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> | |
edce22e1 | 60 | #include <linux/blk-mq.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> |
adc8682e | 66 | #include <net/busy_poll.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
7d67af2c | 78 | #include <linux/splice.h> |
b41e9852 | 79 | #include <linux/task_work.h> |
bcf5a063 | 80 | #include <linux/pagemap.h> |
0f212204 | 81 | #include <linux/io_uring.h> |
ef98eb04 | 82 | #include <linux/tracehook.h> |
5bd2182d | 83 | #include <linux/audit.h> |
cdc1404a | 84 | #include <linux/security.h> |
2b188cc1 | 85 | |
c826bd7a DD |
86 | #define CREATE_TRACE_POINTS |
87 | #include <trace/events/io_uring.h> | |
88 | ||
2b188cc1 JA |
89 | #include <uapi/linux/io_uring.h> |
90 | ||
91 | #include "internal.h" | |
561fb04a | 92 | #include "io-wq.h" |
2b188cc1 | 93 | |
5277deaa | 94 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 95 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
4ce8ad95 | 96 | #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8 |
65e19f54 | 97 | |
187f08c1 | 98 | /* only define max */ |
042b0d85 | 99 | #define IORING_MAX_FIXED_FILES (1U << 15) |
21b55dbc SG |
100 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
101 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 102 | |
187f08c1 | 103 | #define IO_RSRC_TAG_TABLE_SHIFT (PAGE_SHIFT - 3) |
2d091d62 PB |
104 | #define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT) |
105 | #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1) | |
106 | ||
489809e2 PB |
107 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
108 | ||
68fe256a PB |
109 | #define SQE_COMMON_FLAGS (IOSQE_FIXED_FILE | IOSQE_IO_LINK | \ |
110 | IOSQE_IO_HARDLINK | IOSQE_ASYNC) | |
111 | ||
5562a8d7 PB |
112 | #define SQE_VALID_FLAGS (SQE_COMMON_FLAGS | IOSQE_BUFFER_SELECT | \ |
113 | IOSQE_IO_DRAIN | IOSQE_CQE_SKIP_SUCCESS) | |
68fe256a | 114 | |
c854357b | 115 | #define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \ |
d886e185 PB |
116 | REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS | \ |
117 | REQ_F_ASYNC_DATA) | |
b16fed66 | 118 | |
09899b19 PB |
119 | #define IO_TCTX_REFS_CACHE_NR (1U << 10) |
120 | ||
2b188cc1 JA |
121 | struct io_uring { |
122 | u32 head ____cacheline_aligned_in_smp; | |
123 | u32 tail ____cacheline_aligned_in_smp; | |
124 | }; | |
125 | ||
1e84b97b | 126 | /* |
75b28aff HV |
127 | * This data is shared with the application through the mmap at offsets |
128 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
129 | * |
130 | * The offsets to the member fields are published through struct | |
131 | * io_sqring_offsets when calling io_uring_setup. | |
132 | */ | |
75b28aff | 133 | struct io_rings { |
1e84b97b SB |
134 | /* |
135 | * Head and tail offsets into the ring; the offsets need to be | |
136 | * masked to get valid indices. | |
137 | * | |
75b28aff HV |
138 | * The kernel controls head of the sq ring and the tail of the cq ring, |
139 | * and the application controls tail of the sq ring and the head of the | |
140 | * cq ring. | |
1e84b97b | 141 | */ |
75b28aff | 142 | struct io_uring sq, cq; |
1e84b97b | 143 | /* |
75b28aff | 144 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
145 | * ring_entries - 1) |
146 | */ | |
75b28aff HV |
147 | u32 sq_ring_mask, cq_ring_mask; |
148 | /* Ring sizes (constant, power of 2) */ | |
149 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
150 | /* |
151 | * Number of invalid entries dropped by the kernel due to | |
152 | * invalid index stored in array | |
153 | * | |
154 | * Written by the kernel, shouldn't be modified by the | |
155 | * application (i.e. get number of "new events" by comparing to | |
156 | * cached value). | |
157 | * | |
158 | * After a new SQ head value was read by the application this | |
159 | * counter includes all submissions that were dropped reaching | |
160 | * the new SQ head (and possibly more). | |
161 | */ | |
75b28aff | 162 | u32 sq_dropped; |
1e84b97b | 163 | /* |
0d9b5b3a | 164 | * Runtime SQ flags |
1e84b97b SB |
165 | * |
166 | * Written by the kernel, shouldn't be modified by the | |
167 | * application. | |
168 | * | |
169 | * The application needs a full memory barrier before checking | |
170 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
171 | */ | |
75b28aff | 172 | u32 sq_flags; |
0d9b5b3a SG |
173 | /* |
174 | * Runtime CQ flags | |
175 | * | |
176 | * Written by the application, shouldn't be modified by the | |
177 | * kernel. | |
178 | */ | |
fe7e3257 | 179 | u32 cq_flags; |
1e84b97b SB |
180 | /* |
181 | * Number of completion events lost because the queue was full; | |
182 | * this should be avoided by the application by making sure | |
0b4295b5 | 183 | * there are not more requests pending than there is space in |
1e84b97b SB |
184 | * the completion queue. |
185 | * | |
186 | * Written by the kernel, shouldn't be modified by the | |
187 | * application (i.e. get number of "new events" by comparing to | |
188 | * cached value). | |
189 | * | |
190 | * As completion events come in out of order this counter is not | |
191 | * ordered with any other data. | |
192 | */ | |
75b28aff | 193 | u32 cq_overflow; |
1e84b97b SB |
194 | /* |
195 | * Ring buffer of completion events. | |
196 | * | |
197 | * The kernel writes completion events fresh every time they are | |
198 | * produced, so the application is allowed to modify pending | |
199 | * entries. | |
200 | */ | |
75b28aff | 201 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
202 | }; |
203 | ||
45d189c6 | 204 | enum io_uring_cmd_flags { |
51aac424 | 205 | IO_URING_F_COMPLETE_DEFER = 1, |
3b44b371 | 206 | IO_URING_F_UNLOCKED = 2, |
51aac424 PB |
207 | /* int's last bit, sign checks are usually faster than a bit test */ |
208 | IO_URING_F_NONBLOCK = INT_MIN, | |
45d189c6 PB |
209 | }; |
210 | ||
edafccee JA |
211 | struct io_mapped_ubuf { |
212 | u64 ubuf; | |
4751f53d | 213 | u64 ubuf_end; |
edafccee | 214 | unsigned int nr_bvecs; |
de293938 | 215 | unsigned long acct_pages; |
41edf1a5 | 216 | struct bio_vec bvec[]; |
edafccee JA |
217 | }; |
218 | ||
50238531 BM |
219 | struct io_ring_ctx; |
220 | ||
6c2450ae PB |
221 | struct io_overflow_cqe { |
222 | struct io_uring_cqe cqe; | |
223 | struct list_head list; | |
224 | }; | |
225 | ||
a04b0ac0 PB |
226 | struct io_fixed_file { |
227 | /* file * with additional FFS_* flags */ | |
228 | unsigned long file_ptr; | |
229 | }; | |
230 | ||
269bbe5f BM |
231 | struct io_rsrc_put { |
232 | struct list_head list; | |
b60c8dce | 233 | u64 tag; |
50238531 BM |
234 | union { |
235 | void *rsrc; | |
236 | struct file *file; | |
bd54b6fe | 237 | struct io_mapped_ubuf *buf; |
50238531 | 238 | }; |
269bbe5f BM |
239 | }; |
240 | ||
aeca241b | 241 | struct io_file_table { |
042b0d85 | 242 | struct io_fixed_file *files; |
31b51510 JA |
243 | }; |
244 | ||
b895c9a6 | 245 | struct io_rsrc_node { |
05589553 XW |
246 | struct percpu_ref refs; |
247 | struct list_head node; | |
269bbe5f | 248 | struct list_head rsrc_list; |
b895c9a6 | 249 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 250 | struct llist_node llist; |
e297822b | 251 | bool done; |
05589553 XW |
252 | }; |
253 | ||
40ae0ff7 PB |
254 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
255 | ||
b895c9a6 | 256 | struct io_rsrc_data { |
05f3fb3c JA |
257 | struct io_ring_ctx *ctx; |
258 | ||
2d091d62 PB |
259 | u64 **tags; |
260 | unsigned int nr; | |
40ae0ff7 | 261 | rsrc_put_fn *do_put; |
3e942498 | 262 | atomic_t refs; |
05f3fb3c | 263 | struct completion done; |
8bad28d8 | 264 | bool quiesce; |
05f3fb3c JA |
265 | }; |
266 | ||
dbc7d452 JA |
267 | struct io_buffer_list { |
268 | struct list_head list; | |
269 | struct list_head buf_list; | |
270 | __u16 bgid; | |
271 | }; | |
272 | ||
5a2e745d JA |
273 | struct io_buffer { |
274 | struct list_head list; | |
275 | __u64 addr; | |
d1f82808 | 276 | __u32 len; |
5a2e745d | 277 | __u16 bid; |
b1c62645 | 278 | __u16 bgid; |
5a2e745d JA |
279 | }; |
280 | ||
21b55dbc SG |
281 | struct io_restriction { |
282 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
283 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
284 | u8 sqe_flags_allowed; | |
285 | u8 sqe_flags_required; | |
7e84e1c7 | 286 | bool registered; |
21b55dbc SG |
287 | }; |
288 | ||
37d1e2e3 JA |
289 | enum { |
290 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
291 | IO_SQ_THREAD_SHOULD_PARK, | |
292 | }; | |
293 | ||
534ca6d6 JA |
294 | struct io_sq_data { |
295 | refcount_t refs; | |
9e138a48 | 296 | atomic_t park_pending; |
09a6f4ef | 297 | struct mutex lock; |
69fb2131 JA |
298 | |
299 | /* ctx's that are using this sqd */ | |
300 | struct list_head ctx_list; | |
69fb2131 | 301 | |
534ca6d6 JA |
302 | struct task_struct *thread; |
303 | struct wait_queue_head wait; | |
08369246 XW |
304 | |
305 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
306 | int sq_cpu; |
307 | pid_t task_pid; | |
5c2469e0 | 308 | pid_t task_tgid; |
37d1e2e3 JA |
309 | |
310 | unsigned long state; | |
37d1e2e3 | 311 | struct completion exited; |
534ca6d6 JA |
312 | }; |
313 | ||
6dd0be1e | 314 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 315 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 316 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 | 317 | |
a1ab7b35 PB |
318 | struct io_submit_link { |
319 | struct io_kiocb *head; | |
320 | struct io_kiocb *last; | |
321 | }; | |
322 | ||
258b29a9 | 323 | struct io_submit_state { |
5a158c6b PB |
324 | /* inline/task_work completion list, under ->uring_lock */ |
325 | struct io_wq_work_node free_list; | |
326 | /* batch completion logic */ | |
327 | struct io_wq_work_list compl_reqs; | |
a1ab7b35 | 328 | struct io_submit_link link; |
258b29a9 | 329 | |
258b29a9 | 330 | bool plug_started; |
4b628aeb | 331 | bool need_plug; |
3d4aeb9f | 332 | bool flush_cqes; |
5ca7a8b3 | 333 | unsigned short submit_nr; |
5a158c6b | 334 | struct blk_plug plug; |
258b29a9 PB |
335 | }; |
336 | ||
77bc59b4 UA |
337 | struct io_ev_fd { |
338 | struct eventfd_ctx *cq_ev_fd; | |
c75312dd | 339 | unsigned int eventfd_async: 1; |
77bc59b4 UA |
340 | struct rcu_head rcu; |
341 | }; | |
342 | ||
dbc7d452 JA |
343 | #define IO_BUFFERS_HASH_BITS 5 |
344 | ||
2b188cc1 | 345 | struct io_ring_ctx { |
b52ecf8c | 346 | /* const or read-mostly hot data */ |
2b188cc1 JA |
347 | struct { |
348 | struct percpu_ref refs; | |
2b188cc1 | 349 | |
b52ecf8c | 350 | struct io_rings *rings; |
2b188cc1 | 351 | unsigned int flags; |
e1d85334 | 352 | unsigned int compat: 1; |
e1d85334 | 353 | unsigned int drain_next: 1; |
21b55dbc | 354 | unsigned int restricted: 1; |
f18ee4cf | 355 | unsigned int off_timeout_used: 1; |
10c66904 | 356 | unsigned int drain_active: 1; |
5562a8d7 | 357 | unsigned int drain_disabled: 1; |
9aa8dfde | 358 | unsigned int has_evfd: 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; | |
ab409402 | 387 | int rsrc_cached_refs; |
7f1129d2 PB |
388 | struct io_file_table file_table; |
389 | unsigned nr_user_files; | |
390 | unsigned nr_user_bufs; | |
391 | struct io_mapped_ubuf **user_bufs; | |
392 | ||
393 | struct io_submit_state submit_state; | |
5262f567 | 394 | struct list_head timeout_list; |
ef9dd637 | 395 | struct list_head ltimeout_list; |
1d7bb1d5 | 396 | struct list_head cq_overflow_list; |
dbc7d452 | 397 | struct list_head *io_buffers; |
cc3cec83 | 398 | struct list_head io_buffers_cache; |
4d9237e3 | 399 | struct list_head apoll_cache; |
7f1129d2 PB |
400 | struct xarray personalities; |
401 | u32 pers_next; | |
402 | unsigned sq_thread_idle; | |
2b188cc1 JA |
403 | } ____cacheline_aligned_in_smp; |
404 | ||
d0acdee2 | 405 | /* IRQ completion list, under ->completion_lock */ |
c2b6c6bc | 406 | struct io_wq_work_list locked_free_list; |
d0acdee2 | 407 | unsigned int locked_free_nr; |
3c1a2ead | 408 | |
7c30f36a | 409 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
410 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
411 | ||
90554200 | 412 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 413 | struct list_head sqd_list; |
75b28aff | 414 | |
5ed7a37d | 415 | unsigned long check_cq_overflow; |
adc8682e OL |
416 | #ifdef CONFIG_NET_RX_BUSY_POLL |
417 | /* used to track busy poll napi_id */ | |
418 | struct list_head napi_list; | |
419 | spinlock_t napi_lock; /* napi_list lock */ | |
420 | #endif | |
5ed7a37d | 421 | |
206aefde JA |
422 | struct { |
423 | unsigned cached_cq_tail; | |
424 | unsigned cq_entries; | |
77bc59b4 | 425 | struct io_ev_fd __rcu *io_ev_fd; |
0499e582 PB |
426 | struct wait_queue_head cq_wait; |
427 | unsigned cq_extra; | |
428 | atomic_t cq_timeouts; | |
0499e582 | 429 | unsigned cq_last_tm_flush; |
206aefde | 430 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 431 | |
2b188cc1 JA |
432 | struct { |
433 | spinlock_t completion_lock; | |
e94f141b | 434 | |
89850fce JA |
435 | spinlock_t timeout_lock; |
436 | ||
def596e9 | 437 | /* |
540e32a0 | 438 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
439 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
440 | * For SQPOLL, only the single threaded io_sq_thread() will | |
441 | * manipulate the list, hence no extra locking is needed there. | |
442 | */ | |
5eef4e87 | 443 | struct io_wq_work_list iopoll_list; |
78076bb6 JA |
444 | struct hlist_head *cancel_hash; |
445 | unsigned cancel_hash_bits; | |
915b3dde | 446 | bool poll_multi_queue; |
cc3cec83 JA |
447 | |
448 | struct list_head io_buffers_comp; | |
2b188cc1 | 449 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 450 | |
21b55dbc | 451 | struct io_restriction restrictions; |
3c1a2ead | 452 | |
b13a8918 PB |
453 | /* slow path rsrc auxilary data, used by update/register */ |
454 | struct { | |
455 | struct io_rsrc_node *rsrc_backup_node; | |
456 | struct io_mapped_ubuf *dummy_ubuf; | |
457 | struct io_rsrc_data *file_data; | |
458 | struct io_rsrc_data *buf_data; | |
459 | ||
460 | struct delayed_work rsrc_put_work; | |
461 | struct llist_head rsrc_put_llist; | |
462 | struct list_head rsrc_ref_list; | |
463 | spinlock_t rsrc_ref_lock; | |
cc3cec83 JA |
464 | |
465 | struct list_head io_buffers_pages; | |
b13a8918 PB |
466 | }; |
467 | ||
3c1a2ead | 468 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
469 | struct { |
470 | #if defined(CONFIG_UNIX) | |
471 | struct socket *ring_sock; | |
472 | #endif | |
473 | /* hashed buffered write serialization */ | |
474 | struct io_wq_hash *hash_map; | |
475 | ||
476 | /* Only used for accounting purposes */ | |
477 | struct user_struct *user; | |
478 | struct mm_struct *mm_account; | |
479 | ||
480 | /* ctx exit and cancelation */ | |
9011bf9a PB |
481 | struct llist_head fallback_llist; |
482 | struct delayed_work fallback_work; | |
b986af7e PB |
483 | struct work_struct exit_work; |
484 | struct list_head tctx_list; | |
485 | struct completion ref_comp; | |
e139a1ec PB |
486 | u32 iowq_limits[2]; |
487 | bool iowq_limits_set; | |
b986af7e | 488 | }; |
2b188cc1 JA |
489 | }; |
490 | ||
e7a6c00d JA |
491 | /* |
492 | * Arbitrary limit, can be raised if need be | |
493 | */ | |
494 | #define IO_RINGFD_REG_MAX 16 | |
495 | ||
53e043b2 SM |
496 | struct io_uring_task { |
497 | /* submission side */ | |
09899b19 | 498 | int cached_refs; |
53e043b2 SM |
499 | struct xarray xa; |
500 | struct wait_queue_head wait; | |
ee53fb2b SM |
501 | const struct io_ring_ctx *last; |
502 | struct io_wq *io_wq; | |
53e043b2 | 503 | struct percpu_counter inflight; |
b303fe2e | 504 | atomic_t inflight_tracked; |
53e043b2 | 505 | atomic_t in_idle; |
53e043b2 SM |
506 | |
507 | spinlock_t task_lock; | |
508 | struct io_wq_work_list task_list; | |
4813c377 | 509 | struct io_wq_work_list prior_task_list; |
53e043b2 | 510 | struct callback_head task_work; |
e7a6c00d | 511 | struct file **registered_rings; |
6294f368 | 512 | bool task_running; |
53e043b2 SM |
513 | }; |
514 | ||
09bb8394 JA |
515 | /* |
516 | * First field must be the file pointer in all the | |
517 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
518 | */ | |
221c5eb2 JA |
519 | struct io_poll_iocb { |
520 | struct file *file; | |
018043be | 521 | struct wait_queue_head *head; |
221c5eb2 | 522 | __poll_t events; |
392edb45 | 523 | struct wait_queue_entry wait; |
221c5eb2 JA |
524 | }; |
525 | ||
9d805892 | 526 | struct io_poll_update { |
018043be | 527 | struct file *file; |
9d805892 PB |
528 | u64 old_user_data; |
529 | u64 new_user_data; | |
530 | __poll_t events; | |
b69de288 JA |
531 | bool update_events; |
532 | bool update_user_data; | |
018043be PB |
533 | }; |
534 | ||
b5dba59e JA |
535 | struct io_close { |
536 | struct file *file; | |
b5dba59e | 537 | int fd; |
7df778be | 538 | u32 file_slot; |
b5dba59e JA |
539 | }; |
540 | ||
ad8a48ac JA |
541 | struct io_timeout_data { |
542 | struct io_kiocb *req; | |
543 | struct hrtimer timer; | |
544 | struct timespec64 ts; | |
545 | enum hrtimer_mode mode; | |
50c1df2b | 546 | u32 flags; |
ad8a48ac JA |
547 | }; |
548 | ||
8ed8d3c3 JA |
549 | struct io_accept { |
550 | struct file *file; | |
551 | struct sockaddr __user *addr; | |
552 | int __user *addr_len; | |
553 | int flags; | |
aaa4db12 | 554 | u32 file_slot; |
09952e3e | 555 | unsigned long nofile; |
8ed8d3c3 JA |
556 | }; |
557 | ||
558 | struct io_sync { | |
559 | struct file *file; | |
560 | loff_t len; | |
561 | loff_t off; | |
562 | int flags; | |
d63d1b5e | 563 | int mode; |
8ed8d3c3 JA |
564 | }; |
565 | ||
fbf23849 JA |
566 | struct io_cancel { |
567 | struct file *file; | |
568 | u64 addr; | |
569 | }; | |
570 | ||
b29472ee JA |
571 | struct io_timeout { |
572 | struct file *file; | |
bfe68a22 PB |
573 | u32 off; |
574 | u32 target_seq; | |
135fcde8 | 575 | struct list_head list; |
90cd7e42 PB |
576 | /* head of the link, used by linked timeouts only */ |
577 | struct io_kiocb *head; | |
89b263f6 JA |
578 | /* for linked completions */ |
579 | struct io_kiocb *prev; | |
b29472ee JA |
580 | }; |
581 | ||
0bdf7a2d PB |
582 | struct io_timeout_rem { |
583 | struct file *file; | |
584 | u64 addr; | |
9c8e11b3 PB |
585 | |
586 | /* timeout update */ | |
587 | struct timespec64 ts; | |
588 | u32 flags; | |
f1042b6c | 589 | bool ltimeout; |
0bdf7a2d PB |
590 | }; |
591 | ||
9adbd45d JA |
592 | struct io_rw { |
593 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
594 | struct kiocb kiocb; | |
595 | u64 addr; | |
596 | u64 len; | |
597 | }; | |
598 | ||
3fbb51c1 JA |
599 | struct io_connect { |
600 | struct file *file; | |
601 | struct sockaddr __user *addr; | |
602 | int addr_len; | |
603 | }; | |
604 | ||
e47293fd JA |
605 | struct io_sr_msg { |
606 | struct file *file; | |
fddaface | 607 | union { |
4af3417a PB |
608 | struct compat_msghdr __user *umsg_compat; |
609 | struct user_msghdr __user *umsg; | |
610 | void __user *buf; | |
fddaface | 611 | }; |
e47293fd | 612 | int msg_flags; |
bcda7baa | 613 | int bgid; |
fddaface | 614 | size_t len; |
e47293fd JA |
615 | }; |
616 | ||
15b71abe JA |
617 | struct io_open { |
618 | struct file *file; | |
619 | int dfd; | |
b9445598 | 620 | u32 file_slot; |
15b71abe | 621 | struct filename *filename; |
c12cedf2 | 622 | struct open_how how; |
4022e7af | 623 | unsigned long nofile; |
15b71abe JA |
624 | }; |
625 | ||
269bbe5f | 626 | struct io_rsrc_update { |
05f3fb3c JA |
627 | struct file *file; |
628 | u64 arg; | |
629 | u32 nr_args; | |
630 | u32 offset; | |
631 | }; | |
632 | ||
4840e418 JA |
633 | struct io_fadvise { |
634 | struct file *file; | |
635 | u64 offset; | |
636 | u32 len; | |
637 | u32 advice; | |
638 | }; | |
639 | ||
c1ca757b JA |
640 | struct io_madvise { |
641 | struct file *file; | |
642 | u64 addr; | |
643 | u32 len; | |
644 | u32 advice; | |
645 | }; | |
646 | ||
3e4827b0 JA |
647 | struct io_epoll { |
648 | struct file *file; | |
649 | int epfd; | |
650 | int op; | |
651 | int fd; | |
652 | struct epoll_event event; | |
e47293fd JA |
653 | }; |
654 | ||
7d67af2c PB |
655 | struct io_splice { |
656 | struct file *file_out; | |
657 | struct file *file_in; | |
658 | loff_t off_out; | |
659 | loff_t off_in; | |
660 | u64 len; | |
661 | unsigned int flags; | |
662 | }; | |
663 | ||
ddf0322d JA |
664 | struct io_provide_buf { |
665 | struct file *file; | |
666 | __u64 addr; | |
38134ada | 667 | __u32 len; |
ddf0322d JA |
668 | __u32 bgid; |
669 | __u16 nbufs; | |
670 | __u16 bid; | |
671 | }; | |
672 | ||
1d9e1288 BM |
673 | struct io_statx { |
674 | struct file *file; | |
675 | int dfd; | |
676 | unsigned int mask; | |
677 | unsigned int flags; | |
e62753e4 | 678 | const char __user *filename; |
1d9e1288 BM |
679 | struct statx __user *buffer; |
680 | }; | |
681 | ||
36f4fa68 JA |
682 | struct io_shutdown { |
683 | struct file *file; | |
684 | int how; | |
685 | }; | |
686 | ||
80a261fd JA |
687 | struct io_rename { |
688 | struct file *file; | |
689 | int old_dfd; | |
690 | int new_dfd; | |
691 | struct filename *oldpath; | |
692 | struct filename *newpath; | |
693 | int flags; | |
694 | }; | |
695 | ||
14a1143b JA |
696 | struct io_unlink { |
697 | struct file *file; | |
698 | int dfd; | |
699 | int flags; | |
700 | struct filename *filename; | |
701 | }; | |
702 | ||
e34a02dc DK |
703 | struct io_mkdir { |
704 | struct file *file; | |
705 | int dfd; | |
706 | umode_t mode; | |
707 | struct filename *filename; | |
708 | }; | |
709 | ||
7a8721f8 DK |
710 | struct io_symlink { |
711 | struct file *file; | |
712 | int new_dfd; | |
713 | struct filename *oldpath; | |
714 | struct filename *newpath; | |
715 | }; | |
716 | ||
cf30da90 DK |
717 | struct io_hardlink { |
718 | struct file *file; | |
719 | int old_dfd; | |
720 | int new_dfd; | |
721 | struct filename *oldpath; | |
722 | struct filename *newpath; | |
723 | int flags; | |
724 | }; | |
725 | ||
4f57f06c JA |
726 | struct io_msg { |
727 | struct file *file; | |
728 | u64 user_data; | |
729 | u32 len; | |
730 | }; | |
731 | ||
f499a021 JA |
732 | struct io_async_connect { |
733 | struct sockaddr_storage address; | |
734 | }; | |
735 | ||
03b1230c JA |
736 | struct io_async_msghdr { |
737 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
738 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
739 | struct iovec *free_iov; | |
03b1230c JA |
740 | struct sockaddr __user *uaddr; |
741 | struct msghdr msg; | |
b537916c | 742 | struct sockaddr_storage addr; |
03b1230c JA |
743 | }; |
744 | ||
538941e2 | 745 | struct io_rw_state { |
ff6165b2 | 746 | struct iov_iter iter; |
cd658695 | 747 | struct iov_iter_state iter_state; |
c88598a9 | 748 | struct iovec fast_iov[UIO_FASTIOV]; |
538941e2 PB |
749 | }; |
750 | ||
751 | struct io_async_rw { | |
752 | struct io_rw_state s; | |
753 | const struct iovec *free_iovec; | |
227c0c96 | 754 | size_t bytes_done; |
bcf5a063 | 755 | struct wait_page_queue wpq; |
f67676d1 JA |
756 | }; |
757 | ||
6b47ee6e PB |
758 | enum { |
759 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
760 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
761 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
762 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
763 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 764 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
04c76b41 | 765 | REQ_F_CQE_SKIP_BIT = IOSQE_CQE_SKIP_SUCCESS_BIT, |
6b47ee6e | 766 | |
dddca226 | 767 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 768 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
769 | REQ_F_INFLIGHT_BIT, |
770 | REQ_F_CUR_POS_BIT, | |
771 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 772 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 773 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 774 | REQ_F_POLLED_BIT, |
bcda7baa | 775 | REQ_F_BUFFER_SELECTED_BIT, |
e342c807 | 776 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 777 | REQ_F_REISSUE_BIT, |
b8e64b53 | 778 | REQ_F_CREDS_BIT, |
20e60a38 | 779 | REQ_F_REFCOUNT_BIT, |
4d13d1a4 | 780 | REQ_F_ARM_LTIMEOUT_BIT, |
d886e185 | 781 | REQ_F_ASYNC_DATA_BIT, |
04c76b41 | 782 | REQ_F_SKIP_LINK_CQES_BIT, |
91eac1c6 JA |
783 | REQ_F_SINGLE_POLL_BIT, |
784 | REQ_F_DOUBLE_POLL_BIT, | |
7b29f92d | 785 | /* keep async read/write and isreg together and in order */ |
35645ac3 | 786 | REQ_F_SUPPORT_NOWAIT_BIT, |
7b29f92d | 787 | REQ_F_ISREG_BIT, |
84557871 JA |
788 | |
789 | /* not a real bit, just to check we're not overflowing the space */ | |
790 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
791 | }; |
792 | ||
793 | enum { | |
794 | /* ctx owns file */ | |
795 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
796 | /* drain existing IO first */ | |
797 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
798 | /* linked sqes */ | |
799 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
800 | /* doesn't sever on completion < 0 */ | |
801 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
802 | /* IOSQE_ASYNC */ | |
803 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
804 | /* IOSQE_BUFFER_SELECT */ |
805 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
04c76b41 PB |
806 | /* IOSQE_CQE_SKIP_SUCCESS */ |
807 | REQ_F_CQE_SKIP = BIT(REQ_F_CQE_SKIP_BIT), | |
6b47ee6e | 808 | |
6b47ee6e | 809 | /* fail rest of links */ |
93d2bcd2 | 810 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 811 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
812 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
813 | /* read/write uses file position */ | |
814 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
815 | /* must not punt to workers */ | |
816 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 817 | /* has or had linked timeout */ |
6b47ee6e | 818 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
819 | /* needs cleanup */ |
820 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
821 | /* already went through poll handler */ |
822 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
823 | /* buffer already selected */ |
824 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
e342c807 PB |
825 | /* completion is deferred through io_comp_state */ |
826 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
827 | /* caller should reissue async */ |
828 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
35645ac3 PB |
829 | /* supports async reads/writes */ |
830 | REQ_F_SUPPORT_NOWAIT = BIT(REQ_F_SUPPORT_NOWAIT_BIT), | |
7b29f92d JA |
831 | /* regular file */ |
832 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
833 | /* has creds assigned */ |
834 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
20e60a38 PB |
835 | /* skip refcounting if not set */ |
836 | REQ_F_REFCOUNT = BIT(REQ_F_REFCOUNT_BIT), | |
4d13d1a4 PB |
837 | /* there is a linked timeout that has to be armed */ |
838 | REQ_F_ARM_LTIMEOUT = BIT(REQ_F_ARM_LTIMEOUT_BIT), | |
d886e185 PB |
839 | /* ->async_data allocated */ |
840 | REQ_F_ASYNC_DATA = BIT(REQ_F_ASYNC_DATA_BIT), | |
04c76b41 PB |
841 | /* don't post CQEs while failing linked requests */ |
842 | REQ_F_SKIP_LINK_CQES = BIT(REQ_F_SKIP_LINK_CQES_BIT), | |
91eac1c6 JA |
843 | /* single poll may be active */ |
844 | REQ_F_SINGLE_POLL = BIT(REQ_F_SINGLE_POLL_BIT), | |
845 | /* double poll may active */ | |
846 | REQ_F_DOUBLE_POLL = BIT(REQ_F_DOUBLE_POLL_BIT), | |
d7718a9d JA |
847 | }; |
848 | ||
849 | struct async_poll { | |
850 | struct io_poll_iocb poll; | |
807abcb0 | 851 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
852 | }; |
853 | ||
f237c30a | 854 | typedef void (*io_req_tw_func_t)(struct io_kiocb *req, bool *locked); |
5b0a6acc | 855 | |
7cbf1722 | 856 | struct io_task_work { |
5b0a6acc PB |
857 | union { |
858 | struct io_wq_work_node node; | |
859 | struct llist_node fallback_node; | |
860 | }; | |
861 | io_req_tw_func_t func; | |
7cbf1722 JA |
862 | }; |
863 | ||
992da01a PB |
864 | enum { |
865 | IORING_RSRC_FILE = 0, | |
866 | IORING_RSRC_BUFFER = 1, | |
867 | }; | |
868 | ||
09bb8394 JA |
869 | /* |
870 | * NOTE! Each of the iocb union members has the file pointer | |
871 | * as the first entry in their struct definition. So you can | |
872 | * access the file pointer through any of the sub-structs, | |
63c36549 | 873 | * or directly as just 'file' in this struct. |
09bb8394 | 874 | */ |
2b188cc1 | 875 | struct io_kiocb { |
221c5eb2 | 876 | union { |
09bb8394 | 877 | struct file *file; |
9adbd45d | 878 | struct io_rw rw; |
221c5eb2 | 879 | struct io_poll_iocb poll; |
9d805892 | 880 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
881 | struct io_accept accept; |
882 | struct io_sync sync; | |
fbf23849 | 883 | struct io_cancel cancel; |
b29472ee | 884 | struct io_timeout timeout; |
0bdf7a2d | 885 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 886 | struct io_connect connect; |
e47293fd | 887 | struct io_sr_msg sr_msg; |
15b71abe | 888 | struct io_open open; |
b5dba59e | 889 | struct io_close close; |
269bbe5f | 890 | struct io_rsrc_update rsrc_update; |
4840e418 | 891 | struct io_fadvise fadvise; |
c1ca757b | 892 | struct io_madvise madvise; |
3e4827b0 | 893 | struct io_epoll epoll; |
7d67af2c | 894 | struct io_splice splice; |
ddf0322d | 895 | struct io_provide_buf pbuf; |
1d9e1288 | 896 | struct io_statx statx; |
36f4fa68 | 897 | struct io_shutdown shutdown; |
80a261fd | 898 | struct io_rename rename; |
14a1143b | 899 | struct io_unlink unlink; |
e34a02dc | 900 | struct io_mkdir mkdir; |
7a8721f8 | 901 | struct io_symlink symlink; |
cf30da90 | 902 | struct io_hardlink hardlink; |
4f57f06c | 903 | struct io_msg msg; |
221c5eb2 | 904 | }; |
2b188cc1 | 905 | |
d625c6ee | 906 | u8 opcode; |
65a6543d XW |
907 | /* polled IO has completed */ |
908 | u8 iopoll_completed; | |
4f4eeba8 | 909 | u16 buf_index; |
d17e56eb PB |
910 | unsigned int flags; |
911 | ||
912 | u64 user_data; | |
9cf7c104 | 913 | u32 result; |
d17e56eb | 914 | u32 cflags; |
4f4eeba8 | 915 | |
010e8e6b | 916 | struct io_ring_ctx *ctx; |
010e8e6b | 917 | struct task_struct *task; |
d7718a9d | 918 | |
269bbe5f | 919 | struct percpu_ref *fixed_rsrc_refs; |
d886e185 PB |
920 | /* store used ubuf, so we can prevent reloading */ |
921 | struct io_mapped_ubuf *imu; | |
fcb323cc | 922 | |
7e3709d5 | 923 | /* used by request caches, completion batching and iopoll */ |
ef05d9eb | 924 | struct io_wq_work_node comp_list; |
d17e56eb | 925 | atomic_t refs; |
521d61fc | 926 | atomic_t poll_refs; |
7e3709d5 | 927 | struct io_kiocb *link; |
5b0a6acc | 928 | struct io_task_work io_task_work; |
010e8e6b PB |
929 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
930 | struct hlist_node hash_node; | |
7e3709d5 | 931 | /* internal polling, see IORING_FEAT_FAST_POLL */ |
010e8e6b | 932 | struct async_poll *apoll; |
d886e185 PB |
933 | /* opcode allocated if it needs to store data for async defer */ |
934 | void *async_data; | |
7e3709d5 | 935 | /* custom credentials, valid IFF REQ_F_CREDS is set */ |
7e3709d5 | 936 | /* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */ |
30d51dd4 | 937 | struct io_buffer *kbuf; |
521d61fc JA |
938 | const struct cred *creds; |
939 | struct io_wq_work work; | |
2b188cc1 | 940 | }; |
05589553 | 941 | |
13bf43f5 PB |
942 | struct io_tctx_node { |
943 | struct list_head ctx_node; | |
944 | struct task_struct *task; | |
13bf43f5 PB |
945 | struct io_ring_ctx *ctx; |
946 | }; | |
947 | ||
27dc8338 PB |
948 | struct io_defer_entry { |
949 | struct list_head list; | |
950 | struct io_kiocb *req; | |
9cf7c104 | 951 | u32 seq; |
2b188cc1 JA |
952 | }; |
953 | ||
d3656344 | 954 | struct io_op_def { |
d3656344 JA |
955 | /* needs req->file assigned */ |
956 | unsigned needs_file : 1; | |
6d63416d PB |
957 | /* should block plug */ |
958 | unsigned plug : 1; | |
d3656344 JA |
959 | /* hash wq insertion if file is a regular file */ |
960 | unsigned hash_reg_file : 1; | |
961 | /* unbound wq insertion if file is a non-regular file */ | |
962 | unsigned unbound_nonreg_file : 1; | |
8a72758c JA |
963 | /* set if opcode supports polled "wait" */ |
964 | unsigned pollin : 1; | |
965 | unsigned pollout : 1; | |
bcda7baa JA |
966 | /* op supports buffer selection */ |
967 | unsigned buffer_select : 1; | |
26f0505a PB |
968 | /* do prep async if is going to be punted */ |
969 | unsigned needs_async_setup : 1; | |
6d63416d PB |
970 | /* opcode is not supported by this kernel */ |
971 | unsigned not_supported : 1; | |
5bd2182d PM |
972 | /* skip auditing */ |
973 | unsigned audit_skip : 1; | |
e8c2bc1f JA |
974 | /* size of async data needed, if any */ |
975 | unsigned short async_size; | |
d3656344 JA |
976 | }; |
977 | ||
0918682b | 978 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
979 | [IORING_OP_NOP] = {}, |
980 | [IORING_OP_READV] = { | |
d3656344 JA |
981 | .needs_file = 1, |
982 | .unbound_nonreg_file = 1, | |
8a72758c | 983 | .pollin = 1, |
4d954c25 | 984 | .buffer_select = 1, |
26f0505a | 985 | .needs_async_setup = 1, |
27926b68 | 986 | .plug = 1, |
5bd2182d | 987 | .audit_skip = 1, |
e8c2bc1f | 988 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 989 | }, |
0463b6c5 | 990 | [IORING_OP_WRITEV] = { |
d3656344 JA |
991 | .needs_file = 1, |
992 | .hash_reg_file = 1, | |
993 | .unbound_nonreg_file = 1, | |
8a72758c | 994 | .pollout = 1, |
26f0505a | 995 | .needs_async_setup = 1, |
27926b68 | 996 | .plug = 1, |
5bd2182d | 997 | .audit_skip = 1, |
e8c2bc1f | 998 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 999 | }, |
0463b6c5 | 1000 | [IORING_OP_FSYNC] = { |
d3656344 | 1001 | .needs_file = 1, |
5bd2182d | 1002 | .audit_skip = 1, |
d3656344 | 1003 | }, |
0463b6c5 | 1004 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
1005 | .needs_file = 1, |
1006 | .unbound_nonreg_file = 1, | |
8a72758c | 1007 | .pollin = 1, |
27926b68 | 1008 | .plug = 1, |
5bd2182d | 1009 | .audit_skip = 1, |
e8c2bc1f | 1010 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 1011 | }, |
0463b6c5 | 1012 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
1013 | .needs_file = 1, |
1014 | .hash_reg_file = 1, | |
1015 | .unbound_nonreg_file = 1, | |
8a72758c | 1016 | .pollout = 1, |
27926b68 | 1017 | .plug = 1, |
5bd2182d | 1018 | .audit_skip = 1, |
e8c2bc1f | 1019 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 1020 | }, |
0463b6c5 | 1021 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
1022 | .needs_file = 1, |
1023 | .unbound_nonreg_file = 1, | |
5bd2182d PM |
1024 | .audit_skip = 1, |
1025 | }, | |
1026 | [IORING_OP_POLL_REMOVE] = { | |
1027 | .audit_skip = 1, | |
d3656344 | 1028 | }, |
0463b6c5 | 1029 | [IORING_OP_SYNC_FILE_RANGE] = { |
d3656344 | 1030 | .needs_file = 1, |
5bd2182d | 1031 | .audit_skip = 1, |
d3656344 | 1032 | }, |
0463b6c5 | 1033 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
1034 | .needs_file = 1, |
1035 | .unbound_nonreg_file = 1, | |
8a72758c | 1036 | .pollout = 1, |
26f0505a | 1037 | .needs_async_setup = 1, |
e8c2bc1f | 1038 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 1039 | }, |
0463b6c5 | 1040 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
1041 | .needs_file = 1, |
1042 | .unbound_nonreg_file = 1, | |
8a72758c | 1043 | .pollin = 1, |
52de1fe1 | 1044 | .buffer_select = 1, |
26f0505a | 1045 | .needs_async_setup = 1, |
e8c2bc1f | 1046 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 1047 | }, |
0463b6c5 | 1048 | [IORING_OP_TIMEOUT] = { |
5bd2182d | 1049 | .audit_skip = 1, |
e8c2bc1f | 1050 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 1051 | }, |
9c8e11b3 PB |
1052 | [IORING_OP_TIMEOUT_REMOVE] = { |
1053 | /* used by timeout updates' prep() */ | |
5bd2182d | 1054 | .audit_skip = 1, |
9c8e11b3 | 1055 | }, |
0463b6c5 | 1056 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
1057 | .needs_file = 1, |
1058 | .unbound_nonreg_file = 1, | |
8a72758c | 1059 | .pollin = 1, |
d3656344 | 1060 | }, |
5bd2182d PM |
1061 | [IORING_OP_ASYNC_CANCEL] = { |
1062 | .audit_skip = 1, | |
1063 | }, | |
0463b6c5 | 1064 | [IORING_OP_LINK_TIMEOUT] = { |
5bd2182d | 1065 | .audit_skip = 1, |
e8c2bc1f | 1066 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 1067 | }, |
0463b6c5 | 1068 | [IORING_OP_CONNECT] = { |
d3656344 JA |
1069 | .needs_file = 1, |
1070 | .unbound_nonreg_file = 1, | |
8a72758c | 1071 | .pollout = 1, |
26f0505a | 1072 | .needs_async_setup = 1, |
e8c2bc1f | 1073 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 1074 | }, |
0463b6c5 | 1075 | [IORING_OP_FALLOCATE] = { |
d3656344 | 1076 | .needs_file = 1, |
d3656344 | 1077 | }, |
44526bed JA |
1078 | [IORING_OP_OPENAT] = {}, |
1079 | [IORING_OP_CLOSE] = {}, | |
5bd2182d PM |
1080 | [IORING_OP_FILES_UPDATE] = { |
1081 | .audit_skip = 1, | |
1082 | }, | |
1083 | [IORING_OP_STATX] = { | |
1084 | .audit_skip = 1, | |
1085 | }, | |
0463b6c5 | 1086 | [IORING_OP_READ] = { |
3a6820f2 JA |
1087 | .needs_file = 1, |
1088 | .unbound_nonreg_file = 1, | |
8a72758c | 1089 | .pollin = 1, |
bcda7baa | 1090 | .buffer_select = 1, |
27926b68 | 1091 | .plug = 1, |
5bd2182d | 1092 | .audit_skip = 1, |
e8c2bc1f | 1093 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1094 | }, |
0463b6c5 | 1095 | [IORING_OP_WRITE] = { |
3a6820f2 | 1096 | .needs_file = 1, |
7b3188e7 | 1097 | .hash_reg_file = 1, |
3a6820f2 | 1098 | .unbound_nonreg_file = 1, |
8a72758c | 1099 | .pollout = 1, |
27926b68 | 1100 | .plug = 1, |
5bd2182d | 1101 | .audit_skip = 1, |
e8c2bc1f | 1102 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1103 | }, |
0463b6c5 | 1104 | [IORING_OP_FADVISE] = { |
4840e418 | 1105 | .needs_file = 1, |
5bd2182d | 1106 | .audit_skip = 1, |
c1ca757b | 1107 | }, |
44526bed | 1108 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1109 | [IORING_OP_SEND] = { |
fddaface JA |
1110 | .needs_file = 1, |
1111 | .unbound_nonreg_file = 1, | |
8a72758c | 1112 | .pollout = 1, |
5bd2182d | 1113 | .audit_skip = 1, |
fddaface | 1114 | }, |
0463b6c5 | 1115 | [IORING_OP_RECV] = { |
fddaface JA |
1116 | .needs_file = 1, |
