0 | 7873 | file_slot->file_ptr = 0; |
f3bd9dae | 7874 | fput(file); |
c3a31e60 | 7875 | break; |
f3bd9dae | 7876 | } |
c3a31e60 | 7877 | } |
05f3fb3c JA |
7878 | } |
7879 | ||
a7f0ed5a PB |
7880 | if (needs_switch) |
7881 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7882 | return done ? done : err; |
7883 | } | |
05589553 | 7884 | |
5280f7e5 | 7885 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7886 | { |
7887 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7888 | ||
5280f7e5 PB |
7889 | req = io_put_req_find_next(req); |
7890 | return req ? &req->work : NULL; | |
7d723065 JA |
7891 | } |
7892 | ||
685fe7fe JA |
7893 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7894 | struct task_struct *task) | |
24369c2e | 7895 | { |
e941894e | 7896 | struct io_wq_hash *hash; |
24369c2e | 7897 | struct io_wq_data data; |
24369c2e | 7898 | unsigned int concurrency; |
24369c2e | 7899 | |
e941894e JA |
7900 | hash = ctx->hash_map; |
7901 | if (!hash) { | |
7902 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7903 | if (!hash) | |
7904 | return ERR_PTR(-ENOMEM); | |
7905 | refcount_set(&hash->refs, 1); | |
7906 | init_waitqueue_head(&hash->wait); | |
7907 | ctx->hash_map = hash; | |
24369c2e PB |
7908 | } |
7909 | ||
e941894e | 7910 | data.hash = hash; |
685fe7fe | 7911 | data.task = task; |
e9fd9396 | 7912 | data.free_work = io_free_work; |
f5fa38c5 | 7913 | data.do_work = io_wq_submit_work; |
24369c2e | 7914 | |
d25e3a3d JA |
7915 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7916 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7917 | |
5aa75ed5 | 7918 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7919 | } |
7920 | ||
5aa75ed5 JA |
7921 | static int io_uring_alloc_task_context(struct task_struct *task, |
7922 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7923 | { |
7924 | struct io_uring_task *tctx; | |
d8a6df10 | 7925 | int ret; |
0f212204 | 7926 | |
09899b19 | 7927 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
7928 | if (unlikely(!tctx)) |
7929 | return -ENOMEM; | |
7930 | ||
d8a6df10 JA |
7931 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7932 | if (unlikely(ret)) { | |
7933 | kfree(tctx); | |
7934 | return ret; | |
7935 | } | |
7936 | ||
685fe7fe | 7937 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7938 | if (IS_ERR(tctx->io_wq)) { |
7939 | ret = PTR_ERR(tctx->io_wq); | |
7940 | percpu_counter_destroy(&tctx->inflight); | |
7941 | kfree(tctx); | |
7942 | return ret; | |
7943 | } | |
7944 | ||
0f212204 JA |
7945 | xa_init(&tctx->xa); |
7946 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 7947 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7948 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7949 | task->io_uring = tctx; |
7cbf1722 JA |
7950 | spin_lock_init(&tctx->task_lock); |
7951 | INIT_WQ_LIST(&tctx->task_list); | |
7cbf1722 | 7952 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
7953 | return 0; |
7954 | } | |
7955 | ||
7956 | void __io_uring_free(struct task_struct *tsk) | |
7957 | { | |
7958 | struct io_uring_task *tctx = tsk->io_uring; | |
7959 | ||
7960 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 7961 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 7962 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 7963 | |
d8a6df10 | 7964 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7965 | kfree(tctx); |
7966 | tsk->io_uring = NULL; | |
7967 | } | |
7968 | ||
7e84e1c7 SG |
7969 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7970 | struct io_uring_params *p) | |
2b188cc1 JA |
7971 | { |
7972 | int ret; | |
7973 | ||
d25e3a3d JA |
7974 | /* Retain compatibility with failing for an invalid attach attempt */ |
7975 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7976 | IORING_SETUP_ATTACH_WQ) { | |
7977 | struct fd f; | |
7978 | ||
7979 | f = fdget(p->wq_fd); | |
7980 | if (!f.file) | |
7981 | return -ENXIO; | |
d25e3a3d | 7982 | fdput(f); |
f2a48dd0 PB |
7983 | if (f.file->f_op != &io_uring_fops) |
7984 | return -EINVAL; | |
d25e3a3d | 7985 | } |
6c271ce2 | 7986 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7987 | struct task_struct *tsk; |
534ca6d6 | 7988 | struct io_sq_data *sqd; |
26984fbf | 7989 | bool attached; |
534ca6d6 | 7990 | |
26984fbf | 7991 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7992 | if (IS_ERR(sqd)) { |
7993 | ret = PTR_ERR(sqd); | |
7994 | goto err; | |
7995 | } | |
69fb2131 | 7996 | |
7c30f36a | 7997 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7998 | ctx->sq_data = sqd; |
917257da JA |
7999 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8000 | if (!ctx->sq_thread_idle) | |
8001 | ctx->sq_thread_idle = HZ; | |
8002 | ||
78d7f6ba | 8003 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8004 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8005 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8006 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 8007 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
8008 | io_sq_thread_unpark(sqd); |
8009 | ||
de75a3d3 PB |
8010 | if (ret < 0) |
8011 | goto err; | |
8012 | if (attached) | |
5aa75ed5 | 8013 | return 0; |
aa06165d | 8014 | |
6c271ce2 | 8015 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8016 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8017 | |
917257da | 8018 | ret = -EINVAL; |
f2a48dd0 | 8019 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 8020 | goto err_sqpoll; |
37d1e2e3 | 8021 | sqd->sq_cpu = cpu; |
6c271ce2 | 8022 | } else { |
37d1e2e3 | 8023 | sqd->sq_cpu = -1; |
6c271ce2 | 8024 | } |
37d1e2e3 JA |
8025 | |
8026 | sqd->task_pid = current->pid; | |
5c2469e0 | 8027 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8028 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8029 | if (IS_ERR(tsk)) { | |
8030 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8031 | goto err_sqpoll; |
6c271ce2 | 8032 | } |
97a73a0f | 8033 | |
46fe18b1 | 8034 | sqd->thread = tsk; |
97a73a0f | 8035 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8036 | wake_up_new_task(tsk); |
0f212204 JA |
8037 | if (ret) |
8038 | goto err; | |
6c271ce2 JA |
8039 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8040 | /* Can't have SQ_AFF without SQPOLL */ | |
8041 | ret = -EINVAL; | |
8042 | goto err; | |
8043 | } | |
8044 | ||
2b188cc1 | 8045 | return 0; |
f2a48dd0 PB |
8046 | err_sqpoll: |
8047 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 8048 | err: |
37d1e2e3 | 8049 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8050 | return ret; |
8051 | } | |
8052 | ||
a087e2b5 BM |
8053 | static inline void __io_unaccount_mem(struct user_struct *user, |
8054 | unsigned long nr_pages) | |
2b188cc1 JA |
8055 | { |
8056 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8057 | } | |
8058 | ||
a087e2b5 BM |
8059 | static inline int __io_account_mem(struct user_struct *user, |
8060 | unsigned long nr_pages) | |
2b188cc1 JA |
8061 | { |
8062 | unsigned long page_limit, cur_pages, new_pages; | |
8063 | ||
8064 | /* Don't allow more pages than we can safely lock */ | |
8065 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8066 | ||
8067 | do { | |
8068 | cur_pages = atomic_long_read(&user->locked_vm); | |
8069 | new_pages = cur_pages + nr_pages; | |
8070 | if (new_pages > page_limit) | |
8071 | return -ENOMEM; | |
8072 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8073 | new_pages) != cur_pages); | |
8074 | ||
8075 | return 0; | |
8076 | } | |
8077 | ||
26bfa89e | 8078 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8079 | { |
62e398be | 8080 | if (ctx->user) |
a087e2b5 | 8081 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8082 | |
26bfa89e JA |
8083 | if (ctx->mm_account) |
8084 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8085 | } |
8086 | ||
26bfa89e | 8087 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8088 | { |
30975825 BM |
8089 | int ret; |
8090 | ||
62e398be | 8091 | if (ctx->user) { |
30975825 BM |
8092 | ret = __io_account_mem(ctx->user, nr_pages); |
8093 | if (ret) | |
8094 | return ret; | |
8095 | } | |
8096 | ||
26bfa89e JA |
8097 | if (ctx->mm_account) |
8098 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8099 | |
8100 | return 0; | |
8101 | } | |
8102 | ||
2b188cc1 JA |
8103 | static void io_mem_free(void *ptr) |
8104 | { | |
52e04ef4 MR |
8105 | struct page *page; |
8106 | ||
8107 | if (!ptr) | |
8108 | return; | |
2b188cc1 | 8109 | |
52e04ef4 | 8110 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8111 | if (put_page_testzero(page)) |
8112 | free_compound_page(page); | |
8113 | } | |
8114 | ||
8115 | static void *io_mem_alloc(size_t size) | |
8116 | { | |
8117 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8118 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8119 | |
8120 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8121 | } | |
8122 | ||
75b28aff HV |
8123 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8124 | size_t *sq_offset) | |
8125 | { | |
8126 | struct io_rings *rings; | |
8127 | size_t off, sq_array_size; | |
8128 | ||
8129 | off = struct_size(rings, cqes, cq_entries); | |
8130 | if (off == SIZE_MAX) | |
8131 | return SIZE_MAX; | |
8132 | ||
8133 | #ifdef CONFIG_SMP | |
8134 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8135 | if (off == 0) | |
8136 | return SIZE_MAX; | |
8137 | #endif | |
8138 | ||
b36200f5 DV |
8139 | if (sq_offset) |
8140 | *sq_offset = off; | |
8141 | ||
75b28aff HV |
8142 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8143 | if (sq_array_size == SIZE_MAX) | |
8144 | return SIZE_MAX; | |
8145 | ||
8146 | if (check_add_overflow(off, sq_array_size, &off)) | |
8147 | return SIZE_MAX; | |
8148 | ||
75b28aff HV |
8149 | return off; |
8150 | } | |
8151 | ||
41edf1a5 | 8152 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8153 | { |
41edf1a5 | 8154 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8155 | unsigned int i; |
8156 | ||
6224843d PB |
8157 | if (imu != ctx->dummy_ubuf) { |
8158 | for (i = 0; i < imu->nr_bvecs; i++) | |
8159 | unpin_user_page(imu->bvec[i].bv_page); | |
8160 | if (imu->acct_pages) | |
8161 | io_unaccount_mem(ctx, imu->acct_pages); | |
8162 | kvfree(imu); | |
8163 | } | |
41edf1a5 | 8164 | *slot = NULL; |
7f61a1e9 PB |
8165 | } |
8166 | ||
bd54b6fe | 8167 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8168 | { |
634d00df PB |
8169 | io_buffer_unmap(ctx, &prsrc->buf); |
8170 | prsrc->buf = NULL; | |
bd54b6fe | 8171 | } |
edafccee | 8172 | |
bd54b6fe BM |
8173 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8174 | { | |
8175 | unsigned int i; | |
edafccee | 8176 | |
7f61a1e9 PB |
8177 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8178 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8179 | kfree(ctx->user_bufs); |
bb6659cc | 8180 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 8181 | ctx->user_bufs = NULL; |