1117 | .unbound_nonreg_file = 1, | |
8a72758c | 1118 | .pollin = 1, |
bcda7baa | 1119 | .buffer_select = 1, |
5bd2182d | 1120 | .audit_skip = 1, |
fddaface | 1121 | }, |
0463b6c5 | 1122 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1123 | }, |
3e4827b0 JA |
1124 | [IORING_OP_EPOLL_CTL] = { |
1125 | .unbound_nonreg_file = 1, | |
5bd2182d | 1126 | .audit_skip = 1, |
3e4827b0 | 1127 | }, |
7d67af2c PB |
1128 | [IORING_OP_SPLICE] = { |
1129 | .needs_file = 1, | |
1130 | .hash_reg_file = 1, | |
1131 | .unbound_nonreg_file = 1, | |
5bd2182d PM |
1132 | .audit_skip = 1, |
1133 | }, | |
1134 | [IORING_OP_PROVIDE_BUFFERS] = { | |
1135 | .audit_skip = 1, | |
1136 | }, | |
1137 | [IORING_OP_REMOVE_BUFFERS] = { | |
1138 | .audit_skip = 1, | |
ddf0322d | 1139 | }, |
f2a8d5c7 PB |
1140 | [IORING_OP_TEE] = { |
1141 | .needs_file = 1, | |
1142 | .hash_reg_file = 1, | |
1143 | .unbound_nonreg_file = 1, | |
5bd2182d | 1144 | .audit_skip = 1, |
f2a8d5c7 | 1145 | }, |
36f4fa68 JA |
1146 | [IORING_OP_SHUTDOWN] = { |
1147 | .needs_file = 1, | |
1148 | }, | |
44526bed JA |
1149 | [IORING_OP_RENAMEAT] = {}, |
1150 | [IORING_OP_UNLINKAT] = {}, | |
e34a02dc | 1151 | [IORING_OP_MKDIRAT] = {}, |
7a8721f8 | 1152 | [IORING_OP_SYMLINKAT] = {}, |
cf30da90 | 1153 | [IORING_OP_LINKAT] = {}, |
4f57f06c JA |
1154 | [IORING_OP_MSG_RING] = { |
1155 | .needs_file = 1, | |
1156 | }, | |
d3656344 JA |
1157 | }; |
1158 | ||
0756a869 PB |
1159 | /* requests with any of those set should undergo io_disarm_next() */ |
1160 | #define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL) | |
1161 | ||
7a612350 | 1162 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1163 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1164 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1165 | struct task_struct *task, | |
3dd0c97a | 1166 | bool cancel_all); |
78cc687b | 1167 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
1ffc5422 | 1168 | |
913a571a PB |
1169 | static void io_fill_cqe_req(struct io_kiocb *req, s32 res, u32 cflags); |
1170 | ||
ec9c02ad | 1171 | static void io_put_req(struct io_kiocb *req); |
91c2f697 | 1172 | static void io_put_req_deferred(struct io_kiocb *req); |
c7dae4ba | 1173 | static void io_dismantle_req(struct io_kiocb *req); |
94ae5e77 | 1174 | static void io_queue_linked_timeout(struct io_kiocb *req); |
fdecb662 | 1175 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1176 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1177 | unsigned nr_args); |
68fb8979 | 1178 | static void io_clean_op(struct io_kiocb *req); |
ac177053 | 1179 | static struct file *io_file_get(struct io_ring_ctx *ctx, |
8371adf5 | 1180 | struct io_kiocb *req, int fd, bool fixed); |
c5eef2b9 | 1181 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1182 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1183 | |
907d1df3 | 1184 | static void io_req_task_queue(struct io_kiocb *req); |
c450178d | 1185 | static void __io_submit_flush_completions(struct io_ring_ctx *ctx); |
179ae0d1 | 1186 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1187 | |
b9445598 PB |
1188 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
1189 | unsigned int issue_flags, u32 slot_index); | |
7df778be PB |
1190 | static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags); |
1191 | ||
f1042b6c | 1192 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); |
9aa8dfde | 1193 | static void io_eventfd_signal(struct io_ring_ctx *ctx); |
b9445598 | 1194 | |
2b188cc1 JA |
1195 | static struct kmem_cache *req_cachep; |
1196 | ||
0918682b | 1197 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1198 | |
1199 | struct sock *io_uring_get_socket(struct file *file) | |
1200 | { | |
1201 | #if defined(CONFIG_UNIX) | |
1202 | if (file->f_op == &io_uring_fops) { | |
1203 | struct io_ring_ctx *ctx = file->private_data; | |
1204 | ||
1205 | return ctx->ring_sock->sk; | |
1206 | } | |
1207 | #endif | |
1208 | return NULL; | |
1209 | } | |
1210 | EXPORT_SYMBOL(io_uring_get_socket); | |
1211 | ||
f237c30a PB |
1212 | static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) |
1213 | { | |
1214 | if (!*locked) { | |
1215 | mutex_lock(&ctx->uring_lock); | |
1216 | *locked = true; | |
1217 | } | |
1218 | } | |
1219 | ||
f2f87370 PB |
1220 | #define io_for_each_link(pos, head) \ |
1221 | for (pos = (head); pos; pos = pos->link) | |
1222 | ||
21c843d5 PB |
1223 | /* |
1224 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1225 | * see commit f958d7b528b1 for details. | |
1226 | */ | |
1227 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1228 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1229 | ||
1230 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) | |
1231 | { | |
20e60a38 | 1232 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1233 | return atomic_inc_not_zero(&req->refs); |
1234 | } | |
1235 | ||
21c843d5 PB |
1236 | static inline bool req_ref_put_and_test(struct io_kiocb *req) |
1237 | { | |
20e60a38 PB |
1238 | if (likely(!(req->flags & REQ_F_REFCOUNT))) |
1239 | return true; | |
1240 | ||
21c843d5 PB |
1241 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1242 | return atomic_dec_and_test(&req->refs); | |
1243 | } | |
1244 | ||
21c843d5 PB |
1245 | static inline void req_ref_get(struct io_kiocb *req) |
1246 | { | |
20e60a38 | 1247 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1248 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1249 | atomic_inc(&req->refs); | |
1250 | } | |
1251 | ||
c450178d PB |
1252 | static inline void io_submit_flush_completions(struct io_ring_ctx *ctx) |
1253 | { | |
6f33b0bc | 1254 | if (!wq_list_empty(&ctx->submit_state.compl_reqs)) |
c450178d PB |
1255 | __io_submit_flush_completions(ctx); |
1256 | } | |
1257 | ||
48dcd38d | 1258 | static inline void __io_req_set_refcount(struct io_kiocb *req, int nr) |
20e60a38 PB |
1259 | { |
1260 | if (!(req->flags & REQ_F_REFCOUNT)) { | |
1261 | req->flags |= REQ_F_REFCOUNT; | |
48dcd38d | 1262 | atomic_set(&req->refs, nr); |
20e60a38 PB |
1263 | } |
1264 | } | |
1265 | ||
48dcd38d PB |
1266 | static inline void io_req_set_refcount(struct io_kiocb *req) |
1267 | { | |
1268 | __io_req_set_refcount(req, 1); | |
1269 | } | |
1270 | ||
ab409402 PB |
1271 | #define IO_RSRC_REF_BATCH 100 |
1272 | ||
1273 | static inline void io_req_put_rsrc_locked(struct io_kiocb *req, | |
1274 | struct io_ring_ctx *ctx) | |
1275 | __must_hold(&ctx->uring_lock) | |
36f72fe2 | 1276 | { |
ab409402 PB |
1277 | struct percpu_ref *ref = req->fixed_rsrc_refs; |
1278 | ||
1279 | if (ref) { | |
1280 | if (ref == &ctx->rsrc_node->refs) | |
1281 | ctx->rsrc_cached_refs++; | |
1282 | else | |
1283 | percpu_ref_put(ref); | |
1284 | } | |
1285 | } | |
1286 | ||
1287 | static inline void io_req_put_rsrc(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
1288 | { | |
1289 | if (req->fixed_rsrc_refs) | |
1290 | percpu_ref_put(req->fixed_rsrc_refs); | |
1291 | } | |
1292 | ||
1293 | static __cold void io_rsrc_refs_drop(struct io_ring_ctx *ctx) | |
1294 | __must_hold(&ctx->uring_lock) | |
1295 | { | |
1296 | if (ctx->rsrc_cached_refs) { | |
1297 | percpu_ref_put_many(&ctx->rsrc_node->refs, ctx->rsrc_cached_refs); | |
1298 | ctx->rsrc_cached_refs = 0; | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | static void io_rsrc_refs_refill(struct io_ring_ctx *ctx) | |
1303 | __must_hold(&ctx->uring_lock) | |
1304 | { | |
1305 | ctx->rsrc_cached_refs += IO_RSRC_REF_BATCH; | |
1306 | percpu_ref_get_many(&ctx->rsrc_node->refs, IO_RSRC_REF_BATCH); | |
1307 | } | |
36f72fe2 | 1308 | |
a46be971 PB |
1309 | static inline void io_req_set_rsrc_node(struct io_kiocb *req, |
1310 | struct io_ring_ctx *ctx) | |
36f72fe2 | 1311 | { |
269bbe5f | 1312 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1313 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
ab409402 PB |
1314 | ctx->rsrc_cached_refs--; |
1315 | if (unlikely(ctx->rsrc_cached_refs < 0)) | |
1316 | io_rsrc_refs_refill(ctx); | |
36f72fe2 PB |
1317 | } |
1318 | } | |
1319 | ||
cc3cec83 | 1320 | static unsigned int __io_put_kbuf(struct io_kiocb *req, struct list_head *list) |
3648e526 | 1321 | { |
d1fd1c20 | 1322 | struct io_buffer *kbuf = req->kbuf; |
3648e526 HX |
1323 | unsigned int cflags; |
1324 | ||
cc3cec83 | 1325 | cflags = IORING_CQE_F_BUFFER | (kbuf->bid << IORING_CQE_BUFFER_SHIFT); |
3648e526 | 1326 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
cc3cec83 | 1327 | list_add(&kbuf->list, list); |
d1fd1c20 | 1328 | req->kbuf = NULL; |
3648e526 HX |
1329 | return cflags; |
1330 | } | |
1331 | ||
cc3cec83 | 1332 | static inline unsigned int io_put_kbuf_comp(struct io_kiocb *req) |
3648e526 HX |
1333 | { |
1334 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) | |
1335 | return 0; | |
cc3cec83 JA |
1336 | return __io_put_kbuf(req, &req->ctx->io_buffers_comp); |
1337 | } | |
1338 | ||
1339 | static inline unsigned int io_put_kbuf(struct io_kiocb *req, | |
1340 | unsigned issue_flags) | |
1341 | { | |
1342 | unsigned int cflags; | |
1343 | ||
1344 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) | |
1345 | return 0; | |
1346 | ||
1347 | /* | |
1348 | * We can add this buffer back to two lists: | |
1349 | * | |
1350 | * 1) The io_buffers_cache list. This one is protected by the | |
1351 | * ctx->uring_lock. If we already hold this lock, add back to this | |
1352 | * list as we can grab it from issue as well. | |
1353 | * 2) The io_buffers_comp list. This one is protected by the | |
1354 | * ctx->completion_lock. | |
1355 | * | |
1356 | * We migrate buffers from the comp_list to the issue cache list | |
1357 | * when we need one. | |
1358 | */ | |
1359 | if (issue_flags & IO_URING_F_UNLOCKED) { | |
1360 | struct io_ring_ctx *ctx = req->ctx; | |
1361 | ||
1362 | spin_lock(&ctx->completion_lock); | |
1363 | cflags = __io_put_kbuf(req, &ctx->io_buffers_comp); | |
1364 | spin_unlock(&ctx->completion_lock); | |
1365 | } else { | |
1366 | cflags = __io_put_kbuf(req, &req->ctx->io_buffers_cache); | |
1367 | } | |
1368 | ||
1369 | return cflags; | |
3648e526 HX |
1370 | } |
1371 | ||
dbc7d452 JA |
1372 | static struct io_buffer_list *io_buffer_get_list(struct io_ring_ctx *ctx, |
1373 | unsigned int bgid) | |
1374 | { | |
1375 | struct list_head *hash_list; | |
1376 | struct io_buffer_list *bl; | |
1377 | ||
1378 | hash_list = &ctx->io_buffers[hash_32(bgid, IO_BUFFERS_HASH_BITS)]; | |
1379 | list_for_each_entry(bl, hash_list, list) | |
1380 | if (bl->bgid == bgid || bgid == -1U) | |
1381 | return bl; | |
1382 | ||
1383 | return NULL; | |
1384 | } | |
1385 | ||
b1c62645 JA |
1386 | static void io_kbuf_recycle(struct io_kiocb *req) |
1387 | { | |
1388 | struct io_ring_ctx *ctx = req->ctx; | |
dbc7d452 JA |
1389 | struct io_buffer_list *bl; |
1390 | struct io_buffer *buf; | |
b1c62645 JA |
1391 | |
1392 | if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) | |
1393 | return; | |
1394 | ||
1395 | lockdep_assert_held(&ctx->uring_lock); | |
1396 | ||
1397 | buf = req->kbuf; | |
dbc7d452 JA |
1398 | bl = io_buffer_get_list(ctx, buf->bgid); |
1399 | list_add(&buf->list, &bl->buf_list); | |
b1c62645 JA |
1400 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
1401 | req->kbuf = NULL; | |
1402 | } | |
1403 | ||
3dd0c97a PB |
1404 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1405 | bool cancel_all) | |
6af3f48b | 1406 | __must_hold(&req->ctx->timeout_lock) |
08d23634 PB |
1407 | { |
1408 | struct io_kiocb *req; | |
1409 | ||
68207680 | 1410 | if (task && head->task != task) |
08d23634 | 1411 | return false; |
3dd0c97a | 1412 | if (cancel_all) |
08d23634 PB |
1413 | return true; |
1414 | ||
1415 | io_for_each_link(req, head) { | |
b05a1bcd | 1416 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1417 | return true; |
08d23634 PB |
1418 | } |
1419 | return false; | |
1420 | } | |
1421 | ||
6af3f48b PB |
1422 | static bool io_match_linked(struct io_kiocb *head) |
1423 | { | |
1424 | struct io_kiocb *req; | |
1425 | ||
1426 | io_for_each_link(req, head) { | |
1427 | if (req->flags & REQ_F_INFLIGHT) | |
1428 | return true; | |
1429 | } | |
1430 | return false; | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * As io_match_task() but protected against racing with linked timeouts. | |
1435 | * User must not hold timeout_lock. | |
1436 | */ | |
1437 | static bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, | |
1438 | bool cancel_all) | |
1439 | { | |
1440 | bool matched; | |
1441 | ||
1442 | if (task && head->task != task) | |
1443 | return false; | |
1444 | if (cancel_all) | |
1445 | return true; | |
1446 | ||
1447 | if (head->flags & REQ_F_LINK_TIMEOUT) { | |
1448 | struct io_ring_ctx *ctx = head->ctx; | |
1449 | ||
1450 | /* protect against races with linked timeouts */ | |
1451 | spin_lock_irq(&ctx->timeout_lock); | |
1452 | matched = io_match_linked(head); | |
1453 | spin_unlock_irq(&ctx->timeout_lock); | |
1454 | } else { | |
1455 | matched = io_match_linked(head); | |
1456 | } | |
1457 | return matched; | |
1458 | } | |
1459 | ||
d886e185 PB |
1460 | static inline bool req_has_async_data(struct io_kiocb *req) |
1461 | { | |
1462 | return req->flags & REQ_F_ASYNC_DATA; | |
1463 | } | |
1464 | ||
93d2bcd2 | 1465 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1466 | { |
93d2bcd2 | 1467 | req->flags |= REQ_F_FAIL; |
04c76b41 PB |
1468 | if (req->flags & REQ_F_CQE_SKIP) { |
1469 | req->flags &= ~REQ_F_CQE_SKIP; | |
1470 | req->flags |= REQ_F_SKIP_LINK_CQES; | |
1471 | } | |
c40f6379 | 1472 | } |
4a38aed2 | 1473 | |
a8295b98 HX |
1474 | static inline void req_fail_link_node(struct io_kiocb *req, int res) |
1475 | { | |
1476 | req_set_fail(req); | |
1477 | req->result = res; | |
1478 | } | |
1479 | ||
c072481d | 1480 | static __cold void io_ring_ctx_ref_free(struct percpu_ref *ref) |
2b188cc1 JA |
1481 | { |
1482 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1483 | ||
0f158b4c | 1484 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1485 | } |
1486 | ||
8eb7e2d0 PB |
1487 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1488 | { | |
1489 | return !req->timeout.off; | |
1490 | } | |
1491 | ||
c072481d | 1492 | static __cold void io_fallback_req_func(struct work_struct *work) |
f56165e6 PB |
1493 | { |
1494 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
1495 | fallback_work.work); | |
1496 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
1497 | struct io_kiocb *req, *tmp; | |
f237c30a | 1498 | bool locked = false; |
f56165e6 PB |
1499 | |
1500 | percpu_ref_get(&ctx->refs); | |
1501 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) | |
f237c30a | 1502 | req->io_task_work.func(req, &locked); |
5636c00d | 1503 | |
f237c30a | 1504 | if (locked) { |
c450178d | 1505 | io_submit_flush_completions(ctx); |
f237c30a PB |
1506 | mutex_unlock(&ctx->uring_lock); |
1507 | } | |
f56165e6 PB |
1508 | percpu_ref_put(&ctx->refs); |
1509 | } | |
1510 | ||
c072481d | 1511 | static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
2b188cc1 JA |
1512 | { |
1513 | struct io_ring_ctx *ctx; | |
dbc7d452 | 1514 | int i, hash_bits; |
2b188cc1 JA |
1515 | |
1516 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1517 | if (!ctx) | |
1518 | return NULL; | |
1519 | ||
78076bb6 JA |
1520 | /* |
1521 | * Use 5 bits less than the max cq entries, that should give us around | |
1522 | * 32 entries per hash list if totally full and uniformly spread. | |
1523 | */ | |
1524 | hash_bits = ilog2(p->cq_entries); | |
1525 | hash_bits -= 5; | |
1526 | if (hash_bits <= 0) | |
1527 | hash_bits = 1; | |
1528 | ctx->cancel_hash_bits = hash_bits; | |
1529 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1530 | GFP_KERNEL); | |
1531 | if (!ctx->cancel_hash) | |
1532 | goto err; | |
1533 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1534 | ||
6224843d PB |
1535 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1536 | if (!ctx->dummy_ubuf) | |
1537 | goto err; | |
1538 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1539 | ctx->dummy_ubuf->ubuf = -1UL; | |
1540 | ||
dbc7d452 JA |
1541 | ctx->io_buffers = kcalloc(1U << IO_BUFFERS_HASH_BITS, |
1542 | sizeof(struct list_head), GFP_KERNEL); | |
1543 | if (!ctx->io_buffers) | |
1544 | goto err; | |
1545 | for (i = 0; i < (1U << IO_BUFFERS_HASH_BITS); i++) | |
1546 | INIT_LIST_HEAD(&ctx->io_buffers[i]); | |
1547 | ||
21482896 | 1548 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1549 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1550 | goto err; | |
2b188cc1 JA |
1551 | |
1552 | ctx->flags = p->flags; | |
90554200 | 1553 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1554 | INIT_LIST_HEAD(&ctx->sqd_list); |
1d7bb1d5 | 1555 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
cc3cec83 | 1556 | INIT_LIST_HEAD(&ctx->io_buffers_cache); |
4d9237e3 | 1557 | INIT_LIST_HEAD(&ctx->apoll_cache); |
0f158b4c | 1558 | init_completion(&ctx->ref_comp); |
61cf9370 | 1559 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1560 | mutex_init(&ctx->uring_lock); |
311997b3 | 1561 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1562 | spin_lock_init(&ctx->completion_lock); |
89850fce | 1563 | spin_lock_init(&ctx->timeout_lock); |
5eef4e87 | 1564 | INIT_WQ_LIST(&ctx->iopoll_list); |
cc3cec83 JA |
1565 | INIT_LIST_HEAD(&ctx->io_buffers_pages); |
1566 | INIT_LIST_HEAD(&ctx->io_buffers_comp); | |
de0617e4 | 1567 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1568 | INIT_LIST_HEAD(&ctx->timeout_list); |
ef9dd637 | 1569 | INIT_LIST_HEAD(&ctx->ltimeout_list); |
d67d2263 BM |
1570 | spin_lock_init(&ctx->rsrc_ref_lock); |
1571 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1572 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1573 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1574 | INIT_LIST_HEAD(&ctx->tctx_list); |
c2b6c6bc PB |
1575 | ctx->submit_state.free_list.next = NULL; |
1576 | INIT_WQ_LIST(&ctx->locked_free_list); | |
9011bf9a | 1577 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
6f33b0bc | 1578 | INIT_WQ_LIST(&ctx->submit_state.compl_reqs); |
adc8682e OL |
1579 | #ifdef CONFIG_NET_RX_BUSY_POLL |
1580 | INIT_LIST_HEAD(&ctx->napi_list); | |
1581 | spin_lock_init(&ctx->napi_lock); | |
1582 | #endif | |
2b188cc1 | 1583 | return ctx; |
206aefde | 1584 | err: |
6224843d | 1585 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1586 | kfree(ctx->cancel_hash); |
dbc7d452 | 1587 | kfree(ctx->io_buffers); |
206aefde JA |
1588 | kfree(ctx); |
1589 | return NULL; | |
2b188cc1 JA |
1590 | } |
1591 | ||
8f6ed49a PB |
1592 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1593 | { | |
1594 | struct io_rings *r = ctx->rings; | |
1595 | ||
1596 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1597 | ctx->cq_extra--; | |
1598 | } | |
1599 | ||
9cf7c104 | 1600 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1601 | { |
2bc9930e JA |
1602 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1603 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1604 | |
8f6ed49a | 1605 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1606 | } |
de0617e4 | 1607 | |
9d858b21 | 1608 | return false; |
de0617e4 JA |
1609 | } |
1610 | ||
35645ac3 PB |
1611 | #define FFS_NOWAIT 0x1UL |
1612 | #define FFS_ISREG 0x2UL | |
1613 | #define FFS_MASK ~(FFS_NOWAIT|FFS_ISREG) | |
c97d8a0f PB |
1614 | |
1615 | static inline bool io_req_ffs_set(struct io_kiocb *req) | |
1616 | { | |
35645ac3 | 1617 | return req->flags & REQ_F_FIXED_FILE; |
c97d8a0f PB |
1618 | } |
1619 | ||
c072481d | 1620 | static inline void io_req_track_inflight(struct io_kiocb *req) |
ce3d5aae | 1621 | { |
ce3d5aae | 1622 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1623 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1624 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1625 | } |
1626 | } | |
1627 | ||
fd08e530 PB |
1628 | static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req) |
1629 | { | |
906c6caa PB |
1630 | if (WARN_ON_ONCE(!req->link)) |
1631 | return NULL; | |
1632 | ||
4d13d1a4 PB |
1633 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
1634 | req->flags |= REQ_F_LINK_TIMEOUT; | |
fd08e530 PB |
1635 | |
1636 | /* linked timeouts should have two refs once prep'ed */ | |
48dcd38d | 1637 | io_req_set_refcount(req); |
4d13d1a4 PB |
1638 | __io_req_set_refcount(req->link, 2); |
1639 | return req->link; | |
fd08e530 PB |
1640 | } |
1641 | ||
1642 | static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) | |
1643 | { | |
4d13d1a4 | 1644 | if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) |
fd08e530 PB |
1645 | return NULL; |
1646 | return __io_prep_linked_timeout(req); | |
1647 | } | |
1648 | ||
1e6fa521 JA |
1649 | static void io_prep_async_work(struct io_kiocb *req) |
1650 | { | |
1651 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1652 | struct io_ring_ctx *ctx = req->ctx; |
1653 | ||
b8e64b53 PB |
1654 | if (!(req->flags & REQ_F_CREDS)) { |
1655 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 1656 | req->creds = get_current_cred(); |
b8e64b53 | 1657 | } |
003e8dcc | 1658 | |
e1d675df PB |
1659 | req->work.list.next = NULL; |
1660 | req->work.flags = 0; | |
feaadc4f PB |
1661 | if (req->flags & REQ_F_FORCE_ASYNC) |
1662 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1663 | ||
1e6fa521 JA |
1664 | if (req->flags & REQ_F_ISREG) { |
1665 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1666 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1667 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1668 | if (def->unbound_nonreg_file) |
1669 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1670 | } | |
e1d675df PB |
1671 | |
1672 | switch (req->opcode) { | |
1673 | case IORING_OP_SPLICE: | |
1674 | case IORING_OP_TEE: | |
e1d675df PB |
1675 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1676 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1677 | break; | |
1678 | } | |
561fb04a | 1679 | } |
cccf0ee8 | 1680 | |
cbdcb435 | 1681 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1682 | { |
cbdcb435 | 1683 | struct io_kiocb *cur; |
54a91f3b | 1684 | |
44eff40a PB |
1685 | if (req->flags & REQ_F_LINK_TIMEOUT) { |
1686 | struct io_ring_ctx *ctx = req->ctx; | |
1687 | ||
674ee8e1 | 1688 | spin_lock_irq(&ctx->timeout_lock); |
44eff40a PB |
1689 | io_for_each_link(cur, req) |
1690 | io_prep_async_work(cur); | |
674ee8e1 | 1691 | spin_unlock_irq(&ctx->timeout_lock); |
44eff40a PB |
1692 | } else { |
1693 | io_for_each_link(cur, req) | |
1694 | io_prep_async_work(cur); | |
1695 | } | |
561fb04a JA |
1696 | } |
1697 | ||
fff4e40e PB |
1698 | static inline void io_req_add_compl_list(struct io_kiocb *req) |
1699 | { | |
3d4aeb9f | 1700 | struct io_ring_ctx *ctx = req->ctx; |
33ce2aff | 1701 | struct io_submit_state *state = &ctx->submit_state; |
fff4e40e | 1702 | |
3d4aeb9f PB |
1703 | if (!(req->flags & REQ_F_CQE_SKIP)) |
1704 | ctx->submit_state.flush_cqes = true; | |
fff4e40e PB |
1705 | wq_list_add_tail(&req->comp_list, &state->compl_reqs); |
1706 | } | |
1707 | ||
00169246 | 1708 | static void io_queue_async_work(struct io_kiocb *req, bool *dont_use) |
561fb04a | 1709 | { |
a197f664 | 1710 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1711 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1712 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1713 | |
3bfe6106 JA |
1714 | BUG_ON(!tctx); |
1715 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1716 | |
cbdcb435 PB |
1717 | /* init ->work of the whole link before punting */ |
1718 | io_prep_async_link(req); | |
991468dc JA |
1719 | |
1720 | /* | |
1721 | * Not expected to happen, but if we do have a bug where this _can_ | |
1722 | * happen, catch it here and ensure the request is marked as | |
1723 | * canceled. That will make io-wq go through the usual work cancel | |
1724 | * procedure rather than attempt to run this request (or create a new | |
1725 | * worker for it). | |
1726 | */ | |
1727 | if (WARN_ON_ONCE(!same_thread_group(req->task, current))) | |
1728 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1729 | ||
502c87d6 SR |
1730 | trace_io_uring_queue_async_work(ctx, req, req->user_data, req->opcode, req->flags, |
1731 | &req->work, io_wq_is_hashed(&req->work)); | |
ebf93667 | 1732 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1733 | if (link) |
1734 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1735 | } |
1736 | ||
1ee4160c | 1737 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1738 | __must_hold(&req->ctx->completion_lock) |
89850fce | 1739 | __must_hold(&req->ctx->timeout_lock) |
5262f567 | 1740 | { |
e8c2bc1f | 1741 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1742 | |
fd9c7bc5 | 1743 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
2ae2eb9d PB |
1744 | if (status) |
1745 | req_set_fail(req); | |
01cec8c1 PB |
1746 | atomic_set(&req->ctx->cq_timeouts, |
1747 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1748 | list_del_init(&req->timeout.list); |
913a571a | 1749 | io_fill_cqe_req(req, status, 0); |
91c2f697 | 1750 | io_put_req_deferred(req); |
5262f567 JA |
1751 | } |
1752 | } | |
1753 | ||
c072481d | 1754 | static __cold void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1755 | { |
441b8a78 | 1756 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
1757 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1758 | struct io_defer_entry, list); | |
de0617e4 | 1759 | |
9cf7c104 | 1760 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1761 | break; |
27dc8338 | 1762 | list_del_init(&de->list); |
907d1df3 | 1763 | io_req_task_queue(de->req); |
27dc8338 | 1764 | kfree(de); |
441b8a78 | 1765 | } |
04518945 PB |
1766 | } |
1767 | ||
c072481d | 1768 | static __cold void io_flush_timeouts(struct io_ring_ctx *ctx) |
89850fce | 1769 | __must_hold(&ctx->completion_lock) |
de0617e4 | 1770 | { |
441b8a78 | 1771 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
f010505b | 1772 | |
79ebeaee | 1773 | spin_lock_irq(&ctx->timeout_lock); |
f18ee4cf | 1774 | while (!list_empty(&ctx->timeout_list)) { |
f010505b | 1775 | u32 events_needed, events_got; |
360428f8 | 1776 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1777 | struct io_kiocb, timeout.list); |
de0617e4 | 1778 | |
8eb7e2d0 | 1779 | if (io_is_timeout_noseq(req)) |
360428f8 | 1780 | break; |
f010505b MDG |
1781 | |
1782 | /* | |
1783 | * Since seq can easily wrap around over time, subtract | |
1784 | * the last seq at which timeouts were flushed before comparing. | |
1785 | * Assuming not more than 2^31-1 events have happened since, | |
1786 | * these subtractions won't have wrapped, so we can check if | |
1787 | * target is in [last_seq, current_seq] by comparing the two. | |
1788 | */ | |
1789 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1790 | events_got = seq - ctx->cq_last_tm_flush; | |
1791 | if (events_got < events_needed) | |
360428f8 | 1792 | break; |
bfe68a22 | 1793 | |
135fcde8 | 1794 | list_del_init(&req->timeout.list); |
1ee4160c | 1795 | io_kill_timeout(req, 0); |
f18ee4cf | 1796 | } |
f010505b | 1797 | ctx->cq_last_tm_flush = seq; |
79ebeaee | 1798 | spin_unlock_irq(&ctx->timeout_lock); |
360428f8 | 1799 | } |
5262f567 | 1800 | |
9333f6b4 PB |
1801 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) |
1802 | { | |
1803 | /* order cqe stores with ring update */ | |
1804 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
1805 | } | |
1806 | ||
9aa8dfde | 1807 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 1808 | { |
9aa8dfde PB |
1809 | if (ctx->off_timeout_used || ctx->drain_active) { |
1810 | spin_lock(&ctx->completion_lock); | |
1811 | if (ctx->off_timeout_used) | |
1812 | io_flush_timeouts(ctx); | |
1813 | if (ctx->drain_active) | |
1814 | io_queue_deferred(ctx); | |
1815 | io_commit_cqring(ctx); | |
1816 | spin_unlock(&ctx->completion_lock); | |
1817 | } | |
1818 | if (ctx->has_evfd) | |
1819 | io_eventfd_signal(ctx); | |
de0617e4 JA |
1820 | } |
1821 | ||
90554200 JA |
1822 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1823 | { | |
1824 | struct io_rings *r = ctx->rings; | |
1825 | ||
a566c556 | 1826 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
1827 | } |
1828 | ||
888aae2e PB |
1829 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1830 | { | |
1831 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1832 | } | |
1833 | ||
d068b506 | 1834 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) |
2b188cc1 | 1835 | { |
75b28aff | 1836 | struct io_rings *rings = ctx->rings; |
ea5ab3b5 | 1837 | unsigned tail, mask = ctx->cq_entries - 1; |
2b188cc1 | 1838 | |
115e12e5 SB |
1839 | /* |
1840 | * writes to the cq entry need to come after reading head; the | |
1841 | * control dependency is enough as we're using WRITE_ONCE to | |
1842 | * fill the cq entry | |
1843 | */ | |
a566c556 | 1844 | if (__io_cqring_events(ctx) == ctx->cq_entries) |
2b188cc1 JA |
1845 | return NULL; |
1846 | ||
888aae2e | 1847 | tail = ctx->cached_cq_tail++; |
ea5ab3b5 | 1848 | return &rings->cqes[tail & mask]; |
2b188cc1 JA |
1849 | } |
1850 | ||
77bc59b4 | 1851 | static void io_eventfd_signal(struct io_ring_ctx *ctx) |
f2842ab5 | 1852 | { |
77bc59b4 UA |
1853 | struct io_ev_fd *ev_fd; |
1854 | ||
77bc59b4 UA |
1855 | rcu_read_lock(); |
1856 | /* | |
1857 | * rcu_dereference ctx->io_ev_fd once and use it for both for checking | |
1858 | * and eventfd_signal | |
1859 | */ | |
1860 | ev_fd = rcu_dereference(ctx->io_ev_fd); | |
1861 | ||
1862 | /* | |
1863 | * Check again if ev_fd exists incase an io_eventfd_unregister call | |
1864 | * completed between the NULL check of ctx->io_ev_fd at the start of | |
1865 | * the function and rcu_read_lock. | |
1866 | */ | |
1867 | if (unlikely(!ev_fd)) | |
1868 | goto out; | |
7e55a19c | 1869 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
77bc59b4 UA |
1870 | goto out; |
1871 | ||
c75312dd | 1872 | if (!ev_fd->eventfd_async || io_wq_current_is_worker()) |
77bc59b4 | 1873 | eventfd_signal(ev_fd->cq_ev_fd, 1); |
77bc59b4 UA |
1874 | out: |
1875 | rcu_read_unlock(); | |
f2842ab5 JA |
1876 | } |
1877 | ||
9aa8dfde PB |
1878 | static inline void io_cqring_wake(struct io_ring_ctx *ctx) |
1879 | { | |
1880 | /* | |
1881 | * wake_up_all() may seem excessive, but io_wake_function() and | |
1882 | * io_should_wake() handle the termination of the loop and only | |
1883 | * wake as many waiters as we need to. | |
1884 | */ | |
1885 | if (wq_has_sleeper(&ctx->cq_wait)) | |
1886 | wake_up_all(&ctx->cq_wait); | |
1887 | } | |
1888 | ||
2c5d763c JA |
1889 | /* |
1890 | * This should only get called when at least one event has been posted. | |
1891 | * Some applications rely on the eventfd notification count only changing | |
1892 | * IFF a new CQE has been added to the CQ ring. There's no depedency on | |
1893 | * 1:1 relationship between how many times this function is called (and | |
1894 | * hence the eventfd count) and number of CQEs posted to the CQ ring. | |
1895 | */ | |
66fc25ca | 1896 | static inline void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1897 | { |
9aa8dfde PB |
1898 | if (unlikely(ctx->off_timeout_used || ctx->drain_active || |
1899 | ctx->has_evfd)) | |
9333f6b4 PB |
1900 | __io_commit_cqring_flush(ctx); |
1901 | ||
9aa8dfde | 1902 | io_cqring_wake(ctx); |
1d7bb1d5 JA |
1903 | } |
1904 | ||
80c18e4a PB |
1905 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1906 | { | |
9aa8dfde PB |
1907 | if (unlikely(ctx->off_timeout_used || ctx->drain_active || |
1908 | ctx->has_evfd)) | |
9333f6b4 PB |
1909 | __io_commit_cqring_flush(ctx); |
1910 | ||
9aa8dfde PB |
1911 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1912 | io_cqring_wake(ctx); | |
80c18e4a PB |
1913 | } |
1914 | ||
c4a2ed72 | 1915 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1916 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 1917 | { |
b18032bb | 1918 | bool all_flushed, posted; |
1d7bb1d5 | 1919 | |
a566c556 | 1920 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 1921 | return false; |
1d7bb1d5 | 1922 | |
b18032bb | 1923 | posted = false; |
79ebeaee | 1924 | spin_lock(&ctx->completion_lock); |
6c2450ae | 1925 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 1926 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 1927 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 1928 | |
1d7bb1d5 JA |
1929 | if (!cqe && !force) |
1930 | break; | |
6c2450ae PB |
1931 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1932 | struct io_overflow_cqe, list); | |
1933 | if (cqe) | |
1934 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1935 | else | |
8f6ed49a PB |
1936 | io_account_cq_overflow(ctx); |
1937 | ||
b18032bb | 1938 | posted = true; |
6c2450ae PB |
1939 | list_del(&ocqe->list); |
1940 | kfree(ocqe); | |
1d7bb1d5 JA |
1941 | } |
1942 | ||
09e88404 PB |
1943 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1944 | if (all_flushed) { | |
5ed7a37d | 1945 | clear_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
1946 | WRITE_ONCE(ctx->rings->sq_flags, |
1947 | ctx->rings->sq_flags & ~IORING_SQ_CQ_OVERFLOW); | |
09e88404 | 1948 | } |
46930143 | 1949 | |
b18032bb JA |
1950 | if (posted) |
1951 | io_commit_cqring(ctx); | |
79ebeaee | 1952 | spin_unlock(&ctx->completion_lock); |
b18032bb JA |
1953 | if (posted) |
1954 | io_cqring_ev_posted(ctx); | |
09e88404 | 1955 | return all_flushed; |
1d7bb1d5 JA |
1956 | } |
1957 | ||
90f67366 | 1958 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx) |
6c503150 | 1959 | { |
ca0a2651 JA |
1960 | bool ret = true; |
1961 | ||
5ed7a37d | 1962 | if (test_bit(0, &ctx->check_cq_overflow)) { |
6c503150 PB |
1963 | /* iopoll syncs against uring_lock, not completion_lock */ |
1964 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1965 | mutex_lock(&ctx->uring_lock); | |
90f67366 | 1966 | ret = __io_cqring_overflow_flush(ctx, false); |
6c503150 PB |
1967 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1968 | mutex_unlock(&ctx->uring_lock); | |
1969 | } | |
ca0a2651 JA |
1970 | |
1971 | return ret; | |
6c503150 PB |
1972 | } |
1973 | ||
6a290a14 PB |
1974 | /* must to be called somewhat shortly after putting a request */ |
1975 | static inline void io_put_task(struct task_struct *task, int nr) | |
1976 | { | |
1977 | struct io_uring_task *tctx = task->io_uring; | |
1978 | ||
e98e49b2 PB |
1979 | if (likely(task == current)) { |
1980 | tctx->cached_refs += nr; | |
1981 | } else { | |
1982 | percpu_counter_sub(&tctx->inflight, nr); | |
1983 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1984 | wake_up(&tctx->wait); | |
1985 | put_task_struct_many(task, nr); | |
1986 | } | |
6a290a14 PB |
1987 | } |
1988 | ||
9a10867a PB |
1989 | static void io_task_refs_refill(struct io_uring_task *tctx) |
1990 | { | |
1991 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
1992 | ||
1993 | percpu_counter_add(&tctx->inflight, refill); | |
1994 | refcount_add(refill, ¤t->usage); | |
1995 | tctx->cached_refs += refill; | |
1996 | } | |
1997 | ||
1998 | static inline void io_get_task_refs(int nr) | |
1999 | { | |
2000 | struct io_uring_task *tctx = current->io_uring; | |
2001 | ||
2002 | tctx->cached_refs -= nr; | |
2003 | if (unlikely(tctx->cached_refs < 0)) | |
2004 | io_task_refs_refill(tctx); | |
2005 | } | |
2006 | ||
3cc7fdb9 PB |
2007 | static __cold void io_uring_drop_tctx_refs(struct task_struct *task) |
2008 | { | |
2009 | struct io_uring_task *tctx = task->io_uring; | |
2010 | unsigned int refs = tctx->cached_refs; | |
2011 | ||
2012 | if (refs) { | |
2013 | tctx->cached_refs = 0; | |
2014 | percpu_counter_sub(&tctx->inflight, refs); | |
2015 | put_task_struct_many(task, refs); | |
2016 | } | |
2017 | } | |
2018 | ||
d4d19c19 | 2019 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
54daa9b2 | 2020 | s32 res, u32 cflags) |
2b188cc1 | 2021 | { |
cce4b8b0 | 2022 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 2023 | |
cce4b8b0 PB |
2024 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
2025 | if (!ocqe) { | |
2026 | /* | |
2027 | * If we're in ring overflow flush mode, or in task cancel mode, | |
2028 | * or cannot allocate an overflow entry, then we need to drop it | |
2029 | * on the floor. | |
2030 | */ | |
8f6ed49a | 2031 | io_account_cq_overflow(ctx); |
cce4b8b0 | 2032 | return false; |
2b188cc1 | 2033 | } |
cce4b8b0 | 2034 | if (list_empty(&ctx->cq_overflow_list)) { |
5ed7a37d | 2035 | set_bit(0, &ctx->check_cq_overflow); |
20c0b380 NA |
2036 | WRITE_ONCE(ctx->rings->sq_flags, |
2037 | ctx->rings->sq_flags | IORING_SQ_CQ_OVERFLOW); | |
2038 | ||
cce4b8b0 | 2039 | } |
d4d19c19 | 2040 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
2041 | ocqe->cqe.res = res; |
2042 | ocqe->cqe.flags = cflags; | |
2043 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
2044 | return true; | |
2b188cc1 JA |
2045 | } |
2046 | ||
ae4da189 | 2047 | static inline bool __io_fill_cqe(struct io_ring_ctx *ctx, u64 user_data, |
913a571a | 2048 | s32 res, u32 cflags) |
2b188cc1 JA |
2049 | { |
2050 | struct io_uring_cqe *cqe; | |
2051 | ||
2052 | /* | |
2053 | * If we can't get a cq entry, userspace overflowed the | |
2054 | * submission (by quite a lot). Increment the overflow count in | |
2055 | * the ring. | |
2056 | */ | |
d068b506 | 2057 | cqe = io_get_cqe(ctx); |
1d7bb1d5 | 2058 | if (likely(cqe)) { |
d4d19c19 | 2059 | WRITE_ONCE(cqe->user_data, user_data); |
2b188cc1 | 2060 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 2061 | WRITE_ONCE(cqe->flags, cflags); |
8d13326e | 2062 | return true; |
2b188cc1 | 2063 | } |
d4d19c19 | 2064 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
2b188cc1 JA |
2065 | } |
2066 | ||
ae4da189 | 2067 | static inline bool __io_fill_cqe_req(struct io_kiocb *req, s32 res, u32 cflags) |
d5ec1dfa | 2068 | { |
502c87d6 | 2069 | trace_io_uring_complete(req->ctx, req, req->user_data, res, cflags); |