bd54b6fe | 8182 | ctx->buf_data = NULL; |
edafccee | 8183 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8184 | } |
8185 | ||
0a96bbe4 | 8186 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8187 | { |
bd54b6fe | 8188 | int ret; |
edafccee | 8189 | |
bd54b6fe | 8190 | if (!ctx->buf_data) |
edafccee JA |
8191 | return -ENXIO; |
8192 | ||
bd54b6fe BM |
8193 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
8194 | if (!ret) | |
8195 | __io_sqe_buffers_unregister(ctx); | |
8196 | return ret; | |
edafccee JA |
8197 | } |
8198 | ||
8199 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8200 | void __user *arg, unsigned index) | |
8201 | { | |
8202 | struct iovec __user *src; | |
8203 | ||
8204 | #ifdef CONFIG_COMPAT | |
8205 | if (ctx->compat) { | |
8206 | struct compat_iovec __user *ciovs; | |
8207 | struct compat_iovec ciov; | |
8208 | ||
8209 | ciovs = (struct compat_iovec __user *) arg; | |
8210 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8211 | return -EFAULT; | |
8212 | ||
d55e5f5b | 8213 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8214 | dst->iov_len = ciov.iov_len; |
8215 | return 0; | |
8216 | } | |
8217 | #endif | |
8218 | src = (struct iovec __user *) arg; | |
8219 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8220 | return -EFAULT; | |
8221 | return 0; | |
8222 | } | |
8223 | ||
de293938 JA |
8224 | /* |
8225 | * Not super efficient, but this is just a registration time. And we do cache | |
8226 | * the last compound head, so generally we'll only do a full search if we don't | |
8227 | * match that one. | |
8228 | * | |
8229 | * We check if the given compound head page has already been accounted, to | |
8230 | * avoid double accounting it. This allows us to account the full size of the | |
8231 | * page, not just the constituent pages of a huge page. | |
8232 | */ | |
8233 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8234 | int nr_pages, struct page *hpage) | |
8235 | { | |
8236 | int i, j; | |
8237 | ||
8238 | /* check current page array */ | |
8239 | for (i = 0; i < nr_pages; i++) { | |
8240 | if (!PageCompound(pages[i])) | |
8241 | continue; | |
8242 | if (compound_head(pages[i]) == hpage) | |
8243 | return true; | |
8244 | } | |
8245 | ||
8246 | /* check previously registered pages */ | |
8247 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8248 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8249 | |
8250 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8251 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8252 | continue; | |
8253 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8254 | return true; | |
8255 | } | |
8256 | } | |
8257 | ||
8258 | return false; | |
8259 | } | |
8260 | ||
8261 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8262 | int nr_pages, struct io_mapped_ubuf *imu, | |
8263 | struct page **last_hpage) | |
8264 | { | |
8265 | int i, ret; | |
8266 | ||
216e5835 | 8267 | imu->acct_pages = 0; |
de293938 JA |
8268 | for (i = 0; i < nr_pages; i++) { |
8269 | if (!PageCompound(pages[i])) { | |
8270 | imu->acct_pages++; | |
8271 | } else { | |
8272 | struct page *hpage; | |
8273 | ||
8274 | hpage = compound_head(pages[i]); | |
8275 | if (hpage == *last_hpage) | |
8276 | continue; | |
8277 | *last_hpage = hpage; | |
8278 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8279 | continue; | |
8280 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8281 | } | |
8282 | } | |
8283 | ||
8284 | if (!imu->acct_pages) | |
8285 | return 0; | |
8286 | ||
26bfa89e | 8287 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8288 | if (ret) |
8289 | imu->acct_pages = 0; | |
8290 | return ret; | |
8291 | } | |
8292 | ||
0a96bbe4 | 8293 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8294 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8295 | struct page **last_hpage) |
edafccee | 8296 | { |
41edf1a5 | 8297 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8298 | struct vm_area_struct **vmas = NULL; |
8299 | struct page **pages = NULL; | |
0a96bbe4 BM |
8300 | unsigned long off, start, end, ubuf; |
8301 | size_t size; | |
8302 | int ret, pret, nr_pages, i; | |
8303 | ||
6224843d PB |
8304 | if (!iov->iov_base) { |
8305 | *pimu = ctx->dummy_ubuf; | |
8306 | return 0; | |
8307 | } | |
8308 | ||
0a96bbe4 BM |
8309 | ubuf = (unsigned long) iov->iov_base; |
8310 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8311 | start = ubuf >> PAGE_SHIFT; | |
8312 | nr_pages = end - start; | |
8313 | ||
41edf1a5 | 8314 | *pimu = NULL; |
0a96bbe4 BM |
8315 | ret = -ENOMEM; |
8316 | ||
8317 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8318 | if (!pages) | |
8319 | goto done; | |
8320 | ||
8321 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8322 | GFP_KERNEL); | |
8323 | if (!vmas) | |
8324 | goto done; | |
edafccee | 8325 | |
41edf1a5 | 8326 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 8327 | if (!imu) |
0a96bbe4 BM |
8328 | goto done; |
8329 | ||
8330 | ret = 0; | |
8331 | mmap_read_lock(current->mm); | |
8332 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8333 | pages, vmas); | |
8334 | if (pret == nr_pages) { | |
8335 | /* don't support file backed memory */ | |
8336 | for (i = 0; i < nr_pages; i++) { | |
8337 | struct vm_area_struct *vma = vmas[i]; | |
8338 | ||
40dad765 PB |
8339 | if (vma_is_shmem(vma)) |
8340 | continue; | |
0a96bbe4 BM |
8341 | if (vma->vm_file && |
8342 | !is_file_hugepages(vma->vm_file)) { | |
8343 | ret = -EOPNOTSUPP; | |
8344 | break; | |
8345 | } | |
8346 | } | |
8347 | } else { | |
8348 | ret = pret < 0 ? pret : -EFAULT; | |
8349 | } | |
8350 | mmap_read_unlock(current->mm); | |
8351 | if (ret) { | |
8352 | /* | |
8353 | * if we did partial map, or found file backed vmas, | |
8354 | * release any pages we did get | |
8355 | */ | |
8356 | if (pret > 0) | |
8357 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8358 | goto done; |
8359 | } | |
8360 | ||
8361 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8362 | if (ret) { | |
8363 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8364 | goto done; |
8365 | } | |
8366 | ||
8367 | off = ubuf & ~PAGE_MASK; | |
8368 | size = iov->iov_len; | |
8369 | for (i = 0; i < nr_pages; i++) { | |
8370 | size_t vec_len; | |
8371 | ||
8372 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8373 | imu->bvec[i].bv_page = pages[i]; | |
8374 | imu->bvec[i].bv_len = vec_len; | |
8375 | imu->bvec[i].bv_offset = off; | |
8376 | off = 0; | |
8377 | size -= vec_len; | |
8378 | } | |
8379 | /* store original address for later verification */ | |
8380 | imu->ubuf = ubuf; | |
4751f53d | 8381 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8382 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8383 | *pimu = imu; |
0a96bbe4 BM |
8384 | ret = 0; |
8385 | done: | |
41edf1a5 PB |
8386 | if (ret) |
8387 | kvfree(imu); | |
0a96bbe4 BM |
8388 | kvfree(pages); |
8389 | kvfree(vmas); | |
8390 | return ret; | |
8391 | } | |
8392 | ||
2b358604 | 8393 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8394 | { |
87094465 PB |
8395 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8396 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8397 | } |
edafccee | 8398 | |
2b358604 BM |
8399 | static int io_buffer_validate(struct iovec *iov) |
8400 | { | |
50e96989 PB |
8401 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8402 | ||
2b358604 BM |
8403 | /* |
8404 | * Don't impose further limits on the size and buffer | |
8405 | * constraints here, we'll -EINVAL later when IO is | |
8406 | * submitted if they are wrong. | |
8407 | */ | |
6224843d PB |
8408 | if (!iov->iov_base) |
8409 | return iov->iov_len ? -EFAULT : 0; | |
8410 | if (!iov->iov_len) | |
2b358604 | 8411 | return -EFAULT; |
edafccee | 8412 | |
2b358604 BM |
8413 | /* arbitrary limit, but we need something */ |
8414 | if (iov->iov_len > SZ_1G) | |
8415 | return -EFAULT; | |
edafccee | 8416 | |
50e96989 PB |
8417 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8418 | return -EOVERFLOW; | |
8419 | ||
2b358604 BM |
8420 | return 0; |
8421 | } | |
edafccee | 8422 | |
2b358604 | 8423 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 8424 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 8425 | { |
bd54b6fe BM |
8426 | struct page *last_hpage = NULL; |
8427 | struct io_rsrc_data *data; | |
2b358604 BM |
8428 | int i, ret; |
8429 | struct iovec iov; | |
edafccee | 8430 | |
87094465 PB |
8431 | if (ctx->user_bufs) |
8432 | return -EBUSY; | |
489809e2 | 8433 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 8434 | return -EINVAL; |
bd54b6fe | 8435 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8436 | if (ret) |
8437 | return ret; | |
d878c816 PB |
8438 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
8439 | if (ret) | |
8440 | return ret; | |
bd54b6fe BM |
8441 | ret = io_buffers_map_alloc(ctx, nr_args); |
8442 | if (ret) { | |
bb6659cc | 8443 | io_rsrc_data_free(data); |
bd54b6fe BM |
8444 | return ret; |
8445 | } | |
edafccee | 8446 | |
87094465 | 8447 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
8448 | ret = io_copy_iov(ctx, &iov, arg, i); |
8449 | if (ret) | |
0a96bbe4 | 8450 | break; |
2b358604 BM |
8451 | ret = io_buffer_validate(&iov); |
8452 | if (ret) | |
0a96bbe4 | 8453 | break; |
2d091d62 | 8454 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
8455 | ret = -EINVAL; |
8456 | break; | |
8457 | } | |
edafccee | 8458 | |
41edf1a5 PB |
8459 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
8460 | &last_hpage); | |
0a96bbe4 BM |
8461 | if (ret) |
8462 | break; | |
edafccee | 8463 | } |
0a96bbe4 | 8464 | |
bd54b6fe | 8465 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8466 | |
bd54b6fe BM |
8467 | ctx->buf_data = data; |
8468 | if (ret) | |
8469 | __io_sqe_buffers_unregister(ctx); | |
8470 | else | |
8471 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
8472 | return ret; |
8473 | } | |
8474 | ||
634d00df PB |
8475 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
8476 | struct io_uring_rsrc_update2 *up, | |
8477 | unsigned int nr_args) | |
8478 | { | |
8479 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
8480 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
8481 | struct page *last_hpage = NULL; |
8482 | bool needs_switch = false; | |
8483 | __u32 done; | |
8484 | int i, err; | |
8485 | ||
8486 | if (!ctx->buf_data) | |
8487 | return -ENXIO; | |
8488 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
8489 | return -EINVAL; | |
8490 | ||
8491 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
8492 | struct io_mapped_ubuf *imu; |
8493 | int offset = up->offset + done; | |
634d00df PB |
8494 | u64 tag = 0; |
8495 | ||
8496 | err = io_copy_iov(ctx, &iov, iovs, done); | |
8497 | if (err) | |
8498 | break; | |
8499 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
8500 | err = -EFAULT; | |
8501 | break; | |
8502 | } | |
0b8c0e7c PB |
8503 | err = io_buffer_validate(&iov); |
8504 | if (err) | |
8505 | break; | |
cf3770e7 CIK |
8506 | if (!iov.iov_base && tag) { |
8507 | err = -EINVAL; | |
8508 | break; | |
8509 | } | |
0b8c0e7c PB |
8510 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
8511 | if (err) | |
8512 | break; | |
634d00df | 8513 | |
0b8c0e7c | 8514 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 8515 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
8516 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
8517 | ctx->rsrc_node, ctx->user_bufs[i]); | |
8518 | if (unlikely(err)) { | |
8519 | io_buffer_unmap(ctx, &imu); | |
634d00df | 8520 | break; |
0b8c0e7c | 8521 | } |
634d00df PB |
8522 | ctx->user_bufs[i] = NULL; |
8523 | needs_switch = true; | |
8524 | } | |
8525 | ||
0b8c0e7c | 8526 | ctx->user_bufs[i] = imu; |
2d091d62 | 8527 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
8528 | } |
8529 | ||
8530 | if (needs_switch) | |
8531 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
8532 | return done ? done : err; | |
8533 | } | |
8534 | ||
9b402849 JA |
8535 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8536 | { | |
8537 | __s32 __user *fds = arg; | |
8538 | int fd; | |
8539 | ||
8540 | if (ctx->cq_ev_fd) | |
8541 | return -EBUSY; | |
8542 | ||
8543 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8544 | return -EFAULT; | |
8545 | ||
8546 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8547 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8548 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
fe7e3257 | 8549 | |
9b402849 JA |
8550 | ctx->cq_ev_fd = NULL; |
8551 | return ret; | |
8552 | } | |
8553 | ||
8554 | return 0; | |
8555 | } | |
8556 | ||
8557 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8558 | { | |
8559 | if (ctx->cq_ev_fd) { | |
8560 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8561 | ctx->cq_ev_fd = NULL; | |
8562 | return 0; | |
8563 | } | |
8564 | ||
8565 | return -ENXIO; | |
8566 | } | |
8567 | ||
5a2e745d JA |
8568 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8569 | { | |
9e15c3a0 JA |
8570 | struct io_buffer *buf; |
8571 | unsigned long index; | |
8572 | ||
8573 | xa_for_each(&ctx->io_buffers, index, buf) | |
8574 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8575 | } |
8576 | ||
68e68ee6 | 8577 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8578 | { |
68e68ee6 | 8579 | struct io_kiocb *req, *nxt; |
1b4c351f | 8580 | |
68e68ee6 JA |
8581 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8582 | if (tsk && req->task != tsk) | |
8583 | continue; | |
1b4c351f JA |
8584 | list_del(&req->compl.list); |
8585 | kmem_cache_free(req_cachep, req); | |
8586 | } | |
8587 | } | |
8588 | ||
4010fec4 | 8589 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8590 | { |
bf019da7 | 8591 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8592 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8593 | |
9a4fdbd8 JA |
8594 | mutex_lock(&ctx->uring_lock); |
8595 | ||
8e5c66c4 | 8596 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8597 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8598 | submit_state->reqs); | |
8e5c66c4 PB |
8599 | submit_state->free_reqs = 0; |
8600 | } | |
9a4fdbd8 | 8601 | |
dac7a098 | 8602 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8603 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8604 | mutex_unlock(&ctx->uring_lock); |
8605 | } | |
8606 | ||
bd54b6fe | 8607 | static bool io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 8608 | { |
bd54b6fe BM |
8609 | if (!data) |
8610 | return false; | |
8611 | if (!atomic_dec_and_test(&data->refs)) | |
8612 | wait_for_completion(&data->done); | |
8613 | return true; | |
8614 | } | |
04fc6c80 | 8615 | |
2b188cc1 JA |
8616 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8617 | { | |
37d1e2e3 | 8618 | io_sq_thread_finish(ctx); |
2aede0e4 | 8619 | |
37d1e2e3 | 8620 | if (ctx->mm_account) { |
2aede0e4 JA |
8621 | mmdrop(ctx->mm_account); |
8622 | ctx->mm_account = NULL; | |
30975825 | 8623 | } |
def596e9 | 8624 | |
8bad28d8 | 8625 | mutex_lock(&ctx->uring_lock); |
bd54b6fe BM |
8626 | if (io_wait_rsrc_data(ctx->buf_data)) |
8627 | __io_sqe_buffers_unregister(ctx); | |
8628 | if (io_wait_rsrc_data(ctx->file_data)) | |
08480400 | 8629 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
8630 | if (ctx->rings) |
8631 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8632 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8633 | io_eventfd_unregister(ctx); |
5a2e745d | 8634 | io_destroy_buffers(ctx); |
07db298a PB |
8635 | if (ctx->sq_creds) |
8636 | put_cred(ctx->sq_creds); | |
def596e9 | 8637 | |
a7f0ed5a PB |
8638 | /* there are no registered resources left, nobody uses it */ |
8639 | if (ctx->rsrc_node) | |
8640 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8641 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8642 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8643 | flush_delayed_work(&ctx->rsrc_put_work); |
8644 | ||
8645 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8646 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 8647 | |
2b188cc1 | 8648 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8649 | if (ctx->ring_sock) { |
8650 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8651 | sock_release(ctx->ring_sock); |
355e8d26 | 8652 | } |
2b188cc1 JA |
8653 | #endif |
8654 | ||
75b28aff | 8655 | io_mem_free(ctx->rings); |
2b188cc1 | 8656 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8657 | |
8658 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8659 | free_uid(ctx->user); |
4010fec4 | 8660 | io_req_caches_free(ctx); |
e941894e JA |
8661 | if (ctx->hash_map) |
8662 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8663 | kfree(ctx->cancel_hash); |
6224843d | 8664 | kfree(ctx->dummy_ubuf); |
2b188cc1 JA |
8665 | kfree(ctx); |
8666 | } | |
8667 | ||
8668 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8669 | { | |
8670 | struct io_ring_ctx *ctx = file->private_data; | |
8671 | __poll_t mask = 0; | |
8672 | ||
311997b3 | 8673 | poll_wait(file, &ctx->poll_wait, wait); |
4f7067c3 SB |
8674 | /* |
8675 | * synchronizes with barrier from wq_has_sleeper call in | |
8676 | * io_commit_cqring | |
8677 | */ | |
2b188cc1 | 8678 | smp_rmb(); |
90554200 | 8679 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8680 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8681 | |
8682 | /* | |
8683 | * Don't flush cqring overflow list here, just do a simple check. | |
8684 | * Otherwise there could possible be ABBA deadlock: | |
8685 | * CPU0 CPU1 | |
8686 | * ---- ---- | |
8687 | * lock(&ctx->uring_lock); | |
8688 | * lock(&ep->mtx); | |
8689 | * lock(&ctx->uring_lock); | |
8690 | * lock(&ep->mtx); | |
8691 | * | |
8692 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8693 | * pushs them to do the flush. | |
8694 | */ | |
5ed7a37d | 8695 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
8696 | mask |= EPOLLIN | EPOLLRDNORM; |
8697 | ||
8698 | return mask; | |
8699 | } | |
8700 | ||
8701 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8702 | { | |
8703 | struct io_ring_ctx *ctx = file->private_data; | |
8704 | ||
8705 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8706 | } | |
8707 | ||
0bead8cd | 8708 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8709 | { |
4379bf8b | 8710 | const struct cred *creds; |
071698e1 | 8711 | |
61cf9370 | 8712 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8713 | if (creds) { |
8714 | put_cred(creds); | |
0bead8cd | 8715 | return 0; |
1e6fa521 | 8716 | } |
0bead8cd YD |
8717 | |
8718 | return -EINVAL; | |
8719 | } | |
8720 | ||
d56d938b PB |
8721 | struct io_tctx_exit { |
8722 | struct callback_head task_work; | |
8723 | struct completion completion; | |
baf186c4 | 8724 | struct io_ring_ctx *ctx; |
d56d938b PB |
8725 | }; |
8726 | ||
8727 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8728 | { | |
8729 | struct io_uring_task *tctx = current->io_uring; | |
8730 | struct io_tctx_exit *work; | |
8731 | ||
8732 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8733 | /* | |
8734 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8735 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8736 | */ | |
8737 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 8738 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
8739 | complete(&work->completion); |
8740 | } | |
8741 | ||
28090c13 PB |
8742 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8743 | { | |
8744 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8745 | ||
8746 | return req->ctx == data; | |
8747 | } | |
8748 | ||
85faa7b8 JA |
8749 | static void io_ring_exit_work(struct work_struct *work) |
8750 | { | |
d56d938b | 8751 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8752 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8753 | struct io_tctx_exit exit; |
8754 | struct io_tctx_node *node; | |
8755 | int ret; | |
85faa7b8 | 8756 | |
56952e91 JA |
8757 | /* |
8758 | * If we're doing polled IO and end up having requests being | |
8759 | * submitted async (out-of-line), then completions can come in while | |
8760 | * we're waiting for refs to drop. We need to reap these manually, | |
8761 | * as nobody else will be looking for them. | |
8762 | */ | |
b2edc0a7 | 8763 | do { |
3dd0c97a | 8764 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
8765 | if (ctx->sq_data) { |
8766 | struct io_sq_data *sqd = ctx->sq_data; | |
8767 | struct task_struct *tsk; | |
8768 | ||
8769 | io_sq_thread_park(sqd); | |
8770 | tsk = sqd->thread; | |
8771 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
8772 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
8773 | io_cancel_ctx_cb, ctx, true); | |
8774 | io_sq_thread_unpark(sqd); | |
8775 | } | |
b5bb3a24 PB |
8776 | |
8777 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8778 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8779 | |
7f00651a PB |
8780 | init_completion(&exit.completion); |
8781 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8782 | exit.ctx = ctx; | |
89b5066e PB |
8783 | /* |
8784 | * Some may use context even when all refs and requests have been put, | |
8785 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 8786 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
8787 | * this lock/unlock section also waits them to finish. |
8788 | */ | |
d56d938b PB |
8789 | mutex_lock(&ctx->uring_lock); |
8790 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8791 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8792 | ||
d56d938b PB |
8793 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8794 | ctx_node); | |
7f00651a PB |
8795 | /* don't spin on a single task if cancellation failed */ |
8796 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8797 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8798 | if (WARN_ON_ONCE(ret)) | |