ae4da189 | 2070 | return __io_fill_cqe(req->ctx, req->user_data, res, cflags); |
d5ec1dfa SR |
2071 | } |
2072 | ||
913a571a | 2073 | static noinline void io_fill_cqe_req(struct io_kiocb *req, s32 res, u32 cflags) |
bcda7baa | 2074 | { |
04c76b41 | 2075 | if (!(req->flags & REQ_F_CQE_SKIP)) |
ae4da189 | 2076 | __io_fill_cqe_req(req, res, cflags); |
bcda7baa JA |
2077 | } |
2078 | ||
913a571a PB |
2079 | static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data, |
2080 | s32 res, u32 cflags) | |
bcda7baa | 2081 | { |
913a571a | 2082 | ctx->cq_extra++; |
502c87d6 | 2083 | trace_io_uring_complete(ctx, NULL, user_data, res, cflags); |
ae4da189 | 2084 | return __io_fill_cqe(ctx, user_data, res, cflags); |
bcda7baa JA |
2085 | } |
2086 | ||
a37fae8a HX |
2087 | static void __io_req_complete_post(struct io_kiocb *req, s32 res, |
2088 | u32 cflags) | |
2b188cc1 | 2089 | { |
78e19bbe | 2090 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 | 2091 | |
04c76b41 | 2092 | if (!(req->flags & REQ_F_CQE_SKIP)) |
ae4da189 | 2093 | __io_fill_cqe_req(req, res, cflags); |
c7dae4ba JA |
2094 | /* |
2095 | * If we're the last reference to this request, add to our locked | |
2096 | * free_list cache. | |
2097 | */ | |
de9b4cca | 2098 | if (req_ref_put_and_test(req)) { |
7a612350 | 2099 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
0756a869 | 2100 | if (req->flags & IO_DISARM_MASK) |
7a612350 PB |
2101 | io_disarm_next(req); |
2102 | if (req->link) { | |
2103 | io_req_task_queue(req->link); | |
2104 | req->link = NULL; | |
2105 | } | |
2106 | } | |
ab409402 | 2107 | io_req_put_rsrc(req, ctx); |
c7dae4ba JA |
2108 | io_dismantle_req(req); |
2109 | io_put_task(req->task, 1); | |
c2b6c6bc | 2110 | wq_list_add_head(&req->comp_list, &ctx->locked_free_list); |
d0acdee2 | 2111 | ctx->locked_free_nr++; |
180f829f | 2112 | } |
a37fae8a HX |
2113 | } |
2114 | ||
2115 | static void io_req_complete_post(struct io_kiocb *req, s32 res, | |
2116 | u32 cflags) | |
2117 | { | |
2118 | struct io_ring_ctx *ctx = req->ctx; | |
2119 | ||
2120 | spin_lock(&ctx->completion_lock); | |
2121 | __io_req_complete_post(req, res, cflags); | |
7a612350 | 2122 | io_commit_cqring(ctx); |
79ebeaee | 2123 | spin_unlock(&ctx->completion_lock); |
a3f34907 | 2124 | io_cqring_ev_posted(ctx); |
4e3d9ff9 JA |
2125 | } |
2126 | ||
54daa9b2 PB |
2127 | static inline void io_req_complete_state(struct io_kiocb *req, s32 res, |
2128 | u32 cflags) | |
229a7b63 | 2129 | { |
a38d68db | 2130 | req->result = res; |
d17e56eb | 2131 | req->cflags = cflags; |
e342c807 | 2132 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
2133 | } |
2134 | ||
889fca73 | 2135 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
54daa9b2 | 2136 | s32 res, u32 cflags) |
bcda7baa | 2137 | { |
889fca73 PB |
2138 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
2139 | io_req_complete_state(req, res, cflags); | |
a38d68db | 2140 | else |
c7dae4ba | 2141 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
2142 | } |
2143 | ||
54daa9b2 | 2144 | static inline void io_req_complete(struct io_kiocb *req, s32 res) |
0ddf92e8 | 2145 | { |
889fca73 | 2146 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
2147 | } |
2148 | ||
54daa9b2 | 2149 | static void io_req_complete_failed(struct io_kiocb *req, s32 res) |
f41db273 | 2150 | { |
93d2bcd2 | 2151 | req_set_fail(req); |
3b2b78a8 | 2152 | io_req_complete_post(req, res, io_put_kbuf(req, 0)); |
f41db273 PB |
2153 | } |
2154 | ||
c6d3d9cb PB |
2155 | static void io_req_complete_fail_submit(struct io_kiocb *req) |
2156 | { | |
2157 | /* | |
2158 | * We don't submit, fail them all, for that replace hardlinks with | |
2159 | * normal links. Extra REQ_F_LINK is tolerated. | |
2160 | */ | |
2161 | req->flags &= ~REQ_F_HARDLINK; | |
2162 | req->flags |= REQ_F_LINK; | |
2163 | io_req_complete_failed(req, req->result); | |
2164 | } | |
2165 | ||
864ea921 PB |
2166 | /* |
2167 | * Don't initialise the fields below on every allocation, but do that in | |
2168 | * advance and keep them valid across allocations. | |
2169 | */ | |
2170 | static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
2171 | { | |
2172 | req->ctx = ctx; | |
2173 | req->link = NULL; | |
2174 | req->async_data = NULL; | |
2175 | /* not necessary, but safer to zero */ | |
2176 | req->result = 0; | |
2177 | } | |
2178 | ||
dac7a098 | 2179 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
cd0ca2e0 | 2180 | struct io_submit_state *state) |
dac7a098 | 2181 | { |
79ebeaee | 2182 | spin_lock(&ctx->completion_lock); |
c2b6c6bc | 2183 | wq_list_splice(&ctx->locked_free_list, &state->free_list); |
d0acdee2 | 2184 | ctx->locked_free_nr = 0; |
79ebeaee | 2185 | spin_unlock(&ctx->completion_lock); |
dac7a098 PB |
2186 | } |
2187 | ||
dd78f492 | 2188 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 2189 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 2190 | { |
c7dae4ba | 2191 | struct io_submit_state *state = &ctx->submit_state; |
0ddf92e8 | 2192 | |
c7dae4ba JA |
2193 | /* |
2194 | * If we have more than a batch's worth of requests in our IRQ side | |
2195 | * locked cache, grab the lock and move them over to our submission | |
2196 | * side cache. | |
2197 | */ | |
d0acdee2 | 2198 | if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) |
cd0ca2e0 | 2199 | io_flush_cached_locked_reqs(ctx, state); |
c2b6c6bc | 2200 | return !!state->free_list.next; |
0ddf92e8 JA |
2201 | } |
2202 | ||
5d5901a3 PB |
2203 | /* |
2204 | * A request might get retired back into the request caches even before opcode | |
2205 | * handlers and io_issue_sqe() are done with it, e.g. inline completion path. | |
2206 | * Because of that, io_alloc_req() should be called only under ->uring_lock | |
2207 | * and with extra caution to not get a request that is still worked on. | |
2208 | */ | |
c072481d | 2209 | static __cold bool __io_alloc_req_refill(struct io_ring_ctx *ctx) |
5d5901a3 | 2210 | __must_hold(&ctx->uring_lock) |
2b188cc1 | 2211 | { |
e5d1bc0a | 2212 | struct io_submit_state *state = &ctx->submit_state; |
864ea921 | 2213 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
3ab665b7 PB |
2214 | void *reqs[IO_REQ_ALLOC_BATCH]; |
2215 | struct io_kiocb *req; | |
864ea921 | 2216 | int ret, i; |
e5d1bc0a | 2217 | |
c2b6c6bc | 2218 | if (likely(state->free_list.next || io_flush_cached_reqs(ctx))) |
a33ae9ce | 2219 | return true; |
e5d1bc0a | 2220 | |
3ab665b7 | 2221 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, ARRAY_SIZE(reqs), reqs); |
fd6fab2c | 2222 | |
864ea921 PB |
2223 | /* |
2224 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
2225 | * retry single alloc to be on the safe side. | |
2226 | */ | |
2227 | if (unlikely(ret <= 0)) { | |
3ab665b7 PB |
2228 | reqs[0] = kmem_cache_alloc(req_cachep, gfp); |
2229 | if (!reqs[0]) | |
a33ae9ce | 2230 | return false; |
864ea921 | 2231 | ret = 1; |
2b188cc1 | 2232 | } |
864ea921 | 2233 | |
37f0e767 | 2234 | percpu_ref_get_many(&ctx->refs, ret); |
3ab665b7 PB |
2235 | for (i = 0; i < ret; i++) { |
2236 | req = reqs[i]; | |
2237 | ||
2238 | io_preinit_req(req, ctx); | |
c2b6c6bc | 2239 | wq_stack_add_head(&req->comp_list, &state->free_list); |
3ab665b7 | 2240 | } |
a33ae9ce PB |
2241 | return true; |
2242 | } | |
2243 | ||
2244 | static inline bool io_alloc_req_refill(struct io_ring_ctx *ctx) | |
2245 | { | |
2246 | if (unlikely(!ctx->submit_state.free_list.next)) | |
2247 | return __io_alloc_req_refill(ctx); | |
2248 | return true; | |
2249 | } | |
2250 | ||
2251 | static inline struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) | |
2252 | { | |
2253 | struct io_wq_work_node *node; | |
2254 | ||
2255 | node = wq_stack_extract(&ctx->submit_state.free_list); | |
c2b6c6bc | 2256 | return container_of(node, struct io_kiocb, comp_list); |
2b188cc1 JA |
2257 | } |
2258 | ||
e1d767f0 | 2259 | static inline void io_put_file(struct file *file) |
8da11c19 | 2260 | { |
e1d767f0 | 2261 | if (file) |
8da11c19 PB |
2262 | fput(file); |
2263 | } | |
2264 | ||
6b639522 | 2265 | static inline void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2266 | { |
094bae49 | 2267 | unsigned int flags = req->flags; |
929a3af9 | 2268 | |
867f8fa5 | 2269 | if (unlikely(flags & IO_REQ_CLEAN_FLAGS)) |
3a0a6902 | 2270 | io_clean_op(req); |
e1d767f0 PB |
2271 | if (!(flags & REQ_F_FIXED_FILE)) |
2272 | io_put_file(req->file); | |
e65ef56d JA |
2273 | } |
2274 | ||
c072481d | 2275 | static __cold void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2276 | { |
51a4cc11 | 2277 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2278 | |
ab409402 | 2279 | io_req_put_rsrc(req, ctx); |
216578e5 | 2280 | io_dismantle_req(req); |
7c660731 | 2281 | io_put_task(req->task, 1); |
c6ca97b3 | 2282 | |
79ebeaee | 2283 | spin_lock(&ctx->completion_lock); |
c2b6c6bc | 2284 | wq_list_add_head(&req->comp_list, &ctx->locked_free_list); |
c34b025f | 2285 | ctx->locked_free_nr++; |
79ebeaee | 2286 | spin_unlock(&ctx->completion_lock); |
e65ef56d JA |
2287 | } |
2288 | ||
f2f87370 PB |
2289 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2290 | { | |
2291 | struct io_kiocb *nxt = req->link; | |
2292 | ||
2293 | req->link = nxt->link; | |
2294 | nxt->link = NULL; | |
2295 | } | |
2296 | ||
33cc89a9 PB |
2297 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
2298 | __must_hold(&req->ctx->completion_lock) | |
89b263f6 | 2299 | __must_hold(&req->ctx->timeout_lock) |
2665abfd | 2300 | { |
33cc89a9 | 2301 | struct io_kiocb *link = req->link; |
f2f87370 | 2302 | |
b97e736a | 2303 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
c9abd7ad | 2304 | struct io_timeout_data *io = link->async_data; |
7c86ffee | 2305 | |
f2f87370 | 2306 | io_remove_next_linked(req); |
90cd7e42 | 2307 | link->timeout.head = NULL; |
fd9c7bc5 | 2308 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ef9dd637 | 2309 | list_del(&link->timeout.list); |
04c76b41 | 2310 | /* leave REQ_F_CQE_SKIP to io_fill_cqe_req */ |
913a571a | 2311 | io_fill_cqe_req(link, -ECANCELED, 0); |
91c2f697 | 2312 | io_put_req_deferred(link); |
d4729fbd | 2313 | return true; |
c9abd7ad PB |
2314 | } |
2315 | } | |
d4729fbd | 2316 | return false; |
7c86ffee PB |
2317 | } |
2318 | ||
d148ca4b | 2319 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 2320 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 2321 | { |
33cc89a9 | 2322 | struct io_kiocb *nxt, *link = req->link; |
04c76b41 | 2323 | bool ignore_cqes = req->flags & REQ_F_SKIP_LINK_CQES; |
9e645e11 | 2324 | |
f2f87370 | 2325 | req->link = NULL; |
f2f87370 | 2326 | while (link) { |
a8295b98 HX |
2327 | long res = -ECANCELED; |
2328 | ||
2329 | if (link->flags & REQ_F_FAIL) | |
2330 | res = link->result; | |
2331 | ||
f2f87370 PB |
2332 | nxt = link->link; |
2333 | link->link = NULL; | |
2665abfd | 2334 | |
502c87d6 SR |
2335 | trace_io_uring_fail_link(req->ctx, req, req->user_data, |
2336 | req->opcode, link); | |
2337 | ||
04c76b41 PB |
2338 | if (!ignore_cqes) { |
2339 | link->flags &= ~REQ_F_CQE_SKIP; | |
2340 | io_fill_cqe_req(link, res, 0); | |
2341 | } | |
91c2f697 | 2342 | io_put_req_deferred(link); |
f2f87370 | 2343 | link = nxt; |
9e645e11 | 2344 | } |
33cc89a9 | 2345 | } |
9e645e11 | 2346 | |
33cc89a9 PB |
2347 | static bool io_disarm_next(struct io_kiocb *req) |
2348 | __must_hold(&req->ctx->completion_lock) | |
2349 | { | |
2350 | bool posted = false; | |
2351 | ||
0756a869 PB |
2352 | if (req->flags & REQ_F_ARM_LTIMEOUT) { |
2353 | struct io_kiocb *link = req->link; | |
2354 | ||
906c6caa | 2355 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
0756a869 PB |
2356 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
2357 | io_remove_next_linked(req); | |
04c76b41 | 2358 | /* leave REQ_F_CQE_SKIP to io_fill_cqe_req */ |
913a571a | 2359 | io_fill_cqe_req(link, -ECANCELED, 0); |
0756a869 PB |
2360 | io_put_req_deferred(link); |
2361 | posted = true; | |
2362 | } | |
2363 | } else if (req->flags & REQ_F_LINK_TIMEOUT) { | |
89b263f6 JA |
2364 | struct io_ring_ctx *ctx = req->ctx; |
2365 | ||
2366 | spin_lock_irq(&ctx->timeout_lock); | |
33cc89a9 | 2367 | posted = io_kill_linked_timeout(req); |
89b263f6 JA |
2368 | spin_unlock_irq(&ctx->timeout_lock); |
2369 | } | |
93d2bcd2 | 2370 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 2371 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
2372 | posted |= (req->link != NULL); |
2373 | io_fail_links(req); | |
2374 | } | |
2375 | return posted; | |
9e645e11 JA |
2376 | } |
2377 | ||
d81499bf PB |
2378 | static void __io_req_find_next_prep(struct io_kiocb *req) |
2379 | { | |
2380 | struct io_ring_ctx *ctx = req->ctx; | |
2381 | bool posted; | |
2382 | ||
2383 | spin_lock(&ctx->completion_lock); | |
2384 | posted = io_disarm_next(req); | |
2385 | if (posted) | |
33ce2aff | 2386 | io_commit_cqring(ctx); |
d81499bf PB |
2387 | spin_unlock(&ctx->completion_lock); |
2388 | if (posted) | |
2389 | io_cqring_ev_posted(ctx); | |
2390 | } | |
2391 | ||
2392 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) | |
c69f8dbe | 2393 | { |
33cc89a9 | 2394 | struct io_kiocb *nxt; |
944e58bf | 2395 | |
d81499bf PB |
2396 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
2397 | return NULL; | |
9e645e11 JA |
2398 | /* |
2399 | * If LINK is set, we have dependent requests in this chain. If we | |
2400 | * didn't fail this request, queue the first one up, moving any other | |
2401 | * dependencies to the next request. In case of failure, fail the rest | |
2402 | * of the chain. | |
2403 | */ | |
d81499bf PB |
2404 | if (unlikely(req->flags & IO_DISARM_MASK)) |
2405 | __io_req_find_next_prep(req); | |
33cc89a9 PB |
2406 | nxt = req->link; |
2407 | req->link = NULL; | |
2408 | return nxt; | |
4d7dd462 | 2409 | } |
9e645e11 | 2410 | |
f237c30a | 2411 | static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked) |
2c32395d PB |
2412 | { |
2413 | if (!ctx) | |
2414 | return; | |
f237c30a | 2415 | if (*locked) { |
c450178d | 2416 | io_submit_flush_completions(ctx); |
2c32395d | 2417 | mutex_unlock(&ctx->uring_lock); |
f237c30a | 2418 | *locked = false; |
2c32395d PB |
2419 | } |
2420 | percpu_ref_put(&ctx->refs); | |
2421 | } | |
2422 | ||
f28c240e HX |
2423 | static inline void ctx_commit_and_unlock(struct io_ring_ctx *ctx) |
2424 | { | |
2425 | io_commit_cqring(ctx); | |
2426 | spin_unlock(&ctx->completion_lock); | |
2427 | io_cqring_ev_posted(ctx); | |
2428 | } | |
2429 | ||
2430 | static void handle_prev_tw_list(struct io_wq_work_node *node, | |
2431 | struct io_ring_ctx **ctx, bool *uring_locked) | |
2432 | { | |
2433 | if (*ctx && !*uring_locked) | |
2434 | spin_lock(&(*ctx)->completion_lock); | |
2435 | ||
2436 | do { | |
2437 | struct io_wq_work_node *next = node->next; | |
2438 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2439 | io_task_work.node); | |
2440 | ||
2441 | if (req->ctx != *ctx) { | |
2442 | if (unlikely(!*uring_locked && *ctx)) | |
2443 | ctx_commit_and_unlock(*ctx); | |
2444 | ||
2445 | ctx_flush_and_put(*ctx, uring_locked); | |
2446 | *ctx = req->ctx; | |
2447 | /* if not contended, grab and improve batching */ | |
2448 | *uring_locked = mutex_trylock(&(*ctx)->uring_lock); | |
2449 | percpu_ref_get(&(*ctx)->refs); | |
2450 | if (unlikely(!*uring_locked)) | |
2451 | spin_lock(&(*ctx)->completion_lock); | |
2452 | } | |
2453 | if (likely(*uring_locked)) | |
2454 | req->io_task_work.func(req, uring_locked); | |
2455 | else | |
cc3cec83 JA |
2456 | __io_req_complete_post(req, req->result, |
2457 | io_put_kbuf_comp(req)); | |
f28c240e HX |
2458 | node = next; |
2459 | } while (node); | |
2460 | ||
2461 | if (unlikely(!*uring_locked)) | |
2462 | ctx_commit_and_unlock(*ctx); | |
2463 | } | |
2464 | ||
2465 | static void handle_tw_list(struct io_wq_work_node *node, | |
2466 | struct io_ring_ctx **ctx, bool *locked) | |
9f8d032a HX |
2467 | { |
2468 | do { | |
2469 | struct io_wq_work_node *next = node->next; | |
2470 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2471 | io_task_work.node); | |
2472 | ||
2473 | if (req->ctx != *ctx) { | |
2474 | ctx_flush_and_put(*ctx, locked); | |
2475 | *ctx = req->ctx; | |
2476 | /* if not contended, grab and improve batching */ | |
2477 | *locked = mutex_trylock(&(*ctx)->uring_lock); | |
2478 | percpu_ref_get(&(*ctx)->refs); | |
2479 | } | |
2480 | req->io_task_work.func(req, locked); | |
2481 | node = next; | |
2482 | } while (node); | |
2483 | } | |
2484 | ||
7cbf1722 | 2485 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2486 | { |
f28c240e | 2487 | bool uring_locked = false; |
ebd0df2e | 2488 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
2489 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
2490 | task_work); | |
c40f6379 | 2491 | |
16f72070 | 2492 | while (1) { |
f28c240e | 2493 | struct io_wq_work_node *node1, *node2; |
3f18407d | 2494 | |
f28c240e HX |
2495 | if (!tctx->task_list.first && |
2496 | !tctx->prior_task_list.first && uring_locked) | |
8d4ad41e PB |
2497 | io_submit_flush_completions(ctx); |
2498 | ||
3f18407d | 2499 | spin_lock_irq(&tctx->task_lock); |
f28c240e HX |
2500 | node1 = tctx->prior_task_list.first; |
2501 | node2 = tctx->task_list.first; | |
3f18407d | 2502 | INIT_WQ_LIST(&tctx->task_list); |
f28c240e HX |
2503 | INIT_WQ_LIST(&tctx->prior_task_list); |
2504 | if (!node2 && !node1) | |
6294f368 | 2505 | tctx->task_running = false; |
3f18407d | 2506 | spin_unlock_irq(&tctx->task_lock); |
f28c240e | 2507 | if (!node2 && !node1) |
6294f368 | 2508 | break; |
3f18407d | 2509 | |
f28c240e HX |
2510 | if (node1) |
2511 | handle_prev_tw_list(node1, &ctx, &uring_locked); | |
6294f368 | 2512 | |
f28c240e HX |
2513 | if (node2) |
2514 | handle_tw_list(node2, &ctx, &uring_locked); | |
7cbf1722 | 2515 | cond_resched(); |
3f18407d | 2516 | } |
ebd0df2e | 2517 | |
f28c240e | 2518 | ctx_flush_and_put(ctx, &uring_locked); |
3cc7fdb9 PB |
2519 | |
2520 | /* relaxed read is enough as only the task itself sets ->in_idle */ | |
2521 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2522 | io_uring_drop_tctx_refs(current); | |
7cbf1722 JA |
2523 | } |
2524 | ||
4813c377 | 2525 | static void io_req_task_work_add(struct io_kiocb *req, bool priority) |
7cbf1722 | 2526 | { |
c15b79de | 2527 | struct task_struct *tsk = req->task; |
7cbf1722 | 2528 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 2529 | enum task_work_notify_mode notify; |
e09ee510 | 2530 | struct io_wq_work_node *node; |
0b81e80c | 2531 | unsigned long flags; |
6294f368 | 2532 | bool running; |
7cbf1722 JA |
2533 | |
2534 | WARN_ON_ONCE(!tctx); | |
2535 | ||
0b81e80c | 2536 | spin_lock_irqsave(&tctx->task_lock, flags); |
4813c377 HX |
2537 | if (priority) |
2538 | wq_list_add_tail(&req->io_task_work.node, &tctx->prior_task_list); | |
2539 | else | |
2540 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); | |
6294f368 PB |
2541 | running = tctx->task_running; |
2542 | if (!running) | |
2543 | tctx->task_running = true; | |
0b81e80c | 2544 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2545 | |
2546 | /* task_work already pending, we're done */ | |
6294f368 | 2547 | if (running) |
e09ee510 | 2548 | return; |
7cbf1722 | 2549 | |
c15b79de PB |
2550 | /* |
2551 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
2552 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2553 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2554 | * will do the job. | |
2555 | */ | |
2556 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
d97ec623 PB |
2557 | if (likely(!task_work_add(tsk, &tctx->task_work, notify))) { |
2558 | if (notify == TWA_NONE) | |
2559 | wake_up_process(tsk); | |
e09ee510 | 2560 | return; |
c15b79de | 2561 | } |
2215bed9 | 2562 | |
0b81e80c | 2563 | spin_lock_irqsave(&tctx->task_lock, flags); |
6294f368 | 2564 | tctx->task_running = false; |
4813c377 | 2565 | node = wq_list_merge(&tctx->prior_task_list, &tctx->task_list); |
0b81e80c | 2566 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 2567 | |
e09ee510 PB |
2568 | while (node) { |
2569 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
2570 | node = node->next; | |
2571 | if (llist_add(&req->io_task_work.fallback_node, | |
2572 | &req->ctx->fallback_llist)) | |
2573 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
2574 | } | |
eab30c4d PB |
2575 | } |
2576 | ||
f237c30a | 2577 | static void io_req_task_cancel(struct io_kiocb *req, bool *locked) |
c40f6379 | 2578 | { |
87ceb6a6 | 2579 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2580 | |
b18a1a45 | 2581 | /* not needed for normal modes, but SQPOLL depends on it */ |
f237c30a | 2582 | io_tw_lock(ctx, locked); |
2593553a | 2583 | io_req_complete_failed(req, req->result); |
c40f6379 JA |
2584 | } |
2585 | ||
f237c30a | 2586 | static void io_req_task_submit(struct io_kiocb *req, bool *locked) |
c40f6379 JA |
2587 | { |
2588 | struct io_ring_ctx *ctx = req->ctx; | |
2589 | ||
f237c30a | 2590 | io_tw_lock(ctx, locked); |
316319e8 | 2591 | /* req->task == current here, checking PF_EXITING is safe */ |
af066f31 | 2592 | if (likely(!(req->task->flags & PF_EXITING))) |
c5eef2b9 | 2593 | __io_queue_sqe(req); |
81b6d05c | 2594 | else |
2593553a | 2595 | io_req_complete_failed(req, -EFAULT); |
c40f6379 JA |
2596 | } |
2597 | ||
2c4b8eb6 | 2598 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2599 | { |
2c4b8eb6 | 2600 | req->result = ret; |
5b0a6acc | 2601 | req->io_task_work.func = io_req_task_cancel; |
4813c377 | 2602 | io_req_task_work_add(req, false); |
c40f6379 JA |
2603 | } |
2604 | ||
2c4b8eb6 | 2605 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2606 | { |
5b0a6acc | 2607 | req->io_task_work.func = io_req_task_submit; |
4813c377 | 2608 | io_req_task_work_add(req, false); |
a3df7698 PB |
2609 | } |
2610 | ||
773af691 JA |
2611 | static void io_req_task_queue_reissue(struct io_kiocb *req) |
2612 | { | |
2613 | req->io_task_work.func = io_queue_async_work; | |
4813c377 | 2614 | io_req_task_work_add(req, false); |
773af691 JA |
2615 | } |
2616 | ||
f2f87370 | 2617 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2618 | { |
9b5f7bd9 | 2619 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2620 | |
2621 | if (nxt) | |
906a8c3f | 2622 | io_req_task_queue(nxt); |
c69f8dbe JL |
2623 | } |
2624 | ||
c3524383 | 2625 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2626 | { |
c3524383 PB |
2627 | io_queue_next(req); |
2628 | __io_free_req(req); | |
2629 | } | |
8766dd51 | 2630 | |
f237c30a PB |
2631 | static void io_free_req_work(struct io_kiocb *req, bool *locked) |
2632 | { | |
2633 | io_free_req(req); | |
2634 | } | |
2635 | ||
3aa83bfb | 2636 | static void io_free_batch_list(struct io_ring_ctx *ctx, |
1cce17ac | 2637 | struct io_wq_work_node *node) |
3aa83bfb | 2638 | __must_hold(&ctx->uring_lock) |
5af1d13e | 2639 | { |
d4b7a5ef | 2640 | struct task_struct *task = NULL; |
37f0e767 | 2641 | int task_refs = 0; |
5af1d13e | 2642 | |
3aa83bfb PB |
2643 | do { |
2644 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2645 | comp_list); | |
2d6500d4 | 2646 | |
def77acf | 2647 | if (unlikely(req->flags & REQ_F_REFCOUNT)) { |
c1e53a69 | 2648 | node = req->comp_list.next; |
def77acf PB |
2649 | if (!req_ref_put_and_test(req)) |
2650 | continue; | |
c1e53a69 | 2651 | } |
2d6500d4 | 2652 | |
ab409402 | 2653 | io_req_put_rsrc_locked(req, ctx); |
d4b7a5ef PB |
2654 | io_queue_next(req); |
2655 | io_dismantle_req(req); | |
5af1d13e | 2656 | |
d4b7a5ef PB |
2657 | if (req->task != task) { |
2658 | if (task) | |
2659 | io_put_task(task, task_refs); | |
2660 | task = req->task; | |
2661 | task_refs = 0; | |
2662 | } | |
2663 | task_refs++; | |
c1e53a69 | 2664 | node = req->comp_list.next; |
d4b7a5ef | 2665 | wq_stack_add_head(&req->comp_list, &ctx->submit_state.free_list); |
3aa83bfb | 2666 | } while (node); |
d4b7a5ef | 2667 | |
d4b7a5ef PB |
2668 | if (task) |
2669 | io_put_task(task, task_refs); | |
7a743e22 PB |
2670 | } |
2671 | ||
c450178d | 2672 | static void __io_submit_flush_completions(struct io_ring_ctx *ctx) |
a141dd89 | 2673 | __must_hold(&ctx->uring_lock) |
905c172f | 2674 | { |
6f33b0bc | 2675 | struct io_wq_work_node *node, *prev; |
cd0ca2e0 | 2676 | struct io_submit_state *state = &ctx->submit_state; |
905c172f | 2677 | |
3d4aeb9f PB |
2678 | if (state->flush_cqes) { |
2679 | spin_lock(&ctx->completion_lock); | |
2680 | wq_list_for_each(node, prev, &state->compl_reqs) { | |
2681 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
6f33b0bc | 2682 | comp_list); |
5182ed2e | 2683 | |
3d4aeb9f | 2684 | if (!(req->flags & REQ_F_CQE_SKIP)) |
ae4da189 | 2685 | __io_fill_cqe_req(req, req->result, req->cflags); |
4d9237e3 JA |
2686 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
2687 | struct async_poll *apoll = req->apoll; | |
2688 | ||
2689 | if (apoll->double_poll) | |
2690 | kfree(apoll->double_poll); | |
2691 | list_add(&apoll->poll.wait.entry, | |
2692 | &ctx->apoll_cache); | |
2693 | req->flags &= ~REQ_F_POLLED; | |
2694 | } | |
3d4aeb9f PB |
2695 | } |
2696 | ||
2697 | io_commit_cqring(ctx); | |
2698 | spin_unlock(&ctx->completion_lock); | |
2699 | io_cqring_ev_posted(ctx); | |
2700 | state->flush_cqes = false; | |
905c172f | 2701 | } |
5182ed2e | 2702 | |
1cce17ac | 2703 | io_free_batch_list(ctx, state->compl_reqs.first); |
6f33b0bc | 2704 | INIT_WQ_LIST(&state->compl_reqs); |
7a743e22 PB |
2705 | } |
2706 | ||
ba816ad6 JA |
2707 | /* |
2708 | * Drop reference to request, return next in chain (if there is one) if this | |
2709 | * was the last reference to this request. | |
2710 | */ | |
0d85035a | 2711 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2712 | { |
9b5f7bd9 PB |
2713 | struct io_kiocb *nxt = NULL; |
2714 | ||
de9b4cca | 2715 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2716 | nxt = io_req_find_next(req); |
4d7dd462 | 2717 | __io_free_req(req); |
2a44f467 | 2718 | } |
9b5f7bd9 | 2719 | return nxt; |
2b188cc1 JA |
2720 | } |
2721 | ||
0d85035a | 2722 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2723 | { |
de9b4cca | 2724 | if (req_ref_put_and_test(req)) |
e65ef56d | 2725 | io_free_req(req); |
2b188cc1 JA |
2726 | } |
2727 | ||
91c2f697 | 2728 | static inline void io_put_req_deferred(struct io_kiocb *req) |
216578e5 | 2729 | { |
91c2f697 | 2730 | if (req_ref_put_and_test(req)) { |
f237c30a | 2731 | req->io_task_work.func = io_free_req_work; |
4813c377 | 2732 | io_req_task_work_add(req, false); |
543af3a1 | 2733 | } |
216578e5 PB |
2734 | } |
2735 | ||
6c503150 | 2736 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2737 | { |
2738 | /* See comment at the top of this file */ | |
2739 | smp_rmb(); | |
e23de15f | 2740 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2741 | } |
2742 | ||
fb5ccc98 PB |
2743 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2744 | { | |
2745 | struct io_rings *rings = ctx->rings; | |
2746 | ||
2747 | /* make sure SQ entry isn't read before tail */ | |
2748 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2749 | } | |
2750 | ||
4c6e277c JA |
2751 | static inline bool io_run_task_work(void) |
2752 | { | |
ef98eb04 | 2753 | if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) { |
4c6e277c | 2754 | __set_current_state(TASK_RUNNING); |
ef98eb04 | 2755 | tracehook_notify_signal(); |
4c6e277c JA |
2756 | return true; |
2757 | } | |
2758 | ||
2759 | return false; | |
bcda7baa JA |
2760 | } |
2761 | ||
5ba3c874 | 2762 | static int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin) |
def596e9 | 2763 | { |
5eef4e87 | 2764 | struct io_wq_work_node *pos, *start, *prev; |
d729cf9a | 2765 | unsigned int poll_flags = BLK_POLL_NOSLEEP; |
b688f11e | 2766 | DEFINE_IO_COMP_BATCH(iob); |
5ba3c874 | 2767 | int nr_events = 0; |
def596e9 JA |
2768 | |
2769 | /* | |
2770 | * Only spin for completions if we don't have multiple devices hanging | |
87a115fb | 2771 | * off our complete list. |
def596e9 | 2772 | */ |
87a115fb | 2773 | if (ctx->poll_multi_queue || force_nonspin) |
ef99b2d3 | 2774 | poll_flags |= BLK_POLL_ONESHOT; |
def596e9 | 2775 | |
5eef4e87 PB |
2776 | wq_list_for_each(pos, start, &ctx->iopoll_list) { |
2777 | struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list); | |
9adbd45d | 2778 | struct kiocb *kiocb = &req->rw.kiocb; |
a2416e1e | 2779 | int ret; |
def596e9 JA |
2780 | |
2781 | /* | |
581f9810 BM |
2782 | * Move completed and retryable entries to our local lists. |
2783 | * If we find a request that requires polling, break out | |
2784 | * and complete those lists first, if we have entries there. | |
def596e9 | 2785 | */ |
e3f721e6 | 2786 | if (READ_ONCE(req->iopoll_completed)) |
def596e9 JA |
2787 | break; |
2788 | ||
b688f11e | 2789 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, &iob, poll_flags); |
a2416e1e PB |
2790 | if (unlikely(ret < 0)) |
2791 | return ret; | |
2792 | else if (ret) | |
ef99b2d3 | 2793 | poll_flags |= BLK_POLL_ONESHOT; |
def596e9 | 2794 | |
3aadc23e | 2795 | /* iopoll may have completed current req */ |
b688f11e JA |
2796 | if (!rq_list_empty(iob.req_list) || |
2797 | READ_ONCE(req->iopoll_completed)) | |
e3f721e6 | 2798 | break; |
def596e9 JA |
2799 | } |
2800 | ||
b688f11e JA |
2801 | if (!rq_list_empty(iob.req_list)) |
2802 | iob.complete(&iob); | |
5eef4e87 PB |
2803 | else if (!pos) |
2804 | return 0; | |
def596e9 | 2805 | |
5eef4e87 PB |
2806 | prev = start; |
2807 | wq_list_for_each_resume(pos, prev) { | |
2808 | struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list); | |
2809 | ||
b3fa03fd PB |
2810 | /* order with io_complete_rw_iopoll(), e.g. ->result updates */ |
2811 | if (!smp_load_acquire(&req->iopoll_completed)) | |
e3f721e6 | 2812 | break; |
83a13a41 PB |
2813 | if (unlikely(req->flags & REQ_F_CQE_SKIP)) |
2814 | continue; | |
d1fd1c20 | 2815 | |
ae4da189 | 2816 | __io_fill_cqe_req(req, req->result, io_put_kbuf(req, 0)); |
e3f721e6 PB |
2817 | nr_events++; |
2818 | } | |
def596e9 | 2819 | |
f5ed3bcd PB |
2820 | if (unlikely(!nr_events)) |
2821 | return 0; | |
2822 | ||
2823 | io_commit_cqring(ctx); | |
2824 | io_cqring_ev_posted_iopoll(ctx); | |
1cce17ac | 2825 | pos = start ? start->next : ctx->iopoll_list.first; |
5eef4e87 | 2826 | wq_list_cut(&ctx->iopoll_list, prev, start); |
1cce17ac | 2827 | io_free_batch_list(ctx, pos); |
5ba3c874 | 2828 | return nr_events; |
def596e9 JA |
2829 | } |
2830 | ||
def596e9 JA |
2831 | /* |
2832 | * We can't just wait for polled events to come to us, we have to actively | |
2833 | * find and complete them. | |
2834 | */ | |
c072481d | 2835 | static __cold void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2836 | { |
2837 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2838 | return; | |
2839 | ||
2840 | mutex_lock(&ctx->uring_lock); | |
5eef4e87 | 2841 | while (!wq_list_empty(&ctx->iopoll_list)) { |
b2edc0a7 | 2842 | /* let it sleep and repeat later if can't complete a request */ |
5ba3c874 | 2843 | if (io_do_iopoll(ctx, true) == 0) |
b2edc0a7 | 2844 | break; |
08f5439f JA |
2845 | /* |
2846 | * Ensure we allow local-to-the-cpu processing to take place, | |
2847 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2848 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2849 | */ |
3fcee5a6 PB |
2850 | if (need_resched()) { |
2851 | mutex_unlock(&ctx->uring_lock); | |
2852 | cond_resched(); | |
2853 | mutex_lock(&ctx->uring_lock); | |
2854 | } | |
def596e9 JA |
2855 | } |
2856 | mutex_unlock(&ctx->uring_lock); | |
2857 | } | |
2858 | ||
7668b92a | 2859 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2860 | { |
7668b92a | 2861 | unsigned int nr_events = 0; |
e9979b36 | 2862 | int ret = 0; |
500f9fba | 2863 | |
c7849be9 XW |
2864 | /* |
2865 | * We disallow the app entering submit/complete with polling, but we | |
2866 | * still need to lock the ring to prevent racing with polled issue | |
2867 | * that got punted to a workqueue. | |
2868 | */ | |
2869 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2870 | /* |
2871 | * Don't enter poll loop if we already have events pending. | |
2872 | * If we do, we can potentially be spinning for commands that | |
2873 | * already triggered a CQE (eg in error). | |
2874 | */ | |
5ed7a37d | 2875 | if (test_bit(0, &ctx->check_cq_overflow)) |
f39c8a5b PB |
2876 | __io_cqring_overflow_flush(ctx, false); |
2877 | if (io_cqring_events(ctx)) | |
2878 | goto out; | |
def596e9 | 2879 | do { |
500f9fba JA |
2880 | /* |
2881 | * If a submit got punted to a workqueue, we can have the | |
2882 | * application entering polling for a command before it gets | |
2883 | * issued. That app will hold the uring_lock for the duration | |
2884 | * of the poll right here, so we need to take a breather every | |
2885 | * now and then to ensure that the issue has a chance to add | |
2886 | * the poll to the issued list. Otherwise we can spin here | |
2887 | * forever, while the workqueue is stuck trying to acquire the | |
2888 | * very same mutex. | |
2889 | */ | |
5eef4e87 | 2890 | if (wq_list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
2891 | u32 tail = ctx->cached_cq_tail; |
2892 | ||
500f9fba | 2893 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2894 | io_run_task_work(); |
500f9fba | 2895 | mutex_lock(&ctx->uring_lock); |
def596e9 | 2896 | |
8f487ef2 PB |
2897 | /* some requests don't go through iopoll_list */ |
2898 | if (tail != ctx->cached_cq_tail || | |
5eef4e87 | 2899 | wq_list_empty(&ctx->iopoll_list)) |
e9979b36 | 2900 | break; |
500f9fba | 2901 | } |
5ba3c874 PB |
2902 | ret = io_do_iopoll(ctx, !min); |
2903 | if (ret < 0) | |
2904 | break; | |
2905 | nr_events += ret; | |
2906 | ret = 0; | |
2907 | } while (nr_events < min && !need_resched()); | |
f39c8a5b | 2908 | out: |
500f9fba | 2909 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2910 | return ret; |
2911 | } | |
2912 | ||
491381ce | 2913 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2914 | { |
491381ce JA |
2915 | /* |
2916 | * Tell lockdep we inherited freeze protection from submission | |
2917 | * thread. | |
2918 | */ | |
2919 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2920 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2921 | |
1c98679d PB |
2922 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2923 | sb_end_write(sb); | |
2b188cc1 JA |
2924 | } |
2925 | } | |
2926 | ||
b63534c4 | 2927 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2928 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2929 | { |
ab454438 | 2930 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 2931 | |
d886e185 | 2932 | if (!req_has_async_data(req)) |
ab454438 | 2933 | return !io_req_prep_async(req); |
538941e2 | 2934 | iov_iter_restore(&rw->s.iter, &rw->s.iter_state); |
ab454438 | 2935 | return true; |
b63534c4 | 2936 | } |
b63534c4 | 2937 | |
3e6a0d3c | 2938 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2939 | { |
355afaeb | 2940 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2941 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2942 | |
355afaeb JA |
2943 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2944 | return false; | |
3e6a0d3c JA |
2945 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2946 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2947 | return false; |
7c977a58 JA |
2948 | /* |
2949 | * If ref is dying, we might be running poll reap from the exit work. | |
2950 | * Don't attempt to reissue from that path, just let it fail with | |
2951 | * -EAGAIN. | |
2952 | */ | |
3e6a0d3c JA |
2953 | if (percpu_ref_is_dying(&ctx->refs)) |
2954 | return false; | |
ef046888 JA |
2955 | /* |
2956 | * Play it safe and assume not safe to re-import and reissue if we're | |
2957 | * not in the original thread group (or in task context). | |
2958 | */ | |
2959 | if (!same_thread_group(req->task, current) || !in_task()) | |
2960 | return false; | |
3e6a0d3c JA |
2961 | return true; |
2962 | } | |
e82ad485 | 2963 | #else |
a1ff1e3f | 2964 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
2965 | { |
2966 | return false; | |
2967 | } | |
e82ad485 | 2968 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 2969 | { |
b63534c4 JA |
2970 | return false; |
2971 | } | |
3e6a0d3c | 2972 | #endif |
b63534c4 | 2973 | |
8ef12efe | 2974 | static bool __io_complete_rw_common(struct io_kiocb *req, long res) |
a1d7c393 | 2975 | { |
f63cf519 | 2976 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) { |
b65c128f | 2977 | kiocb_end_write(req); |
f63cf519 JA |
2978 | fsnotify_modify(req->file); |
2979 | } else { | |
2980 | fsnotify_access(req->file); | |
2981 | } | |
258f3a7f | 2982 | if (unlikely(res != req->result)) { |
9532b99b PB |
2983 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && |
2984 | io_rw_should_reissue(req)) { | |
2985 | req->flags |= REQ_F_REISSUE; | |
8ef12efe | 2986 | return true; |
9532b99b | 2987 | } |
93d2bcd2 | 2988 | req_set_fail(req); |
8ef12efe | 2989 | req->result = res; |
9532b99b | 2990 | } |
8ef12efe JA |
2991 | return false; |
2992 | } | |
2993 | ||
cc8e9ba7 | 2994 | static inline void io_req_task_complete(struct io_kiocb *req, bool *locked) |
8ef12efe | 2995 | { |
54daa9b2 | 2996 | int res = req->result; |
126180b9 PB |
2997 | |
2998 | if (*locked) { | |
cc3cec83 | 2999 | io_req_complete_state(req, res, io_put_kbuf(req, 0)); |
fff4e40e | 3000 | io_req_add_compl_list(req); |
126180b9 | 3001 | } else { |
cc3cec83 JA |
3002 | io_req_complete_post(req, res, |
3003 | io_put_kbuf(req, IO_URING_F_UNLOCKED)); | |
126180b9 | 3004 | } |
8ef12efe JA |
3005 | } |
3006 | ||
00f6e68b | 3007 | static void __io_complete_rw(struct io_kiocb *req, long res, |
8ef12efe JA |
3008 | unsigned int issue_flags) |
3009 | { | |
3010 | if (__io_complete_rw_common(req, res)) | |
3011 | return; | |
cc3cec83 JA |
3012 | __io_req_complete(req, issue_flags, req->result, |
3013 | io_put_kbuf(req, issue_flags)); | |
ba816ad6 JA |
3014 | } |
3015 | ||