8799 | continue; | |
8800 | wake_up_process(node->task); | |
8801 | ||
8802 | mutex_unlock(&ctx->uring_lock); | |
8803 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8804 | mutex_lock(&ctx->uring_lock); |
8805 | } | |
8806 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8807 | spin_lock_irq(&ctx->completion_lock); |
8808 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8809 | |
85faa7b8 JA |
8810 | io_ring_ctx_free(ctx); |
8811 | } | |
8812 | ||
80c4cbdb PB |
8813 | /* Returns true if we found and killed one or more timeouts */ |
8814 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
3dd0c97a | 8815 | bool cancel_all) |
80c4cbdb PB |
8816 | { |
8817 | struct io_kiocb *req, *tmp; | |
8818 | int canceled = 0; | |
8819 | ||
8820 | spin_lock_irq(&ctx->completion_lock); | |
8821 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
3dd0c97a | 8822 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
8823 | io_kill_timeout(req, -ECANCELED); |
8824 | canceled++; | |
8825 | } | |
8826 | } | |
51520426 PB |
8827 | if (canceled != 0) |
8828 | io_commit_cqring(ctx); | |
80c4cbdb | 8829 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8830 | if (canceled != 0) |
8831 | io_cqring_ev_posted(ctx); | |
8832 | return canceled != 0; | |
8833 | } | |
8834 | ||
2b188cc1 JA |
8835 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8836 | { | |
61cf9370 MWO |
8837 | unsigned long index; |
8838 | struct creds *creds; | |
8839 | ||
2b188cc1 JA |
8840 | mutex_lock(&ctx->uring_lock); |
8841 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8842 | if (ctx->rings) |
6c2450ae | 8843 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8844 | xa_for_each(&ctx->personalities, index, creds) |
8845 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8846 | mutex_unlock(&ctx->uring_lock); |
8847 | ||
3dd0c97a PB |
8848 | io_kill_timeouts(ctx, NULL, true); |
8849 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 8850 | |
15dff286 | 8851 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8852 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8853 | |
85faa7b8 | 8854 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8855 | /* |
8856 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8857 | * if we're exiting a ton of rings at the same time. It just adds | |
8858 | * noise and overhead, there's no discernable change in runtime | |
8859 | * over using system_wq. | |
8860 | */ | |
8861 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8862 | } |
8863 | ||
8864 | static int io_uring_release(struct inode *inode, struct file *file) | |
8865 | { | |
8866 | struct io_ring_ctx *ctx = file->private_data; | |
8867 | ||
8868 | file->private_data = NULL; | |
8869 | io_ring_ctx_wait_and_kill(ctx); | |
8870 | return 0; | |
8871 | } | |
8872 | ||
f6edbabb PB |
8873 | struct io_task_cancel { |
8874 | struct task_struct *task; | |
3dd0c97a | 8875 | bool all; |
f6edbabb | 8876 | }; |
f254ac04 | 8877 | |
f6edbabb | 8878 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8879 | { |
9a472ef7 | 8880 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8881 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8882 | bool ret; |
8883 | ||
3dd0c97a | 8884 | if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8885 | unsigned long flags; |
8886 | struct io_ring_ctx *ctx = req->ctx; | |
8887 | ||
8888 | /* protect against races with linked timeouts */ | |
8889 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
3dd0c97a | 8890 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8891 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8892 | } else { | |
3dd0c97a | 8893 | ret = io_match_task(req, cancel->task, cancel->all); |
9a472ef7 PB |
8894 | } |
8895 | return ret; | |
b711d4ea JA |
8896 | } |
8897 | ||
e1915f76 | 8898 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
3dd0c97a | 8899 | struct task_struct *task, bool cancel_all) |
b7ddce3c | 8900 | { |
e1915f76 | 8901 | struct io_defer_entry *de; |
b7ddce3c PB |
8902 | LIST_HEAD(list); |
8903 | ||
8904 | spin_lock_irq(&ctx->completion_lock); | |
8905 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
3dd0c97a | 8906 | if (io_match_task(de->req, task, cancel_all)) { |
b7ddce3c PB |
8907 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8908 | break; | |
8909 | } | |
8910 | } | |
8911 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8912 | if (list_empty(&list)) |
8913 | return false; | |
b7ddce3c PB |
8914 | |
8915 | while (!list_empty(&list)) { | |
8916 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8917 | list_del_init(&de->list); | |
f41db273 | 8918 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8919 | kfree(de); |
8920 | } | |
e1915f76 | 8921 | return true; |
b7ddce3c PB |
8922 | } |
8923 | ||
1b00764f PB |
8924 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
8925 | { | |
8926 | struct io_tctx_node *node; | |
8927 | enum io_wq_cancel cret; | |
8928 | bool ret = false; | |
8929 | ||
8930 | mutex_lock(&ctx->uring_lock); | |
8931 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8932 | struct io_uring_task *tctx = node->task->io_uring; | |
8933 | ||
8934 | /* | |
8935 | * io_wq will stay alive while we hold uring_lock, because it's | |
8936 | * killed after ctx nodes, which requires to take the lock. | |
8937 | */ | |
8938 | if (!tctx || !tctx->io_wq) | |
8939 | continue; | |
8940 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8941 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8942 | } | |
8943 | mutex_unlock(&ctx->uring_lock); | |
8944 | ||
8945 | return ret; | |
8946 | } | |
8947 | ||
9936c7c2 PB |
8948 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8949 | struct task_struct *task, | |
3dd0c97a | 8950 | bool cancel_all) |
9936c7c2 | 8951 | { |
3dd0c97a | 8952 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 8953 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8954 | |
8955 | while (1) { | |
8956 | enum io_wq_cancel cret; | |
8957 | bool ret = false; | |
8958 | ||
1b00764f PB |
8959 | if (!task) { |
8960 | ret |= io_uring_try_cancel_iowq(ctx); | |
8961 | } else if (tctx && tctx->io_wq) { | |
8962 | /* | |
8963 | * Cancels requests of all rings, not only @ctx, but | |
8964 | * it's fine as the task is in exit/exec. | |
8965 | */ | |
5aa75ed5 | 8966 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8967 | &cancel, true); |
8968 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8969 | } | |
8970 | ||
8971 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 8972 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 8973 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
9936c7c2 PB |
8974 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8975 | io_iopoll_try_reap_events(ctx); | |
8976 | ret = true; | |
8977 | } | |
8978 | } | |
8979 | ||
3dd0c97a PB |
8980 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
8981 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
8982 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
8983 | if (task) |
8984 | ret |= io_run_task_work(); | |
9936c7c2 PB |
8985 | if (!ret) |
8986 | break; | |
8987 | cond_resched(); | |
8988 | } | |
8989 | } | |
8990 | ||
eef51daa | 8991 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 8992 | { |
236434c3 | 8993 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8994 | struct io_tctx_node *node; |
a528b04e | 8995 | int ret; |
236434c3 MWO |
8996 | |
8997 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8998 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8999 | if (unlikely(ret)) |
9000 | return ret; | |
236434c3 | 9001 | tctx = current->io_uring; |
0f212204 | 9002 | } |
cf27f3b1 PB |
9003 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
9004 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
9005 | if (!node) | |
9006 | return -ENOMEM; | |
9007 | node->ctx = ctx; | |
9008 | node->task = current; | |
13bf43f5 | 9009 | |
cf27f3b1 PB |
9010 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
9011 | node, GFP_KERNEL)); | |
9012 | if (ret) { | |
9013 | kfree(node); | |
9014 | return ret; | |
0f212204 | 9015 | } |
cf27f3b1 PB |
9016 | |
9017 | mutex_lock(&ctx->uring_lock); | |
9018 | list_add(&node->ctx_node, &ctx->tctx_list); | |
9019 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 9020 | } |
cf27f3b1 | 9021 | tctx->last = ctx; |
0f212204 JA |
9022 | return 0; |
9023 | } | |
9024 | ||
cf27f3b1 PB |
9025 | /* |
9026 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9027 | */ | |
eef51daa | 9028 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
9029 | { |
9030 | struct io_uring_task *tctx = current->io_uring; | |
9031 | ||
9032 | if (likely(tctx && tctx->last == ctx)) | |
9033 | return 0; | |
eef51daa | 9034 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
9035 | } |
9036 | ||
0f212204 JA |
9037 | /* |
9038 | * Remove this io_uring_file -> task mapping. | |
9039 | */ | |
eef51daa | 9040 | static void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
9041 | { |
9042 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9043 | struct io_tctx_node *node; |
2941267b | 9044 | |
eebd2e37 PB |
9045 | if (!tctx) |
9046 | return; | |
13bf43f5 PB |
9047 | node = xa_erase(&tctx->xa, index); |
9048 | if (!node) | |
2941267b | 9049 | return; |
0f212204 | 9050 | |
13bf43f5 PB |
9051 | WARN_ON_ONCE(current != node->task); |
9052 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
9053 | ||
9054 | mutex_lock(&node->ctx->uring_lock); | |
9055 | list_del(&node->ctx_node); | |
9056 | mutex_unlock(&node->ctx->uring_lock); | |
9057 | ||
baf186c4 | 9058 | if (tctx->last == node->ctx) |
0f212204 | 9059 | tctx->last = NULL; |
13bf43f5 | 9060 | kfree(node); |
0f212204 JA |
9061 | } |
9062 | ||
8452d4a6 | 9063 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 9064 | { |
ba5ef6dc | 9065 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 9066 | struct io_tctx_node *node; |
de7f1d9e PB |
9067 | unsigned long index; |
9068 | ||
13bf43f5 | 9069 | xa_for_each(&tctx->xa, index, node) |
eef51daa | 9070 | io_uring_del_tctx_node(index); |
b16ef427 ME |
9071 | if (wq) { |
9072 | /* | |
9073 | * Must be after io_uring_del_task_file() (removes nodes under | |
9074 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). | |
9075 | */ | |
9076 | tctx->io_wq = NULL; | |
ba5ef6dc | 9077 | io_wq_put_and_exit(wq); |
b16ef427 | 9078 | } |
de7f1d9e PB |
9079 | } |
9080 | ||
3f48cf18 | 9081 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 9082 | { |
3f48cf18 PB |
9083 | if (tracked) |
9084 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
9085 | return percpu_counter_sum(&tctx->inflight); |
9086 | } | |
9087 | ||
09899b19 PB |
9088 | static void io_uring_drop_tctx_refs(struct task_struct *task) |
9089 | { | |
9090 | struct io_uring_task *tctx = task->io_uring; | |
9091 | unsigned int refs = tctx->cached_refs; | |
9092 | ||
9093 | tctx->cached_refs = 0; | |
9094 | percpu_counter_sub(&tctx->inflight, refs); | |
9095 | put_task_struct_many(task, refs); | |
9096 | } | |
9097 | ||
78cc687b PB |
9098 | /* |
9099 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
9100 | * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation. | |
9101 | */ | |
9102 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) | |
0e9ddb39 | 9103 | { |
521d6a73 | 9104 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9105 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9106 | s64 inflight; |
9107 | DEFINE_WAIT(wait); | |
fdaf083c | 9108 | |
78cc687b PB |
9109 | WARN_ON_ONCE(sqd && sqd->thread != current); |
9110 | ||
6d042ffb PO |
9111 | if (!current->io_uring) |
9112 | return; | |
17a91051 PB |
9113 | if (tctx->io_wq) |
9114 | io_wq_exit_start(tctx->io_wq); | |
9115 | ||
09899b19 | 9116 | io_uring_drop_tctx_refs(current); |
0e9ddb39 PB |
9117 | atomic_inc(&tctx->in_idle); |
9118 | do { | |
9119 | /* read completions before cancelations */ | |
78cc687b | 9120 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
9121 | if (!inflight) |
9122 | break; | |
fdaf083c | 9123 | |
78cc687b PB |
9124 | if (!sqd) { |
9125 | struct io_tctx_node *node; | |
9126 | unsigned long index; | |
0f212204 | 9127 | |
78cc687b PB |
9128 | xa_for_each(&tctx->xa, index, node) { |
9129 | /* sqpoll task will cancel all its requests */ | |
9130 | if (node->ctx->sq_data) | |
9131 | continue; | |
9132 | io_uring_try_cancel_requests(node->ctx, current, | |
9133 | cancel_all); | |
9134 | } | |
9135 | } else { | |
9136 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
9137 | io_uring_try_cancel_requests(ctx, current, | |
9138 | cancel_all); | |
9139 | } | |
17a91051 | 9140 | |
0f212204 | 9141 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
0f212204 | 9142 | /* |
a1bb3cd5 PB |
9143 | * If we've seen completions, retry without waiting. This |
9144 | * avoids a race where a completion comes in before we did | |
9145 | * prepare_to_wait(). | |
0f212204 | 9146 | */ |
3dd0c97a | 9147 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 9148 | schedule(); |
f57555ed | 9149 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9150 | } while (1); |
fdaf083c | 9151 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9152 | |
8452d4a6 | 9153 | io_uring_clean_tctx(tctx); |
3dd0c97a | 9154 | if (cancel_all) { |
3f48cf18 PB |
9155 | /* for exec all current's requests should be gone, kill tctx */ |
9156 | __io_uring_free(current); | |
9157 | } | |
44e728b8 PB |
9158 | } |
9159 | ||
78cc687b PB |
9160 | void __io_uring_cancel(struct files_struct *files) |
9161 | { | |
9162 | io_uring_cancel_generic(!files, NULL); | |
9163 | } | |
9164 | ||
6c5c240e RP |
9165 | static void *io_uring_validate_mmap_request(struct file *file, |
9166 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9167 | { |
2b188cc1 | 9168 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9169 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9170 | struct page *page; |
9171 | void *ptr; | |
9172 | ||
9173 | switch (offset) { | |
9174 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9175 | case IORING_OFF_CQ_RING: |
9176 | ptr = ctx->rings; | |
2b188cc1 JA |
9177 | break; |
9178 | case IORING_OFF_SQES: | |
9179 | ptr = ctx->sq_sqes; | |
9180 | break; | |
2b188cc1 | 9181 | default: |
6c5c240e | 9182 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9183 | } |
9184 | ||
9185 | page = virt_to_head_page(ptr); | |
a50b854e | 9186 | if (sz > page_size(page)) |
6c5c240e RP |
9187 | return ERR_PTR(-EINVAL); |
9188 | ||
9189 | return ptr; | |
9190 | } | |
9191 | ||
9192 | #ifdef CONFIG_MMU | |
9193 | ||
9194 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9195 | { | |
9196 | size_t sz = vma->vm_end - vma->vm_start; | |
9197 | unsigned long pfn; | |
9198 | void *ptr; | |
9199 | ||
9200 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9201 | if (IS_ERR(ptr)) | |
9202 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9203 | |
9204 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9205 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9206 | } | |
9207 | ||
6c5c240e RP |
9208 | #else /* !CONFIG_MMU */ |
9209 | ||
9210 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9211 | { | |
9212 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9213 | } | |
9214 | ||
9215 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9216 | { | |
9217 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9218 | } | |
9219 | ||
9220 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9221 | unsigned long addr, unsigned long len, | |
9222 | unsigned long pgoff, unsigned long flags) | |
9223 | { | |
9224 | void *ptr; | |
9225 | ||
9226 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9227 | if (IS_ERR(ptr)) | |
9228 | return PTR_ERR(ptr); | |
9229 | ||
9230 | return (unsigned long) ptr; | |
9231 | } | |
9232 | ||
9233 | #endif /* !CONFIG_MMU */ | |
9234 | ||
d9d05217 | 9235 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9236 | { |
9237 | DEFINE_WAIT(wait); | |
9238 | ||
9239 | do { | |
9240 | if (!io_sqring_full(ctx)) | |
9241 | break; | |
90554200 JA |
9242 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9243 | ||
9244 | if (!io_sqring_full(ctx)) | |
9245 | break; | |
90554200 JA |
9246 | schedule(); |
9247 | } while (!signal_pending(current)); | |
9248 | ||
9249 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9250 | return 0; |
90554200 JA |
9251 | } |
9252 | ||
c73ebb68 HX |
9253 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9254 | struct __kernel_timespec __user **ts, | |
9255 | const sigset_t __user **sig) | |
9256 | { | |
9257 | struct io_uring_getevents_arg arg; | |
9258 | ||
9259 | /* | |
9260 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9261 | * is just a pointer to the sigset_t. | |
9262 | */ | |
9263 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9264 | *sig = (const sigset_t __user *) argp; | |
9265 | *ts = NULL; | |
9266 | return 0; | |
9267 | } | |
9268 | ||
9269 | /* | |
9270 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9271 | * timespec and sigset_t pointers if good. | |
9272 | */ | |
9273 | if (*argsz != sizeof(arg)) | |
9274 | return -EINVAL; | |
9275 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9276 | return -EFAULT; | |
9277 | *sig = u64_to_user_ptr(arg.sigmask); | |
9278 | *argsz = arg.sigmask_sz; | |
9279 | *ts = u64_to_user_ptr(arg.ts); | |
9280 | return 0; | |
9281 | } | |
9282 | ||
2b188cc1 | 9283 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9284 | u32, min_complete, u32, flags, const void __user *, argp, |
9285 | size_t, argsz) | |
2b188cc1 JA |
9286 | { |
9287 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9288 | int submitted = 0; |
9289 | struct fd f; | |
33f993da | 9290 | long ret; |
2b188cc1 | 9291 | |
4c6e277c | 9292 | io_run_task_work(); |
b41e9852 | 9293 | |
33f993da PB |
9294 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9295 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9296 | return -EINVAL; |
9297 | ||
9298 | f = fdget(fd); | |
33f993da | 9299 | if (unlikely(!f.file)) |
2b188cc1 JA |
9300 | return -EBADF; |
9301 | ||
9302 | ret = -EOPNOTSUPP; | |
33f993da | 9303 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9304 | goto out_fput; |
9305 | ||
9306 | ret = -ENXIO; | |
9307 | ctx = f.file->private_data; | |
33f993da | 9308 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9309 | goto out_fput; |
9310 | ||
7e84e1c7 | 9311 | ret = -EBADFD; |
33f993da | 9312 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9313 | goto out; |
9314 | ||
6c271ce2 JA |
9315 | /* |
9316 | * For SQ polling, the thread will do all submissions and completions. | |
9317 | * Just return the requested submit count, and wake the thread if | |
9318 | * we were asked to. | |
9319 | */ | |
b2a9eada | 9320 | ret = 0; |
6c271ce2 | 9321 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9322 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9323 | |
d9d05217 | 9324 | ret = -EOWNERDEAD; |
fe7e3257 | 9325 | if (unlikely(ctx->sq_data->thread == NULL)) |
04147488 | 9326 | goto out; |
6c271ce2 | 9327 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9328 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9329 | if (flags & IORING_ENTER_SQ_WAIT) { |
9330 | ret = io_sqpoll_wait_sq(ctx); | |
9331 | if (ret) | |
9332 | goto out; | |
9333 | } | |
6c271ce2 | 9334 | submitted = to_submit; |
b2a9eada | 9335 | } else if (to_submit) { |
eef51daa | 9336 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
9337 | if (unlikely(ret)) |
9338 | goto out; | |
2b188cc1 | 9339 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9340 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9341 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9342 | |
9343 | if (submitted != to_submit) | |
9344 | goto out; | |
2b188cc1 JA |
9345 | } |
9346 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9347 | const sigset_t __user *sig; |
9348 | struct __kernel_timespec __user *ts; | |
9349 | ||
9350 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9351 | if (unlikely(ret)) | |
9352 | goto out; | |
9353 | ||
2b188cc1 JA |
9354 | min_complete = min(min_complete, ctx->cq_entries); |
9355 | ||
32b2244a XW |
9356 | /* |
9357 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9358 | * space applications don't need to do io completion events | |
9359 | * polling again, they can rely on io_sq_thread to do polling | |
9360 | * work, which can reduce cpu usage and uring_lock contention. | |
9361 | */ | |
9362 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9363 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9364 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9365 | } else { |
c73ebb68 | 9366 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9367 | } |
2b188cc1 JA |
9368 | } |
9369 | ||
7c504e65 | 9370 | out: |
6805b32e | 9371 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9372 | out_fput: |
9373 | fdput(f); | |
9374 | return submitted ? submitted : ret; | |
9375 | } | |
9376 | ||
bebdb65e | 9377 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9378 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9379 | const struct cred *cred) | |
87ce955b | 9380 | { |
87ce955b JA |
9381 | struct user_namespace *uns = seq_user_ns(m); |
9382 | struct group_info *gi; | |
9383 | kernel_cap_t cap; | |
9384 | unsigned __capi; | |
9385 | int g; | |
9386 | ||
9387 | seq_printf(m, "%5d\n", id); | |
9388 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9389 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9390 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9391 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9392 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9393 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9394 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9395 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9396 | seq_puts(m, "\n\tGroups:\t"); | |