6b19b766 | 3016 | static void io_complete_rw(struct kiocb *kiocb, long res) |
ba816ad6 | 3017 | { |
9adbd45d | 3018 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 3019 | |
8ef12efe JA |
3020 | if (__io_complete_rw_common(req, res)) |
3021 | return; | |
3022 | req->result = res; | |
3023 | req->io_task_work.func = io_req_task_complete; | |
f28c240e | 3024 | io_req_task_work_add(req, !!(req->ctx->flags & IORING_SETUP_SQPOLL)); |
2b188cc1 JA |
3025 | } |
3026 | ||
6b19b766 | 3027 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res) |
def596e9 | 3028 | { |
9adbd45d | 3029 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 3030 | |
491381ce JA |
3031 | if (kiocb->ki_flags & IOCB_WRITE) |
3032 | kiocb_end_write(req); | |
9532b99b | 3033 | if (unlikely(res != req->result)) { |
b66ceaf3 PB |
3034 | if (res == -EAGAIN && io_rw_should_reissue(req)) { |
3035 | req->flags |= REQ_F_REISSUE; | |
3036 | return; | |
9532b99b | 3037 | } |
258f3a7f | 3038 | req->result = res; |
8c130827 | 3039 | } |
bbde017a | 3040 | |
b3fa03fd PB |
3041 | /* order with io_iopoll_complete() checking ->iopoll_completed */ |
3042 | smp_store_release(&req->iopoll_completed, 1); | |
def596e9 JA |
3043 | } |
3044 | ||
3045 | /* | |
3046 | * After the iocb has been issued, it's safe to be found on the poll list. | |
3047 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 3048 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
3049 | * accessing the kiocb cookie. |
3050 | */ | |
9882131c | 3051 | static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags) |
def596e9 JA |
3052 | { |
3053 | struct io_ring_ctx *ctx = req->ctx; | |
3b44b371 | 3054 | const bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
cb3d8972 PB |
3055 | |
3056 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
3b44b371 | 3057 | if (unlikely(needs_lock)) |
cb3d8972 | 3058 | mutex_lock(&ctx->uring_lock); |
def596e9 JA |
3059 | |
3060 | /* | |
3061 | * Track whether we have multiple files in our lists. This will impact | |
3062 | * how we do polling eventually, not spinning if we're on potentially | |
3063 | * different devices. | |
3064 | */ | |
5eef4e87 | 3065 | if (wq_list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
3066 | ctx->poll_multi_queue = false; |
3067 | } else if (!ctx->poll_multi_queue) { | |
def596e9 JA |
3068 | struct io_kiocb *list_req; |
3069 | ||
5eef4e87 PB |
3070 | list_req = container_of(ctx->iopoll_list.first, struct io_kiocb, |
3071 | comp_list); | |
30da1b45 | 3072 | if (list_req->file != req->file) |
915b3dde | 3073 | ctx->poll_multi_queue = true; |
def596e9 JA |
3074 | } |
3075 | ||
3076 | /* | |
3077 | * For fast devices, IO may have already completed. If it has, add | |
3078 | * it to the front so we find it first. | |
3079 | */ | |
65a6543d | 3080 | if (READ_ONCE(req->iopoll_completed)) |
5eef4e87 | 3081 | wq_list_add_head(&req->comp_list, &ctx->iopoll_list); |
def596e9 | 3082 | else |
5eef4e87 | 3083 | wq_list_add_tail(&req->comp_list, &ctx->iopoll_list); |
bdcd3eab | 3084 | |
3b44b371 | 3085 | if (unlikely(needs_lock)) { |
cb3d8972 PB |
3086 | /* |
3087 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
3088 | * in sq thread task context or in io worker task context. If | |
3089 | * current task context is sq thread, we don't need to check | |
3090 | * whether should wake up sq thread. | |
3091 | */ | |
3092 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
3093 | wq_has_sleeper(&ctx->sq_data->wait)) | |
3094 | wake_up(&ctx->sq_data->wait); | |
3095 | ||
3096 | mutex_unlock(&ctx->uring_lock); | |
3097 | } | |
def596e9 JA |
3098 | } |
3099 | ||
4503b767 JA |
3100 | static bool io_bdev_nowait(struct block_device *bdev) |
3101 | { | |
9ba0d0c8 | 3102 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
3103 | } |
3104 | ||
2b188cc1 JA |
3105 | /* |
3106 | * If we tracked the file through the SCM inflight mechanism, we could support | |
3107 | * any file. For now, just ensure that anything potentially problematic is done | |
3108 | * inline. | |
3109 | */ | |
88459b50 | 3110 | static bool __io_file_supports_nowait(struct file *file, umode_t mode) |
2b188cc1 | 3111 | { |
4503b767 | 3112 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
3113 | if (IS_ENABLED(CONFIG_BLOCK) && |
3114 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
3115 | return true; |
3116 | return false; | |
3117 | } | |
976517f1 | 3118 | if (S_ISSOCK(mode)) |
2b188cc1 | 3119 | return true; |
4503b767 | 3120 | if (S_ISREG(mode)) { |
4e7b5671 CH |
3121 | if (IS_ENABLED(CONFIG_BLOCK) && |
3122 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
3123 | file->f_op != &io_uring_fops) |
3124 | return true; | |
3125 | return false; | |
3126 | } | |
2b188cc1 | 3127 | |
c5b85625 JA |
3128 | /* any ->read/write should understand O_NONBLOCK */ |
3129 | if (file->f_flags & O_NONBLOCK) | |
3130 | return true; | |
35645ac3 | 3131 | return file->f_mode & FMODE_NOWAIT; |
2b188cc1 | 3132 | } |
c5b85625 | 3133 | |
88459b50 PB |
3134 | /* |
3135 | * If we tracked the file through the SCM inflight mechanism, we could support | |
3136 | * any file. For now, just ensure that anything potentially problematic is done | |
3137 | * inline. | |
3138 | */ | |
3139 | static unsigned int io_file_get_flags(struct file *file) | |
3140 | { | |
3141 | umode_t mode = file_inode(file)->i_mode; | |
3142 | unsigned int res = 0; | |
af197f50 | 3143 | |
88459b50 PB |
3144 | if (S_ISREG(mode)) |
3145 | res |= FFS_ISREG; | |
3146 | if (__io_file_supports_nowait(file, mode)) | |
3147 | res |= FFS_NOWAIT; | |
3148 | return res; | |
2b188cc1 JA |
3149 | } |
3150 | ||
35645ac3 | 3151 | static inline bool io_file_supports_nowait(struct io_kiocb *req) |
7b29f92d | 3152 | { |
88459b50 | 3153 | return req->flags & REQ_F_SUPPORT_NOWAIT; |
7b29f92d JA |
3154 | } |
3155 | ||
b9a6b8f9 | 3156 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 3157 | { |
def596e9 | 3158 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 3159 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 3160 | struct file *file = req->file; |
09bb8394 JA |
3161 | unsigned ioprio; |
3162 | int ret; | |
2b188cc1 | 3163 | |
88459b50 PB |
3164 | if (!io_req_ffs_set(req)) |
3165 | req->flags |= io_file_get_flags(file) << REQ_F_SUPPORT_NOWAIT_BIT; | |
491381ce | 3166 | |
2b188cc1 | 3167 | kiocb->ki_pos = READ_ONCE(sqe->off); |
5cb03d63 | 3168 | kiocb->ki_flags = iocb_flags(file); |
3e577dcd PB |
3169 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); |
3170 | if (unlikely(ret)) | |
3171 | return ret; | |
2b188cc1 | 3172 | |
5d329e12 JA |
3173 | /* |
3174 | * If the file is marked O_NONBLOCK, still allow retry for it if it | |
3175 | * supports async. Otherwise it's impossible to use O_NONBLOCK files | |
3176 | * reliably. If not, or it IOCB_NOWAIT is set, don't retry. | |
3177 | */ | |
3178 | if ((kiocb->ki_flags & IOCB_NOWAIT) || | |
35645ac3 | 3179 | ((file->f_flags & O_NONBLOCK) && !io_file_supports_nowait(req))) |
75c668cd PB |
3180 | req->flags |= REQ_F_NOWAIT; |
3181 | ||
def596e9 | 3182 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
5cb03d63 | 3183 | if (!(kiocb->ki_flags & IOCB_DIRECT) || !file->f_op->iopoll) |
09bb8394 | 3184 | return -EOPNOTSUPP; |
2b188cc1 | 3185 | |
394918eb | 3186 | kiocb->ki_flags |= IOCB_HIPRI | IOCB_ALLOC_CACHE; |
def596e9 | 3187 | kiocb->ki_complete = io_complete_rw_iopoll; |
65a6543d | 3188 | req->iopoll_completed = 0; |
def596e9 | 3189 | } else { |
09bb8394 JA |
3190 | if (kiocb->ki_flags & IOCB_HIPRI) |
3191 | return -EINVAL; | |
def596e9 JA |
3192 | kiocb->ki_complete = io_complete_rw; |
3193 | } | |
9adbd45d | 3194 | |
fb27274a PB |
3195 | ioprio = READ_ONCE(sqe->ioprio); |
3196 | if (ioprio) { | |
3197 | ret = ioprio_check_cap(ioprio); | |
3198 | if (ret) | |
3199 | return ret; | |
3200 | ||
3201 | kiocb->ki_ioprio = ioprio; | |
3202 | } else { | |
3203 | kiocb->ki_ioprio = get_current_ioprio(); | |
eae071c9 PB |
3204 | } |
3205 | ||
578c0ee2 | 3206 | req->imu = NULL; |
3529d8c2 JA |
3207 | req->rw.addr = READ_ONCE(sqe->addr); |
3208 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 3209 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 3210 | return 0; |
2b188cc1 JA |
3211 | } |
3212 | ||
3213 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
3214 | { | |
3215 | switch (ret) { | |
3216 | case -EIOCBQUEUED: | |
3217 | break; | |
3218 | case -ERESTARTSYS: | |
3219 | case -ERESTARTNOINTR: | |
3220 | case -ERESTARTNOHAND: | |
3221 | case -ERESTART_RESTARTBLOCK: | |
3222 | /* | |
3223 | * We can't just restart the syscall, since previously | |
3224 | * submitted sqes may already be in progress. Just fail this | |
3225 | * IO with EINTR. | |
3226 | */ | |
3227 | ret = -EINTR; | |
df561f66 | 3228 | fallthrough; |
2b188cc1 | 3229 | default: |
6b19b766 | 3230 | kiocb->ki_complete(kiocb, ret); |
2b188cc1 JA |
3231 | } |
3232 | } | |
3233 | ||
b4aec400 | 3234 | static inline loff_t *io_kiocb_update_pos(struct io_kiocb *req) |
d34e1e5b DY |
3235 | { |
3236 | struct kiocb *kiocb = &req->rw.kiocb; | |
b4aec400 | 3237 | bool is_stream = req->file->f_mode & FMODE_STREAM; |
d34e1e5b DY |
3238 | |
3239 | if (kiocb->ki_pos == -1) { | |
b4aec400 | 3240 | if (!is_stream) { |
d34e1e5b DY |
3241 | req->flags |= REQ_F_CUR_POS; |
3242 | kiocb->ki_pos = req->file->f_pos; | |
b4aec400 | 3243 | return &kiocb->ki_pos; |
d34e1e5b DY |
3244 | } else { |
3245 | kiocb->ki_pos = 0; | |
b4aec400 | 3246 | return NULL; |
d34e1e5b DY |
3247 | } |
3248 | } | |
b4aec400 | 3249 | return is_stream ? NULL : &kiocb->ki_pos; |
d34e1e5b DY |
3250 | } |
3251 | ||
2ea537ca | 3252 | static void kiocb_done(struct io_kiocb *req, ssize_t ret, |
889fca73 | 3253 | unsigned int issue_flags) |
ba816ad6 | 3254 | { |
e8c2bc1f | 3255 | struct io_async_rw *io = req->async_data; |
ba04291e | 3256 | |
227c0c96 | 3257 | /* add previously done IO, if any */ |
d886e185 | 3258 | if (req_has_async_data(req) && io->bytes_done > 0) { |
227c0c96 | 3259 | if (ret < 0) |
e8c2bc1f | 3260 | ret = io->bytes_done; |
227c0c96 | 3261 | else |
e8c2bc1f | 3262 | ret += io->bytes_done; |
227c0c96 JA |
3263 | } |
3264 | ||
ba04291e | 3265 | if (req->flags & REQ_F_CUR_POS) |
2ea537ca PB |
3266 | req->file->f_pos = req->rw.kiocb.ki_pos; |
3267 | if (ret >= 0 && (req->rw.kiocb.ki_complete == io_complete_rw)) | |
00f6e68b | 3268 | __io_complete_rw(req, ret, issue_flags); |
ba816ad6 | 3269 | else |
2ea537ca | 3270 | io_rw_done(&req->rw.kiocb, ret); |
97284637 | 3271 | |
b66ceaf3 | 3272 | if (req->flags & REQ_F_REISSUE) { |
97284637 | 3273 | req->flags &= ~REQ_F_REISSUE; |
b91ef187 | 3274 | if (io_resubmit_prep(req)) |
773af691 | 3275 | io_req_task_queue_reissue(req); |
b91ef187 PB |
3276 | else |
3277 | io_req_task_queue_fail(req, ret); | |
97284637 | 3278 | } |
ba816ad6 JA |
3279 | } |
3280 | ||
eae071c9 PB |
3281 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
3282 | struct io_mapped_ubuf *imu) | |
edafccee | 3283 | { |
9adbd45d | 3284 | size_t len = req->rw.len; |
75769e3f | 3285 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 3286 | size_t offset; |
edafccee | 3287 | |
75769e3f | 3288 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
3289 | return -EFAULT; |
3290 | /* not inside the mapped region */ | |
4751f53d | 3291 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
3292 | return -EFAULT; |
3293 | ||
3294 | /* | |
3295 | * May not be a start of buffer, set size appropriately | |
3296 | * and advance us to the beginning. | |
3297 | */ | |
3298 | offset = buf_addr - imu->ubuf; | |
3299 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3300 | |
3301 | if (offset) { | |
3302 | /* | |
3303 | * Don't use iov_iter_advance() here, as it's really slow for | |
3304 | * using the latter parts of a big fixed buffer - it iterates | |
3305 | * over each segment manually. We can cheat a bit here, because | |
3306 | * we know that: | |
3307 | * | |
3308 | * 1) it's a BVEC iter, we set it up | |
3309 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3310 | * first and last bvec | |
3311 | * | |
3312 | * So just find our index, and adjust the iterator afterwards. | |
3313 | * If the offset is within the first bvec (or the whole first | |
3314 | * bvec, just use iov_iter_advance(). This makes it easier | |
3315 | * since we can just skip the first segment, which may not | |
3316 | * be PAGE_SIZE aligned. | |
3317 | */ | |
3318 | const struct bio_vec *bvec = imu->bvec; | |
3319 | ||
3320 | if (offset <= bvec->bv_len) { | |
3321 | iov_iter_advance(iter, offset); | |
3322 | } else { | |
3323 | unsigned long seg_skip; | |
3324 | ||
3325 | /* skip first vec */ | |
3326 | offset -= bvec->bv_len; | |
3327 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3328 | ||
3329 | iter->bvec = bvec + seg_skip; | |
3330 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3331 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3332 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3333 | } |
3334 | } | |
3335 | ||
847595de | 3336 | return 0; |
edafccee JA |
3337 | } |
3338 | ||
eae071c9 PB |
3339 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
3340 | { | |
eae071c9 PB |
3341 | struct io_mapped_ubuf *imu = req->imu; |
3342 | u16 index, buf_index = req->buf_index; | |
3343 | ||
3344 | if (likely(!imu)) { | |
578c0ee2 PB |
3345 | struct io_ring_ctx *ctx = req->ctx; |
3346 | ||
eae071c9 PB |
3347 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
3348 | return -EFAULT; | |
578c0ee2 | 3349 | io_req_set_rsrc_node(req, ctx); |
eae071c9 PB |
3350 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
3351 | imu = READ_ONCE(ctx->user_bufs[index]); | |
3352 | req->imu = imu; | |
3353 | } | |
3354 | return __io_import_fixed(req, rw, iter, imu); | |
3355 | } | |
3356 | ||
bcda7baa JA |
3357 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3358 | { | |
3359 | if (needs_lock) | |
3360 | mutex_unlock(&ctx->uring_lock); | |
3361 | } | |
3362 | ||
3363 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3364 | { | |
3365 | /* | |
3366 | * "Normal" inline submissions always hold the uring_lock, since we | |
3367 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3368 | * The only exception is when we've detached the request and issue it | |
3369 | * from an async worker thread, grab the lock for that case. | |
3370 | */ | |
3371 | if (needs_lock) | |
3372 | mutex_lock(&ctx->uring_lock); | |
3373 | } | |
3374 | ||
dbc7d452 JA |
3375 | static void io_buffer_add_list(struct io_ring_ctx *ctx, |
3376 | struct io_buffer_list *bl, unsigned int bgid) | |
3377 | { | |
3378 | struct list_head *list; | |
3379 | ||
3380 | list = &ctx->io_buffers[hash_32(bgid, IO_BUFFERS_HASH_BITS)]; | |
3381 | INIT_LIST_HEAD(&bl->buf_list); | |
3382 | bl->bgid = bgid; | |
3383 | list_add(&bl->list, list); | |
3384 | } | |
3385 | ||
bcda7baa | 3386 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, |
51aac424 | 3387 | int bgid, unsigned int issue_flags) |
bcda7baa | 3388 | { |
30d51dd4 | 3389 | struct io_buffer *kbuf = req->kbuf; |
3b44b371 | 3390 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
dbc7d452 JA |
3391 | struct io_ring_ctx *ctx = req->ctx; |
3392 | struct io_buffer_list *bl; | |
bcda7baa JA |
3393 | |
3394 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3395 | return kbuf; | |
3396 | ||
dbc7d452 | 3397 | io_ring_submit_lock(ctx, needs_lock); |
bcda7baa | 3398 | |
dbc7d452 | 3399 | lockdep_assert_held(&ctx->uring_lock); |
bcda7baa | 3400 | |
dbc7d452 JA |
3401 | bl = io_buffer_get_list(ctx, bgid); |
3402 | if (bl && !list_empty(&bl->buf_list)) { | |
3403 | kbuf = list_first_entry(&bl->buf_list, struct io_buffer, list); | |
3404 | list_del(&kbuf->list); | |
bcda7baa JA |
3405 | if (*len > kbuf->len) |
3406 | *len = kbuf->len; | |
30d51dd4 PB |
3407 | req->flags |= REQ_F_BUFFER_SELECTED; |
3408 | req->kbuf = kbuf; | |
bcda7baa JA |
3409 | } else { |
3410 | kbuf = ERR_PTR(-ENOBUFS); | |
3411 | } | |
3412 | ||
3413 | io_ring_submit_unlock(req->ctx, needs_lock); | |
bcda7baa JA |
3414 | return kbuf; |
3415 | } | |
3416 | ||
4d954c25 | 3417 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
51aac424 | 3418 | unsigned int issue_flags) |
4d954c25 JA |
3419 | { |
3420 | struct io_buffer *kbuf; | |
4f4eeba8 | 3421 | u16 bgid; |
4d954c25 | 3422 | |
4f4eeba8 | 3423 | bgid = req->buf_index; |
51aac424 | 3424 | kbuf = io_buffer_select(req, len, bgid, issue_flags); |
4d954c25 JA |
3425 | if (IS_ERR(kbuf)) |
3426 | return kbuf; | |
4d954c25 JA |
3427 | return u64_to_user_ptr(kbuf->addr); |
3428 | } | |
3429 | ||
3430 | #ifdef CONFIG_COMPAT | |
3431 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
51aac424 | 3432 | unsigned int issue_flags) |
4d954c25 JA |
3433 | { |
3434 | struct compat_iovec __user *uiov; | |
3435 | compat_ssize_t clen; | |
3436 | void __user *buf; | |
3437 | ssize_t len; | |
3438 | ||
3439 | uiov = u64_to_user_ptr(req->rw.addr); | |
3440 | if (!access_ok(uiov, sizeof(*uiov))) | |
3441 | return -EFAULT; | |
3442 | if (__get_user(clen, &uiov->iov_len)) | |
3443 | return -EFAULT; | |
3444 | if (clen < 0) | |
3445 | return -EINVAL; | |
3446 | ||
3447 | len = clen; | |
51aac424 | 3448 | buf = io_rw_buffer_select(req, &len, issue_flags); |
4d954c25 JA |
3449 | if (IS_ERR(buf)) |
3450 | return PTR_ERR(buf); | |
3451 | iov[0].iov_base = buf; | |
3452 | iov[0].iov_len = (compat_size_t) len; | |
3453 | return 0; | |
3454 | } | |
3455 | #endif | |
3456 | ||
3457 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
51aac424 | 3458 | unsigned int issue_flags) |
4d954c25 JA |
3459 | { |
3460 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3461 | void __user *buf; | |
3462 | ssize_t len; | |
3463 | ||
3464 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3465 | return -EFAULT; | |
3466 | ||
3467 | len = iov[0].iov_len; | |
3468 | if (len < 0) | |
3469 | return -EINVAL; | |
51aac424 | 3470 | buf = io_rw_buffer_select(req, &len, issue_flags); |
4d954c25 JA |
3471 | if (IS_ERR(buf)) |
3472 | return PTR_ERR(buf); | |
3473 | iov[0].iov_base = buf; | |
3474 | iov[0].iov_len = len; | |
3475 | return 0; | |
3476 | } | |
3477 | ||
3478 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
51aac424 | 3479 | unsigned int issue_flags) |
4d954c25 | 3480 | { |
dddb3e26 | 3481 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
30d51dd4 | 3482 | struct io_buffer *kbuf = req->kbuf; |
dddb3e26 | 3483 | |
dddb3e26 JA |
3484 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
3485 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3486 | return 0; |
dddb3e26 | 3487 | } |
dd201662 | 3488 | if (req->rw.len != 1) |
4d954c25 JA |
3489 | return -EINVAL; |
3490 | ||
3491 | #ifdef CONFIG_COMPAT | |
3492 | if (req->ctx->compat) | |
51aac424 | 3493 | return io_compat_import(req, iov, issue_flags); |
4d954c25 JA |
3494 | #endif |
3495 | ||
51aac424 | 3496 | return __io_iov_buffer_select(req, iov, issue_flags); |
4d954c25 JA |
3497 | } |
3498 | ||
caa8fe6e PB |
3499 | static struct iovec *__io_import_iovec(int rw, struct io_kiocb *req, |
3500 | struct io_rw_state *s, | |
3501 | unsigned int issue_flags) | |
2b188cc1 | 3502 | { |
5e49c973 | 3503 | struct iov_iter *iter = &s->iter; |
847595de | 3504 | u8 opcode = req->opcode; |
caa8fe6e | 3505 | struct iovec *iovec; |
d1d681b0 PB |
3506 | void __user *buf; |
3507 | size_t sqe_len; | |
4d954c25 | 3508 | ssize_t ret; |
edafccee | 3509 | |
f3251183 PB |
3510 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
3511 | ret = io_import_fixed(req, rw, iter); | |
3512 | if (ret) | |
3513 | return ERR_PTR(ret); | |
3514 | return NULL; | |
3515 | } | |
2b188cc1 | 3516 | |
bcda7baa | 3517 | /* buffer index only valid with fixed read/write, or buffer select */ |
d1d681b0 | 3518 | if (unlikely(req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))) |
caa8fe6e | 3519 | return ERR_PTR(-EINVAL); |
9adbd45d | 3520 | |
d1d681b0 PB |
3521 | buf = u64_to_user_ptr(req->rw.addr); |
3522 | sqe_len = req->rw.len; | |
9adbd45d | 3523 | |
3a6820f2 | 3524 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3525 | if (req->flags & REQ_F_BUFFER_SELECT) { |
51aac424 | 3526 | buf = io_rw_buffer_select(req, &sqe_len, issue_flags); |
867a23ea | 3527 | if (IS_ERR(buf)) |
898df244 | 3528 | return ERR_CAST(buf); |
3f9d6441 | 3529 | req->rw.len = sqe_len; |
bcda7baa JA |
3530 | } |
3531 | ||
5e49c973 | 3532 | ret = import_single_range(rw, buf, sqe_len, s->fast_iov, iter); |
f3251183 PB |
3533 | if (ret) |
3534 | return ERR_PTR(ret); | |
3535 | return NULL; | |
3a6820f2 JA |
3536 | } |
3537 | ||
caa8fe6e | 3538 | iovec = s->fast_iov; |
4d954c25 | 3539 | if (req->flags & REQ_F_BUFFER_SELECT) { |
caa8fe6e | 3540 | ret = io_iov_buffer_select(req, iovec, issue_flags); |
f3251183 PB |
3541 | if (ret) |
3542 | return ERR_PTR(ret); | |
3543 | iov_iter_init(iter, rw, iovec, 1, iovec->iov_len); | |
3544 | return NULL; | |
4d954c25 JA |
3545 | } |
3546 | ||
caa8fe6e | 3547 | ret = __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, &iovec, iter, |
89cd35c5 | 3548 | req->ctx->compat); |
caa8fe6e PB |
3549 | if (unlikely(ret < 0)) |
3550 | return ERR_PTR(ret); | |
3551 | return iovec; | |
2b188cc1 JA |
3552 | } |
3553 | ||
5e49c973 PB |
3554 | static inline int io_import_iovec(int rw, struct io_kiocb *req, |
3555 | struct iovec **iovec, struct io_rw_state *s, | |
3556 | unsigned int issue_flags) | |
3557 | { | |
caa8fe6e PB |
3558 | *iovec = __io_import_iovec(rw, req, s, issue_flags); |
3559 | if (unlikely(IS_ERR(*iovec))) | |
3560 | return PTR_ERR(*iovec); | |
5e49c973 | 3561 | |
5e49c973 | 3562 | iov_iter_save_state(&s->iter, &s->iter_state); |
caa8fe6e | 3563 | return 0; |
2b188cc1 JA |
3564 | } |
3565 | ||
0fef9483 JA |
3566 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3567 | { | |
5b09e37e | 3568 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3569 | } |
3570 | ||
31b51510 | 3571 | /* |
32960613 JA |
3572 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3573 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3574 | */ |
4017eb91 | 3575 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3576 | { |
4017eb91 JA |
3577 | struct kiocb *kiocb = &req->rw.kiocb; |
3578 | struct file *file = req->file; | |
32960613 | 3579 | ssize_t ret = 0; |
af9c45ec | 3580 | loff_t *ppos; |
32960613 JA |
3581 | |
3582 | /* | |
3583 | * Don't support polled IO through this interface, and we can't | |
3584 | * support non-blocking either. For the latter, this just causes | |
3585 | * the kiocb to be handled from an async context. | |
3586 | */ | |
3587 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3588 | return -EOPNOTSUPP; | |
35645ac3 PB |
3589 | if ((kiocb->ki_flags & IOCB_NOWAIT) && |
3590 | !(kiocb->ki_filp->f_flags & O_NONBLOCK)) | |
32960613 JA |
3591 | return -EAGAIN; |
3592 | ||
af9c45ec DY |
3593 | ppos = io_kiocb_ppos(kiocb); |
3594 | ||
32960613 | 3595 | while (iov_iter_count(iter)) { |
311ae9e1 | 3596 | struct iovec iovec; |
32960613 JA |
3597 | ssize_t nr; |
3598 | ||
311ae9e1 PB |
3599 | if (!iov_iter_is_bvec(iter)) { |
3600 | iovec = iov_iter_iovec(iter); | |
3601 | } else { | |
4017eb91 JA |
3602 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3603 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3604 | } |
3605 | ||
32960613 JA |
3606 | if (rw == READ) { |
3607 | nr = file->f_op->read(file, iovec.iov_base, | |
af9c45ec | 3608 | iovec.iov_len, ppos); |
32960613 JA |
3609 | } else { |
3610 | nr = file->f_op->write(file, iovec.iov_base, | |
af9c45ec | 3611 | iovec.iov_len, ppos); |
32960613 JA |
3612 | } |
3613 | ||
3614 | if (nr < 0) { | |
3615 | if (!ret) | |
3616 | ret = nr; | |
3617 | break; | |
3618 | } | |
5e929367 | 3619 | ret += nr; |
16c8d2df JA |
3620 | if (!iov_iter_is_bvec(iter)) { |
3621 | iov_iter_advance(iter, nr); | |
3622 | } else { | |
16c8d2df | 3623 | req->rw.addr += nr; |
5e929367 JA |
3624 | req->rw.len -= nr; |
3625 | if (!req->rw.len) | |
3626 | break; | |
16c8d2df | 3627 | } |
32960613 JA |
3628 | if (nr != iovec.iov_len) |
3629 | break; | |
32960613 JA |
3630 | } |
3631 | ||
3632 | return ret; | |
3633 | } | |
3634 | ||
ff6165b2 JA |
3635 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3636 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3637 | { |
e8c2bc1f | 3638 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3639 | |
538941e2 | 3640 | memcpy(&rw->s.iter, iter, sizeof(*iter)); |
afb87658 | 3641 | rw->free_iovec = iovec; |
227c0c96 | 3642 | rw->bytes_done = 0; |
ff6165b2 | 3643 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3644 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3645 | return; |
b64e3444 | 3646 | if (!iovec) { |
ff6165b2 JA |
3647 | unsigned iov_off = 0; |
3648 | ||
538941e2 | 3649 | rw->s.iter.iov = rw->s.fast_iov; |
ff6165b2 JA |
3650 | if (iter->iov != fast_iov) { |
3651 | iov_off = iter->iov - fast_iov; | |
538941e2 | 3652 | rw->s.iter.iov += iov_off; |
ff6165b2 | 3653 | } |
538941e2 PB |
3654 | if (rw->s.fast_iov != fast_iov) |
3655 | memcpy(rw->s.fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3656 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3657 | } else { |
3658 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3659 | } |
3660 | } | |
3661 | ||
8d4af685 | 3662 | static inline bool io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3663 | { |
e8c2bc1f JA |
3664 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3665 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
d886e185 PB |
3666 | if (req->async_data) { |
3667 | req->flags |= REQ_F_ASYNC_DATA; | |
3668 | return false; | |
3669 | } | |
3670 | return true; | |
3d9932a8 XW |
3671 | } |
3672 | ||
ff6165b2 | 3673 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
c88598a9 | 3674 | struct io_rw_state *s, bool force) |
b7bb4f7d | 3675 | { |
26f0505a | 3676 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3677 | return 0; |
d886e185 | 3678 | if (!req_has_async_data(req)) { |
cd658695 JA |
3679 | struct io_async_rw *iorw; |
3680 | ||
6cb78689 | 3681 | if (io_alloc_async_data(req)) { |
6bf985dc | 3682 | kfree(iovec); |
5d204bcf | 3683 | return -ENOMEM; |
6bf985dc | 3684 | } |
b7bb4f7d | 3685 | |
c88598a9 | 3686 | io_req_map_rw(req, iovec, s->fast_iov, &s->iter); |
cd658695 JA |
3687 | iorw = req->async_data; |
3688 | /* we've copied and mapped the iter, ensure state is saved */ | |
538941e2 | 3689 | iov_iter_save_state(&iorw->s.iter, &iorw->s.iter_state); |
5d204bcf | 3690 | } |
b7bb4f7d | 3691 | return 0; |
f67676d1 JA |
3692 | } |
3693 | ||
73debe68 | 3694 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3695 | { |
e8c2bc1f | 3696 | struct io_async_rw *iorw = req->async_data; |
5e49c973 | 3697 | struct iovec *iov; |
847595de | 3698 | int ret; |
c3e330a4 | 3699 | |
51aac424 | 3700 | /* submission path, ->uring_lock should already be taken */ |
3b44b371 | 3701 | ret = io_import_iovec(rw, req, &iov, &iorw->s, 0); |
c3e330a4 PB |
3702 | if (unlikely(ret < 0)) |
3703 | return ret; | |
3704 | ||
ab0b196c PB |
3705 | iorw->bytes_done = 0; |
3706 | iorw->free_iovec = iov; | |
3707 | if (iov) | |
3708 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3709 | return 0; |
3710 | } | |
3711 | ||
73debe68 | 3712 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3713 | { |
3529d8c2 JA |
3714 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3715 | return -EBADF; | |
b9a6b8f9 | 3716 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3717 | } |
3718 | ||
c1dd91d1 | 3719 | /* |
ffdc8dab | 3720 | * This is our waitqueue callback handler, registered through __folio_lock_async() |
c1dd91d1 JA |
3721 | * when we initially tried to do the IO with the iocb armed our waitqueue. |
3722 | * This gets called when the page is unlocked, and we generally expect that to | |
3723 | * happen when the page IO is completed and the page is now uptodate. This will | |
3724 | * queue a task_work based retry of the operation, attempting to copy the data | |
3725 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3726 | * do a thread based blocking retry of the operation. That's the unexpected | |
3727 | * slow path. | |
3728 | */ | |
bcf5a063 JA |
3729 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3730 | int sync, void *arg) | |
3731 | { | |
3732 | struct wait_page_queue *wpq; | |
3733 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3734 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3735 | |
3736 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3737 | ||
cdc8fcb4 LT |
3738 | if (!wake_page_match(wpq, key)) |
3739 | return 0; | |
3740 | ||
c8d317aa | 3741 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 | 3742 | list_del_init(&wait->entry); |
921b9054 | 3743 | io_req_task_queue(req); |
bcf5a063 JA |
3744 | return 1; |
3745 | } | |
3746 | ||
c1dd91d1 JA |
3747 | /* |
3748 | * This controls whether a given IO request should be armed for async page | |
3749 | * based retry. If we return false here, the request is handed to the async | |
3750 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3751 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3752 | * will either succeed because the page is now uptodate and unlocked, or it | |
3753 | * will register a callback when the page is unlocked at IO completion. Through | |
3754 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3755 | * That retry will attempt the buffered read again. The retry will generally | |
3756 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3757 | * async worker threads for a blocking retry. | |
3758 | */ | |
227c0c96 | 3759 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3760 | { |
e8c2bc1f JA |
3761 | struct io_async_rw *rw = req->async_data; |
3762 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3763 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3764 | |
bcf5a063 JA |
3765 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3766 | if (req->flags & REQ_F_NOWAIT) | |
3767 | return false; | |
f67676d1 | 3768 | |
227c0c96 | 3769 | /* Only for buffered IO */ |
3b2a4439 | 3770 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3771 | return false; |
3b2a4439 | 3772 | |
bcf5a063 JA |
3773 | /* |
3774 | * just use poll if we can, and don't attempt if the fs doesn't | |
3775 | * support callback based unlocks | |
3776 | */ | |
3777 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3778 | return false; | |
f67676d1 | 3779 | |
3b2a4439 JA |
3780 | wait->wait.func = io_async_buf_func; |
3781 | wait->wait.private = req; | |
3782 | wait->wait.flags = 0; | |
3783 | INIT_LIST_HEAD(&wait->wait.entry); | |
3784 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3785 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3786 | kiocb->ki_waitq = wait; |
3b2a4439 | 3787 | return true; |
bcf5a063 JA |
3788 | } |
3789 | ||
aeab9506 | 3790 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 | 3791 | { |
607b6fb8 | 3792 | if (likely(req->file->f_op->read_iter)) |
bcf5a063 | 3793 | return call_read_iter(req->file, &req->rw.kiocb, iter); |
2dd2111d | 3794 | else if (req->file->f_op->read) |
4017eb91 | 3795 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3796 | else |
3797 | return -EINVAL; | |
f67676d1 JA |
3798 | } |
3799 | ||
7db30437 ML |
3800 | static bool need_read_all(struct io_kiocb *req) |
3801 | { | |
3802 | return req->flags & REQ_F_ISREG || | |
3803 | S_ISBLK(file_inode(req->file)->i_mode); | |
3804 | } | |
3805 | ||
889fca73 | 3806 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 3807 | { |
607b6fb8 | 3808 | struct io_rw_state __s, *s = &__s; |
c88598a9 | 3809 | struct iovec *iovec; |
9adbd45d | 3810 | struct kiocb *kiocb = &req->rw.kiocb; |
45d189c6 | 3811 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
d886e185 | 3812 | struct io_async_rw *rw; |
cd658695 | 3813 | ssize_t ret, ret2; |
b4aec400 | 3814 | loff_t *ppos; |
ff6165b2 | 3815 | |
607b6fb8 PB |
3816 | if (!req_has_async_data(req)) { |
3817 | ret = io_import_iovec(READ, req, &iovec, s, issue_flags); | |
3818 | if (unlikely(ret < 0)) | |
3819 | return ret; | |
3820 | } else { | |
2be2eb02 JA |
3821 | /* |
3822 | * Safe and required to re-import if we're using provided | |
3823 | * buffers, as we dropped the selected one before retry. | |
3824 | */ | |
3825 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3826 | ret = io_import_iovec(READ, req, &iovec, s, issue_flags); | |
3827 | if (unlikely(ret < 0)) | |
3828 | return ret; | |
3829 | } | |
3830 | ||
d886e185 | 3831 | rw = req->async_data; |
c88598a9 | 3832 | s = &rw->s; |
cd658695 JA |
3833 | /* |
3834 | * We come here from an earlier attempt, restore our state to | |
3835 | * match in case it doesn't. It's cheap enough that we don't | |
3836 | * need to make this conditional. | |
3837 | */ | |
c88598a9 | 3838 | iov_iter_restore(&s->iter, &s->iter_state); |
2846c481 | 3839 | iovec = NULL; |
2846c481 | 3840 | } |
c88598a9 | 3841 | req->result = iov_iter_count(&s->iter); |
2b188cc1 | 3842 | |
607b6fb8 PB |
3843 | if (force_nonblock) { |
3844 | /* If the file doesn't support async, just async punt */ | |
35645ac3 | 3845 | if (unlikely(!io_file_supports_nowait(req))) { |
607b6fb8 PB |
3846 | ret = io_setup_async_rw(req, iovec, s, true); |
3847 | return ret ?: -EAGAIN; | |
3848 | } | |
a88fc400 | 3849 | kiocb->ki_flags |= IOCB_NOWAIT; |
607b6fb8 PB |
3850 | } else { |
3851 | /* Ensure we clear previously set non-block flag */ | |
3852 | kiocb->ki_flags &= ~IOCB_NOWAIT; | |
6713e7a6 | 3853 | } |
9e645e11 | 3854 | |
b4aec400 | 3855 | ppos = io_kiocb_update_pos(req); |
d34e1e5b | 3856 | |
b4aec400 | 3857 | ret = rw_verify_area(READ, req->file, ppos, req->result); |
5ea5dd45 PB |
3858 | if (unlikely(ret)) { |
3859 | kfree(iovec); | |
3860 | return ret; | |
3861 | } | |
2b188cc1 | 3862 | |
c88598a9 | 3863 | ret = io_iter_do_read(req, &s->iter); |
32960613 | 3864 | |
230d50d4 | 3865 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3866 | req->flags &= ~REQ_F_REISSUE; |
9af177ee JA |
3867 | /* if we can poll, just do that */ |
3868 | if (req->opcode == IORING_OP_READ && file_can_poll(req->file)) | |
3869 | return -EAGAIN; | |
eefdf30f JA |
3870 | /* IOPOLL retry should happen for io-wq threads */ |
3871 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3872 | goto done; |
75c668cd PB |
3873 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3874 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3875 | goto done; |
f38c7e3a | 3876 | ret = 0; |
230d50d4 JA |
3877 | } else if (ret == -EIOCBQUEUED) { |
3878 | goto out_free; | |
f80a50a6 | 3879 | } else if (ret == req->result || ret <= 0 || !force_nonblock || |
7db30437 | 3880 | (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { |
7335e3bf | 3881 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3882 | goto done; |
227c0c96 JA |
3883 | } |
3884 | ||
cd658695 JA |
3885 | /* |
3886 | * Don't depend on the iter state matching what was consumed, or being | |
3887 | * untouched in case of error. Restore it and we'll advance it | |
3888 | * manually if we need to. | |
3889 | */ | |
c88598a9 | 3890 | iov_iter_restore(&s->iter, &s->iter_state); |
cd658695 | 3891 | |
c88598a9 | 3892 | ret2 = io_setup_async_rw(req, iovec, s, true); |
6bf985dc PB |
3893 | if (ret2) |
3894 | return ret2; | |
3895 | ||
fe1cdd55 | 3896 | iovec = NULL; |
e8c2bc1f | 3897 | rw = req->async_data; |
c88598a9 | 3898 | s = &rw->s; |
cd658695 JA |
3899 | /* |
3900 | * Now use our persistent iterator and state, if we aren't already. | |
3901 | * We've restored and mapped the iter to match. | |
3902 | */ | |
227c0c96 | 3903 | |
b23df91b | 3904 | do { |
cd658695 JA |
3905 | /* |
3906 | * We end up here because of a partial read, either from | |
3907 | * above or inside this loop. Advance the iter by the bytes | |
3908 | * that were consumed. | |
3909 | */ | |
c88598a9 PB |
3910 | iov_iter_advance(&s->iter, ret); |
3911 | if (!iov_iter_count(&s->iter)) | |
cd658695 | 3912 | break; |
b23df91b | 3913 | rw->bytes_done += ret; |
c88598a9 | 3914 | iov_iter_save_state(&s->iter, &s->iter_state); |
cd658695 | 3915 | |
b23df91b PB |
3916 | /* if we can retry, do so with the callbacks armed */ |
3917 | if (!io_rw_should_retry(req)) { | |
3918 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3919 | return -EAGAIN; | |
3920 | } | |
3921 | ||
3922 | /* | |
3923 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3924 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3925 | * desired page gets unlocked. We can also get a partial read | |
3926 | * here, and if we do, then just retry at the new offset. | |
3927 | */ | |
c88598a9 | 3928 | ret = io_iter_do_read(req, &s->iter); |
b23df91b PB |
3929 | if (ret == -EIOCBQUEUED) |
3930 | return 0; | |
227c0c96 | 3931 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3932 | kiocb->ki_flags &= ~IOCB_WAITQ; |
c88598a9 | 3933 | iov_iter_restore(&s->iter, &s->iter_state); |
cd658695 | 3934 | } while (ret > 0); |
227c0c96 | 3935 | done: |
2ea537ca | 3936 | kiocb_done(req, ret, issue_flags); |
fe1cdd55 PB |
3937 | out_free: |
3938 | /* it's faster to check here then delegate to kfree */ | |
3939 | if (iovec) | |
3940 | kfree(iovec); | |
5ea5dd45 | 3941 | return 0; |
2b188cc1 JA |
3942 | } |
3943 | ||
73debe68 | 3944 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3945 | { |
3529d8c2 JA |
3946 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3947 | return -EBADF; | |
3884b83d | 3948 | req->rw.kiocb.ki_hint = ki_hint_validate(file_write_hint(req->file)); |
b9a6b8f9 | 3949 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3950 | } |
3951 | ||
889fca73 | 3952 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 3953 | { |
607b6fb8 | 3954 | struct io_rw_state __s, *s = &__s; |
c88598a9 | 3955 | struct iovec *iovec; |
9adbd45d | 3956 | struct kiocb *kiocb = &req->rw.kiocb; |