9397 | gi = cred->group_info; | |
9398 | for (g = 0; g < gi->ngroups; g++) { | |
9399 | seq_put_decimal_ull(m, g ? " " : "", | |
9400 | from_kgid_munged(uns, gi->gid[g])); | |
9401 | } | |
9402 | seq_puts(m, "\n\tCapEff:\t"); | |
9403 | cap = cred->cap_effective; | |
9404 | CAP_FOR_EACH_U32(__capi) | |
9405 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9406 | seq_putc(m, '\n'); | |
9407 | return 0; | |
9408 | } | |
9409 | ||
9410 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9411 | { | |
dbbe9c64 | 9412 | struct io_sq_data *sq = NULL; |
fad8e0de | 9413 | bool has_lock; |
87ce955b JA |
9414 | int i; |
9415 | ||
fad8e0de JA |
9416 | /* |
9417 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9418 | * since fdinfo case grabs it in the opposite direction of normal use | |
9419 | * cases. If we fail to get the lock, we just don't iterate any | |
9420 | * structures that could be going away outside the io_uring mutex. | |
9421 | */ | |
9422 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9423 | ||
5f3f26f9 | 9424 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9425 | sq = ctx->sq_data; |
5f3f26f9 JA |
9426 | if (!sq->thread) |
9427 | sq = NULL; | |
9428 | } | |
dbbe9c64 JQ |
9429 | |
9430 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9431 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9432 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9433 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9434 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9435 | |
87ce955b JA |
9436 | if (f) |
9437 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9438 | else | |
9439 | seq_printf(m, "%5u: <none>\n", i); | |
9440 | } | |
9441 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9442 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9443 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9444 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9445 | |
4751f53d | 9446 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9447 | } |
61cf9370 MWO |
9448 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9449 | unsigned long index; | |
9450 | const struct cred *cred; | |
9451 | ||
87ce955b | 9452 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9453 | xa_for_each(&ctx->personalities, index, cred) |
9454 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9455 | } |
d7718a9d JA |
9456 | seq_printf(m, "PollList:\n"); |
9457 | spin_lock_irq(&ctx->completion_lock); | |
9458 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9459 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9460 | struct io_kiocb *req; | |
9461 | ||
9462 | hlist_for_each_entry(req, list, hash_node) | |
9463 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9464 | req->task->task_works != NULL); | |
9465 | } | |
9466 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9467 | if (has_lock) |
9468 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9469 | } |
9470 | ||
9471 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9472 | { | |
9473 | struct io_ring_ctx *ctx = f->private_data; | |
9474 | ||
9475 | if (percpu_ref_tryget(&ctx->refs)) { | |
9476 | __io_uring_show_fdinfo(ctx, m); | |
9477 | percpu_ref_put(&ctx->refs); | |
9478 | } | |
9479 | } | |
bebdb65e | 9480 | #endif |
87ce955b | 9481 | |
2b188cc1 JA |
9482 | static const struct file_operations io_uring_fops = { |
9483 | .release = io_uring_release, | |
9484 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9485 | #ifndef CONFIG_MMU |
9486 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9487 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9488 | #endif | |
2b188cc1 JA |
9489 | .poll = io_uring_poll, |
9490 | .fasync = io_uring_fasync, | |
bebdb65e | 9491 | #ifdef CONFIG_PROC_FS |
87ce955b | 9492 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9493 | #endif |
2b188cc1 JA |
9494 | }; |
9495 | ||
9496 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9497 | struct io_uring_params *p) | |
9498 | { | |
75b28aff HV |
9499 | struct io_rings *rings; |
9500 | size_t size, sq_array_offset; | |
2b188cc1 | 9501 | |
bd740481 JA |
9502 | /* make sure these are sane, as we already accounted them */ |
9503 | ctx->sq_entries = p->sq_entries; | |
9504 | ctx->cq_entries = p->cq_entries; | |
9505 | ||
75b28aff HV |
9506 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9507 | if (size == SIZE_MAX) | |
9508 | return -EOVERFLOW; | |
9509 | ||
9510 | rings = io_mem_alloc(size); | |
9511 | if (!rings) | |
2b188cc1 JA |
9512 | return -ENOMEM; |
9513 | ||
75b28aff HV |
9514 | ctx->rings = rings; |
9515 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9516 | rings->sq_ring_mask = p->sq_entries - 1; | |
9517 | rings->cq_ring_mask = p->cq_entries - 1; | |
9518 | rings->sq_ring_entries = p->sq_entries; | |
9519 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
9520 | |
9521 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9522 | if (size == SIZE_MAX) { |
9523 | io_mem_free(ctx->rings); | |
9524 | ctx->rings = NULL; | |
2b188cc1 | 9525 | return -EOVERFLOW; |
eb065d30 | 9526 | } |
2b188cc1 JA |
9527 | |
9528 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9529 | if (!ctx->sq_sqes) { |
9530 | io_mem_free(ctx->rings); | |
9531 | ctx->rings = NULL; | |
2b188cc1 | 9532 | return -ENOMEM; |
eb065d30 | 9533 | } |
2b188cc1 | 9534 | |
2b188cc1 JA |
9535 | return 0; |
9536 | } | |
9537 | ||
9faadcc8 PB |
9538 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9539 | { | |
9540 | int ret, fd; | |
9541 | ||
9542 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9543 | if (fd < 0) | |
9544 | return fd; | |
9545 | ||
eef51daa | 9546 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
9547 | if (ret) { |
9548 | put_unused_fd(fd); | |
9549 | return ret; | |
9550 | } | |
9551 | fd_install(fd, file); | |
9552 | return fd; | |
9553 | } | |
9554 | ||
2b188cc1 JA |
9555 | /* |
9556 | * Allocate an anonymous fd, this is what constitutes the application | |
9557 | * visible backing of an io_uring instance. The application mmaps this | |
9558 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9559 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9560 | */ | |
9faadcc8 | 9561 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9562 | { |
9563 | struct file *file; | |
9faadcc8 | 9564 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9565 | int ret; |
9566 | ||
2b188cc1 JA |
9567 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9568 | &ctx->ring_sock); | |
9569 | if (ret) | |
9faadcc8 | 9570 | return ERR_PTR(ret); |
2b188cc1 JA |
9571 | #endif |
9572 | ||
2b188cc1 JA |
9573 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9574 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9575 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9576 | if (IS_ERR(file)) { |
9577 | sock_release(ctx->ring_sock); | |
9578 | ctx->ring_sock = NULL; | |
9579 | } else { | |
9580 | ctx->ring_sock->file = file; | |
0f212204 | 9581 | } |
2b188cc1 | 9582 | #endif |
9faadcc8 | 9583 | return file; |
2b188cc1 JA |
9584 | } |
9585 | ||
7f13657d XW |
9586 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9587 | struct io_uring_params __user *params) | |
2b188cc1 | 9588 | { |
2b188cc1 | 9589 | struct io_ring_ctx *ctx; |
9faadcc8 | 9590 | struct file *file; |
2b188cc1 JA |
9591 | int ret; |
9592 | ||
8110c1a6 | 9593 | if (!entries) |
2b188cc1 | 9594 | return -EINVAL; |
8110c1a6 JA |
9595 | if (entries > IORING_MAX_ENTRIES) { |
9596 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9597 | return -EINVAL; | |
9598 | entries = IORING_MAX_ENTRIES; | |
9599 | } | |
2b188cc1 JA |
9600 | |
9601 | /* | |
9602 | * Use twice as many entries for the CQ ring. It's possible for the | |
9603 | * application to drive a higher depth than the size of the SQ ring, | |
9604 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9605 | * some flexibility in overcommitting a bit. If the application has |
9606 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9607 | * of CQ ring entries manually. | |
2b188cc1 JA |
9608 | */ |
9609 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9610 | if (p->flags & IORING_SETUP_CQSIZE) { |
9611 | /* | |
9612 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9613 | * to a power-of-two, if it isn't already. We do NOT impose | |
9614 | * any cq vs sq ring sizing. | |
9615 | */ | |
eb2667b3 | 9616 | if (!p->cq_entries) |
33a107f0 | 9617 | return -EINVAL; |
8110c1a6 JA |
9618 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9619 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9620 | return -EINVAL; | |
9621 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9622 | } | |
eb2667b3 JQ |
9623 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9624 | if (p->cq_entries < p->sq_entries) | |
9625 | return -EINVAL; | |
33a107f0 JA |
9626 | } else { |
9627 | p->cq_entries = 2 * p->sq_entries; | |
9628 | } | |
2b188cc1 | 9629 | |
2b188cc1 | 9630 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9631 | if (!ctx) |
2b188cc1 | 9632 | return -ENOMEM; |
2b188cc1 | 9633 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9634 | if (!capable(CAP_IPC_LOCK)) |
9635 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9636 | |
9637 | /* | |
9638 | * This is just grabbed for accounting purposes. When a process exits, | |
9639 | * the mm is exited and dropped before the files, hence we need to hang | |
9640 | * on to this mm purely for the purposes of being able to unaccount | |
9641 | * memory (locked/pinned vm). It's not used for anything else. | |
9642 | */ | |
6b7898eb | 9643 | mmgrab(current->mm); |
2aede0e4 | 9644 | ctx->mm_account = current->mm; |
6b7898eb | 9645 | |
2b188cc1 JA |
9646 | ret = io_allocate_scq_urings(ctx, p); |
9647 | if (ret) | |
9648 | goto err; | |
9649 | ||
7e84e1c7 | 9650 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9651 | if (ret) |
9652 | goto err; | |
eae071c9 | 9653 | /* always set a rsrc node */ |
47b228ce PB |
9654 | ret = io_rsrc_node_switch_start(ctx); |
9655 | if (ret) | |
9656 | goto err; | |
eae071c9 | 9657 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 9658 | |
2b188cc1 | 9659 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9660 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9661 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9662 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9663 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9664 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9665 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9666 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9667 | |