45d189c6 | 3957 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
cd658695 | 3958 | ssize_t ret, ret2; |
b4aec400 | 3959 | loff_t *ppos; |
2b188cc1 | 3960 | |
607b6fb8 | 3961 | if (!req_has_async_data(req)) { |
5e49c973 PB |
3962 | ret = io_import_iovec(WRITE, req, &iovec, s, issue_flags); |
3963 | if (unlikely(ret < 0)) | |
2846c481 | 3964 | return ret; |
607b6fb8 PB |
3965 | } else { |
3966 | struct io_async_rw *rw = req->async_data; | |
3967 | ||
3968 | s = &rw->s; | |
3969 | iov_iter_restore(&s->iter, &s->iter_state); | |
2846c481 | 3970 | iovec = NULL; |
2846c481 | 3971 | } |
c88598a9 | 3972 | req->result = iov_iter_count(&s->iter); |
2b188cc1 | 3973 | |
607b6fb8 PB |
3974 | if (force_nonblock) { |
3975 | /* If the file doesn't support async, just async punt */ | |
35645ac3 | 3976 | if (unlikely(!io_file_supports_nowait(req))) |
607b6fb8 | 3977 | goto copy_iov; |
fd6c2e4c | 3978 | |
607b6fb8 PB |
3979 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3980 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3981 | (req->flags & REQ_F_ISREG)) | |
3982 | goto copy_iov; | |
31b51510 | 3983 | |
607b6fb8 PB |
3984 | kiocb->ki_flags |= IOCB_NOWAIT; |
3985 | } else { | |
3986 | /* Ensure we clear previously set non-block flag */ | |
3987 | kiocb->ki_flags &= ~IOCB_NOWAIT; | |
3988 | } | |
31b51510 | 3989 | |
b4aec400 | 3990 | ppos = io_kiocb_update_pos(req); |
d34e1e5b | 3991 | |
b4aec400 | 3992 | ret = rw_verify_area(WRITE, req->file, ppos, req->result); |
fa15bafb PB |
3993 | if (unlikely(ret)) |
3994 | goto out_free; | |
4ed734b0 | 3995 | |
fa15bafb PB |
3996 | /* |
3997 | * Open-code file_start_write here to grab freeze protection, | |
3998 | * which will be released by another thread in | |
3999 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
4000 | * released so that it doesn't complain about the held lock when | |
4001 | * we return to userspace. | |
4002 | */ | |
4003 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 4004 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
4005 | __sb_writers_release(file_inode(req->file)->i_sb, |
4006 | SB_FREEZE_WRITE); | |
4007 | } | |
4008 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 4009 | |
35645ac3 | 4010 | if (likely(req->file->f_op->write_iter)) |
c88598a9 | 4011 | ret2 = call_write_iter(req->file, kiocb, &s->iter); |
2dd2111d | 4012 | else if (req->file->f_op->write) |
c88598a9 | 4013 | ret2 = loop_rw_iter(WRITE, req, &s->iter); |
2dd2111d GH |
4014 | else |
4015 | ret2 = -EINVAL; | |
4ed734b0 | 4016 | |
6ad7f233 PB |
4017 | if (req->flags & REQ_F_REISSUE) { |
4018 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 4019 | ret2 = -EAGAIN; |
6ad7f233 | 4020 | } |
230d50d4 | 4021 | |
fa15bafb PB |
4022 | /* |
4023 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
4024 | * retry them without IOCB_NOWAIT. | |
4025 | */ | |
4026 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
4027 | ret2 = -EAGAIN; | |
75c668cd PB |
4028 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
4029 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 4030 | goto done; |
fa15bafb | 4031 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f | 4032 | /* IOPOLL retry should happen for io-wq threads */ |
b10841c9 | 4033 | if (ret2 == -EAGAIN && (req->ctx->flags & IORING_SETUP_IOPOLL)) |
eefdf30f | 4034 | goto copy_iov; |
355afaeb | 4035 | done: |
2ea537ca | 4036 | kiocb_done(req, ret2, issue_flags); |
fa15bafb | 4037 | } else { |
f67676d1 | 4038 | copy_iov: |
c88598a9 PB |
4039 | iov_iter_restore(&s->iter, &s->iter_state); |
4040 | ret = io_setup_async_rw(req, iovec, s, false); | |
6bf985dc | 4041 | return ret ?: -EAGAIN; |
2b188cc1 | 4042 | } |
31b51510 | 4043 | out_free: |
f261c168 | 4044 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 4045 | if (iovec) |
6f2cc166 | 4046 | kfree(iovec); |
2b188cc1 JA |
4047 | return ret; |
4048 | } | |
4049 | ||
80a261fd JA |
4050 | static int io_renameat_prep(struct io_kiocb *req, |
4051 | const struct io_uring_sqe *sqe) | |
4052 | { | |
4053 | struct io_rename *ren = &req->rename; | |
4054 | const char __user *oldf, *newf; | |
4055 | ||
ed7eb259 JA |
4056 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4057 | return -EINVAL; | |
26578cda | 4058 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
ed7eb259 | 4059 | return -EINVAL; |
80a261fd JA |
4060 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
4061 | return -EBADF; | |
4062 | ||
4063 | ren->old_dfd = READ_ONCE(sqe->fd); | |
4064 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4065 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4066 | ren->new_dfd = READ_ONCE(sqe->len); | |
4067 | ren->flags = READ_ONCE(sqe->rename_flags); | |
4068 | ||
4069 | ren->oldpath = getname(oldf); | |
4070 | if (IS_ERR(ren->oldpath)) | |
4071 | return PTR_ERR(ren->oldpath); | |
4072 | ||
4073 | ren->newpath = getname(newf); | |
4074 | if (IS_ERR(ren->newpath)) { | |
4075 | putname(ren->oldpath); | |
4076 | return PTR_ERR(ren->newpath); | |
4077 | } | |
4078 | ||
4079 | req->flags |= REQ_F_NEED_CLEANUP; | |
4080 | return 0; | |
4081 | } | |
4082 | ||
45d189c6 | 4083 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
4084 | { |
4085 | struct io_rename *ren = &req->rename; | |
4086 | int ret; | |
4087 | ||
45d189c6 | 4088 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
4089 | return -EAGAIN; |
4090 | ||
4091 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
4092 | ren->newpath, ren->flags); | |
4093 | ||
4094 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4095 | if (ret < 0) | |
93d2bcd2 | 4096 | req_set_fail(req); |
80a261fd JA |
4097 | io_req_complete(req, ret); |
4098 | return 0; | |
4099 | } | |
4100 | ||
14a1143b JA |
4101 | static int io_unlinkat_prep(struct io_kiocb *req, |
4102 | const struct io_uring_sqe *sqe) | |
4103 | { | |
4104 | struct io_unlink *un = &req->unlink; | |
4105 | const char __user *fname; | |
4106 | ||
22634bc5 JA |
4107 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4108 | return -EINVAL; | |
26578cda PB |
4109 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || |
4110 | sqe->splice_fd_in) | |
22634bc5 | 4111 | return -EINVAL; |
14a1143b JA |
4112 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
4113 | return -EBADF; | |
4114 | ||
4115 | un->dfd = READ_ONCE(sqe->fd); | |
4116 | ||
4117 | un->flags = READ_ONCE(sqe->unlink_flags); | |
4118 | if (un->flags & ~AT_REMOVEDIR) | |
4119 | return -EINVAL; | |
4120 | ||
4121 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4122 | un->filename = getname(fname); | |
4123 | if (IS_ERR(un->filename)) | |
4124 | return PTR_ERR(un->filename); | |
4125 | ||
4126 | req->flags |= REQ_F_NEED_CLEANUP; | |
4127 | return 0; | |
4128 | } | |
4129 | ||
45d189c6 | 4130 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
4131 | { |
4132 | struct io_unlink *un = &req->unlink; | |
4133 | int ret; | |
4134 | ||
45d189c6 | 4135 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
4136 | return -EAGAIN; |
4137 | ||
4138 | if (un->flags & AT_REMOVEDIR) | |
4139 | ret = do_rmdir(un->dfd, un->filename); | |
4140 | else | |
4141 | ret = do_unlinkat(un->dfd, un->filename); | |
4142 | ||
4143 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4144 | if (ret < 0) | |
93d2bcd2 | 4145 | req_set_fail(req); |
14a1143b JA |
4146 | io_req_complete(req, ret); |
4147 | return 0; | |
4148 | } | |
4149 | ||
e34a02dc DK |
4150 | static int io_mkdirat_prep(struct io_kiocb *req, |
4151 | const struct io_uring_sqe *sqe) | |
4152 | { | |
4153 | struct io_mkdir *mkd = &req->mkdir; | |
4154 | const char __user *fname; | |
4155 | ||
4156 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4157 | return -EINVAL; | |
4158 | if (sqe->ioprio || sqe->off || sqe->rw_flags || sqe->buf_index || | |
4159 | sqe->splice_fd_in) | |
4160 | return -EINVAL; | |
4161 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
4162 | return -EBADF; | |
4163 | ||
4164 | mkd->dfd = READ_ONCE(sqe->fd); | |
4165 | mkd->mode = READ_ONCE(sqe->len); | |
4166 | ||
4167 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4168 | mkd->filename = getname(fname); | |
4169 | if (IS_ERR(mkd->filename)) | |
4170 | return PTR_ERR(mkd->filename); | |
4171 | ||
4172 | req->flags |= REQ_F_NEED_CLEANUP; | |
4173 | return 0; | |
4174 | } | |
4175 | ||
04f34081 | 4176 | static int io_mkdirat(struct io_kiocb *req, unsigned int issue_flags) |
e34a02dc DK |
4177 | { |
4178 | struct io_mkdir *mkd = &req->mkdir; | |
4179 | int ret; | |
4180 | ||
4181 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4182 | return -EAGAIN; | |
4183 | ||
4184 | ret = do_mkdirat(mkd->dfd, mkd->filename, mkd->mode); | |
4185 | ||
4186 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4187 | if (ret < 0) | |
4188 | req_set_fail(req); | |
4189 | io_req_complete(req, ret); | |
4190 | return 0; | |
4191 | } | |
4192 | ||
7a8721f8 DK |
4193 | static int io_symlinkat_prep(struct io_kiocb *req, |
4194 | const struct io_uring_sqe *sqe) | |
4195 | { | |
4196 | struct io_symlink *sl = &req->symlink; | |
4197 | const char __user *oldpath, *newpath; | |
4198 | ||
4199 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4200 | return -EINVAL; | |
4201 | if (sqe->ioprio || sqe->len || sqe->rw_flags || sqe->buf_index || | |
4202 | sqe->splice_fd_in) | |
4203 | return -EINVAL; | |
4204 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
4205 | return -EBADF; | |
4206 | ||
4207 | sl->new_dfd = READ_ONCE(sqe->fd); | |
4208 | oldpath = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4209 | newpath = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4210 | ||
4211 | sl->oldpath = getname(oldpath); | |
4212 | if (IS_ERR(sl->oldpath)) | |
4213 | return PTR_ERR(sl->oldpath); | |
4214 | ||
4215 | sl->newpath = getname(newpath); | |
4216 | if (IS_ERR(sl->newpath)) { | |
4217 | putname(sl->oldpath); | |
4218 | return PTR_ERR(sl->newpath); | |
4219 | } | |
4220 | ||
4221 | req->flags |= REQ_F_NEED_CLEANUP; | |
4222 | return 0; | |
4223 | } | |
4224 | ||
04f34081 | 4225 | static int io_symlinkat(struct io_kiocb *req, unsigned int issue_flags) |
7a8721f8 DK |
4226 | { |
4227 | struct io_symlink *sl = &req->symlink; | |
4228 | int ret; | |
4229 | ||
4230 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4231 | return -EAGAIN; | |
4232 | ||
4233 | ret = do_symlinkat(sl->oldpath, sl->new_dfd, sl->newpath); | |
4234 | ||
4235 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4236 | if (ret < 0) | |
4237 | req_set_fail(req); | |
4238 | io_req_complete(req, ret); | |
4239 | return 0; | |
4240 | } | |
4241 | ||
cf30da90 DK |
4242 | static int io_linkat_prep(struct io_kiocb *req, |
4243 | const struct io_uring_sqe *sqe) | |
4244 | { | |
4245 | struct io_hardlink *lnk = &req->hardlink; | |
4246 | const char __user *oldf, *newf; | |
4247 | ||
4248 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4249 | return -EINVAL; | |
4250 | if (sqe->ioprio || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) | |
4251 | return -EINVAL; | |
4252 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
4253 | return -EBADF; | |
4254 | ||
4255 | lnk->old_dfd = READ_ONCE(sqe->fd); | |
4256 | lnk->new_dfd = READ_ONCE(sqe->len); | |
4257 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4258 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4259 | lnk->flags = READ_ONCE(sqe->hardlink_flags); | |
4260 | ||
4261 | lnk->oldpath = getname(oldf); | |
4262 | if (IS_ERR(lnk->oldpath)) | |
4263 | return PTR_ERR(lnk->oldpath); | |
4264 | ||
4265 | lnk->newpath = getname(newf); | |
4266 | if (IS_ERR(lnk->newpath)) { | |
4267 | putname(lnk->oldpath); | |
4268 | return PTR_ERR(lnk->newpath); | |
4269 | } | |
4270 | ||
4271 | req->flags |= REQ_F_NEED_CLEANUP; | |
4272 | return 0; | |
4273 | } | |
4274 | ||
04f34081 | 4275 | static int io_linkat(struct io_kiocb *req, unsigned int issue_flags) |
cf30da90 DK |
4276 | { |
4277 | struct io_hardlink *lnk = &req->hardlink; | |
4278 | int ret; | |
4279 | ||
4280 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4281 | return -EAGAIN; | |
4282 | ||
4283 | ret = do_linkat(lnk->old_dfd, lnk->oldpath, lnk->new_dfd, | |
4284 | lnk->newpath, lnk->flags); | |
4285 | ||
4286 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4287 | if (ret < 0) | |
4288 | req_set_fail(req); | |
4289 | io_req_complete(req, ret); | |
4290 | return 0; | |
4291 | } | |
4292 | ||
36f4fa68 JA |
4293 | static int io_shutdown_prep(struct io_kiocb *req, |
4294 | const struct io_uring_sqe *sqe) | |
4295 | { | |
4296 | #if defined(CONFIG_NET) | |
4297 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4298 | return -EINVAL; | |
26578cda PB |
4299 | if (unlikely(sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || |
4300 | sqe->buf_index || sqe->splice_fd_in)) | |
36f4fa68 JA |
4301 | return -EINVAL; |
4302 | ||
4303 | req->shutdown.how = READ_ONCE(sqe->len); | |
4304 | return 0; | |
4305 | #else | |
4306 | return -EOPNOTSUPP; | |
4307 | #endif | |
4308 | } | |
4309 | ||
45d189c6 | 4310 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
4311 | { |
4312 | #if defined(CONFIG_NET) | |
4313 | struct socket *sock; | |
4314 | int ret; | |
4315 | ||
45d189c6 | 4316 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
4317 | return -EAGAIN; |
4318 | ||
48aba79b | 4319 | sock = sock_from_file(req->file); |
36f4fa68 | 4320 | if (unlikely(!sock)) |
48aba79b | 4321 | return -ENOTSOCK; |
36f4fa68 JA |
4322 | |
4323 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d | 4324 | if (ret < 0) |
93d2bcd2 | 4325 | req_set_fail(req); |
36f4fa68 JA |
4326 | io_req_complete(req, ret); |
4327 | return 0; | |
4328 | #else | |
4329 | return -EOPNOTSUPP; | |
4330 | #endif | |
4331 | } | |
4332 | ||
f2a8d5c7 PB |
4333 | static int __io_splice_prep(struct io_kiocb *req, |
4334 | const struct io_uring_sqe *sqe) | |
7d67af2c | 4335 | { |
fe7e3257 | 4336 | struct io_splice *sp = &req->splice; |
7d67af2c | 4337 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 4338 | |
3232dd02 PB |
4339 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4340 | return -EINVAL; | |
7d67af2c PB |
4341 | |
4342 | sp->file_in = NULL; | |
7d67af2c PB |
4343 | sp->len = READ_ONCE(sqe->len); |
4344 | sp->flags = READ_ONCE(sqe->splice_flags); | |
4345 | ||
4346 | if (unlikely(sp->flags & ~valid_flags)) | |
4347 | return -EINVAL; | |
4348 | ||
62906e89 | 4349 | sp->file_in = io_file_get(req->ctx, req, READ_ONCE(sqe->splice_fd_in), |
8371adf5 PB |
4350 | (sp->flags & SPLICE_F_FD_IN_FIXED)); |
4351 | if (!sp->file_in) | |
4352 | return -EBADF; | |
7d67af2c | 4353 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
4354 | return 0; |
4355 | } | |
4356 | ||
f2a8d5c7 PB |
4357 | static int io_tee_prep(struct io_kiocb *req, |
4358 | const struct io_uring_sqe *sqe) | |
4359 | { | |
4360 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
4361 | return -EINVAL; | |
4362 | return __io_splice_prep(req, sqe); | |
4363 | } | |
4364 | ||
45d189c6 | 4365 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
4366 | { |
4367 | struct io_splice *sp = &req->splice; | |
4368 | struct file *in = sp->file_in; | |
4369 | struct file *out = sp->file_out; | |
4370 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
4371 | long ret = 0; | |
4372 | ||
45d189c6 | 4373 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
4374 | return -EAGAIN; |
4375 | if (sp->len) | |
4376 | ret = do_tee(in, out, sp->len, flags); | |
4377 | ||
e1d767f0 PB |
4378 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4379 | io_put_file(in); | |
f2a8d5c7 PB |
4380 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4381 | ||
f2a8d5c7 | 4382 | if (ret != sp->len) |
93d2bcd2 | 4383 | req_set_fail(req); |
e1e16097 | 4384 | io_req_complete(req, ret); |
f2a8d5c7 PB |
4385 | return 0; |
4386 | } | |
4387 | ||
4388 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4389 | { | |
fe7e3257 | 4390 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
4391 | |
4392 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
4393 | sp->off_out = READ_ONCE(sqe->off); | |
4394 | return __io_splice_prep(req, sqe); | |
4395 | } | |
4396 | ||
45d189c6 | 4397 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
4398 | { |
4399 | struct io_splice *sp = &req->splice; | |
4400 | struct file *in = sp->file_in; | |
4401 | struct file *out = sp->file_out; | |
4402 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
4403 | loff_t *poff_in, *poff_out; | |
c9687426 | 4404 | long ret = 0; |
7d67af2c | 4405 | |
45d189c6 | 4406 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 4407 | return -EAGAIN; |
7d67af2c PB |
4408 | |
4409 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
4410 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 4411 | |
948a7749 | 4412 | if (sp->len) |
c9687426 | 4413 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 4414 | |
e1d767f0 PB |
4415 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
4416 | io_put_file(in); | |
7d67af2c PB |
4417 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4418 | ||
7d67af2c | 4419 | if (ret != sp->len) |
93d2bcd2 | 4420 | req_set_fail(req); |
e1e16097 | 4421 | io_req_complete(req, ret); |
7d67af2c PB |
4422 | return 0; |
4423 | } | |
4424 | ||
2b188cc1 JA |
4425 | /* |
4426 | * IORING_OP_NOP just posts a completion event, nothing else. | |
4427 | */ | |
889fca73 | 4428 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
4429 | { |
4430 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4431 | |
def596e9 JA |
4432 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4433 | return -EINVAL; | |
4434 | ||
889fca73 | 4435 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4436 | return 0; |
4437 | } | |
4438 | ||
4f57f06c JA |
4439 | static int io_msg_ring_prep(struct io_kiocb *req, |
4440 | const struct io_uring_sqe *sqe) | |
4441 | { | |
f3b6a41e JA |
4442 | if (unlikely(sqe->addr || sqe->ioprio || sqe->rw_flags || |
4443 | sqe->splice_fd_in || sqe->buf_index || sqe->personality)) | |
4f57f06c JA |
4444 | return -EINVAL; |
4445 | ||
4446 | if (req->file->f_op != &io_uring_fops) | |
4447 | return -EBADFD; | |
4448 | ||
4449 | req->msg.user_data = READ_ONCE(sqe->off); | |
4450 | req->msg.len = READ_ONCE(sqe->len); | |
4451 | return 0; | |
4452 | } | |
4453 | ||
4454 | static int io_msg_ring(struct io_kiocb *req, unsigned int issue_flags) | |
4455 | { | |
4456 | struct io_ring_ctx *target_ctx; | |
4457 | struct io_msg *msg = &req->msg; | |
4458 | int ret = -EOVERFLOW; | |
4459 | bool filled; | |
4460 | ||
4461 | target_ctx = req->file->private_data; | |
4462 | ||
4463 | spin_lock(&target_ctx->completion_lock); | |
4464 | filled = io_fill_cqe_aux(target_ctx, msg->user_data, msg->len, | |
4465 | IORING_CQE_F_MSG); | |
4466 | io_commit_cqring(target_ctx); | |
4467 | spin_unlock(&target_ctx->completion_lock); | |
4468 | ||
4469 | if (filled) { | |
4470 | io_cqring_ev_posted(target_ctx); | |
4471 | ret = 0; | |
4472 | } | |
4473 | ||
4474 | __io_req_complete(req, issue_flags, ret, 0); | |
4475 | return 0; | |
4476 | } | |
4477 | ||
1155c76a | 4478 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4479 | { |
6b06314c | 4480 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4481 | |
09bb8394 JA |
4482 | if (!req->file) |
4483 | return -EBADF; | |
c992fe29 | 4484 | |
6b06314c | 4485 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4486 | return -EINVAL; |
26578cda PB |
4487 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
4488 | sqe->splice_fd_in)) | |
c992fe29 CH |
4489 | return -EINVAL; |
4490 | ||
8ed8d3c3 JA |
4491 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4492 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4493 | return -EINVAL; | |
4494 | ||
4495 | req->sync.off = READ_ONCE(sqe->off); | |
4496 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4497 | return 0; |
4498 | } | |
4499 | ||
45d189c6 | 4500 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4501 | { |
8ed8d3c3 | 4502 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4503 | int ret; |
4504 | ||
ac45abc0 | 4505 | /* fsync always requires a blocking context */ |
45d189c6 | 4506 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4507 | return -EAGAIN; |
4508 | ||
9adbd45d | 4509 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4510 | end > 0 ? end : LLONG_MAX, |
4511 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4512 | if (ret < 0) | |
93d2bcd2 | 4513 | req_set_fail(req); |
e1e16097 | 4514 | io_req_complete(req, ret); |
c992fe29 CH |
4515 | return 0; |
4516 | } | |
4517 | ||
d63d1b5e JA |
4518 | static int io_fallocate_prep(struct io_kiocb *req, |
4519 | const struct io_uring_sqe *sqe) | |
4520 | { | |
26578cda PB |
4521 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags || |
4522 | sqe->splice_fd_in) | |
d63d1b5e | 4523 | return -EINVAL; |
3232dd02 PB |
4524 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4525 | return -EINVAL; | |
d63d1b5e JA |
4526 | |
4527 | req->sync.off = READ_ONCE(sqe->off); | |
4528 | req->sync.len = READ_ONCE(sqe->addr); | |
4529 | req->sync.mode = READ_ONCE(sqe->len); | |
4530 | return 0; | |
4531 | } | |
4532 | ||
45d189c6 | 4533 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4534 | { |
ac45abc0 PB |
4535 | int ret; |
4536 | ||
d63d1b5e | 4537 | /* fallocate always requiring blocking context */ |
45d189c6 | 4538 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4539 | return -EAGAIN; |
ac45abc0 PB |
4540 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4541 | req->sync.len); | |
ac45abc0 | 4542 | if (ret < 0) |
93d2bcd2 | 4543 | req_set_fail(req); |
f63cf519 JA |
4544 | else |
4545 | fsnotify_modify(req->file); | |
e1e16097 | 4546 | io_req_complete(req, ret); |
5d17b4a4 JA |
4547 | return 0; |
4548 | } | |
4549 | ||
ec65fea5 | 4550 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4551 | { |
f8748881 | 4552 | const char __user *fname; |
15b71abe | 4553 | int ret; |
b7bb4f7d | 4554 | |
d3fddf6d PB |
4555 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4556 | return -EINVAL; | |
b9445598 | 4557 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4558 | return -EINVAL; |
ec65fea5 | 4559 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4560 | return -EBADF; |
03b1230c | 4561 | |
ec65fea5 PB |
4562 | /* open.how should be already initialised */ |
4563 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4564 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4565 | |
25e72d10 PB |
4566 | req->open.dfd = READ_ONCE(sqe->fd); |
4567 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4568 | req->open.filename = getname(fname); |
15b71abe JA |
4569 | if (IS_ERR(req->open.filename)) { |
4570 | ret = PTR_ERR(req->open.filename); | |
4571 | req->open.filename = NULL; | |
4572 | return ret; | |
4573 | } | |
b9445598 PB |
4574 | |
4575 | req->open.file_slot = READ_ONCE(sqe->file_index); | |
4576 | if (req->open.file_slot && (req->open.how.flags & O_CLOEXEC)) | |
4577 | return -EINVAL; | |
4578 | ||
4022e7af | 4579 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4580 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4581 | return 0; |
03b1230c JA |
4582 | } |
4583 | ||
ec65fea5 PB |
4584 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4585 | { | |
d3fddf6d PB |
4586 | u64 mode = READ_ONCE(sqe->len); |
4587 | u64 flags = READ_ONCE(sqe->open_flags); | |
ec65fea5 | 4588 | |
ec65fea5 PB |
4589 | req->open.how = build_open_how(flags, mode); |
4590 | return __io_openat_prep(req, sqe); | |
4591 | } | |
4592 | ||
cebdb986 | 4593 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4594 | { |
cebdb986 | 4595 | struct open_how __user *how; |
cebdb986 | 4596 | size_t len; |
0fa03c62 JA |
4597 | int ret; |
4598 | ||
cebdb986 JA |
4599 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4600 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4601 | if (len < OPEN_HOW_SIZE_VER0) |
4602 | return -EINVAL; | |
3529d8c2 | 4603 | |
cebdb986 JA |
4604 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4605 | len); | |
4606 | if (ret) | |
4607 | return ret; | |
3529d8c2 | 4608 | |
ec65fea5 | 4609 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4610 | } |
4611 | ||
45d189c6 | 4612 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4613 | { |
4614 | struct open_flags op; | |
15b71abe | 4615 | struct file *file; |
b9445598 PB |
4616 | bool resolve_nonblock, nonblock_set; |
4617 | bool fixed = !!req->open.file_slot; | |
15b71abe JA |
4618 | int ret; |
4619 | ||
cebdb986 | 4620 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4621 | if (ret) |
4622 | goto err; | |
3a81fd02 JA |
4623 | nonblock_set = op.open_flag & O_NONBLOCK; |
4624 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4625 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4626 | /* |
4627 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4628 | * it'll always -EAGAIN | |
4629 | */ | |
4630 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4631 | return -EAGAIN; | |
4632 | op.lookup_flags |= LOOKUP_CACHED; | |
4633 | op.open_flag |= O_NONBLOCK; | |
4634 | } | |
15b71abe | 4635 | |
b9445598 PB |
4636 | if (!fixed) { |
4637 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); | |
4638 | if (ret < 0) | |
4639 | goto err; | |
4640 | } | |
15b71abe JA |
4641 | |
4642 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 4643 | if (IS_ERR(file)) { |
944d1444 | 4644 | /* |
12dcb58a PB |
4645 | * We could hang on to this 'fd' on retrying, but seems like |
4646 | * marginal gain for something that is now known to be a slower | |
4647 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 4648 | */ |
b9445598 PB |
4649 | if (!fixed) |
4650 | put_unused_fd(ret); | |
3a81fd02 | 4651 | |
15b71abe | 4652 | ret = PTR_ERR(file); |
12dcb58a PB |
4653 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
4654 | if (ret == -EAGAIN && | |
4655 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
4656 | return -EAGAIN; | |
4657 | goto err; | |
15b71abe | 4658 | } |
12dcb58a PB |
4659 | |
4660 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
4661 | file->f_flags &= ~O_NONBLOCK; | |
4662 | fsnotify_open(file); | |
b9445598 PB |
4663 | |
4664 | if (!fixed) | |
4665 | fd_install(ret, file); | |
4666 | else | |
4667 | ret = io_install_fixed_file(req, file, issue_flags, | |
4668 | req->open.file_slot - 1); | |
15b71abe JA |
4669 | err: |
4670 | putname(req->open.filename); | |
8fef80bf | 4671 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 4672 | if (ret < 0) |
93d2bcd2 | 4673 | req_set_fail(req); |
0bdf3398 | 4674 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
4675 | return 0; |
4676 | } | |
4677 | ||
45d189c6 | 4678 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4679 | { |
e45cff58 | 4680 | return io_openat2(req, issue_flags); |
cebdb986 JA |
4681 | } |
4682 | ||
067524e9 JA |
4683 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4684 | const struct io_uring_sqe *sqe) | |
4685 | { | |
4686 | struct io_provide_buf *p = &req->pbuf; | |
4687 | u64 tmp; | |
4688 | ||
26578cda PB |
4689 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off || |
4690 | sqe->splice_fd_in) | |
067524e9 JA |
4691 | return -EINVAL; |
4692 | ||
4693 | tmp = READ_ONCE(sqe->fd); | |
4694 | if (!tmp || tmp > USHRT_MAX) | |
4695 | return -EINVAL; | |
4696 | ||
4697 | memset(p, 0, sizeof(*p)); | |
4698 | p->nbufs = tmp; | |
4699 | p->bgid = READ_ONCE(sqe->buf_group); | |
4700 | return 0; | |
4701 | } | |
4702 | ||
dbc7d452 JA |
4703 | static int __io_remove_buffers(struct io_ring_ctx *ctx, |
4704 | struct io_buffer_list *bl, unsigned nbufs) | |
067524e9 JA |
4705 | { |
4706 | unsigned i = 0; | |
4707 | ||
4708 | /* shouldn't happen */ | |
4709 | if (!nbufs) | |
4710 | return 0; | |
4711 | ||
4712 | /* the head kbuf is the list itself */ | |
dbc7d452 | 4713 | while (!list_empty(&bl->buf_list)) { |
067524e9 JA |
4714 | struct io_buffer *nxt; |
4715 | ||
dbc7d452 | 4716 | nxt = list_first_entry(&bl->buf_list, struct io_buffer, list); |
067524e9 | 4717 | list_del(&nxt->list); |
067524e9 JA |
4718 | if (++i == nbufs) |
4719 | return i; | |
1d0254e6 | 4720 | cond_resched(); |
067524e9 JA |
4721 | } |
4722 | i++; | |
067524e9 JA |
4723 | |
4724 | return i; | |
4725 | } | |
4726 | ||
889fca73 | 4727 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4728 | { |
4729 | struct io_provide_buf *p = &req->pbuf; | |
4730 | struct io_ring_ctx *ctx = req->ctx; | |
dbc7d452 | 4731 | struct io_buffer_list *bl; |
067524e9 | 4732 | int ret = 0; |
3b44b371 | 4733 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
067524e9 | 4734 | |
3b44b371 | 4735 | io_ring_submit_lock(ctx, needs_lock); |
067524e9 JA |
4736 | |
4737 | lockdep_assert_held(&ctx->uring_lock); | |
4738 | ||
4739 | ret = -ENOENT; | |
dbc7d452 JA |
4740 | bl = io_buffer_get_list(ctx, p->bgid); |
4741 | if (bl) | |
4742 | ret = __io_remove_buffers(ctx, bl, p->nbufs); | |
067524e9 | 4743 | if (ret < 0) |
93d2bcd2 | 4744 | req_set_fail(req); |
067524e9 | 4745 | |
9fb8cb49 PB |
4746 | /* complete before unlock, IOPOLL may need the lock */ |
4747 | __io_req_complete(req, issue_flags, ret, 0); | |
3b44b371 | 4748 | io_ring_submit_unlock(ctx, needs_lock); |
067524e9 JA |
4749 | return 0; |
4750 | } | |
4751 | ||
ddf0322d JA |
4752 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4753 | const struct io_uring_sqe *sqe) | |
4754 | { | |
38134ada | 4755 | unsigned long size, tmp_check; |
ddf0322d JA |
4756 | struct io_provide_buf *p = &req->pbuf; |
4757 | u64 tmp; | |
4758 | ||
26578cda | 4759 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
ddf0322d JA |
4760 | return -EINVAL; |
4761 | ||
4762 | tmp = READ_ONCE(sqe->fd); | |
4763 | if (!tmp || tmp > USHRT_MAX) | |
4764 | return -E2BIG; | |
4765 | p->nbufs = tmp; | |
4766 | p->addr = READ_ONCE(sqe->addr); | |
4767 | p->len = READ_ONCE(sqe->len); | |
4768 | ||
38134ada PB |
4769 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
4770 | &size)) | |
4771 | return -EOVERFLOW; | |
4772 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
4773 | return -EOVERFLOW; | |
4774 | ||
d81269fe PB |
4775 | size = (unsigned long)p->len * p->nbufs; |
4776 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
4777 | return -EFAULT; |
4778 | ||
4779 | p->bgid = READ_ONCE(sqe->buf_group); | |
4780 | tmp = READ_ONCE(sqe->off); | |
4781 | if (tmp > USHRT_MAX) | |
4782 | return -E2BIG; | |
4783 | p->bid = tmp; | |
4784 | return 0; | |
4785 | } | |
4786 | ||
cc3cec83 JA |
4787 | static int io_refill_buffer_cache(struct io_ring_ctx *ctx) |
4788 | { | |
4789 | struct io_buffer *buf; | |
4790 | struct page *page; | |
4791 | int bufs_in_page; | |
4792 | ||
4793 | /* | |
4794 | * Completions that don't happen inline (eg not under uring_lock) will | |
4795 | * add to ->io_buffers_comp. If we don't have any free buffers, check | |
4796 | * the completion list and splice those entries first. | |
4797 | */ | |
4798 | if (!list_empty_careful(&ctx->io_buffers_comp)) { | |
4799 | spin_lock(&ctx->completion_lock); | |
4800 | if (!list_empty(&ctx->io_buffers_comp)) { | |
4801 | list_splice_init(&ctx->io_buffers_comp, | |
4802 | &ctx->io_buffers_cache); | |
4803 | spin_unlock(&ctx->completion_lock); | |
4804 | return 0; | |
4805 | } | |
4806 | spin_unlock(&ctx->completion_lock); | |
4807 | } | |
4808 | ||
4809 | /* | |
4810 | * No free buffers and no completion entries either. Allocate a new | |
4811 | * page worth of buffer entries and add those to our freelist. | |
4812 | */ | |
4813 | page = alloc_page(GFP_KERNEL_ACCOUNT); | |
4814 | if (!page) | |
4815 | return -ENOMEM; | |
4816 | ||
4817 | list_add(&page->lru, &ctx->io_buffers_pages); | |
4818 | ||
4819 | buf = page_address(page); | |
4820 | bufs_in_page = PAGE_SIZE / sizeof(*buf); | |
4821 | while (bufs_in_page) { | |
4822 | list_add_tail(&buf->list, &ctx->io_buffers_cache); | |
4823 | buf++; | |
4824 | bufs_in_page--; | |
4825 | } | |
4826 | ||
4827 | return 0; | |
4828 | } | |
4829 | ||
4830 | static int io_add_buffers(struct io_ring_ctx *ctx, struct io_provide_buf *pbuf, | |
dbc7d452 | 4831 | struct io_buffer_list *bl) |
ddf0322d JA |
4832 | { |
4833 | struct io_buffer *buf; | |
4834 | u64 addr = pbuf->addr; | |
4835 | int i, bid = pbuf->bid; | |
4836 | ||
4837 | for (i = 0; i < pbuf->nbufs; i++) { | |
cc3cec83 JA |
4838 | if (list_empty(&ctx->io_buffers_cache) && |
4839 | io_refill_buffer_cache(ctx)) | |
ddf0322d | 4840 | break; |
cc3cec83 JA |
4841 | buf = list_first_entry(&ctx->io_buffers_cache, struct io_buffer, |
4842 | list); | |
dbc7d452 | 4843 | list_move_tail(&buf->list, &bl->buf_list); |
ddf0322d | 4844 | buf->addr = addr; |
d1f82808 | 4845 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d | 4846 | buf->bid = bid; |
b1c62645 | 4847 | buf->bgid = pbuf->bgid; |
ddf0322d JA |
4848 | addr += pbuf->len; |
4849 | bid++; | |
f240762f | 4850 | cond_resched(); |
ddf0322d JA |
4851 | } |
4852 | ||
dbc7d452 | 4853 | return i ? 0 : -ENOMEM; |
ddf0322d JA |
4854 | } |
4855 | ||
889fca73 | 4856 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4857 | { |
4858 | struct io_provide_buf *p = &req->pbuf; | |
4859 | struct io_ring_ctx *ctx = req->ctx; | |
dbc7d452 | 4860 | struct io_buffer_list *bl; |
ddf0322d | 4861 | int ret = 0; |
3b44b371 | 4862 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
ddf0322d | 4863 | |
3b44b371 | 4864 | io_ring_submit_lock(ctx, needs_lock); |
ddf0322d JA |
4865 | |
4866 | lockdep_assert_held(&ctx->uring_lock); | |
4867 | ||
dbc7d452 JA |
4868 | bl = io_buffer_get_list(ctx, p->bgid); |
4869 | if (unlikely(!bl)) { | |
4870 | bl = kmalloc(sizeof(*bl), GFP_KERNEL); | |
4871 | if (!bl) { | |
4872 | ret = -ENOMEM; | |
4873 | goto err; | |
4874 | } | |
4875 | io_buffer_add_list(ctx, bl, p->bgid); | |
ddf0322d | 4876 | } |
dbc7d452 JA |
4877 | |
4878 | ret = io_add_buffers(ctx, p, bl); | |
4879 | err: | |
ddf0322d | 4880 | if (ret < 0) |
93d2bcd2 | 4881 | req_set_fail(req); |
9fb8cb49 PB |
4882 | /* complete before unlock, IOPOLL may need the lock */ |
4883 | __io_req_complete(req, issue_flags, ret, 0); | |
3b44b371 | 4884 | io_ring_submit_unlock(ctx, needs_lock); |
ddf0322d | 4885 | return 0; |
cebdb986 JA |
4886 | } |
4887 | ||
3e4827b0 JA |
4888 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4889 | const struct io_uring_sqe *sqe) | |
4890 | { | |
4891 | #if defined(CONFIG_EPOLL) | |
26578cda | 4892 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
3e4827b0 | 4893 | return -EINVAL; |
2d74d042 | 4894 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4895 | return -EINVAL; |
3e4827b0 JA |
4896 | |
4897 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4898 | req->epoll.op = READ_ONCE(sqe->len); | |
4899 | req->epoll.fd = READ_ONCE(sqe->off); | |
4900 | ||
4901 | if (ep_op_has_event(req->epoll.op)) { | |
4902 | struct epoll_event __user *ev; | |
4903 | ||
4904 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4905 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4906 | return -EFAULT; | |
4907 | } | |
4908 | ||
4909 | return 0; | |
4910 | #else | |
4911 | return -EOPNOTSUPP; | |
4912 | #endif | |
4913 | } | |
4914 | ||
889fca73 | 4915 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4916 | { |
4917 | #if defined(CONFIG_EPOLL) | |
4918 | struct io_epoll *ie = &req->epoll; | |
4919 | int ret; | |
45d189c6 | 4920 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4921 | |
4922 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4923 | if (force_nonblock && ret == -EAGAIN) | |
4924 | return -EAGAIN; | |
4925 | ||
4926 | if (ret < 0) | |
93d2bcd2 | 4927 | req_set_fail(req); |
889fca73 | 4928 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4929 | return 0; |
4930 | #else | |
4931 | return -EOPNOTSUPP; | |
4932 | #endif | |
4933 | } | |
4934 | ||
c1ca757b JA |
4935 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4936 | { | |
4937 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
26578cda | 4938 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->splice_fd_in) |
c1ca757b | 4939 | return -EINVAL; |
3232dd02 PB |
4940 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4941 | return -EINVAL; | |
c1ca757b JA |
4942 | |
4943 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4944 | req->madvise.len = READ_ONCE(sqe->len); | |
4945 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4946 | return 0; | |
4947 | #else | |
4948 | return -EOPNOTSUPP; | |
4949 | #endif | |
4950 | } | |
4951 | ||
45d189c6 | 4952 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4953 | { |
4954 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4955 | struct io_madvise *ma = &req->madvise; | |
4956 | int ret; | |
4957 | ||
45d189c6 | 4958 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4959 | return -EAGAIN; |
4960 | ||
0726b01e | 4961 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b | 4962 | if (ret < 0) |
93d2bcd2 | 4963 | req_set_fail(req); |
e1e16097 | 4964 | io_req_complete(req, ret); |
c1ca757b JA |
4965 | return 0; |
4966 | #else | |
4967 | return -EOPNOTSUPP; | |
4968 | #endif | |
4969 | } | |
4970 | ||
4840e418 JA |
4971 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4972 | { | |
26578cda | 4973 | if (sqe->ioprio || sqe->buf_index || sqe->addr || sqe->splice_fd_in) |
4840e418 | 4974 | return -EINVAL; |
3232dd02 PB |
4975 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4976 | return -EINVAL; | |
4840e418 JA |
4977 | |
4978 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4979 | req->fadvise.len = READ_ONCE(sqe->len); | |
4980 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4981 | return 0; | |
4982 | } | |
4983 | ||
45d189c6 | 4984 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4985 | { |
4986 | struct io_fadvise *fa = &req->fadvise; | |
4987 | int ret; | |
4988 | ||
45d189c6 | 4989 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4990 | switch (fa->advice) { |