9668 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9669 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9670 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9671 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9672 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9673 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9674 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9675 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9676 | |
7f13657d XW |
9677 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9678 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9679 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9680 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e PB |
9681 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
9682 | IORING_FEAT_RSRC_TAGS; | |
7f13657d XW |
9683 | |
9684 | if (copy_to_user(params, p, sizeof(*p))) { | |
9685 | ret = -EFAULT; | |
9686 | goto err; | |
9687 | } | |
d1719f70 | 9688 | |
9faadcc8 PB |
9689 | file = io_uring_get_file(ctx); |
9690 | if (IS_ERR(file)) { | |
9691 | ret = PTR_ERR(file); | |
9692 | goto err; | |
9693 | } | |
9694 | ||
044c1ab3 JA |
9695 | /* |
9696 | * Install ring fd as the very last thing, so we don't risk someone | |
9697 | * having closed it before we finish setup | |
9698 | */ | |
9faadcc8 PB |
9699 | ret = io_uring_install_fd(ctx, file); |
9700 | if (ret < 0) { | |
9701 | /* fput will clean it up */ | |
9702 | fput(file); | |
9703 | return ret; | |
9704 | } | |
044c1ab3 | 9705 | |
c826bd7a | 9706 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9707 | return ret; |
9708 | err: | |
9709 | io_ring_ctx_wait_and_kill(ctx); | |
9710 | return ret; | |
9711 | } | |
9712 | ||
9713 | /* | |
9714 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9715 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9716 | * params structure passed in. | |
9717 | */ | |
9718 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9719 | { | |
9720 | struct io_uring_params p; | |
2b188cc1 JA |
9721 | int i; |
9722 | ||
9723 | if (copy_from_user(&p, params, sizeof(p))) | |
9724 | return -EFAULT; | |
9725 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9726 | if (p.resv[i]) | |
9727 | return -EINVAL; | |
9728 | } | |
9729 | ||
6c271ce2 | 9730 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9731 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9732 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9733 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9734 | return -EINVAL; |
9735 | ||
7f13657d | 9736 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9737 | } |
9738 | ||
9739 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9740 | struct io_uring_params __user *, params) | |
9741 | { | |
9742 | return io_uring_setup(entries, params); | |
9743 | } | |
9744 | ||
66f4af93 JA |
9745 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9746 | { | |
9747 | struct io_uring_probe *p; | |
9748 | size_t size; | |
9749 | int i, ret; | |
9750 | ||
9751 | size = struct_size(p, ops, nr_args); | |
9752 | if (size == SIZE_MAX) | |
9753 | return -EOVERFLOW; | |
9754 | p = kzalloc(size, GFP_KERNEL); | |
9755 | if (!p) | |
9756 | return -ENOMEM; | |
9757 | ||
9758 | ret = -EFAULT; | |
9759 | if (copy_from_user(p, arg, size)) | |
9760 | goto out; | |
9761 | ret = -EINVAL; | |
9762 | if (memchr_inv(p, 0, size)) | |
9763 | goto out; | |
9764 | ||
9765 | p->last_op = IORING_OP_LAST - 1; | |
9766 | if (nr_args > IORING_OP_LAST) | |
9767 | nr_args = IORING_OP_LAST; | |
9768 | ||
9769 | for (i = 0; i < nr_args; i++) { | |
9770 | p->ops[i].op = i; | |
9771 | if (!io_op_defs[i].not_supported) | |
9772 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9773 | } | |
9774 | p->ops_len = i; | |
9775 | ||
9776 | ret = 0; | |
9777 | if (copy_to_user(arg, p, size)) | |
9778 | ret = -EFAULT; | |
9779 | out: | |
9780 | kfree(p); | |
9781 | return ret; | |
9782 | } | |
9783 | ||
071698e1 JA |
9784 | static int io_register_personality(struct io_ring_ctx *ctx) |
9785 | { | |
4379bf8b | 9786 | const struct cred *creds; |
61cf9370 | 9787 | u32 id; |
1e6fa521 | 9788 | int ret; |
071698e1 | 9789 | |
4379bf8b | 9790 | creds = get_current_cred(); |
1e6fa521 | 9791 | |
61cf9370 MWO |
9792 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9793 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9794 | if (!ret) | |
9795 | return id; | |
9796 | put_cred(creds); | |
1e6fa521 | 9797 | return ret; |
071698e1 JA |
9798 | } |
9799 | ||
21b55dbc SG |
9800 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9801 | unsigned int nr_args) | |
9802 | { | |
9803 | struct io_uring_restriction *res; | |
9804 | size_t size; | |
9805 | int i, ret; | |
9806 | ||
7e84e1c7 SG |
9807 | /* Restrictions allowed only if rings started disabled */ |
9808 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9809 | return -EBADFD; | |
9810 | ||
21b55dbc | 9811 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9812 | if (ctx->restrictions.registered) |
21b55dbc SG |
9813 | return -EBUSY; |
9814 | ||
9815 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9816 | return -EINVAL; | |
9817 | ||
9818 | size = array_size(nr_args, sizeof(*res)); | |
9819 | if (size == SIZE_MAX) | |
9820 | return -EOVERFLOW; | |
9821 | ||
9822 | res = memdup_user(arg, size); | |
9823 | if (IS_ERR(res)) | |
9824 | return PTR_ERR(res); | |
9825 | ||
9826 | ret = 0; | |
9827 | ||
9828 | for (i = 0; i < nr_args; i++) { | |
9829 | switch (res[i].opcode) { | |
9830 | case IORING_RESTRICTION_REGISTER_OP: | |
9831 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9832 | ret = -EINVAL; | |
9833 | goto out; | |
9834 | } | |
9835 | ||
9836 | __set_bit(res[i].register_op, | |
9837 | ctx->restrictions.register_op); | |
9838 | break; | |
9839 | case IORING_RESTRICTION_SQE_OP: | |
9840 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9841 | ret = -EINVAL; | |
9842 | goto out; | |
9843 | } | |
9844 | ||
9845 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9846 | break; | |
9847 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9848 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9849 | break; | |
9850 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9851 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9852 | break; | |
9853 | default: | |
9854 | ret = -EINVAL; | |
9855 | goto out; | |
9856 | } | |
9857 | } | |
9858 | ||
9859 | out: | |
9860 | /* Reset all restrictions if an error happened */ | |
9861 | if (ret != 0) | |
9862 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9863 | else | |
7e84e1c7 | 9864 | ctx->restrictions.registered = true; |
21b55dbc SG |
9865 | |
9866 | kfree(res); | |
9867 | return ret; | |
9868 | } | |
9869 | ||
7e84e1c7 SG |
9870 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9871 | { | |
9872 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9873 | return -EBADFD; | |
9874 | ||
9875 | if (ctx->restrictions.registered) | |
9876 | ctx->restricted = 1; | |
9877 | ||
0298ef96 PB |
9878 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9879 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9880 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9881 | return 0; |
9882 | } | |
9883 | ||
fdecb662 | 9884 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 9885 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
9886 | unsigned nr_args) |
9887 | { | |
9888 | __u32 tmp; | |
9889 | int err; | |
9890 | ||
c3bdad02 PB |
9891 | if (up->resv) |
9892 | return -EINVAL; | |
98f0b3b4 PB |
9893 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
9894 | return -EOVERFLOW; | |
9895 | err = io_rsrc_node_switch_start(ctx); | |
9896 | if (err) | |
9897 | return err; | |
9898 | ||
fdecb662 PB |
9899 | switch (type) { |
9900 | case IORING_RSRC_FILE: | |
98f0b3b4 | 9901 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
9902 | case IORING_RSRC_BUFFER: |
9903 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
9904 | } |
9905 | return -EINVAL; | |
9906 | } | |
9907 | ||
c3bdad02 PB |
9908 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
9909 | unsigned nr_args) | |
98f0b3b4 | 9910 | { |
c3bdad02 | 9911 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
9912 | |
9913 | if (!nr_args) | |
9914 | return -EINVAL; | |
c3bdad02 PB |
9915 | memset(&up, 0, sizeof(up)); |
9916 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
9917 | return -EFAULT; | |
9918 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
9919 | } | |
9920 | ||
9921 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 9922 | unsigned size, unsigned type) |
c3bdad02 PB |
9923 | { |
9924 | struct io_uring_rsrc_update2 up; | |
9925 | ||
9926 | if (size != sizeof(up)) | |
9927 | return -EINVAL; | |
98f0b3b4 PB |
9928 | if (copy_from_user(&up, arg, sizeof(up))) |
9929 | return -EFAULT; | |
992da01a | 9930 | if (!up.nr || up.resv) |
98f0b3b4 | 9931 | return -EINVAL; |
992da01a | 9932 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
9933 | } |
9934 | ||
792e3582 | 9935 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 9936 | unsigned int size, unsigned int type) |
792e3582 PB |
9937 | { |
9938 | struct io_uring_rsrc_register rr; | |
9939 | ||
9940 | /* keep it extendible */ | |
9941 | if (size != sizeof(rr)) | |
9942 | return -EINVAL; | |
9943 | ||
9944 | memset(&rr, 0, sizeof(rr)); | |
9945 | if (copy_from_user(&rr, arg, size)) | |
9946 | return -EFAULT; | |
992da01a | 9947 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
9948 | return -EINVAL; |
9949 | ||
992da01a | 9950 | switch (type) { |
792e3582 PB |
9951 | case IORING_RSRC_FILE: |
9952 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
9953 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
9954 | case IORING_RSRC_BUFFER: |
9955 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
9956 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
9957 | } |
9958 | return -EINVAL; | |
9959 | } | |
9960 | ||
fe76421d JA |
9961 | static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg, |
9962 | unsigned len) | |
9963 | { | |
9964 | struct io_uring_task *tctx = current->io_uring; | |
9965 | cpumask_var_t new_mask; | |
9966 | int ret; | |
9967 | ||
9968 | if (!tctx || !tctx->io_wq) | |
9969 | return -EINVAL; | |
9970 | ||
9971 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
9972 | return -ENOMEM; | |
9973 | ||
9974 | cpumask_clear(new_mask); | |
9975 | if (len > cpumask_size()) | |
9976 | len = cpumask_size(); | |
9977 | ||
9978 | if (copy_from_user(new_mask, arg, len)) { | |
9979 | free_cpumask_var(new_mask); | |
9980 | return -EFAULT; | |
9981 | } | |
9982 | ||
9983 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
9984 | free_cpumask_var(new_mask); | |
9985 | return ret; | |
9986 | } | |
9987 | ||
9988 | static int io_unregister_iowq_aff(struct io_ring_ctx *ctx) | |
9989 | { | |
9990 | struct io_uring_task *tctx = current->io_uring; | |
9991 | ||