4991 | case POSIX_FADV_NORMAL: | |
4992 | case POSIX_FADV_RANDOM: | |
4993 | case POSIX_FADV_SEQUENTIAL: | |
4994 | break; | |
4995 | default: | |
4996 | return -EAGAIN; | |
4997 | } | |
4998 | } | |
4840e418 JA |
4999 | |
5000 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
5001 | if (ret < 0) | |
93d2bcd2 | 5002 | req_set_fail(req); |
0bdf3398 | 5003 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
5004 | return 0; |
5005 | } | |
5006 | ||
eddc7ef5 JA |
5007 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5008 | { | |
2d74d042 | 5009 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 5010 | return -EINVAL; |
26578cda | 5011 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
eddc7ef5 | 5012 | return -EINVAL; |
9c280f90 | 5013 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 5014 | return -EBADF; |
eddc7ef5 | 5015 | |
1d9e1288 BM |
5016 | req->statx.dfd = READ_ONCE(sqe->fd); |
5017 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 5018 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
5019 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
5020 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
5021 | |
5022 | return 0; | |
5023 | } | |
5024 | ||
45d189c6 | 5025 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 5026 | { |
1d9e1288 | 5027 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
5028 | int ret; |
5029 | ||
59d70013 | 5030 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
5031 | return -EAGAIN; |
5032 | ||
e62753e4 BM |
5033 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
5034 | ctx->buffer); | |
eddc7ef5 | 5035 | |
eddc7ef5 | 5036 | if (ret < 0) |
93d2bcd2 | 5037 | req_set_fail(req); |
e1e16097 | 5038 | io_req_complete(req, ret); |
eddc7ef5 JA |
5039 | return 0; |
5040 | } | |
5041 | ||
b5dba59e JA |
5042 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5043 | { | |
14587a46 | 5044 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 5045 | return -EINVAL; |
b5dba59e | 5046 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
7df778be | 5047 | sqe->rw_flags || sqe->buf_index) |
b5dba59e | 5048 | return -EINVAL; |
9c280f90 | 5049 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 5050 | return -EBADF; |
b5dba59e JA |
5051 | |
5052 | req->close.fd = READ_ONCE(sqe->fd); | |
7df778be PB |
5053 | req->close.file_slot = READ_ONCE(sqe->file_index); |
5054 | if (req->close.file_slot && req->close.fd) | |
5055 | return -EINVAL; | |
5056 | ||
b5dba59e | 5057 | return 0; |
b5dba59e JA |
5058 | } |
5059 | ||
889fca73 | 5060 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 5061 | { |
9eac1904 | 5062 | struct files_struct *files = current->files; |
3af73b28 | 5063 | struct io_close *close = &req->close; |
9eac1904 | 5064 | struct fdtable *fdt; |
a1fde923 PB |
5065 | struct file *file = NULL; |
5066 | int ret = -EBADF; | |
b5dba59e | 5067 | |
7df778be PB |
5068 | if (req->close.file_slot) { |
5069 | ret = io_close_fixed(req, issue_flags); | |
5070 | goto err; | |
5071 | } | |
5072 | ||
9eac1904 JA |
5073 | spin_lock(&files->file_lock); |
5074 | fdt = files_fdtable(files); | |
5075 | if (close->fd >= fdt->max_fds) { | |
5076 | spin_unlock(&files->file_lock); | |
5077 | goto err; | |
5078 | } | |
5079 | file = fdt->fd[close->fd]; | |
a1fde923 | 5080 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
5081 | spin_unlock(&files->file_lock); |
5082 | file = NULL; | |
5083 | goto err; | |
3af73b28 | 5084 | } |
b5dba59e JA |
5085 | |
5086 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 5087 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 5088 | spin_unlock(&files->file_lock); |
0bf0eefd | 5089 | return -EAGAIN; |
a2100672 | 5090 | } |
b5dba59e | 5091 | |
9eac1904 JA |
5092 | ret = __close_fd_get_file(close->fd, &file); |
5093 | spin_unlock(&files->file_lock); | |
5094 | if (ret < 0) { | |
5095 | if (ret == -ENOENT) | |
5096 | ret = -EBADF; | |
5097 | goto err; | |
5098 | } | |
5099 | ||
3af73b28 | 5100 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
5101 | ret = filp_close(file, current->files); |
5102 | err: | |
3af73b28 | 5103 | if (ret < 0) |
93d2bcd2 | 5104 | req_set_fail(req); |
9eac1904 JA |
5105 | if (file) |
5106 | fput(file); | |
889fca73 | 5107 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 5108 | return 0; |
b5dba59e JA |
5109 | } |
5110 | ||
1155c76a | 5111 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
5112 | { |
5113 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 5114 | |
5d17b4a4 JA |
5115 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
5116 | return -EINVAL; | |
26578cda PB |
5117 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || |
5118 | sqe->splice_fd_in)) | |
5d17b4a4 JA |
5119 | return -EINVAL; |
5120 | ||
8ed8d3c3 JA |
5121 | req->sync.off = READ_ONCE(sqe->off); |
5122 | req->sync.len = READ_ONCE(sqe->len); | |
5123 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
5124 | return 0; |
5125 | } | |
5126 | ||
45d189c6 | 5127 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 5128 | { |
8ed8d3c3 JA |
5129 | int ret; |
5130 | ||
ac45abc0 | 5131 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 5132 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
5133 | return -EAGAIN; |
5134 | ||
9adbd45d | 5135 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
5136 | req->sync.flags); |
5137 | if (ret < 0) | |
93d2bcd2 | 5138 | req_set_fail(req); |
e1e16097 | 5139 | io_req_complete(req, ret); |
5d17b4a4 JA |
5140 | return 0; |
5141 | } | |
5142 | ||
469956e8 | 5143 | #if defined(CONFIG_NET) |
02d27d89 PB |
5144 | static int io_setup_async_msg(struct io_kiocb *req, |
5145 | struct io_async_msghdr *kmsg) | |
5146 | { | |
e8c2bc1f JA |
5147 | struct io_async_msghdr *async_msg = req->async_data; |
5148 | ||
5149 | if (async_msg) | |
02d27d89 | 5150 | return -EAGAIN; |
e8c2bc1f | 5151 | if (io_alloc_async_data(req)) { |
257e84a5 | 5152 | kfree(kmsg->free_iov); |
02d27d89 PB |
5153 | return -ENOMEM; |
5154 | } | |
e8c2bc1f | 5155 | async_msg = req->async_data; |
02d27d89 | 5156 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 5157 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 5158 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
5159 | /* if were using fast_iov, set it to the new one */ |
5160 | if (!async_msg->free_iov) | |
5161 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
5162 | ||
02d27d89 PB |
5163 | return -EAGAIN; |
5164 | } | |
5165 | ||
2ae523ed PB |
5166 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
5167 | struct io_async_msghdr *iomsg) | |
5168 | { | |
2ae523ed | 5169 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 5170 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 5171 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 5172 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
5173 | } |
5174 | ||
93642ef8 PB |
5175 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
5176 | { | |
5177 | int ret; | |
5178 | ||
93642ef8 PB |
5179 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
5180 | if (!ret) | |
5181 | req->flags |= REQ_F_NEED_CLEANUP; | |
5182 | return ret; | |
5183 | } | |
5184 | ||
3529d8c2 | 5185 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 5186 | { |
e47293fd | 5187 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 5188 | |
d2b6f48b PB |
5189 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5190 | return -EINVAL; | |
5191 | ||
270a5940 | 5192 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 5193 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
5194 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
5195 | if (sr->msg_flags & MSG_DONTWAIT) | |
5196 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 5197 | |
d8768362 JA |
5198 | #ifdef CONFIG_COMPAT |
5199 | if (req->ctx->compat) | |
5200 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
5201 | #endif | |
93642ef8 | 5202 | return 0; |
03b1230c JA |
5203 | } |
5204 | ||
889fca73 | 5205 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 5206 | { |
6b754c8b | 5207 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 5208 | struct socket *sock; |
7a7cacba | 5209 | unsigned flags; |
0031275d | 5210 | int min_ret = 0; |
0fa03c62 JA |
5211 | int ret; |
5212 | ||
dba4a925 | 5213 | sock = sock_from_file(req->file); |
7a7cacba | 5214 | if (unlikely(!sock)) |
dba4a925 | 5215 | return -ENOTSOCK; |
3529d8c2 | 5216 | |
d886e185 PB |
5217 | if (req_has_async_data(req)) { |
5218 | kmsg = req->async_data; | |
5219 | } else { | |
7a7cacba PB |
5220 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
5221 | if (ret) | |
5222 | return ret; | |
5223 | kmsg = &iomsg; | |
0fa03c62 | 5224 | } |
0fa03c62 | 5225 | |
04411806 PB |
5226 | flags = req->sr_msg.msg_flags; |
5227 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 5228 | flags |= MSG_DONTWAIT; |
0031275d SM |
5229 | if (flags & MSG_WAITALL) |
5230 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
5231 | ||
7a7cacba | 5232 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
0fa03c62 | 5233 | |
7297ce3d PB |
5234 | if (ret < min_ret) { |
5235 | if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) | |
5236 | return io_setup_async_msg(req, kmsg); | |
5237 | if (ret == -ERESTARTSYS) | |
5238 | ret = -EINTR; | |
5239 | req_set_fail(req); | |
5240 | } | |
257e84a5 PB |
5241 | /* fast path, check for non-NULL to avoid function call */ |
5242 | if (kmsg->free_iov) | |
5243 | kfree(kmsg->free_iov); | |
99bc4c38 | 5244 | req->flags &= ~REQ_F_NEED_CLEANUP; |
889fca73 | 5245 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 5246 | return 0; |
03b1230c | 5247 | } |
aa1fa28f | 5248 | |
889fca73 | 5249 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 5250 | { |
7a7cacba PB |
5251 | struct io_sr_msg *sr = &req->sr_msg; |
5252 | struct msghdr msg; | |
5253 | struct iovec iov; | |
fddaface | 5254 | struct socket *sock; |
7a7cacba | 5255 | unsigned flags; |
0031275d | 5256 | int min_ret = 0; |
fddaface JA |
5257 | int ret; |
5258 | ||
dba4a925 | 5259 | sock = sock_from_file(req->file); |
7a7cacba | 5260 | if (unlikely(!sock)) |
dba4a925 | 5261 | return -ENOTSOCK; |
fddaface | 5262 | |
7a7cacba PB |
5263 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
5264 | if (unlikely(ret)) | |
14db8411 | 5265 | return ret; |
fddaface | 5266 | |
7a7cacba PB |
5267 | msg.msg_name = NULL; |
5268 | msg.msg_control = NULL; | |
5269 | msg.msg_controllen = 0; | |
5270 | msg.msg_namelen = 0; | |
fddaface | 5271 | |
04411806 PB |
5272 | flags = req->sr_msg.msg_flags; |
5273 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 5274 | flags |= MSG_DONTWAIT; |
0031275d SM |
5275 | if (flags & MSG_WAITALL) |
5276 | min_ret = iov_iter_count(&msg.msg_iter); | |
5277 | ||
7a7cacba PB |
5278 | msg.msg_flags = flags; |
5279 | ret = sock_sendmsg(sock, &msg); | |
7297ce3d PB |
5280 | if (ret < min_ret) { |
5281 | if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) | |
5282 | return -EAGAIN; | |
5283 | if (ret == -ERESTARTSYS) | |
5284 | ret = -EINTR; | |
93d2bcd2 | 5285 | req_set_fail(req); |
7297ce3d | 5286 | } |
889fca73 | 5287 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 5288 | return 0; |
fddaface JA |
5289 | } |
5290 | ||
1400e697 PB |
5291 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
5292 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
5293 | { |
5294 | struct io_sr_msg *sr = &req->sr_msg; | |
5295 | struct iovec __user *uiov; | |
5296 | size_t iov_len; | |
5297 | int ret; | |
5298 | ||
1400e697 PB |
5299 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
5300 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
5301 | if (ret) |
5302 | return ret; | |
5303 | ||
5304 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
5305 | if (iov_len > 1) | |
5306 | return -EINVAL; | |
5476dfed | 5307 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 5308 | return -EFAULT; |
5476dfed | 5309 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 5310 | iomsg->free_iov = NULL; |
52de1fe1 | 5311 | } else { |
257e84a5 | 5312 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 5313 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 5314 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 5315 | false); |
52de1fe1 JA |
5316 | if (ret > 0) |
5317 | ret = 0; | |
5318 | } | |
5319 | ||
5320 | return ret; | |
5321 | } | |
5322 | ||
5323 | #ifdef CONFIG_COMPAT | |
5324 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 5325 | struct io_async_msghdr *iomsg) |
52de1fe1 | 5326 | { |
52de1fe1 JA |
5327 | struct io_sr_msg *sr = &req->sr_msg; |
5328 | struct compat_iovec __user *uiov; | |
5329 | compat_uptr_t ptr; | |
5330 | compat_size_t len; | |
5331 | int ret; | |
5332 | ||
4af3417a PB |
5333 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
5334 | &ptr, &len); | |
52de1fe1 JA |
5335 | if (ret) |
5336 | return ret; | |
5337 | ||
5338 | uiov = compat_ptr(ptr); | |
5339 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
5340 | compat_ssize_t clen; | |
5341 | ||
5342 | if (len > 1) | |
5343 | return -EINVAL; | |
5344 | if (!access_ok(uiov, sizeof(*uiov))) | |
5345 | return -EFAULT; | |
5346 | if (__get_user(clen, &uiov->iov_len)) | |
5347 | return -EFAULT; | |
5348 | if (clen < 0) | |
5349 | return -EINVAL; | |
2d280bc8 | 5350 | sr->len = clen; |
257e84a5 | 5351 | iomsg->free_iov = NULL; |
52de1fe1 | 5352 | } else { |
257e84a5 | 5353 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 5354 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 5355 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 5356 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
5357 | if (ret < 0) |
5358 | return ret; | |
5359 | } | |
5360 | ||
5361 | return 0; | |
5362 | } | |
5363 | #endif | |
5364 | ||
1400e697 PB |
5365 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
5366 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 5367 | { |
1400e697 | 5368 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
5369 | |
5370 | #ifdef CONFIG_COMPAT | |
5371 | if (req->ctx->compat) | |
1400e697 | 5372 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 5373 | #endif |
52de1fe1 | 5374 | |
1400e697 | 5375 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
5376 | } |
5377 | ||
bcda7baa | 5378 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
51aac424 | 5379 | unsigned int issue_flags) |
bcda7baa JA |
5380 | { |
5381 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 5382 | |
51aac424 | 5383 | return io_buffer_select(req, &sr->len, sr->bgid, issue_flags); |
fddaface JA |
5384 | } |
5385 | ||
93642ef8 | 5386 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 5387 | { |
99bc4c38 | 5388 | int ret; |
3529d8c2 | 5389 | |
93642ef8 PB |
5390 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
5391 | if (!ret) | |
5392 | req->flags |= REQ_F_NEED_CLEANUP; | |
5393 | return ret; | |
5394 | } | |
5395 | ||
5396 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5397 | { | |
5398 | struct io_sr_msg *sr = &req->sr_msg; | |
5399 | ||
d2b6f48b PB |
5400 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5401 | return -EINVAL; | |
5402 | ||
270a5940 | 5403 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 5404 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 5405 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
5406 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
5407 | if (sr->msg_flags & MSG_DONTWAIT) | |
5408 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 5409 | |
d8768362 JA |
5410 | #ifdef CONFIG_COMPAT |
5411 | if (req->ctx->compat) | |
5412 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
5413 | #endif | |
93642ef8 | 5414 | return 0; |
aa1fa28f JA |
5415 | } |
5416 | ||
889fca73 | 5417 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 5418 | { |
6b754c8b | 5419 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 5420 | struct socket *sock; |
7fbb1b54 | 5421 | struct io_buffer *kbuf; |
7a7cacba | 5422 | unsigned flags; |
d1fd1c20 | 5423 | int ret, min_ret = 0; |
45d189c6 | 5424 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 5425 | |
dba4a925 | 5426 | sock = sock_from_file(req->file); |
7a7cacba | 5427 | if (unlikely(!sock)) |
dba4a925 | 5428 | return -ENOTSOCK; |
3529d8c2 | 5429 | |
d886e185 PB |
5430 | if (req_has_async_data(req)) { |
5431 | kmsg = req->async_data; | |
5432 | } else { | |
7a7cacba PB |
5433 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
5434 | if (ret) | |
681fda8d | 5435 | return ret; |
7a7cacba PB |
5436 | kmsg = &iomsg; |
5437 | } | |
03b1230c | 5438 | |
bc02ef33 | 5439 | if (req->flags & REQ_F_BUFFER_SELECT) { |
51aac424 | 5440 | kbuf = io_recv_buffer_select(req, issue_flags); |
bc02ef33 | 5441 | if (IS_ERR(kbuf)) |
52de1fe1 | 5442 | return PTR_ERR(kbuf); |
7a7cacba | 5443 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
5444 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
5445 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
5446 | 1, req->sr_msg.len); |
5447 | } | |
52de1fe1 | 5448 | |
04411806 PB |
5449 | flags = req->sr_msg.msg_flags; |
5450 | if (force_nonblock) | |
7a7cacba | 5451 | flags |= MSG_DONTWAIT; |
0031275d SM |
5452 | if (flags & MSG_WAITALL) |
5453 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
5454 | ||
7a7cacba PB |
5455 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
5456 | kmsg->uaddr, flags); | |
7297ce3d PB |
5457 | if (ret < min_ret) { |
5458 | if (ret == -EAGAIN && force_nonblock) | |
5459 | return io_setup_async_msg(req, kmsg); | |
5460 | if (ret == -ERESTARTSYS) | |
5461 | ret = -EINTR; | |
5462 | req_set_fail(req); | |
5463 | } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) { | |
5464 | req_set_fail(req); | |
5465 | } | |
03b1230c | 5466 | |
257e84a5 PB |
5467 | /* fast path, check for non-NULL to avoid function call */ |
5468 | if (kmsg->free_iov) | |
5469 | kfree(kmsg->free_iov); | |
99bc4c38 | 5470 | req->flags &= ~REQ_F_NEED_CLEANUP; |
cc3cec83 | 5471 | __io_req_complete(req, issue_flags, ret, io_put_kbuf(req, issue_flags)); |
03b1230c | 5472 | return 0; |
0fa03c62 | 5473 | } |
5d17b4a4 | 5474 | |
889fca73 | 5475 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 5476 | { |
6b754c8b | 5477 | struct io_buffer *kbuf; |
7a7cacba PB |
5478 | struct io_sr_msg *sr = &req->sr_msg; |
5479 | struct msghdr msg; | |
5480 | void __user *buf = sr->buf; | |
fddaface | 5481 | struct socket *sock; |
7a7cacba PB |
5482 | struct iovec iov; |
5483 | unsigned flags; | |
d1fd1c20 | 5484 | int ret, min_ret = 0; |
45d189c6 | 5485 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 5486 | |
dba4a925 | 5487 | sock = sock_from_file(req->file); |
7a7cacba | 5488 | if (unlikely(!sock)) |
dba4a925 | 5489 | return -ENOTSOCK; |
fddaface | 5490 | |
bc02ef33 | 5491 | if (req->flags & REQ_F_BUFFER_SELECT) { |
51aac424 | 5492 | kbuf = io_recv_buffer_select(req, issue_flags); |
bcda7baa JA |
5493 | if (IS_ERR(kbuf)) |
5494 | return PTR_ERR(kbuf); | |
7a7cacba | 5495 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 5496 | } |
bcda7baa | 5497 | |
7a7cacba | 5498 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
5499 | if (unlikely(ret)) |
5500 | goto out_free; | |
fddaface | 5501 | |
7a7cacba PB |
5502 | msg.msg_name = NULL; |
5503 | msg.msg_control = NULL; | |
5504 | msg.msg_controllen = 0; | |
5505 | msg.msg_namelen = 0; | |
5506 | msg.msg_iocb = NULL; | |
5507 | msg.msg_flags = 0; | |
fddaface | 5508 | |
04411806 PB |
5509 | flags = req->sr_msg.msg_flags; |
5510 | if (force_nonblock) | |
7a7cacba | 5511 | flags |= MSG_DONTWAIT; |
0031275d SM |
5512 | if (flags & MSG_WAITALL) |
5513 | min_ret = iov_iter_count(&msg.msg_iter); | |
5514 | ||
7a7cacba | 5515 | ret = sock_recvmsg(sock, &msg, flags); |
7297ce3d PB |
5516 | if (ret < min_ret) { |
5517 | if (ret == -EAGAIN && force_nonblock) | |
5518 | return -EAGAIN; | |
5519 | if (ret == -ERESTARTSYS) | |
5520 | ret = -EINTR; | |
5521 | req_set_fail(req); | |
5522 | } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) { | |
0d7c1153 | 5523 | out_free: |
93d2bcd2 | 5524 | req_set_fail(req); |
7297ce3d | 5525 | } |
cc3cec83 JA |
5526 | |
5527 | __io_req_complete(req, issue_flags, ret, io_put_kbuf(req, issue_flags)); | |
fddaface | 5528 | return 0; |
fddaface JA |
5529 | } |
5530 | ||
3529d8c2 | 5531 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5532 | { |
8ed8d3c3 JA |
5533 | struct io_accept *accept = &req->accept; |
5534 | ||
14587a46 | 5535 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5536 | return -EINVAL; |
aaa4db12 | 5537 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5538 | return -EINVAL; |
5539 | ||
d55e5f5b JA |
5540 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5541 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5542 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5543 | accept->nofile = rlimit(RLIMIT_NOFILE); |
a7083ad5 | 5544 | |
aaa4db12 | 5545 | accept->file_slot = READ_ONCE(sqe->file_index); |
adf3a9e9 | 5546 | if (accept->file_slot && (accept->flags & SOCK_CLOEXEC)) |
aaa4db12 | 5547 | return -EINVAL; |
a7083ad5 PB |
5548 | if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) |
5549 | return -EINVAL; | |
5550 | if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) | |
5551 | accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; | |
8ed8d3c3 | 5552 | return 0; |
8ed8d3c3 | 5553 | } |
17f2fe35 | 5554 | |
889fca73 | 5555 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5556 | { |
5557 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5558 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5559 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
aaa4db12 | 5560 | bool fixed = !!accept->file_slot; |
a7083ad5 PB |
5561 | struct file *file; |
5562 | int ret, fd; | |
8ed8d3c3 | 5563 | |
e697deed JX |
5564 | if (req->file->f_flags & O_NONBLOCK) |
5565 | req->flags |= REQ_F_NOWAIT; | |
5566 | ||
aaa4db12 PB |
5567 | if (!fixed) { |
5568 | fd = __get_unused_fd_flags(accept->flags, accept->nofile); | |
5569 | if (unlikely(fd < 0)) | |
5570 | return fd; | |
5571 | } | |
a7083ad5 PB |
5572 | file = do_accept(req->file, file_flags, accept->addr, accept->addr_len, |
5573 | accept->flags); | |
5574 | if (IS_ERR(file)) { | |
aaa4db12 PB |
5575 | if (!fixed) |
5576 | put_unused_fd(fd); | |
a7083ad5 PB |
5577 | ret = PTR_ERR(file); |
5578 | if (ret == -EAGAIN && force_nonblock) | |
5579 | return -EAGAIN; | |
ac45abc0 PB |
5580 | if (ret == -ERESTARTSYS) |
5581 | ret = -EINTR; | |
93d2bcd2 | 5582 | req_set_fail(req); |
aaa4db12 | 5583 | } else if (!fixed) { |
a7083ad5 PB |
5584 | fd_install(fd, file); |
5585 | ret = fd; | |
aaa4db12 PB |
5586 | } else { |
5587 | ret = io_install_fixed_file(req, file, issue_flags, | |
5588 | accept->file_slot - 1); | |
ac45abc0 | 5589 | } |
889fca73 | 5590 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5591 | return 0; |
8ed8d3c3 JA |
5592 | } |
5593 | ||
93642ef8 PB |
5594 | static int io_connect_prep_async(struct io_kiocb *req) |
5595 | { | |
5596 | struct io_async_connect *io = req->async_data; | |
5597 | struct io_connect *conn = &req->connect; | |
5598 | ||
5599 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
5600 | } | |
5601 | ||
3529d8c2 | 5602 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5603 | { |
3529d8c2 | 5604 | struct io_connect *conn = &req->connect; |
f499a021 | 5605 | |
14587a46 | 5606 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 | 5607 | return -EINVAL; |
26578cda PB |
5608 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags || |
5609 | sqe->splice_fd_in) | |
3fbb51c1 JA |
5610 | return -EINVAL; |
5611 | ||
3529d8c2 JA |
5612 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5613 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 5614 | return 0; |
f499a021 JA |
5615 | } |
5616 | ||
889fca73 | 5617 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5618 | { |
e8c2bc1f | 5619 | struct io_async_connect __io, *io; |
f8e85cf2 | 5620 | unsigned file_flags; |
3fbb51c1 | 5621 | int ret; |
45d189c6 | 5622 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5623 | |
d886e185 | 5624 | if (req_has_async_data(req)) { |
e8c2bc1f | 5625 | io = req->async_data; |
f499a021 | 5626 | } else { |
3529d8c2 JA |
5627 | ret = move_addr_to_kernel(req->connect.addr, |
5628 | req->connect.addr_len, | |
e8c2bc1f | 5629 | &__io.address); |
f499a021 JA |
5630 | if (ret) |
5631 | goto out; | |
5632 | io = &__io; | |
5633 | } | |
5634 | ||
3fbb51c1 JA |
5635 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5636 | ||
e8c2bc1f | 5637 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5638 | req->connect.addr_len, file_flags); |
87f80d62 | 5639 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
d886e185 | 5640 | if (req_has_async_data(req)) |
b7bb4f7d | 5641 | return -EAGAIN; |
e8c2bc1f | 5642 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5643 | ret = -ENOMEM; |
5644 | goto out; | |
5645 | } | |
e8c2bc1f | 5646 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 5647 | return -EAGAIN; |
f499a021 | 5648 | } |
f8e85cf2 JA |
5649 | if (ret == -ERESTARTSYS) |
5650 | ret = -EINTR; | |
f499a021 | 5651 | out: |
4e88d6e7 | 5652 | if (ret < 0) |
93d2bcd2 | 5653 | req_set_fail(req); |
889fca73 | 5654 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5655 | return 0; |
469956e8 Y |
5656 | } |
5657 | #else /* !CONFIG_NET */ | |
99a10081 JA |
5658 | #define IO_NETOP_FN(op) \ |
5659 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
5660 | { \ | |
5661 | return -EOPNOTSUPP; \ | |
5662 | } | |
5663 | ||
5664 | #define IO_NETOP_PREP(op) \ | |
5665 | IO_NETOP_FN(op) \ | |
5666 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
5667 | { \ | |
5668 | return -EOPNOTSUPP; \ | |
5669 | } \ | |
5670 | ||
5671 | #define IO_NETOP_PREP_ASYNC(op) \ | |
5672 | IO_NETOP_PREP(op) \ | |
5673 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
5674 | { \ | |
5675 | return -EOPNOTSUPP; \ | |
5676 | } | |
5677 | ||
5678 | IO_NETOP_PREP_ASYNC(sendmsg); | |
5679 | IO_NETOP_PREP_ASYNC(recvmsg); | |
5680 | IO_NETOP_PREP_ASYNC(connect); | |
5681 | IO_NETOP_PREP(accept); | |
5682 | IO_NETOP_FN(send); | |
5683 | IO_NETOP_FN(recv); | |
469956e8 | 5684 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5685 | |
adc8682e OL |
5686 | #ifdef CONFIG_NET_RX_BUSY_POLL |
5687 | ||
5688 | #define NAPI_TIMEOUT (60 * SEC_CONVERSION) | |
5689 | ||
5690 | struct napi_entry { | |
5691 | struct list_head list; | |
5692 | unsigned int napi_id; | |
5693 | unsigned long timeout; | |
5694 | }; | |
5695 | ||
5696 | /* | |
5697 | * Add busy poll NAPI ID from sk. | |
5698 | */ | |
5699 | static void io_add_napi(struct file *file, struct io_ring_ctx *ctx) | |
5700 | { | |
5701 | unsigned int napi_id; | |
5702 | struct socket *sock; | |
5703 | struct sock *sk; | |
5704 | struct napi_entry *ne; | |
5705 | ||
5706 | if (!net_busy_loop_on()) | |
5707 | return; | |
5708 | ||
5709 | sock = sock_from_file(file); | |
5710 | if (!sock) | |
5711 | return; | |
5712 | ||
5713 | sk = sock->sk; | |
5714 | if (!sk) | |
5715 | return; | |
5716 | ||
5717 | napi_id = READ_ONCE(sk->sk_napi_id); | |
5718 | ||
5719 | /* Non-NAPI IDs can be rejected */ | |
5720 | if (napi_id < MIN_NAPI_ID) | |
5721 | return; | |
5722 | ||
5723 | spin_lock(&ctx->napi_lock); | |
5724 | list_for_each_entry(ne, &ctx->napi_list, list) { | |
5725 | if (ne->napi_id == napi_id) { | |
5726 | ne->timeout = jiffies + NAPI_TIMEOUT; | |
5727 | goto out; | |
5728 | } | |
5729 | } | |
5730 | ||
5731 | ne = kmalloc(sizeof(*ne), GFP_NOWAIT); | |
5732 | if (!ne) | |
5733 | goto out; | |
5734 | ||
5735 | ne->napi_id = napi_id; | |
5736 | ne->timeout = jiffies + NAPI_TIMEOUT; | |
5737 | list_add_tail(&ne->list, &ctx->napi_list); | |
5738 | out: | |
5739 | spin_unlock(&ctx->napi_lock); | |
5740 | } | |
5741 | ||
5742 | static inline void io_check_napi_entry_timeout(struct napi_entry *ne) | |
5743 | { | |
5744 | if (time_after(jiffies, ne->timeout)) { | |
5745 | list_del(&ne->list); | |
5746 | kfree(ne); | |
5747 | } | |
5748 | } | |
5749 | ||
5750 | /* | |
5751 | * Busy poll if globally on and supporting sockets found | |
5752 | */ | |
5753 | static bool io_napi_busy_loop(struct list_head *napi_list) | |
5754 | { | |
5755 | struct napi_entry *ne, *n; | |
5756 | ||
5757 | list_for_each_entry_safe(ne, n, napi_list, list) { | |
5758 | napi_busy_loop(ne->napi_id, NULL, NULL, true, | |
5759 | BUSY_POLL_BUDGET); | |
5760 | io_check_napi_entry_timeout(ne); | |
5761 | } | |
5762 | return !list_empty(napi_list); | |
5763 | } | |
5764 | ||
5765 | static void io_free_napi_list(struct io_ring_ctx *ctx) | |
5766 | { | |
5767 | spin_lock(&ctx->napi_lock); | |
5768 | while (!list_empty(&ctx->napi_list)) { | |
5769 | struct napi_entry *ne = | |
5770 | list_first_entry(&ctx->napi_list, struct napi_entry, | |
5771 | list); | |
5772 | ||
5773 | list_del(&ne->list); | |
5774 | kfree(ne); | |
5775 | } | |
5776 | spin_unlock(&ctx->napi_lock); | |
5777 | } | |
5778 | #else | |
5779 | static inline void io_add_napi(struct file *file, struct io_ring_ctx *ctx) | |
5780 | { | |
5781 | } | |
5782 | ||
5783 | static inline void io_free_napi_list(struct io_ring_ctx *ctx) | |
5784 | { | |
5785 | } | |
5786 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
5787 | ||
d7718a9d JA |
5788 | struct io_poll_table { |
5789 | struct poll_table_struct pt; | |
5790 | struct io_kiocb *req; | |
68b11e8b | 5791 | int nr_entries; |
d7718a9d JA |
5792 | int error; |
5793 | }; | |
ce593a6c | 5794 | |
aa43477b PB |
5795 | #define IO_POLL_CANCEL_FLAG BIT(31) |
5796 | #define IO_POLL_REF_MASK ((1u << 20)-1) | |
6d816e08 | 5797 | |
aa43477b PB |
5798 | /* |
5799 | * If refs part of ->poll_refs (see IO_POLL_REF_MASK) is 0, it's free. We can | |
5800 | * bump it and acquire ownership. It's disallowed to modify requests while not | |
5801 | * owning it, that prevents from races for enqueueing task_work's and b/w | |
5802 | * arming poll and wakeups. | |
5803 | */ | |
5804 | static inline bool io_poll_get_ownership(struct io_kiocb *req) | |
5805 | { | |
5806 | return !(atomic_fetch_inc(&req->poll_refs) & IO_POLL_REF_MASK); | |
d7718a9d JA |
5807 | } |
5808 | ||
aa43477b | 5809 | static void io_poll_mark_cancelled(struct io_kiocb *req) |
74ce6ce4 | 5810 | { |
aa43477b | 5811 | atomic_or(IO_POLL_CANCEL_FLAG, &req->poll_refs); |
74ce6ce4 JA |
5812 | } |
5813 | ||
d4e7cd36 | 5814 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5815 | { |
e8c2bc1f | 5816 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5817 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5818 | return req->async_data; |
d4e7cd36 JA |
5819 | return req->apoll->double_poll; |
5820 | } | |
5821 | ||
5822 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5823 | { | |
5824 | if (req->opcode == IORING_OP_POLL_ADD) | |
5825 | return &req->poll; | |
5826 | return &req->apoll->poll; | |
5827 | } | |
5828 | ||
5641897a | 5829 | static void io_poll_req_insert(struct io_kiocb *req) |
d4e7cd36 | 5830 | { |
5641897a PB |
5831 | struct io_ring_ctx *ctx = req->ctx; |
5832 | struct hlist_head *list; | |
18bceab1 | 5833 | |
5641897a PB |
5834 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; |
5835 | hlist_add_head(&req->hash_node, list); | |
18bceab1 JA |
5836 | } |
5837 | ||
5641897a PB |
5838 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, |
5839 | wait_queue_func_t wake_func) | |
18bceab1 | 5840 | { |
5641897a | 5841 | poll->head = NULL; |
5641897a PB |
5842 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5843 | /* mask in events that we always want/need */ | |
5844 | poll->events = events | IO_POLL_UNMASK; | |
5845 | INIT_LIST_HEAD(&poll->wait.entry); | |
5846 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
18bceab1 JA |
5847 | } |
5848 | ||
aa43477b | 5849 | static inline void io_poll_remove_entry(struct io_poll_iocb *poll) |
18bceab1 | 5850 | { |
791f3465 | 5851 | struct wait_queue_head *head = smp_load_acquire(&poll->head); |
18bceab1 | 5852 | |
791f3465 PB |
5853 | if (head) { |
5854 | spin_lock_irq(&head->lock); | |
5855 | list_del_init(&poll->wait.entry); | |
5856 | poll->head = NULL; | |
5857 | spin_unlock_irq(&head->lock); | |
5858 | } | |
aa43477b | 5859 | } |
18bceab1 | 5860 | |
aa43477b PB |
5861 | static void io_poll_remove_entries(struct io_kiocb *req) |
5862 | { | |
91eac1c6 JA |
5863 | /* |
5864 | * Nothing to do if neither of those flags are set. Avoid dipping | |
5865 | * into the poll/apoll/double cachelines if we can. | |
5866 | */ | |
5867 | if (!(req->flags & (REQ_F_SINGLE_POLL | REQ_F_DOUBLE_POLL))) | |
5868 | return; | |
18bceab1 | 5869 | |
791f3465 PB |
5870 | /* |
5871 | * While we hold the waitqueue lock and the waitqueue is nonempty, | |
5872 | * wake_up_pollfree() will wait for us. However, taking the waitqueue | |
5873 | * lock in the first place can race with the waitqueue being freed. | |
5874 | * | |
5875 | * We solve this as eventpoll does: by taking advantage of the fact that | |
5876 | * all users of wake_up_pollfree() will RCU-delay the actual free. If | |
5877 | * we enter rcu_read_lock() and see that the pointer to the queue is | |
5878 | * non-NULL, we can then lock it without the memory being freed out from | |
5879 | * under us. | |
5880 | * | |
5881 | * Keep holding rcu_read_lock() as long as we hold the queue lock, in | |
5882 | * case the caller deletes the entry from the queue, leaving it empty. | |
5883 | * In that case, only RCU prevents the queue memory from being freed. | |
5884 | */ | |
5885 | rcu_read_lock(); | |
91eac1c6 JA |
5886 | if (req->flags & REQ_F_SINGLE_POLL) |
5887 | io_poll_remove_entry(io_poll_get_single(req)); | |
5888 | if (req->flags & REQ_F_DOUBLE_POLL) | |
5889 | io_poll_remove_entry(io_poll_get_double(req)); | |
791f3465 | 5890 | rcu_read_unlock(); |
18bceab1 JA |
5891 | } |
5892 | ||
aa43477b PB |
5893 | /* |
5894 | * All poll tw should go through this. Checks for poll events, manages | |
5895 | * references, does rewait, etc. | |
5896 | * | |
5897 | * Returns a negative error on failure. >0 when no action require, which is | |
5898 | * either spurious wakeup or multishot CQE is served. 0 when it's done with | |
5899 | * the request, then the mask is stored in req->result. | |
5900 | */ | |
5901 | static int io_poll_check_events(struct io_kiocb *req) | |
18bceab1 | 5902 | { |
74ce6ce4 | 5903 | struct io_ring_ctx *ctx = req->ctx; |
d4e7cd36 | 5904 | struct io_poll_iocb *poll = io_poll_get_single(req); |
aa43477b | 5905 | int v; |
18bceab1 | 5906 | |
316319e8 | 5907 | /* req->task == current here, checking PF_EXITING is safe */ |
e09ee510 | 5908 | if (unlikely(req->task->flags & PF_EXITING)) |
aa43477b | 5909 | io_poll_mark_cancelled(req); |
18bceab1 | 5910 | |
aa43477b PB |
5911 | do { |
5912 | v = atomic_read(&req->poll_refs); | |
74ce6ce4 | 5913 | |
aa43477b PB |
5914 | /* tw handler should be the owner, and so have some references */ |
5915 | if (WARN_ON_ONCE(!(v & IO_POLL_REF_MASK))) | |
5916 | return 0; | |
5917 | if (v & IO_POLL_CANCEL_FLAG) | |
5918 | return -ECANCELED; | |
8706e04e | 5919 | |
aa43477b | 5920 | if (!req->result) { |
81459350 | 5921 | struct poll_table_struct pt = { ._key = req->cflags }; |
18bceab1 | 5922 | |
81459350 | 5923 | req->result = vfs_poll(req->file, &pt) & req->cflags; |
c8b5e260 | 5924 | } |
74ce6ce4 | 5925 | |
aa43477b | 5926 | /* multishot, just fill an CQE and proceed */ |
81459350 | 5927 | if (req->result && !(req->cflags & EPOLLONESHOT)) { |
aa43477b PB |
5928 | __poll_t mask = mangle_poll(req->result & poll->events); |
5929 | bool filled; | |
18bceab1 | 5930 | |
aa43477b PB |
5931 | spin_lock(&ctx->completion_lock); |
5932 | filled = io_fill_cqe_aux(ctx, req->user_data, mask, | |
5933 | IORING_CQE_F_MORE); | |
5934 | io_commit_cqring(ctx); | |
5935 | spin_unlock(&ctx->completion_lock); | |
5936 | if (unlikely(!filled)) | |
5937 | return -ECANCELED; | |
5938 | io_cqring_ev_posted(ctx); | |
adc8682e | 5939 | io_add_napi(req->file, ctx); |
aa43477b PB |
5940 | } else if (req->result) { |
5941 | return 0; | |
5942 | } | |
18bceab1 | 5943 | |
aa43477b PB |
5944 | /* |
5945 | * Release all references, retry if someone tried to restart | |
5946 | * task_work while we were executing it. | |
5947 | */ | |
5948 | } while (atomic_sub_return(v & IO_POLL_REF_MASK, &req->poll_refs)); | |
18bceab1 | 5949 | |
18bceab1 JA |
5950 | return 1; |
5951 | } | |
5952 | ||
aa43477b | 5953 | static void io_poll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 | 5954 | { |
18bceab1 | 5955 | struct io_ring_ctx *ctx = req->ctx; |
aa43477b | 5956 | int ret; |
18bceab1 | 5957 | |
aa43477b PB |
5958 | ret = io_poll_check_events(req); |
5959 | if (ret > 0) | |
5960 | return; | |
5961 | ||
5962 | if (!ret) { | |
5963 | req->result = mangle_poll(req->result & req->poll.events); | |
e27414be | 5964 | } else { |
aa43477b PB |
5965 | req->result = ret; |
5966 | req_set_fail(req); | |
a62682f9 | 5967 | } |
aa43477b PB |
5968 | |
5969 | io_poll_remove_entries(req); | |
5970 | spin_lock(&ctx->completion_lock); | |
5971 | hash_del(&req->hash_node); | |
5972 | __io_req_complete_post(req, req->result, 0); | |
5973 | io_commit_cqring(ctx); | |
5974 | spin_unlock(&ctx->completion_lock); | |
5975 | io_cqring_ev_posted(ctx); | |
18bceab1 JA |
5976 | } |
5977 | ||
aa43477b | 5978 | static void io_apoll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 JA |
5979 | { |
5980 | struct io_ring_ctx *ctx = req->ctx; | |
aa43477b | 5981 | int ret; |
18bceab1 | 5982 | |
aa43477b PB |
5983 | ret = io_poll_check_events(req); |
5984 | if (ret > 0) | |
5985 | return; | |
18bceab1 | 5986 | |
aa43477b PB |
5987 | io_poll_remove_entries(req); |
5988 | spin_lock(&ctx->completion_lock); | |
5989 | hash_del(&req->hash_node); | |
5990 | spin_unlock(&ctx->completion_lock); | |