9992 | if (!tctx || !tctx->io_wq) | |
9993 | return -EINVAL; | |
9994 | ||
9995 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
9996 | } | |
9997 | ||
071698e1 JA |
9998 | static bool io_register_op_must_quiesce(int op) |
9999 | { | |
10000 | switch (op) { | |
bd54b6fe BM |
10001 | case IORING_REGISTER_BUFFERS: |
10002 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 10003 | case IORING_REGISTER_FILES: |
071698e1 JA |
10004 | case IORING_UNREGISTER_FILES: |
10005 | case IORING_REGISTER_FILES_UPDATE: | |
10006 | case IORING_REGISTER_PROBE: | |
10007 | case IORING_REGISTER_PERSONALITY: | |
10008 | case IORING_UNREGISTER_PERSONALITY: | |
992da01a PB |
10009 | case IORING_REGISTER_FILES2: |
10010 | case IORING_REGISTER_FILES_UPDATE2: | |
10011 | case IORING_REGISTER_BUFFERS2: | |
10012 | case IORING_REGISTER_BUFFERS_UPDATE: | |
fe76421d JA |
10013 | case IORING_REGISTER_IOWQ_AFF: |
10014 | case IORING_UNREGISTER_IOWQ_AFF: | |
071698e1 JA |
10015 | return false; |
10016 | default: | |
10017 | return true; | |
10018 | } | |
10019 | } | |
10020 | ||
edafccee JA |
10021 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10022 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10023 | __releases(ctx->uring_lock) |
10024 | __acquires(ctx->uring_lock) | |
edafccee JA |
10025 | { |
10026 | int ret; | |
10027 | ||
35fa71a0 JA |
10028 | /* |
10029 | * We're inside the ring mutex, if the ref is already dying, then | |
10030 | * someone else killed the ctx or is already going through | |
10031 | * io_uring_register(). | |
10032 | */ | |
10033 | if (percpu_ref_is_dying(&ctx->refs)) | |
10034 | return -ENXIO; | |
10035 | ||
75c4021a PB |
10036 | if (ctx->restricted) { |
10037 | if (opcode >= IORING_REGISTER_LAST) | |
10038 | return -EINVAL; | |
10039 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
10040 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
10041 | return -EACCES; | |
10042 | } | |
10043 | ||
071698e1 | 10044 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10045 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10046 | |
05f3fb3c JA |
10047 | /* |
10048 | * Drop uring mutex before waiting for references to exit. If | |
10049 | * another thread is currently inside io_uring_enter() it might | |
10050 | * need to grab the uring_lock to make progress. If we hold it | |
10051 | * here across the drain wait, then we can deadlock. It's safe | |
10052 | * to drop the mutex here, since no new references will come in | |
10053 | * after we've killed the percpu ref. | |
10054 | */ | |
10055 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10056 | do { |
10057 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10058 | if (!ret) | |
10059 | break; | |
ed6930c9 JA |
10060 | ret = io_run_task_work_sig(); |
10061 | if (ret < 0) | |
10062 | break; | |
af9c1a44 | 10063 | } while (1); |
05f3fb3c | 10064 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10065 | |
c150368b | 10066 | if (ret) { |
f70865db PB |
10067 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
10068 | return ret; | |
c150368b | 10069 | } |
05f3fb3c | 10070 | } |
edafccee JA |
10071 | |
10072 | switch (opcode) { | |
10073 | case IORING_REGISTER_BUFFERS: | |
634d00df | 10074 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
10075 | break; |
10076 | case IORING_UNREGISTER_BUFFERS: | |
10077 | ret = -EINVAL; | |
10078 | if (arg || nr_args) | |
10079 | break; | |
0a96bbe4 | 10080 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10081 | break; |
6b06314c | 10082 | case IORING_REGISTER_FILES: |
792e3582 | 10083 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
10084 | break; |
10085 | case IORING_UNREGISTER_FILES: | |
10086 | ret = -EINVAL; | |
10087 | if (arg || nr_args) | |
10088 | break; | |
10089 | ret = io_sqe_files_unregister(ctx); | |
10090 | break; | |
c3a31e60 | 10091 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 10092 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 10093 | break; |
9b402849 | 10094 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10095 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10096 | ret = -EINVAL; |
10097 | if (nr_args != 1) | |
10098 | break; | |
10099 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10100 | if (ret) |
10101 | break; | |
10102 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10103 | ctx->eventfd_async = 1; | |
10104 | else | |
10105 | ctx->eventfd_async = 0; | |
9b402849 JA |
10106 | break; |
10107 | case IORING_UNREGISTER_EVENTFD: | |
10108 | ret = -EINVAL; | |
10109 | if (arg || nr_args) | |
10110 | break; | |
10111 | ret = io_eventfd_unregister(ctx); | |
10112 | break; | |
66f4af93 JA |
10113 | case IORING_REGISTER_PROBE: |
10114 | ret = -EINVAL; | |
10115 | if (!arg || nr_args > 256) | |
10116 | break; | |
10117 | ret = io_probe(ctx, arg, nr_args); | |
10118 | break; | |
071698e1 JA |
10119 | case IORING_REGISTER_PERSONALITY: |
10120 | ret = -EINVAL; | |
10121 | if (arg || nr_args) | |
10122 | break; | |
10123 | ret = io_register_personality(ctx); | |
10124 | break; | |
10125 | case IORING_UNREGISTER_PERSONALITY: | |
10126 | ret = -EINVAL; | |
10127 | if (arg) | |
10128 | break; | |
10129 | ret = io_unregister_personality(ctx, nr_args); | |
10130 | break; | |
7e84e1c7 SG |
10131 | case IORING_REGISTER_ENABLE_RINGS: |
10132 | ret = -EINVAL; | |
10133 | if (arg || nr_args) | |
10134 | break; | |
10135 | ret = io_register_enable_rings(ctx); | |
10136 | break; | |
21b55dbc SG |
10137 | case IORING_REGISTER_RESTRICTIONS: |
10138 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10139 | break; | |
992da01a PB |
10140 | case IORING_REGISTER_FILES2: |
10141 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
10142 | break; | |
10143 | case IORING_REGISTER_FILES_UPDATE2: | |
10144 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10145 | IORING_RSRC_FILE); | |
10146 | break; | |
10147 | case IORING_REGISTER_BUFFERS2: | |
10148 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 10149 | break; |
992da01a PB |
10150 | case IORING_REGISTER_BUFFERS_UPDATE: |
10151 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
10152 | IORING_RSRC_BUFFER); | |
c3bdad02 | 10153 | break; |
fe76421d JA |
10154 | case IORING_REGISTER_IOWQ_AFF: |
10155 | ret = -EINVAL; | |
10156 | if (!arg || !nr_args) | |
10157 | break; | |
10158 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
10159 | break; | |
10160 | case IORING_UNREGISTER_IOWQ_AFF: | |
10161 | ret = -EINVAL; | |
10162 | if (arg || nr_args) | |
10163 | break; | |
10164 | ret = io_unregister_iowq_aff(ctx); | |
10165 | break; | |
edafccee JA |
10166 | default: |
10167 | ret = -EINVAL; | |
10168 | break; | |
10169 | } | |
10170 | ||
071698e1 | 10171 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10172 | /* bring the ctx back to life */ |
05f3fb3c | 10173 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10174 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10175 | } |
edafccee JA |
10176 | return ret; |
10177 | } | |
10178 | ||
10179 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10180 | void __user *, arg, unsigned int, nr_args) | |
10181 | { | |
10182 | struct io_ring_ctx *ctx; | |
10183 | long ret = -EBADF; | |
10184 | struct fd f; | |
10185 | ||
10186 | f = fdget(fd); | |
10187 | if (!f.file) | |
10188 | return -EBADF; | |
10189 | ||
10190 | ret = -EOPNOTSUPP; | |
10191 | if (f.file->f_op != &io_uring_fops) | |
10192 | goto out_fput; | |
10193 | ||
10194 | ctx = f.file->private_data; | |
10195 | ||
b6c23dd5 PB |
10196 | io_run_task_work(); |
10197 | ||
edafccee JA |
10198 | mutex_lock(&ctx->uring_lock); |
10199 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10200 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10201 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10202 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10203 | out_fput: |
10204 | fdput(f); | |
10205 | return ret; | |
10206 | } | |
10207 | ||
2b188cc1 JA |
10208 | static int __init io_uring_init(void) |
10209 | { | |
d7f62e82 SM |
10210 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10211 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10212 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10213 | } while (0) | |
10214 | ||
10215 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10216 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10217 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10218 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10219 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10220 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10221 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10222 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10223 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10224 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10225 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10226 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10227 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10228 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10229 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10230 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10231 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10232 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10233 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10234 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10235 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10236 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10237 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10238 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10239 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10240 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10241 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10242 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10243 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 10244 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 10245 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 10246 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10247 | |
b0d658ec PB |
10248 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
10249 | sizeof(struct io_uring_rsrc_update)); | |
10250 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
10251 | sizeof(struct io_uring_rsrc_update2)); | |
10252 | /* should fit into one byte */ | |
10253 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
10254 | ||
d3656344 | 10255 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10256 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
16340eab | 10257 | |
91f245d5 JA |
10258 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10259 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10260 | return 0; |
10261 | }; | |
10262 | __initcall(io_uring_init); |