18bceab1 | 5991 | |
aa43477b PB |
5992 | if (!ret) |
5993 | io_req_task_submit(req, locked); | |
5994 | else | |
5995 | io_req_complete_failed(req, ret); | |
18bceab1 JA |
5996 | } |
5997 | ||
81459350 | 5998 | static void __io_poll_execute(struct io_kiocb *req, int mask, int events) |
aa43477b PB |
5999 | { |
6000 | req->result = mask; | |
81459350 JA |
6001 | /* |
6002 | * This is useful for poll that is armed on behalf of another | |
6003 | * request, and where the wakeup path could be on a different | |
6004 | * CPU. We want to avoid pulling in req->apoll->events for that | |
6005 | * case. | |
6006 | */ | |
6007 | req->cflags = events; | |
aa43477b PB |
6008 | if (req->opcode == IORING_OP_POLL_ADD) |
6009 | req->io_task_work.func = io_poll_task_func; | |
6010 | else | |
6011 | req->io_task_work.func = io_apoll_task_func; | |
6012 | ||
502c87d6 | 6013 | trace_io_uring_task_add(req->ctx, req, req->user_data, req->opcode, mask); |
aa43477b PB |
6014 | io_req_task_work_add(req, false); |
6015 | } | |
6016 | ||
81459350 | 6017 | static inline void io_poll_execute(struct io_kiocb *req, int res, int events) |
aa43477b PB |
6018 | { |
6019 | if (io_poll_get_ownership(req)) | |
81459350 | 6020 | __io_poll_execute(req, res, events); |
aa43477b PB |
6021 | } |
6022 | ||
6023 | static void io_poll_cancel_req(struct io_kiocb *req) | |
6024 | { | |
6025 | io_poll_mark_cancelled(req); | |
6026 | /* kick tw, which should complete the request */ | |
81459350 | 6027 | io_poll_execute(req, 0, 0); |
aa43477b PB |
6028 | } |
6029 | ||
6030 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
6031 | void *key) | |
18bceab1 JA |
6032 | { |
6033 | struct io_kiocb *req = wait->private; | |
aa43477b PB |
6034 | struct io_poll_iocb *poll = container_of(wait, struct io_poll_iocb, |
6035 | wait); | |
18bceab1 JA |
6036 | __poll_t mask = key_to_poll(key); |
6037 | ||
791f3465 PB |
6038 | if (unlikely(mask & POLLFREE)) { |
6039 | io_poll_mark_cancelled(req); | |
6040 | /* we have to kick tw in case it's not already */ | |
81459350 | 6041 | io_poll_execute(req, 0, poll->events); |
791f3465 PB |
6042 | |
6043 | /* | |
6044 | * If the waitqueue is being freed early but someone is already | |
6045 | * holds ownership over it, we have to tear down the request as | |
6046 | * best we can. That means immediately removing the request from | |
6047 | * its waitqueue and preventing all further accesses to the | |
6048 | * waitqueue via the request. | |
6049 | */ | |
6050 | list_del_init(&poll->wait.entry); | |
6051 | ||
6052 | /* | |
6053 | * Careful: this *must* be the last step, since as soon | |
6054 | * as req->head is NULL'ed out, the request can be | |
6055 | * completed and freed, since aio_poll_complete_work() | |
6056 | * will no longer need to take the waitqueue lock. | |
6057 | */ | |
6058 | smp_store_release(&poll->head, NULL); | |
6059 | return 1; | |
6060 | } | |
6061 | ||
aa43477b | 6062 | /* for instances that support it check for an event match first */ |
18bceab1 JA |
6063 | if (mask && !(mask & poll->events)) |
6064 | return 0; | |
6065 | ||
eb0089d6 PB |
6066 | if (io_poll_get_ownership(req)) { |
6067 | /* optional, saves extra locking for removal in tw handler */ | |
6068 | if (mask && poll->events & EPOLLONESHOT) { | |
6069 | list_del_init(&poll->wait.entry); | |
6070 | poll->head = NULL; | |
91eac1c6 | 6071 | req->flags &= ~REQ_F_SINGLE_POLL; |
eb0089d6 | 6072 | } |
81459350 | 6073 | __io_poll_execute(req, mask, poll->events); |
eb0089d6 | 6074 | } |
18bceab1 | 6075 | return 1; |
18bceab1 JA |
6076 | } |
6077 | ||
6078 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
6079 | struct wait_queue_head *head, |
6080 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
6081 | { |
6082 | struct io_kiocb *req = pt->req; | |
6083 | ||
6084 | /* | |
68b11e8b PB |
6085 | * The file being polled uses multiple waitqueues for poll handling |
6086 | * (e.g. one for read, one for write). Setup a separate io_poll_iocb | |
6087 | * if this happens. | |
18bceab1 | 6088 | */ |
68b11e8b | 6089 | if (unlikely(pt->nr_entries)) { |
aa43477b | 6090 | struct io_poll_iocb *first = poll; |
58852d4d | 6091 | |
23a65db8 | 6092 | /* double add on the same waitqueue head, ignore */ |
aa43477b | 6093 | if (first->head == head) |
23a65db8 | 6094 | return; |
18bceab1 | 6095 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 6096 | if (*poll_ptr) { |
23a65db8 PB |
6097 | if ((*poll_ptr)->head == head) |
6098 | return; | |
18bceab1 JA |
6099 | pt->error = -EINVAL; |
6100 | return; | |
6101 | } | |
aa43477b | 6102 | |
18bceab1 JA |
6103 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
6104 | if (!poll) { | |
6105 | pt->error = -ENOMEM; | |
6106 | return; | |
6107 | } | |
91eac1c6 | 6108 | req->flags |= REQ_F_DOUBLE_POLL; |
aa43477b | 6109 | io_init_poll_iocb(poll, first->events, first->wait.func); |
807abcb0 | 6110 | *poll_ptr = poll; |
d886e185 PB |
6111 | if (req->opcode == IORING_OP_POLL_ADD) |
6112 | req->flags |= REQ_F_ASYNC_DATA; | |
18bceab1 JA |
6113 | } |
6114 | ||
91eac1c6 | 6115 | req->flags |= REQ_F_SINGLE_POLL; |
68b11e8b | 6116 | pt->nr_entries++; |
18bceab1 | 6117 | poll->head = head; |
aa43477b | 6118 | poll->wait.private = req; |
a31eb4a2 JX |
6119 | |
6120 | if (poll->events & EPOLLEXCLUSIVE) | |
6121 | add_wait_queue_exclusive(head, &poll->wait); | |
6122 | else | |
6123 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
6124 | } |
6125 | ||
aa43477b | 6126 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
18bceab1 JA |
6127 | struct poll_table_struct *p) |
6128 | { | |
6129 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
d7718a9d | 6130 | |
aa43477b PB |
6131 | __io_queue_proc(&pt->req->poll, pt, head, |
6132 | (struct io_poll_iocb **) &pt->req->async_data); | |
d7718a9d JA |
6133 | } |
6134 | ||
aa43477b PB |
6135 | static int __io_arm_poll_handler(struct io_kiocb *req, |
6136 | struct io_poll_iocb *poll, | |
6137 | struct io_poll_table *ipt, __poll_t mask) | |
d7718a9d JA |
6138 | { |
6139 | struct io_ring_ctx *ctx = req->ctx; | |
aa43477b | 6140 | int v; |
d7718a9d | 6141 | |
4d52f338 | 6142 | INIT_HLIST_NODE(&req->hash_node); |
aa43477b | 6143 | io_init_poll_iocb(poll, mask, io_poll_wake); |
b90cd197 | 6144 | poll->file = req->file; |
18bceab1 | 6145 | poll->wait.private = req; |
d7718a9d JA |
6146 | |
6147 | ipt->pt._key = mask; | |
6148 | ipt->req = req; | |
68b11e8b PB |
6149 | ipt->error = 0; |
6150 | ipt->nr_entries = 0; | |
d7718a9d | 6151 | |
aa43477b PB |
6152 | /* |
6153 | * Take the ownership to delay any tw execution up until we're done | |
6154 | * with poll arming. see io_poll_get_ownership(). | |
6155 | */ | |
6156 | atomic_set(&req->poll_refs, 1); | |
d7718a9d | 6157 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
aa43477b PB |
6158 | |
6159 | if (mask && (poll->events & EPOLLONESHOT)) { | |
6160 | io_poll_remove_entries(req); | |
6161 | /* no one else has access to the req, forget about the ref */ | |
6162 | return mask; | |
6163 | } | |
6164 | if (!mask && unlikely(ipt->error || !ipt->nr_entries)) { | |
6165 | io_poll_remove_entries(req); | |
6166 | if (!ipt->error) | |
6167 | ipt->error = -EINVAL; | |
6168 | return 0; | |
6169 | } | |
d7718a9d | 6170 | |
79ebeaee | 6171 | spin_lock(&ctx->completion_lock); |
aa43477b PB |
6172 | io_poll_req_insert(req); |
6173 | spin_unlock(&ctx->completion_lock); | |
6174 | ||
6175 | if (mask) { | |
6176 | /* can't multishot if failed, just queue the event we've got */ | |
6177 | if (unlikely(ipt->error || !ipt->nr_entries)) | |
6178 | poll->events |= EPOLLONESHOT; | |
81459350 | 6179 | __io_poll_execute(req, mask, poll->events); |
aa43477b | 6180 | return 0; |
d7718a9d | 6181 | } |
adc8682e | 6182 | io_add_napi(req->file, req->ctx); |
d7718a9d | 6183 | |
aa43477b PB |
6184 | /* |
6185 | * Release ownership. If someone tried to queue a tw while it was | |
6186 | * locked, kick it off for them. | |
6187 | */ | |
6188 | v = atomic_dec_return(&req->poll_refs); | |
6189 | if (unlikely(v & IO_POLL_REF_MASK)) | |
81459350 | 6190 | __io_poll_execute(req, 0, poll->events); |
aa43477b PB |
6191 | return 0; |
6192 | } | |
6193 | ||
6194 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
6195 | struct poll_table_struct *p) | |
6196 | { | |
6197 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
6198 | struct async_poll *apoll = pt->req->apoll; | |
6199 | ||
6200 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); | |
d7718a9d JA |
6201 | } |
6202 | ||
59b735ae OL |
6203 | enum { |
6204 | IO_APOLL_OK, | |
6205 | IO_APOLL_ABORTED, | |
6206 | IO_APOLL_READY | |
6207 | }; | |
6208 | ||
4d9237e3 | 6209 | static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags) |
d7718a9d JA |
6210 | { |
6211 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
6212 | struct io_ring_ctx *ctx = req->ctx; | |
6213 | struct async_poll *apoll; | |
6214 | struct io_poll_table ipt; | |
aa43477b PB |
6215 | __poll_t mask = EPOLLONESHOT | POLLERR | POLLPRI; |
6216 | int ret; | |
d7718a9d | 6217 | |
b2d9c3da PB |
6218 | if (!def->pollin && !def->pollout) |
6219 | return IO_APOLL_ABORTED; | |
658d0a40 PB |
6220 | if (!file_can_poll(req->file) || (req->flags & REQ_F_POLLED)) |
6221 | return IO_APOLL_ABORTED; | |
b2d9c3da PB |
6222 | |
6223 | if (def->pollin) { | |
b2d9c3da PB |
6224 | mask |= POLLIN | POLLRDNORM; |
6225 | ||
6226 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
6227 | if ((req->opcode == IORING_OP_RECVMSG) && | |
6228 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
6229 | mask &= ~POLLIN; | |
6230 | } else { | |
b2d9c3da PB |
6231 | mask |= POLLOUT | POLLWRNORM; |
6232 | } | |
6233 | ||
4d9237e3 JA |
6234 | if (!(issue_flags & IO_URING_F_UNLOCKED) && |
6235 | !list_empty(&ctx->apoll_cache)) { | |
6236 | apoll = list_first_entry(&ctx->apoll_cache, struct async_poll, | |
6237 | poll.wait.entry); | |
6238 | list_del_init(&apoll->poll.wait.entry); | |
6239 | } else { | |
6240 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
6241 | if (unlikely(!apoll)) | |
6242 | return IO_APOLL_ABORTED; | |
6243 | } | |
807abcb0 | 6244 | apoll->double_poll = NULL; |
d7718a9d | 6245 | req->apoll = apoll; |
b2d9c3da | 6246 | req->flags |= REQ_F_POLLED; |
d7718a9d JA |
6247 | ipt.pt._qproc = io_async_queue_proc; |
6248 | ||
abdad709 JA |
6249 | io_kbuf_recycle(req); |
6250 | ||
aa43477b | 6251 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask); |
41a5169c HX |
6252 | if (ret || ipt.error) |
6253 | return ret ? IO_APOLL_READY : IO_APOLL_ABORTED; | |
6254 | ||
502c87d6 | 6255 | trace_io_uring_poll_arm(ctx, req, req->user_data, req->opcode, |
236daeae | 6256 | mask, apoll->poll.events); |
59b735ae | 6257 | return IO_APOLL_OK; |
d7718a9d JA |
6258 | } |
6259 | ||
76e1b642 JA |
6260 | /* |
6261 | * Returns true if we found and killed one or more poll requests | |
6262 | */ | |
c072481d PB |
6263 | static __cold bool io_poll_remove_all(struct io_ring_ctx *ctx, |
6264 | struct task_struct *tsk, bool cancel_all) | |
221c5eb2 | 6265 | { |
78076bb6 | 6266 | struct hlist_node *tmp; |
221c5eb2 | 6267 | struct io_kiocb *req; |
aa43477b PB |
6268 | bool found = false; |
6269 | int i; | |
221c5eb2 | 6270 | |
79ebeaee | 6271 | spin_lock(&ctx->completion_lock); |
78076bb6 JA |
6272 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
6273 | struct hlist_head *list; | |
6274 | ||
6275 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 6276 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
42a7b4ed | 6277 | if (io_match_task_safe(req, tsk, cancel_all)) { |
aa43477b PB |
6278 | io_poll_cancel_req(req); |
6279 | found = true; | |
6280 | } | |
f3606e3a | 6281 | } |
221c5eb2 | 6282 | } |
79ebeaee | 6283 | spin_unlock(&ctx->completion_lock); |
aa43477b | 6284 | return found; |
221c5eb2 JA |
6285 | } |
6286 | ||
9ba5fac8 PB |
6287 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
6288 | bool poll_only) | |
e07785b0 | 6289 | __must_hold(&ctx->completion_lock) |
47f46768 | 6290 | { |
78076bb6 | 6291 | struct hlist_head *list; |
47f46768 JA |
6292 | struct io_kiocb *req; |
6293 | ||
78076bb6 JA |
6294 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
6295 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
6296 | if (sqe_addr != req->user_data) |
6297 | continue; | |
9ba5fac8 PB |
6298 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
6299 | continue; | |
b2cb805f | 6300 | return req; |
47f46768 | 6301 | } |
b2cb805f JA |
6302 | return NULL; |
6303 | } | |
6304 | ||
aa43477b PB |
6305 | static bool io_poll_disarm(struct io_kiocb *req) |
6306 | __must_hold(&ctx->completion_lock) | |
6307 | { | |
6308 | if (!io_poll_get_ownership(req)) | |
6309 | return false; | |
6310 | io_poll_remove_entries(req); | |
6311 | hash_del(&req->hash_node); | |
6312 | return true; | |
6313 | } | |
6314 | ||
9ba5fac8 PB |
6315 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
6316 | bool poll_only) | |
e07785b0 | 6317 | __must_hold(&ctx->completion_lock) |
b2cb805f | 6318 | { |
aa43477b | 6319 | struct io_kiocb *req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f | 6320 | |
b2cb805f JA |
6321 | if (!req) |
6322 | return -ENOENT; | |
aa43477b PB |
6323 | io_poll_cancel_req(req); |
6324 | return 0; | |
47f46768 JA |
6325 | } |
6326 | ||
9096af3e PB |
6327 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
6328 | unsigned int flags) | |
6329 | { | |
6330 | u32 events; | |
47f46768 | 6331 | |
9096af3e PB |
6332 | events = READ_ONCE(sqe->poll32_events); |
6333 | #ifdef __BIG_ENDIAN | |
6334 | events = swahw32(events); | |
6335 | #endif | |
6336 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
6337 | events |= EPOLLONESHOT; | |
6338 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
6339 | } |
6340 | ||
c5de0036 | 6341 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 6342 | const struct io_uring_sqe *sqe) |
0969e783 | 6343 | { |
c5de0036 PB |
6344 | struct io_poll_update *upd = &req->poll_update; |
6345 | u32 flags; | |
6346 | ||
0969e783 JA |
6347 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
6348 | return -EINVAL; | |
26578cda | 6349 | if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) |
c5de0036 PB |
6350 | return -EINVAL; |
6351 | flags = READ_ONCE(sqe->len); | |
6352 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
6353 | IORING_POLL_ADD_MULTI)) | |
6354 | return -EINVAL; | |
6355 | /* meaningless without update */ | |
6356 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
6357 | return -EINVAL; |
6358 | ||
c5de0036 PB |
6359 | upd->old_user_data = READ_ONCE(sqe->addr); |
6360 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
6361 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 6362 | |
c5de0036 PB |
6363 | upd->new_user_data = READ_ONCE(sqe->off); |
6364 | if (!upd->update_user_data && upd->new_user_data) | |
6365 | return -EINVAL; | |
6366 | if (upd->update_events) | |
6367 | upd->events = io_poll_parse_events(sqe, flags); | |
6368 | else if (sqe->poll32_events) | |
6369 | return -EINVAL; | |
221c5eb2 | 6370 | |
221c5eb2 JA |
6371 | return 0; |
6372 | } | |
6373 | ||
3529d8c2 | 6374 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
6375 | { |
6376 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 6377 | u32 flags; |
221c5eb2 JA |
6378 | |
6379 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
6380 | return -EINVAL; | |
c5de0036 | 6381 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
6382 | return -EINVAL; |
6383 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 6384 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 | 6385 | return -EINVAL; |
04c76b41 PB |
6386 | if ((flags & IORING_POLL_ADD_MULTI) && (req->flags & REQ_F_CQE_SKIP)) |
6387 | return -EINVAL; | |
221c5eb2 | 6388 | |
48dcd38d | 6389 | io_req_set_refcount(req); |
81459350 | 6390 | req->cflags = poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
6391 | return 0; |
6392 | } | |
6393 | ||
61e98203 | 6394 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
6395 | { |
6396 | struct io_poll_iocb *poll = &req->poll; | |
0969e783 | 6397 | struct io_poll_table ipt; |
aa43477b | 6398 | int ret; |
0969e783 | 6399 | |
d7718a9d | 6400 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 6401 | |
aa43477b PB |
6402 | ret = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events); |
6403 | ret = ret ?: ipt.error; | |
6404 | if (ret) | |
6405 | __io_req_complete(req, issue_flags, ret, 0); | |
6406 | return 0; | |
221c5eb2 JA |
6407 | } |
6408 | ||
c5de0036 | 6409 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
6410 | { |
6411 | struct io_ring_ctx *ctx = req->ctx; | |
6412 | struct io_kiocb *preq; | |
2bbb146d | 6413 | int ret2, ret = 0; |
cc8e9ba7 | 6414 | bool locked; |
b69de288 | 6415 | |
79ebeaee | 6416 | spin_lock(&ctx->completion_lock); |
9ba5fac8 | 6417 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
aa43477b | 6418 | if (!preq || !io_poll_disarm(preq)) { |
79ebeaee | 6419 | spin_unlock(&ctx->completion_lock); |
aa43477b | 6420 | ret = preq ? -EALREADY : -ENOENT; |
2bbb146d | 6421 | goto out; |
b69de288 | 6422 | } |
79ebeaee | 6423 | spin_unlock(&ctx->completion_lock); |
cb3b200e | 6424 | |
2bbb146d PB |
6425 | if (req->poll_update.update_events || req->poll_update.update_user_data) { |
6426 | /* only mask one event flags, keep behavior flags */ | |
6427 | if (req->poll_update.update_events) { | |
6428 | preq->poll.events &= ~0xffff; | |
6429 | preq->poll.events |= req->poll_update.events & 0xffff; | |
6430 | preq->poll.events |= IO_POLL_UNMASK; | |
cb3b200e | 6431 | } |
2bbb146d PB |
6432 | if (req->poll_update.update_user_data) |
6433 | preq->user_data = req->poll_update.new_user_data; | |
b69de288 | 6434 | |
2bbb146d PB |
6435 | ret2 = io_poll_add(preq, issue_flags); |
6436 | /* successfully updated, don't complete poll request */ | |
6437 | if (!ret2) | |
6438 | goto out; | |
b69de288 | 6439 | } |
6224590d | 6440 | |
2bbb146d | 6441 | req_set_fail(preq); |
cc8e9ba7 PB |
6442 | preq->result = -ECANCELED; |
6443 | locked = !(issue_flags & IO_URING_F_UNLOCKED); | |
6444 | io_req_task_complete(preq, &locked); | |
2bbb146d PB |
6445 | out: |
6446 | if (ret < 0) | |
6224590d | 6447 | req_set_fail(req); |
2bbb146d | 6448 | /* complete update request, we're done with it */ |
cc8e9ba7 | 6449 | __io_req_complete(req, issue_flags, ret, 0); |
b69de288 | 6450 | return 0; |
89850fce JA |
6451 | } |
6452 | ||
5262f567 JA |
6453 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
6454 | { | |
ad8a48ac JA |
6455 | struct io_timeout_data *data = container_of(timer, |
6456 | struct io_timeout_data, timer); | |
6457 | struct io_kiocb *req = data->req; | |
6458 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
6459 | unsigned long flags; |
6460 | ||
89850fce | 6461 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
a71976f3 | 6462 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
6463 | atomic_set(&req->ctx->cq_timeouts, |
6464 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
89850fce | 6465 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); |
01cec8c1 | 6466 | |
a90c8bf6 PB |
6467 | if (!(data->flags & IORING_TIMEOUT_ETIME_SUCCESS)) |
6468 | req_set_fail(req); | |
6469 | ||
6470 | req->result = -ETIME; | |
6471 | req->io_task_work.func = io_req_task_complete; | |
4813c377 | 6472 | io_req_task_work_add(req, false); |
5262f567 JA |
6473 | return HRTIMER_NORESTART; |
6474 | } | |
6475 | ||
fbd15848 PB |
6476 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
6477 | __u64 user_data) | |
89850fce | 6478 | __must_hold(&ctx->timeout_lock) |
f254ac04 | 6479 | { |
fbd15848 | 6480 | struct io_timeout_data *io; |
47f46768 | 6481 | struct io_kiocb *req; |
fd9c7bc5 | 6482 | bool found = false; |
f254ac04 | 6483 | |
135fcde8 | 6484 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
6485 | found = user_data == req->user_data; |
6486 | if (found) | |
47f46768 | 6487 | break; |
47f46768 | 6488 | } |
fd9c7bc5 PB |
6489 | if (!found) |
6490 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
6491 | |
6492 | io = req->async_data; | |
fd9c7bc5 | 6493 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 6494 | return ERR_PTR(-EALREADY); |
a71976f3 | 6495 | list_del_init(&req->timeout.list); |
fbd15848 PB |
6496 | return req; |
6497 | } | |
47f46768 | 6498 | |
fbd15848 | 6499 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
ec3c3d0f | 6500 | __must_hold(&ctx->completion_lock) |
89850fce | 6501 | __must_hold(&ctx->timeout_lock) |
fbd15848 PB |
6502 | { |
6503 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
6504 | ||
6505 | if (IS_ERR(req)) | |
6506 | return PTR_ERR(req); | |
6695490d | 6507 | io_req_task_queue_fail(req, -ECANCELED); |
f254ac04 JA |
6508 | return 0; |
6509 | } | |
6510 | ||
50c1df2b JA |
6511 | static clockid_t io_timeout_get_clock(struct io_timeout_data *data) |
6512 | { | |
6513 | switch (data->flags & IORING_TIMEOUT_CLOCK_MASK) { | |
6514 | case IORING_TIMEOUT_BOOTTIME: | |
6515 | return CLOCK_BOOTTIME; | |
6516 | case IORING_TIMEOUT_REALTIME: | |
6517 | return CLOCK_REALTIME; | |
6518 | default: | |
6519 | /* can't happen, vetted at prep time */ | |
6520 | WARN_ON_ONCE(1); | |
6521 | fallthrough; | |
6522 | case 0: | |
6523 | return CLOCK_MONOTONIC; | |
6524 | } | |
6525 | } | |
6526 | ||
f1042b6c PB |
6527 | static int io_linked_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
6528 | struct timespec64 *ts, enum hrtimer_mode mode) | |
6529 | __must_hold(&ctx->timeout_lock) | |
6530 | { | |
6531 | struct io_timeout_data *io; | |
6532 | struct io_kiocb *req; | |
6533 | bool found = false; | |
6534 | ||
6535 | list_for_each_entry(req, &ctx->ltimeout_list, timeout.list) { | |
6536 | found = user_data == req->user_data; | |
6537 | if (found) | |
6538 | break; | |
6539 | } | |
6540 | if (!found) | |
6541 | return -ENOENT; | |
6542 | ||
6543 | io = req->async_data; | |
6544 | if (hrtimer_try_to_cancel(&io->timer) == -1) | |
6545 | return -EALREADY; | |
6546 | hrtimer_init(&io->timer, io_timeout_get_clock(io), mode); | |
6547 | io->timer.function = io_link_timeout_fn; | |
6548 | hrtimer_start(&io->timer, timespec64_to_ktime(*ts), mode); | |
6549 | return 0; | |
6550 | } | |
6551 | ||
9c8e11b3 PB |
6552 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
6553 | struct timespec64 *ts, enum hrtimer_mode mode) | |
89850fce | 6554 | __must_hold(&ctx->timeout_lock) |
47f46768 | 6555 | { |
9c8e11b3 PB |
6556 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
6557 | struct io_timeout_data *data; | |
47f46768 | 6558 | |
9c8e11b3 PB |
6559 | if (IS_ERR(req)) |
6560 | return PTR_ERR(req); | |
47f46768 | 6561 | |
9c8e11b3 PB |
6562 | req->timeout.off = 0; /* noseq */ |
6563 | data = req->async_data; | |
6564 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
50c1df2b | 6565 | hrtimer_init(&data->timer, io_timeout_get_clock(data), mode); |
9c8e11b3 PB |
6566 | data->timer.function = io_timeout_fn; |
6567 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
6568 | return 0; | |
47f46768 JA |
6569 | } |
6570 | ||
3529d8c2 JA |
6571 | static int io_timeout_remove_prep(struct io_kiocb *req, |
6572 | const struct io_uring_sqe *sqe) | |
b29472ee | 6573 | { |
9c8e11b3 PB |
6574 | struct io_timeout_rem *tr = &req->timeout_rem; |
6575 | ||
b29472ee JA |
6576 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
6577 | return -EINVAL; | |
61710e43 DA |
6578 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6579 | return -EINVAL; | |
26578cda | 6580 | if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->splice_fd_in) |
b29472ee JA |
6581 | return -EINVAL; |
6582 | ||
f1042b6c | 6583 | tr->ltimeout = false; |
9c8e11b3 PB |
6584 | tr->addr = READ_ONCE(sqe->addr); |
6585 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
f1042b6c PB |
6586 | if (tr->flags & IORING_TIMEOUT_UPDATE_MASK) { |
6587 | if (hweight32(tr->flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
6588 | return -EINVAL; | |
6589 | if (tr->flags & IORING_LINK_TIMEOUT_UPDATE) | |
6590 | tr->ltimeout = true; | |
6591 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE_MASK|IORING_TIMEOUT_ABS)) | |
9c8e11b3 PB |
6592 | return -EINVAL; |
6593 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
6594 | return -EFAULT; | |
2087009c YB |
6595 | if (tr->ts.tv_sec < 0 || tr->ts.tv_nsec < 0) |
6596 | return -EINVAL; | |
9c8e11b3 PB |
6597 | } else if (tr->flags) { |
6598 | /* timeout removal doesn't support flags */ | |
b29472ee | 6599 | return -EINVAL; |
9c8e11b3 | 6600 | } |
b29472ee | 6601 | |
b29472ee JA |
6602 | return 0; |
6603 | } | |
6604 | ||
8662daec PB |
6605 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
6606 | { | |
6607 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
6608 | : HRTIMER_MODE_REL; | |
6609 | } | |
6610 | ||
11365043 JA |
6611 | /* |
6612 | * Remove or update an existing timeout command | |
6613 | */ | |
61e98203 | 6614 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 6615 | { |
9c8e11b3 | 6616 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 6617 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 6618 | int ret; |
11365043 | 6619 | |
ec3c3d0f PB |
6620 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { |
6621 | spin_lock(&ctx->completion_lock); | |
6622 | spin_lock_irq(&ctx->timeout_lock); | |
9c8e11b3 | 6623 | ret = io_timeout_cancel(ctx, tr->addr); |
ec3c3d0f PB |
6624 | spin_unlock_irq(&ctx->timeout_lock); |
6625 | spin_unlock(&ctx->completion_lock); | |
6626 | } else { | |
f1042b6c PB |
6627 | enum hrtimer_mode mode = io_translate_timeout_mode(tr->flags); |
6628 | ||
ec3c3d0f | 6629 | spin_lock_irq(&ctx->timeout_lock); |
f1042b6c PB |
6630 | if (tr->ltimeout) |
6631 | ret = io_linked_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
6632 | else | |
6633 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
ec3c3d0f PB |
6634 | spin_unlock_irq(&ctx->timeout_lock); |
6635 | } | |
11365043 | 6636 | |
4e88d6e7 | 6637 | if (ret < 0) |
93d2bcd2 | 6638 | req_set_fail(req); |
505657bc | 6639 | io_req_complete_post(req, ret, 0); |
11365043 | 6640 | return 0; |
5262f567 JA |
6641 | } |
6642 | ||
3529d8c2 | 6643 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 6644 | bool is_timeout_link) |
5262f567 | 6645 | { |
ad8a48ac | 6646 | struct io_timeout_data *data; |
a41525ab | 6647 | unsigned flags; |
56080b02 | 6648 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 6649 | |
ad8a48ac | 6650 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 6651 | return -EINVAL; |
26578cda PB |
6652 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1 || |
6653 | sqe->splice_fd_in) | |
a41525ab | 6654 | return -EINVAL; |
56080b02 | 6655 | if (off && is_timeout_link) |
2d28390a | 6656 | return -EINVAL; |
a41525ab | 6657 | flags = READ_ONCE(sqe->timeout_flags); |
6224590d PB |
6658 | if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_CLOCK_MASK | |
6659 | IORING_TIMEOUT_ETIME_SUCCESS)) | |
50c1df2b JA |
6660 | return -EINVAL; |
6661 | /* more than one clock specified is invalid, obviously */ | |
6662 | if (hweight32(flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
5262f567 | 6663 | return -EINVAL; |
bdf20073 | 6664 | |
ef9dd637 | 6665 | INIT_LIST_HEAD(&req->timeout.list); |
bfe68a22 | 6666 | req->timeout.off = off; |
f18ee4cf PB |
6667 | if (unlikely(off && !req->ctx->off_timeout_used)) |
6668 | req->ctx->off_timeout_used = true; | |
26a61679 | 6669 | |
d6a644a7 PB |
6670 | if (WARN_ON_ONCE(req_has_async_data(req))) |
6671 | return -EFAULT; | |
6672 | if (io_alloc_async_data(req)) | |
26a61679 JA |
6673 | return -ENOMEM; |
6674 | ||
e8c2bc1f | 6675 | data = req->async_data; |
ad8a48ac | 6676 | data->req = req; |
50c1df2b | 6677 | data->flags = flags; |
ad8a48ac JA |
6678 | |
6679 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
6680 | return -EFAULT; |
6681 | ||
f6223ff7 YB |
6682 | if (data->ts.tv_sec < 0 || data->ts.tv_nsec < 0) |
6683 | return -EINVAL; | |
6684 | ||
8662daec | 6685 | data->mode = io_translate_timeout_mode(flags); |
50c1df2b | 6686 | hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode); |
b97e736a PB |
6687 | |
6688 | if (is_timeout_link) { | |
6689 | struct io_submit_link *link = &req->ctx->submit_state.link; | |
6690 | ||
6691 | if (!link->head) | |
6692 | return -EINVAL; | |
6693 | if (link->last->opcode == IORING_OP_LINK_TIMEOUT) | |
6694 | return -EINVAL; | |
4d13d1a4 PB |
6695 | req->timeout.head = link->last; |
6696 | link->last->flags |= REQ_F_ARM_LTIMEOUT; | |
b97e736a | 6697 | } |
ad8a48ac JA |
6698 | return 0; |
6699 | } | |
6700 | ||
61e98203 | 6701 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 6702 | { |
ad8a48ac | 6703 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 6704 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 6705 | struct list_head *entry; |
bfe68a22 | 6706 | u32 tail, off = req->timeout.off; |
ad8a48ac | 6707 | |
89850fce | 6708 | spin_lock_irq(&ctx->timeout_lock); |
93bd25bb | 6709 | |
5262f567 JA |
6710 | /* |
6711 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
6712 | * timeout event to be satisfied. If it isn't set, then this is |
6713 | * a pure timeout request, sequence isn't used. | |
5262f567 | 6714 | */ |
8eb7e2d0 | 6715 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
6716 | entry = ctx->timeout_list.prev; |
6717 | goto add; | |
6718 | } | |
5262f567 | 6719 | |
bfe68a22 PB |
6720 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
6721 | req->timeout.target_seq = tail + off; | |
5262f567 | 6722 | |
f010505b MDG |
6723 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
6724 | * This is safe because ->completion_lock is held, and submissions | |
6725 | * and completions are never mixed in the same ->completion_lock section. | |
6726 | */ | |
6727 | ctx->cq_last_tm_flush = tail; | |
6728 | ||
5262f567 JA |
6729 | /* |
6730 | * Insertion sort, ensuring the first entry in the list is always | |
6731 | * the one we need first. | |
6732 | */ | |
5262f567 | 6733 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
6734 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
6735 | timeout.list); | |
5262f567 | 6736 | |
8eb7e2d0 | 6737 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 6738 | continue; |
bfe68a22 PB |
6739 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
6740 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
6741 | break; |
6742 | } | |
93bd25bb | 6743 | add: |
135fcde8 | 6744 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
6745 | data->timer.function = io_timeout_fn; |
6746 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
89850fce | 6747 | spin_unlock_irq(&ctx->timeout_lock); |
5262f567 JA |
6748 | return 0; |
6749 | } | |
5262f567 | 6750 | |
f458dd84 PB |
6751 | struct io_cancel_data { |
6752 | struct io_ring_ctx *ctx; | |
6753 | u64 user_data; | |
6754 | }; | |
6755 | ||
62755e35 JA |
6756 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
6757 | { | |
6758 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 6759 | struct io_cancel_data *cd = data; |
62755e35 | 6760 | |
f458dd84 | 6761 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
6762 | } |
6763 | ||
f458dd84 PB |
6764 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
6765 | struct io_ring_ctx *ctx) | |
62755e35 | 6766 | { |
f458dd84 | 6767 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 6768 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
6769 | int ret = 0; |
6770 | ||
f458dd84 | 6771 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
6772 | return -ENOENT; |
6773 | ||
f458dd84 | 6774 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
6775 | switch (cancel_ret) { |
6776 | case IO_WQ_CANCEL_OK: | |
6777 | ret = 0; | |
6778 | break; | |
6779 | case IO_WQ_CANCEL_RUNNING: | |
6780 | ret = -EALREADY; | |
6781 | break; | |
6782 | case IO_WQ_CANCEL_NOTFOUND: | |
6783 | ret = -ENOENT; | |
6784 | break; | |
6785 | } | |
6786 | ||
e977d6d3 JA |
6787 | return ret; |
6788 | } | |
6789 | ||
8cb01fac | 6790 | static int io_try_cancel_userdata(struct io_kiocb *req, u64 sqe_addr) |
47f46768 | 6791 | { |
8cb01fac | 6792 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 JA |
6793 | int ret; |
6794 | ||
dadebc35 | 6795 | WARN_ON_ONCE(!io_wq_current_is_worker() && req->task != current); |
8cb01fac | 6796 | |
f458dd84 | 6797 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
ccbf7261 JA |
6798 | /* |
6799 | * Fall-through even for -EALREADY, as we may have poll armed | |
6800 | * that need unarming. | |
6801 | */ | |
6802 | if (!ret) | |
6803 | return 0; | |
505657bc PB |
6804 | |
6805 | spin_lock(&ctx->completion_lock); | |
ccbf7261 JA |
6806 | ret = io_poll_cancel(ctx, sqe_addr, false); |
6807 | if (ret != -ENOENT) | |
6808 | goto out; | |
6809 | ||
79ebeaee | 6810 | spin_lock_irq(&ctx->timeout_lock); |
47f46768 | 6811 | ret = io_timeout_cancel(ctx, sqe_addr); |
79ebeaee | 6812 | spin_unlock_irq(&ctx->timeout_lock); |
505657bc PB |
6813 | out: |
6814 | spin_unlock(&ctx->completion_lock); | |
6815 | return ret; | |
47f46768 JA |
6816 | } |
6817 | ||
3529d8c2 JA |
6818 | static int io_async_cancel_prep(struct io_kiocb *req, |
6819 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6820 | { |
fbf23849 | 6821 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6822 | return -EINVAL; |
61710e43 DA |
6823 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6824 | return -EINVAL; | |
26578cda PB |
6825 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags || |
6826 | sqe->splice_fd_in) | |
e977d6d3 JA |
6827 | return -EINVAL; |
6828 | ||
fbf23849 JA |
6829 | req->cancel.addr = READ_ONCE(sqe->addr); |
6830 | return 0; | |
6831 | } | |
6832 | ||
61e98203 | 6833 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6834 | { |
6835 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 | 6836 | u64 sqe_addr = req->cancel.addr; |
3b44b371 | 6837 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
58f99373 PB |
6838 | struct io_tctx_node *node; |
6839 | int ret; | |
6840 | ||
8cb01fac | 6841 | ret = io_try_cancel_userdata(req, sqe_addr); |
58f99373 PB |
6842 | if (ret != -ENOENT) |
6843 | goto done; | |
58f99373 PB |
6844 | |
6845 | /* slow path, try all io-wq's */ | |
3b44b371 | 6846 | io_ring_submit_lock(ctx, needs_lock); |
58f99373 PB |
6847 | ret = -ENOENT; |
6848 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
6849 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 6850 | |
58f99373 PB |
6851 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
6852 | if (ret != -ENOENT) | |
6853 | break; | |
6854 | } | |
3b44b371 | 6855 | io_ring_submit_unlock(ctx, needs_lock); |
58f99373 | 6856 | done: |
58f99373 | 6857 | if (ret < 0) |
93d2bcd2 | 6858 | req_set_fail(req); |
505657bc | 6859 | io_req_complete_post(req, ret, 0); |
5262f567 JA |
6860 | return 0; |
6861 | } | |
6862 | ||
269bbe5f | 6863 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6864 | const struct io_uring_sqe *sqe) |
6865 | { | |
61710e43 DA |
6866 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6867 | return -EINVAL; | |
26578cda | 6868 | if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) |
05f3fb3c JA |
6869 | return -EINVAL; |
6870 | ||
269bbe5f BM |
6871 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6872 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6873 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6874 | return -EINVAL; |
269bbe5f | 6875 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6876 | return 0; |
6877 | } | |
6878 | ||
889fca73 | 6879 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6880 | { |
6881 | struct io_ring_ctx *ctx = req->ctx; | |
3b44b371 | 6882 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
c3bdad02 | 6883 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 6884 | int ret; |
fbf23849 | 6885 | |
269bbe5f BM |
6886 | up.offset = req->rsrc_update.offset; |
6887 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
6888 | up.nr = 0; |
6889 | up.tags = 0; | |
615cee49 | 6890 | up.resv = 0; |
05f3fb3c | 6891 | |
3b44b371 | 6892 | io_ring_submit_lock(ctx, needs_lock); |
fdecb662 | 6893 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 6894 | &up, req->rsrc_update.nr_args); |
3b44b371 | 6895 | io_ring_submit_unlock(ctx, needs_lock); |
05f3fb3c JA |
6896 | |
6897 | if (ret < 0) | |
93d2bcd2 | 6898 | req_set_fail(req); |
889fca73 | 6899 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6900 | return 0; |
6901 | } | |
6902 | ||
bfe76559 | 6903 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6904 | { |
d625c6ee | 6905 | switch (req->opcode) { |
e781573e | 6906 | case IORING_OP_NOP: |
bfe76559 | 6907 | return 0; |
f67676d1 JA |
6908 | case IORING_OP_READV: |
6909 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6910 | case IORING_OP_READ: |
bfe76559 | 6911 | return io_read_prep(req, sqe); |
f67676d1 JA |
6912 | case IORING_OP_WRITEV: |
6913 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6914 | case IORING_OP_WRITE: |
bfe76559 | 6915 | return io_write_prep(req, sqe); |
0969e783 | 6916 | case IORING_OP_POLL_ADD: |
bfe76559 | 6917 | return io_poll_add_prep(req, sqe); |
0969e783 | 6918 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 6919 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 6920 | case IORING_OP_FSYNC: |
1155c76a | 6921 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6922 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6923 | return io_sfr_prep(req, sqe); |
03b1230c | 6924 | case IORING_OP_SENDMSG: |
fddaface | 6925 | case IORING_OP_SEND: |
bfe76559 | 6926 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6927 | case IORING_OP_RECVMSG: |
fddaface | 6928 | case IORING_OP_RECV: |
bfe76559 | 6929 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6930 | case IORING_OP_CONNECT: |
bfe76559 | 6931 | return io_connect_prep(req, sqe); |
2d28390a | 6932 | case IORING_OP_TIMEOUT: |
bfe76559 | 6933 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6934 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6935 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6936 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6937 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6938 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6939 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6940 | case IORING_OP_ACCEPT: |
bfe76559 | 6941 | return io_accept_prep(req, sqe); |
d63d1b5e | 6942 | case IORING_OP_FALLOCATE: |
bfe76559 | 6943 | return io_fallocate_prep(req, sqe); |
15b71abe | 6944 | case IORING_OP_OPENAT: |
bfe76559 | 6945 | return io_openat_prep(req, sqe); |
b5dba59e | 6946 | case IORING_OP_CLOSE: |
bfe76559 | 6947 | return io_close_prep(req, sqe); |
05f3fb3c | 6948 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6949 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6950 | case IORING_OP_STATX: |
bfe76559 | 6951 | return io_statx_prep(req, sqe); |
4840e418 | 6952 | case IORING_OP_FADVISE: |
bfe76559 | 6953 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6954 | case IORING_OP_MADVISE: |
bfe76559 | 6955 | return io_madvise_prep(req, sqe); |
cebdb986 | 6956 | case IORING_OP_OPENAT2: |
bfe76559 | 6957 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6958 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6959 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6960 | case IORING_OP_SPLICE: |
bfe76559 | 6961 | return io_splice_prep(req, sqe); |
ddf0322d | 6962 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6963 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6964 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6965 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6966 | case IORING_OP_TEE: |
bfe76559 | 6967 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6968 | case IORING_OP_SHUTDOWN: |
6969 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6970 | case IORING_OP_RENAMEAT: |
6971 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6972 | case IORING_OP_UNLINKAT: |
6973 | return io_unlinkat_prep(req, sqe); | |
e34a02dc DK |
6974 | case IORING_OP_MKDIRAT: |
6975 | return io_mkdirat_prep(req, sqe); | |
7a8721f8 DK |
6976 | case IORING_OP_SYMLINKAT: |
6977 | return io_symlinkat_prep(req, sqe); | |
cf30da90 DK |
6978 | case IORING_OP_LINKAT: |
6979 | return io_linkat_prep(req, sqe); | |
4f57f06c JA |
6980 | case IORING_OP_MSG_RING: |
6981 | return io_msg_ring_prep(req, sqe); | |
f67676d1 JA |
6982 | } |
6983 | ||
bfe76559 PB |
6984 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6985 | req->opcode); | |
bd54b6fe | 6986 | return -EINVAL; |
bfe76559 PB |
6987 | } |
6988 | ||
93642ef8 | 6989 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 6990 | { |
b7e298d2 PB |
6991 | if (!io_op_defs[req->opcode].needs_async_setup) |
6992 | return 0; | |
d886e185 | 6993 | if (WARN_ON_ONCE(req_has_async_data(req))) |
b7e298d2 PB |
6994 | return -EFAULT; |
6995 | if (io_alloc_async_data(req)) | |
6996 | return -EAGAIN; | |
6997 | ||
93642ef8 PB |
6998 | switch (req->opcode) { |
6999 | case IORING_OP_READV: | |
93642ef8 PB |
7000 | return io_rw_prep_async(req, READ); |
7001 | case IORING_OP_WRITEV: | |
93642ef8 PB |
7002 | return io_rw_prep_async(req, WRITE); |
7003 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
7004 | return io_sendmsg_prep_async(req); |
7005 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
7006 | return io_recvmsg_prep_async(req); |
7007 | case IORING_OP_CONNECT: | |
7008 | return io_connect_prep_async(req); | |
7009 | } | |
b7e298d2 PB |
7010 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
7011 | req->opcode); | |
7012 | return -EFAULT; | |
f67676d1 JA |
7013 | } |
7014 | ||
9cf7c104 PB |
7015 | static u32 io_get_sequence(struct io_kiocb *req) |
7016 | { | |
a3dbdf54 | 7017 | u32 seq = req->ctx->cached_sq_head; |
9cf7c104 | 7018 | |
a3dbdf54 PB |
7019 | /* need original cached_sq_head, but it was increased for each req */ |
7020 | io_for_each_link(req, req) | |
7021 | seq--; | |
7022 | return seq; | |
9cf7c104 PB |
7023 | } |
7024 | ||
c072481d | 7025 | static __cold void io_drain_req(struct io_kiocb *req) |
de0617e4 | 7026 | { |
a197f664 | 7027 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 7028 | struct io_defer_entry *de; |
f67676d1 | 7029 | int ret; |
e0eb71dc | 7030 | u32 seq = io_get_sequence(req); |
3c19966d | 7031 | |
9d858b21 | 7032 | /* Still need defer if there is pending req in defer list. */ |
e302f104 | 7033 | spin_lock(&ctx->completion_lock); |
5e371265 | 7034 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) { |
e302f104 | 7035 | spin_unlock(&ctx->completion_lock); |
e0eb71dc | 7036 | queue: |
10c66904 | 7037 | ctx->drain_active = false; |
e0eb71dc PB |
7038 | io_req_task_queue(req); |
7039 | return; | |
10c66904 | 7040 | } |
e302f104 | 7041 | spin_unlock(&ctx->completion_lock); |
9cf7c104 | 7042 | |
b7e298d2 | 7043 | ret = io_req_prep_async(req); |
e0eb71dc PB |
7044 | if (ret) { |
7045 | fail: | |
7046 | io_req_complete_failed(req, ret); | |
7047 | return; | |
7048 | } | |
cbdcb435 | 7049 | io_prep_async_link(req); |
27dc8338 | 7050 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 7051 | if (!de) { |
1b48773f | 7052 | ret = -ENOMEM; |
e0eb71dc | 7053 | goto fail; |
76cc33d7 | 7054 | } |
2d28390a | 7055 | |
79ebeaee | 7056 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 7057 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
79ebeaee | 7058 | spin_unlock(&ctx->completion_lock); |
27dc8338 | 7059 | kfree(de); |
e0eb71dc | 7060 | goto queue; |
de0617e4 JA |
7061 | } |
7062 | ||
502c87d6 | 7063 | trace_io_uring_defer(ctx, req, req->user_data, req->opcode); |
27dc8338 | 7064 | de->req = req; |
9cf7c104 | 7065 | de->seq = seq; |
27dc8338 | 7066 | list_add_tail(&de->list, &ctx->defer_list); |
79ebeaee | 7067 | spin_unlock(&ctx->completion_lock); |
de0617e4 JA |
7068 | } |
7069 | ||
68fb8979 | 7070 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 7071 | { |
d1fd1c20 | 7072 | if (req->flags & REQ_F_BUFFER_SELECTED) |
cc3cec83 | 7073 | io_put_kbuf_comp(req); |
99bc4c38 | 7074 | |
0e1b6fe3 PB |
7075 | if (req->flags & REQ_F_NEED_CLEANUP) { |
7076 | switch (req->opcode) { | |
7077 | case IORING_OP_READV: | |
7078 | case IORING_OP_READ_FIXED: | |
7079 | case IORING_OP_READ: | |
7080 | case IORING_OP_WRITEV: | |
7081 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
7082 | case IORING_OP_WRITE: { |
7083 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
7084 | |
7085 | kfree(io->free_iovec); | |
0e1b6fe3 | 7086 | break; |
e8c2bc1f | 7087 | } |
0e1b6fe3 | 7088 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
7089 | case IORING_OP_SENDMSG: { |
7090 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
7091 | |
7092 | kfree(io->free_iov); | |
0e1b6fe3 | 7093 | break; |
e8c2bc1f | 7094 | } |
0e1b6fe3 PB |
7095 | case IORING_OP_SPLICE: |
7096 | case IORING_OP_TEE: | |
e1d767f0 PB |
7097 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
7098 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 7099 | break; |
f3cd4850 JA |
7100 | case IORING_OP_OPENAT: |
7101 | case IORING_OP_OPENAT2: | |
7102 | if (req->open.filename) | |
7103 | putname(req->open.filename); | |
7104 | break; | |
80a261fd JA |
7105 | case IORING_OP_RENAMEAT: |
7106 | putname(req->rename.oldpath); | |
7107 | putname(req->rename.newpath); | |
7108 | break; | |
14a1143b JA |
7109 | case IORING_OP_UNLINKAT: |
7110 | putname(req->unlink.filename); | |
7111 | break; | |
e34a02dc DK |
7112 | case IORING_OP_MKDIRAT: |
7113 | putname(req->mkdir.filename); | |
7114 | break; | |
7a8721f8 DK |
7115 | case IORING_OP_SYMLINKAT: |
7116 | putname(req->symlink.oldpath); | |
7117 | putname(req->symlink.newpath); | |
7118 | break; | |
cf30da90 DK |
7119 | case IORING_OP_LINKAT: |
7120 | putname(req->hardlink.oldpath); | |
7121 | putname(req->hardlink.newpath); | |
7122 | break; | |
0e1b6fe3 | 7123 | } |
99bc4c38 | 7124 | } |
75652a30 JA |
7125 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
7126 | kfree(req->apoll->double_poll); | |
7127 | kfree(req->apoll); | |
7128 | req->apoll = NULL; | |
7129 | } | |
3a0a6902 PB |
7130 | if (req->flags & REQ_F_INFLIGHT) { |
7131 | struct io_uring_task *tctx = req->task->io_uring; | |
7132 | ||
7133 | atomic_dec(&tctx->inflight_tracked); | |
3a0a6902 | 7134 | } |
c854357b | 7135 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 7136 | put_cred(req->creds); |
d886e185 PB |
7137 | if (req->flags & REQ_F_ASYNC_DATA) { |
7138 | kfree(req->async_data); | |
7139 | req->async_data = NULL; | |
7140 | } | |
c854357b | 7141 | req->flags &= ~IO_REQ_CLEAN_FLAGS; |
99bc4c38 PB |
7142 | } |
7143 | ||
889fca73 | 7144 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 7145 | { |
5730b27e | 7146 | const struct cred *creds = NULL; |
d625c6ee | 7147 | int ret; |
2b188cc1 | 7148 | |
6878b40e | 7149 | if (unlikely((req->flags & REQ_F_CREDS) && req->creds != current_cred())) |
c10d1f98 | 7150 | creds = override_creds(req->creds); |
5730b27e | 7151 | |
5bd2182d PM |
7152 | if (!io_op_defs[req->opcode].audit_skip) |
7153 | audit_uring_entry(req->opcode); | |
7154 | ||
d625c6ee | 7155 | switch (req->opcode) { |
2b188cc1 | 7156 | case IORING_OP_NOP: |
889fca73 | 7157 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
7158 | break; |
7159 | case IORING_OP_READV: | |
edafccee | 7160 | case IORING_OP_READ_FIXED: |
3a6820f2 | 7161 | case IORING_OP_READ: |
889fca73 | 7162 | ret = io_read(req, issue_flags); |
edafccee | 7163 | break; |
3529d8c2 | 7164 | case IORING_OP_WRITEV: |
edafccee | 7165 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 7166 | case IORING_OP_WRITE: |
889fca73 | 7167 | ret = io_write(req, issue_flags); |
2b188cc1 | 7168 | break; |
c992fe29 | 7169 | case IORING_OP_FSYNC: |
45d189c6 | 7170 | ret = io_fsync(req, issue_flags); |
c992fe29 | 7171 | break; |
221c5eb2 | 7172 | case IORING_OP_POLL_ADD: |
61e98203 | 7173 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
7174 | break; |
7175 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 7176 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 7177 | break; |
5d17b4a4 | 7178 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 7179 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 7180 | break; |
0fa03c62 | 7181 | case IORING_OP_SENDMSG: |
889fca73 | 7182 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 7183 | break; |
fddaface | 7184 | case IORING_OP_SEND: |
889fca73 | 7185 | ret = io_send(req, issue_flags); |
0fa03c62 | 7186 | break; |
aa1fa28f | 7187 | case IORING_OP_RECVMSG: |
889fca73 | 7188 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 7189 | break; |
fddaface | 7190 | case IORING_OP_RECV: |
889fca73 | 7191 | ret = io_recv(req, issue_flags); |
aa1fa28f | 7192 | break; |
5262f567 | 7193 | case IORING_OP_TIMEOUT: |
61e98203 | 7194 | ret = io_timeout(req, issue_flags); |
5262f567 | 7195 | break; |
11365043 | 7196 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 7197 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 7198 | break; |
17f2fe35 | 7199 | case IORING_OP_ACCEPT: |
889fca73 | 7200 | ret = io_accept(req, issue_flags); |
17f2fe35 | 7201 | break; |
f8e85cf2 | 7202 | case IORING_OP_CONNECT: |
889fca73 | 7203 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 7204 | break; |
62755e35 | 7205 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 7206 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 7207 | break; |
d63d1b5e | 7208 | case IORING_OP_FALLOCATE: |
45d189c6 | 7209 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 7210 | break; |
15b71abe | 7211 | case IORING_OP_OPENAT: |
45d189c6 | 7212 | ret = io_openat(req, issue_flags); |
15b71abe | 7213 | break; |
b5dba59e | 7214 | case IORING_OP_CLOSE: |
889fca73 | 7215 | ret = io_close(req, issue_flags); |
b5dba59e | 7216 | break; |
05f3fb3c | 7217 | case IORING_OP_FILES_UPDATE: |
889fca73 | 7218 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 7219 | break; |
eddc7ef5 | 7220 | case IORING_OP_STATX: |
45d189c6 | 7221 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 7222 | break; |
4840e418 | 7223 | case IORING_OP_FADVISE: |
45d189c6 | 7224 | ret = io_fadvise(req, issue_flags); |
4840e418 | 7225 | break; |
c1ca757b | 7226 | case IORING_OP_MADVISE: |
45d189c6 | 7227 | ret = io_madvise(req, issue_flags); |
c1ca757b | 7228 | break; |
cebdb986 | 7229 | case IORING_OP_OPENAT2: |
45d189c6 | 7230 | ret = io_openat2(req, issue_flags); |
cebdb986 | 7231 | break; |
3e4827b0 | 7232 | case IORING_OP_EPOLL_CTL: |
889fca73 | 7233 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 7234 | break; |
7d67af2c | 7235 | case IORING_OP_SPLICE: |
45d189c6 | 7236 | ret = io_splice(req, issue_flags); |
7d67af2c | 7237 | break; |
ddf0322d | 7238 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 7239 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 7240 | break; |
067524e9 | 7241 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 7242 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 7243 | break; |
f2a8d5c7 | 7244 | case IORING_OP_TEE: |
45d189c6 | 7245 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 7246 | break; |
36f4fa68 | 7247 | case IORING_OP_SHUTDOWN: |
45d189c6 | 7248 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 7249 | break; |
80a261fd | 7250 | case IORING_OP_RENAMEAT: |
45d189c6 | 7251 | ret = io_renameat(req, issue_flags); |
80a261fd | 7252 | break; |
14a1143b | 7253 | case IORING_OP_UNLINKAT: |
45d189c6 | 7254 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 7255 | break; |
e34a02dc DK |
7256 | case IORING_OP_MKDIRAT: |
7257 | ret = io_mkdirat(req, issue_flags); | |
7258 | break; | |
7a8721f8 DK |
7259 | case IORING_OP_SYMLINKAT: |
7260 | ret = io_symlinkat(req, issue_flags); | |
7261 | break; | |
cf30da90 DK |
7262 | case IORING_OP_LINKAT: |
7263 | ret = io_linkat(req, issue_flags); | |
7264 | break; | |
4f57f06c JA |
7265 | case IORING_OP_MSG_RING: |
7266 | ret = io_msg_ring(req, issue_flags); | |
7267 | break; | |
2b188cc1 JA |
7268 | default: |
7269 | ret = -EINVAL; | |
7270 | break; | |
7271 | } | |
7272 | ||
5bd2182d PM |
7273 | if (!io_op_defs[req->opcode].audit_skip) |
7274 | audit_uring_exit(!ret, ret); | |
7275 | ||
5730b27e JA |
7276 | if (creds) |
7277 | revert_creds(creds); | |
def596e9 JA |
7278 | if (ret) |
7279 | return ret; | |
b532576e | 7280 | /* If the op doesn't have a file, we're not polling for it */ |
9983028e | 7281 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
9882131c | 7282 | io_iopoll_req_issued(req, issue_flags); |
def596e9 JA |
7283 | |
7284 | return 0; | |
2b188cc1 JA |
7285 | } |
7286 | ||
ebc11b6c PB |
7287 | static struct io_wq_work *io_wq_free_work(struct io_wq_work *work) |
7288 | { | |
7289 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7290 | ||
7291 | req = io_put_req_find_next(req); | |
7292 | return req ? &req->work : NULL; | |
7293 | } | |
7294 | ||
5280f7e5 | 7295 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
7296 | { |
7297 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
d01905db PB |
7298 | unsigned int issue_flags = IO_URING_F_UNLOCKED; |
7299 | bool needs_poll = false; | |
6df1db6b | 7300 | struct io_kiocb *timeout; |
561fb04a | 7301 | int ret = 0; |
2b188cc1 | 7302 | |
48dcd38d PB |
7303 | /* one will be dropped by ->io_free_work() after returning to io-wq */ |
7304 | if (!(req->flags & REQ_F_REFCOUNT)) | |
7305 | __io_req_set_refcount(req, 2); | |
7306 | else | |
7307 | req_ref_get(req); | |
5d5901a3 | 7308 | |
6df1db6b PB |
7309 | timeout = io_prep_linked_timeout(req); |
7310 | if (timeout) | |
7311 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 7312 | |
dadebc35 | 7313 | /* either cancelled or io-wq is dying, so don't touch tctx->iowq */ |
d01905db PB |
7314 | if (work->flags & IO_WQ_WORK_CANCEL) { |
7315 | io_req_task_queue_fail(req, -ECANCELED); | |
7316 | return; | |
7317 | } | |
31b51510 | 7318 | |
d01905db | 7319 | if (req->flags & REQ_F_FORCE_ASYNC) { |
afb7f56f PB |
7320 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
7321 | bool opcode_poll = def->pollin || def->pollout; | |
7322 | ||
7323 | if (opcode_poll && file_can_poll(req->file)) { | |
7324 | needs_poll = true; | |
d01905db | 7325 | issue_flags |= IO_URING_F_NONBLOCK; |
afb7f56f | 7326 | } |
561fb04a | 7327 | } |
31b51510 | 7328 | |
d01905db PB |
7329 | do { |
7330 | ret = io_issue_sqe(req, issue_flags); | |
7331 | if (ret != -EAGAIN) | |
7332 | break; | |
7333 | /* | |
7334 | * We can get EAGAIN for iopolled IO even though we're | |
7335 | * forcing a sync submission from here, since we can't | |
7336 | * wait for request slots on the block side. | |
7337 | */ | |
7338 | if (!needs_poll) { | |
7339 | cond_resched(); | |
7340 | continue; | |
90fa0288 HX |
7341 | } |
7342 | ||
4d9237e3 | 7343 | if (io_arm_poll_handler(req, issue_flags) == IO_APOLL_OK) |
d01905db PB |
7344 | return; |
7345 | /* aborted or ready, in either case retry blocking */ | |
7346 | needs_poll = false; | |
7347 | issue_flags &= ~IO_URING_F_NONBLOCK; | |
7348 | } while (1); | |
31b51510 | 7349 | |
a3df7698 | 7350 | /* avoid locking problems by failing it from a clean context */ |
5d5901a3 | 7351 | if (ret) |
a3df7698 | 7352 | io_req_task_queue_fail(req, ret); |
2b188cc1 JA |
7353 | } |
7354 | ||
aeca241b | 7355 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
042b0d85 | 7356 | unsigned i) |
65e19f54 | 7357 | { |
042b0d85 | 7358 | return &table->files[i]; |
dafecf19 PB |
7359 | } |
7360 | ||
65e19f54 JA |
7361 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
7362 | int index) | |
7363 | { | |
aeca241b | 7364 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 7365 | |
a04b0ac0 | 7366 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
7367 | } |
7368 | ||
a04b0ac0 | 7369 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
7370 | { |
7371 | unsigned long file_ptr = (unsigned long) file; | |
7372 | ||
88459b50 | 7373 | file_ptr |= io_file_get_flags(file); |
a04b0ac0 | 7374 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
7375 | } |
7376 | ||
ac177053 PB |
7377 | static inline struct file *io_file_get_fixed(struct io_ring_ctx *ctx, |
7378 | struct io_kiocb *req, int fd) | |
09bb8394 | 7379 | { |
8da11c19 | 7380 | struct file *file; |
ac177053 | 7381 | unsigned long file_ptr; |
09bb8394 | 7382 | |
ac177053 PB |
7383 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
7384 | return NULL; | |
7385 | fd = array_index_nospec(fd, ctx->nr_user_files); | |
7386 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; | |
7387 | file = (struct file *) (file_ptr & FFS_MASK); | |
7388 | file_ptr &= ~FFS_MASK; | |
7389 | /* mask in overlapping REQ_F and FFS bits */ | |
35645ac3 | 7390 | req->flags |= (file_ptr << REQ_F_SUPPORT_NOWAIT_BIT); |
a46be971 | 7391 | io_req_set_rsrc_node(req, ctx); |
ac177053 PB |
7392 | return file; |
7393 | } | |
d44f554e | 7394 | |
ac177053 | 7395 | static struct file *io_file_get_normal(struct io_ring_ctx *ctx, |
ac177053 PB |
7396 | struct io_kiocb *req, int fd) |
7397 | { | |
62906e89 | 7398 | struct file *file = fget(fd); |
ac177053 | 7399 | |
502c87d6 | 7400 | trace_io_uring_file_get(ctx, req, req->user_data, fd); |
09bb8394 | 7401 | |
ac177053 PB |
7402 | /* we don't allow fixed io_uring files */ |
7403 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
7404 | io_req_track_inflight(req); | |
8371adf5 | 7405 | return file; |
09bb8394 JA |
7406 | } |
7407 | ||
ac177053 | 7408 | static inline struct file *io_file_get(struct io_ring_ctx *ctx, |
ac177053 PB |
7409 | struct io_kiocb *req, int fd, bool fixed) |
7410 | { | |
7411 | if (fixed) | |
7412 | return io_file_get_fixed(ctx, req, fd); | |
7413 | else | |
62906e89 | 7414 | return io_file_get_normal(ctx, req, fd); |
ac177053 PB |
7415 | } |
7416 | ||
f237c30a | 7417 | static void io_req_task_link_timeout(struct io_kiocb *req, bool *locked) |
89b263f6 JA |
7418 | { |
7419 | struct io_kiocb *prev = req->timeout.prev; | |
617a8948 | 7420 | int ret = -ENOENT; |
89b263f6 JA |
7421 | |
7422 | if (prev) { | |
617a8948 PB |
7423 | if (!(req->task->flags & PF_EXITING)) |
7424 | ret = io_try_cancel_userdata(req, prev->user_data); | |
505657bc | 7425 | io_req_complete_post(req, ret ?: -ETIME, 0); |
89b263f6 | 7426 | io_put_req(prev); |
89b263f6 JA |
7427 | } else { |
7428 | io_req_complete_post(req, -ETIME, 0); | |
7429 | } | |
7430 | } | |
7431 | ||
2665abfd | 7432 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 7433 | { |
ad8a48ac JA |
7434 | struct io_timeout_data *data = container_of(timer, |
7435 | struct io_timeout_data, timer); | |
90cd7e42 | 7436 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 7437 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 7438 | unsigned long flags; |
2665abfd | 7439 | |
89b263f6 | 7440 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
90cd7e42 PB |
7441 | prev = req->timeout.head; |
7442 | req->timeout.head = NULL; | |
2665abfd JA |
7443 | |
7444 | /* | |
7445 | * We don't expect the list to be empty, that will only happen if we | |
7446 | * race with the completion of the linked work. | |
7447 | */ | |
447c19f3 | 7448 | if (prev) { |
f2f87370 | 7449 | io_remove_next_linked(prev); |
447c19f3 PB |
7450 | if (!req_ref_inc_not_zero(prev)) |
7451 | prev = NULL; | |
7452 | } | |
ef9dd637 | 7453 | list_del(&req->timeout.list); |
89b263f6 JA |
7454 | req->timeout.prev = prev; |
7455 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); | |
2665abfd | 7456 | |
89b263f6 | 7457 | req->io_task_work.func = io_req_task_link_timeout; |
4813c377 | 7458 | io_req_task_work_add(req, false); |
2665abfd JA |
7459 | return HRTIMER_NORESTART; |
7460 | } | |
7461 | ||
de968c18 | 7462 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 7463 | { |
de968c18 PB |
7464 | struct io_ring_ctx *ctx = req->ctx; |
7465 | ||
89b263f6 | 7466 | spin_lock_irq(&ctx->timeout_lock); |
76a46e06 | 7467 | /* |
f2f87370 PB |
7468 | * If the back reference is NULL, then our linked request finished |
7469 | * before we got a chance to setup the timer | |
76a46e06 | 7470 | */ |
90cd7e42 | 7471 | if (req->timeout.head) { |
e8c2bc1f | 7472 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 7473 | |
ad8a48ac JA |
7474 | data->timer.function = io_link_timeout_fn; |
7475 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
7476 | data->mode); | |
ef9dd637 | 7477 | list_add_tail(&req->timeout.list, &ctx->ltimeout_list); |
2665abfd | 7478 | } |
89b263f6 | 7479 | spin_unlock_irq(&ctx->timeout_lock); |
2665abfd | 7480 | /* drop submission reference */ |
76a46e06 JA |
7481 | io_put_req(req); |
7482 | } | |
2665abfd | 7483 | |
d475a9a6 PB |
7484 | static void io_queue_sqe_arm_apoll(struct io_kiocb *req) |
7485 | __must_hold(&req->ctx->uring_lock) | |
7486 | { | |
7487 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); | |
7488 | ||
4d9237e3 | 7489 | switch (io_arm_poll_handler(req, 0)) { |
d475a9a6 | 7490 | case IO_APOLL_READY: |
d475a9a6 PB |
7491 | io_req_task_queue(req); |
7492 | break; | |
7493 | case IO_APOLL_ABORTED: | |
7494 | /* | |
7495 | * Queued up for async execution, worker will release | |
7496 | * submit reference when the iocb is actually submitted. | |
7497 | */ | |
b1c62645 | 7498 | io_kbuf_recycle(req); |
d475a9a6 PB |
7499 | io_queue_async_work(req, NULL); |
7500 | break; | |
b1c62645 | 7501 | case IO_APOLL_OK: |
b1c62645 | 7502 | break; |
d475a9a6 PB |
7503 | } |
7504 | ||
7505 | if (linked_timeout) | |
7506 | io_queue_linked_timeout(linked_timeout); | |
7507 | } | |
7508 | ||
7509 | static inline void __io_queue_sqe(struct io_kiocb *req) | |
282cdc86 | 7510 | __must_hold(&req->ctx->uring_lock) |
2b188cc1 | 7511 | { |
906c6caa | 7512 | struct io_kiocb *linked_timeout; |
e0c5c576 | 7513 | int ret; |
2b188cc1 | 7514 | |
c5eef2b9 | 7515 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 7516 | |
fff4e40e PB |
7517 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
7518 | io_req_add_compl_list(req); | |
d9f9d284 | 7519 | return; |
fff4e40e | 7520 | } |
491381ce JA |
7521 | /* |
7522 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
7523 | * doesn't support non-blocking read/write attempts | |
7524 | */ | |
1840038e | 7525 | if (likely(!ret)) { |
906c6caa PB |
7526 | linked_timeout = io_prep_linked_timeout(req); |
7527 | if (linked_timeout) | |
7528 | io_queue_linked_timeout(linked_timeout); | |
1840038e | 7529 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
d475a9a6 | 7530 | io_queue_sqe_arm_apoll(req); |
0d63c148 | 7531 | } else { |
f41db273 | 7532 | io_req_complete_failed(req, ret); |
9e645e11 | 7533 | } |
2b188cc1 JA |
7534 | } |
7535 | ||
4652fe3f | 7536 | static void io_queue_sqe_fallback(struct io_kiocb *req) |
282cdc86 | 7537 | __must_hold(&req->ctx->uring_lock) |
4fe2c963 | 7538 | { |
4652fe3f | 7539 | if (req->flags & REQ_F_FAIL) { |
c6d3d9cb | 7540 | io_req_complete_fail_submit(req); |
e0eb71dc PB |
7541 | } else if (unlikely(req->ctx->drain_active)) { |
7542 | io_drain_req(req); | |
76cc33d7 PB |
7543 | } else { |
7544 | int ret = io_req_prep_async(req); | |
7545 | ||
7546 | if (unlikely(ret)) | |
7547 | io_req_complete_failed(req, ret); | |
7548 | else | |
f237c30a | 7549 | io_queue_async_work(req, NULL); |
ce35a47a | 7550 | } |
4fe2c963 JL |
7551 | } |
7552 | ||
4652fe3f PB |
7553 | static inline void io_queue_sqe(struct io_kiocb *req) |
7554 | __must_hold(&req->ctx->uring_lock) | |
7555 | { | |
7556 | if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)))) | |
7557 | __io_queue_sqe(req); | |
7558 | else | |
7559 | io_queue_sqe_fallback(req); | |
7560 | } | |
7561 | ||
b16fed66 PB |
7562 | /* |
7563 | * Check SQE restrictions (opcode and flags). | |
7564 | * | |
7565 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
7566 | */ | |
7567 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
7568 | struct io_kiocb *req, | |
7569 | unsigned int sqe_flags) | |
4fe2c963 | 7570 | { |
b16fed66 PB |
7571 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) |
7572 | return false; | |
7573 | ||
7574 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
7575 | ctx->restrictions.sqe_flags_required) | |
7576 | return false; | |
7577 | ||
7578 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
7579 | ctx->restrictions.sqe_flags_required)) | |
7580 | return false; | |
7581 | ||
7582 | return true; | |
4fe2c963 JL |
7583 | } |
7584 | ||
22b2ca31 PB |
7585 | static void io_init_req_drain(struct io_kiocb *req) |
7586 | { | |
7587 | struct io_ring_ctx *ctx = req->ctx; | |
7588 | struct io_kiocb *head = ctx->submit_state.link.head; | |
7589 | ||
7590 | ctx->drain_active = true; | |
7591 | if (head) { | |
7592 | /* | |
7593 | * If we need to drain a request in the middle of a link, drain | |
7594 | * the head request and the next request/link after the current | |
7595 | * link. Considering sequential execution of links, | |
b6c7db32 | 7596 | * REQ_F_IO_DRAIN will be maintained for every request of our |
22b2ca31 PB |
7597 | * link. |
7598 | */ | |
b6c7db32 | 7599 | head->flags |= REQ_F_IO_DRAIN | REQ_F_FORCE_ASYNC; |
22b2ca31 PB |
7600 | ctx->drain_next = true; |
7601 | } | |
7602 | } | |
7603 | ||
b16fed66 PB |
7604 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
7605 | const struct io_uring_sqe *sqe) | |
282cdc86 | 7606 | __must_hold(&ctx->uring_lock) |
b16fed66 | 7607 | { |
b16fed66 | 7608 | unsigned int sqe_flags; |
fc0ae024 | 7609 | int personality; |
4a04d1d1 | 7610 | u8 opcode; |
b16fed66 | 7611 | |
864ea921 | 7612 | /* req is partially pre-initialised, see io_preinit_req() */ |
4a04d1d1 | 7613 | req->opcode = opcode = READ_ONCE(sqe->opcode); |
b16fed66 PB |
7614 | /* same numerical values with corresponding REQ_F_*, safe to copy */ |
7615 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
7616 | req->user_data = READ_ONCE(sqe->user_data); | |
b16fed66 | 7617 | req->file = NULL; |
b16fed66 | 7618 | req->fixed_rsrc_refs = NULL; |
b16fed66 | 7619 | req->task = current; |
b16fed66 | 7620 | |
4a04d1d1 PB |
7621 | if (unlikely(opcode >= IORING_OP_LAST)) { |
7622 | req->opcode = 0; | |
b16fed66 | 7623 | return -EINVAL; |
4a04d1d1 | 7624 | } |
68fe256a PB |
7625 | if (unlikely(sqe_flags & ~SQE_COMMON_FLAGS)) { |
7626 | /* enforce forwards compatibility on users */ | |
7627 | if (sqe_flags & ~SQE_VALID_FLAGS) | |
7628 | return -EINVAL; | |
7629 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
4a04d1d1 | 7630 | !io_op_defs[opcode].buffer_select) |
68fe256a | 7631 | return -EOPNOTSUPP; |
5562a8d7 PB |
7632 | if (sqe_flags & IOSQE_CQE_SKIP_SUCCESS) |
7633 | ctx->drain_disabled = true; | |
7634 | if (sqe_flags & IOSQE_IO_DRAIN) { | |
7635 | if (ctx->drain_disabled) | |
7636 | return -EOPNOTSUPP; | |
22b2ca31 | 7637 | io_init_req_drain(req); |
5562a8d7 | 7638 | } |
2a56a9bd PB |
7639 | } |
7640 | if (unlikely(ctx->restricted || ctx->drain_active || ctx->drain_next)) { | |
7641 | if (ctx->restricted && !io_check_restriction(ctx, req, sqe_flags)) | |
7642 | return -EACCES; | |
7643 | /* knock it to the slow queue path, will be drained there */ | |
7644 | if (ctx->drain_active) | |
7645 | req->flags |= REQ_F_FORCE_ASYNC; | |
7646 | /* if there is no link, we're at "next" request and need to drain */ | |
7647 | if (unlikely(ctx->drain_next) && !ctx->submit_state.link.head) { | |
7648 | ctx->drain_next = false; | |
7649 | ctx->drain_active = true; | |
b6c7db32 | 7650 | req->flags |= REQ_F_IO_DRAIN | REQ_F_FORCE_ASYNC; |
2a56a9bd | 7651 | } |
68fe256a | 7652 | } |
b16fed66 | 7653 | |
4a04d1d1 | 7654 | if (io_op_defs[opcode].needs_file) { |
6d63416d PB |
7655 | struct io_submit_state *state = &ctx->submit_state; |
7656 | ||
7657 | /* | |
7658 | * Plug now if we have more than 2 IO left after this, and the | |
7659 | * target is potentially a read/write to block based storage. | |
7660 | */ | |
4a04d1d1 | 7661 | if (state->need_plug && io_op_defs[opcode].plug) { |
6d63416d PB |
7662 | state->plug_started = true; |
7663 | state->need_plug = false; | |
5ca7a8b3 | 7664 | blk_start_plug_nr_ios(&state->plug, state->submit_nr); |
6d63416d PB |
7665 | } |
7666 | ||
62906e89 | 7667 | req->file = io_file_get(ctx, req, READ_ONCE(sqe->fd), |
ac177053 | 7668 | (sqe_flags & IOSQE_FIXED_FILE)); |
b16fed66 | 7669 | if (unlikely(!req->file)) |
fc0ae024 | 7670 | return -EBADF; |
b16fed66 | 7671 | } |
863e0560 | 7672 | |
003e8dcc JA |
7673 | personality = READ_ONCE(sqe->personality); |
7674 | if (personality) { | |
cdab10bf LT |
7675 | int ret; |
7676 | ||
c10d1f98 PB |
7677 | req->creds = xa_load(&ctx->personalities, personality); |
7678 | if (!req->creds) | |
003e8dcc | 7679 | return -EINVAL; |
c10d1f98 | 7680 | get_cred(req->creds); |
cdc1404a PM |
7681 | ret = security_uring_override_creds(req->creds); |
7682 | if (ret) { | |
7683 | put_cred(req->creds); | |
7684 | return ret; | |
7685 | } | |
b8e64b53 | 7686 | req->flags |= REQ_F_CREDS; |
003e8dcc | 7687 | } |
b16fed66 | 7688 | |
fc0ae024 | 7689 | return io_req_prep(req, sqe); |
b16fed66 PB |
7690 | } |
7691 | ||
a6b8cadc | 7692 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 7693 | const struct io_uring_sqe *sqe) |
282cdc86 | 7694 | __must_hold(&ctx->uring_lock) |
9e645e11 | 7695 | { |
a1ab7b35 | 7696 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 7697 | int ret; |
9e645e11 | 7698 | |
a6b8cadc PB |
7699 | ret = io_init_req(ctx, req, sqe); |
7700 | if (unlikely(ret)) { | |
502c87d6 | 7701 | trace_io_uring_req_failed(sqe, ctx, req, ret); |
a87acfde | 7702 | |
a8295b98 | 7703 | /* fail even hard links since we don't submit */ |
de59bc10 | 7704 | if (link->head) { |
a8295b98 HX |
7705 | /* |
7706 | * we can judge a link req is failed or cancelled by if | |
7707 | * REQ_F_FAIL is set, but the head is an exception since | |
7708 | * it may be set REQ_F_FAIL because of other req's failure | |
7709 | * so let's leverage req->result to distinguish if a head | |
7710 | * is set REQ_F_FAIL because of its failure or other req's | |
7711 | * failure so that we can set the correct ret code for it. | |
7712 | * init result here to avoid affecting the normal path. | |
7713 | */ | |
7714 | if (!(link->head->flags & REQ_F_FAIL)) | |
7715 | req_fail_link_node(link->head, -ECANCELED); | |
7716 | } else if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { | |
7717 | /* | |
7718 | * the current req is a normal req, we should return | |
7719 | * error and thus break the submittion loop. | |
7720 | */ | |
7721 | io_req_complete_failed(req, ret); | |
7722 | return ret; | |
de59bc10 | 7723 | } |
a8295b98 | 7724 | req_fail_link_node(req, ret); |
a6b8cadc | 7725 | } |
441b8a78 | 7726 | |
be7053b7 | 7727 | /* don't need @sqe from now on */ |
502c87d6 | 7728 | trace_io_uring_submit_sqe(ctx, req, req->user_data, req->opcode, |
236daeae OL |
7729 | req->flags, true, |
7730 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 7731 | |
9e645e11 JA |
7732 | /* |
7733 | * If we already have a head request, queue this one for async | |
7734 | * submittal once the head completes. If we don't have a head but | |
7735 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
7736 | * submitted sync once the chain is complete. If none of those | |
7737 | * conditions are true (normal request), then just queue it. | |
7738 | */ | |
863e0560 PB |
7739 | if (link->head) { |
7740 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 7741 | |
a8295b98 HX |
7742 | if (!(req->flags & REQ_F_FAIL)) { |
7743 | ret = io_req_prep_async(req); | |
7744 | if (unlikely(ret)) { | |
7745 | req_fail_link_node(req, ret); | |
7746 | if (!(head->flags & REQ_F_FAIL)) | |
7747 | req_fail_link_node(head, -ECANCELED); | |
7748 | } | |
7749 | } | |
9d76377f | 7750 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 7751 | link->last->link = req; |
863e0560 | 7752 | link->last = req; |
32fe525b | 7753 | |
f15a3431 PB |
7754 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) |
7755 | return 0; | |
32fe525b | 7756 | /* last request of a link, enqueue the link */ |
f15a3431 PB |
7757 | link->head = NULL; |
7758 | req = head; | |
7759 | } else if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { | |
7760 | link->head = req; | |
7761 | link->last = req; | |
7762 | return 0; | |
9e645e11 | 7763 | } |
2e6e1fde | 7764 | |
f15a3431 | 7765 | io_queue_sqe(req); |
1d4240cc | 7766 | return 0; |
9e645e11 JA |
7767 | } |
7768 | ||
9a56a232 JA |
7769 | /* |
7770 | * Batched submission is done, ensure local IO is flushed out. | |
7771 | */ | |
553deffd | 7772 | static void io_submit_state_end(struct io_ring_ctx *ctx) |
9a56a232 | 7773 | { |
553deffd PB |
7774 | struct io_submit_state *state = &ctx->submit_state; |
7775 | ||
a1ab7b35 | 7776 | if (state->link.head) |
de59bc10 | 7777 | io_queue_sqe(state->link.head); |
553deffd | 7778 | /* flush only after queuing links as they can generate completions */ |
c450178d | 7779 | io_submit_flush_completions(ctx); |
27926b68 JA |
7780 | if (state->plug_started) |
7781 | blk_finish_plug(&state->plug); | |
9a56a232 JA |
7782 | } |
7783 | ||
7784 | /* | |
7785 | * Start submission side cache. | |
7786 | */ | |
7787 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 7788 | unsigned int max_ios) |
9a56a232 | 7789 | { |
27926b68 | 7790 | state->plug_started = false; |
4b628aeb | 7791 | state->need_plug = max_ios > 2; |
5ca7a8b3 | 7792 | state->submit_nr = max_ios; |
a1ab7b35 PB |
7793 | /* set only head, no need to init link_last in advance */ |
7794 | state->link.head = NULL; | |
9a56a232 JA |
7795 | } |
7796 | ||
2b188cc1 JA |
7797 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
7798 | { | |
75b28aff | 7799 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 7800 | |
caf582c6 PB |
7801 | /* |
7802 | * Ensure any loads from the SQEs are done at this point, | |
7803 | * since once we write the new head, the application could | |
7804 | * write new data to them. | |
7805 | */ | |
7806 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
7807 | } |
7808 | ||
2b188cc1 | 7809 | /* |
dd9ae8a0 | 7810 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
7811 | * that is mapped by userspace. This means that care needs to be taken to |
7812 | * ensure that reads are stable, as we cannot rely on userspace always | |
7813 | * being a good citizen. If members of the sqe are validated and then later | |
7814 | * used, it's important that those reads are done through READ_ONCE() to | |
7815 | * prevent a re-load down the line. | |
7816 | */ | |
709b302f | 7817 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 7818 | { |
ea5ab3b5 | 7819 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 7820 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
7821 | |
7822 | /* | |
7823 | * The cached sq head (or cq tail) serves two purposes: | |
7824 | * | |
7825 | * 1) allows us to batch the cost of updating the user visible | |
7826 | * head updates. | |
7827 | * 2) allows the kernel side to track the head on its own, even | |
7828 | * though the application is the one updating it. | |
7829 | */ | |
17d3aeb3 | 7830 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
709b302f PB |
7831 | if (likely(head < ctx->sq_entries)) |
7832 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
7833 | |
7834 | /* drop invalid entries */ | |
15641e42 PB |
7835 | ctx->cq_extra--; |
7836 | WRITE_ONCE(ctx->rings->sq_dropped, | |
7837 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
7838 | return NULL; |
7839 | } | |
7840 | ||
0f212204 | 7841 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
282cdc86 | 7842 | __must_hold(&ctx->uring_lock) |
6c271ce2 | 7843 | { |
69629809 | 7844 | unsigned int entries = io_sqring_entries(ctx); |
46c4e16a | 7845 | int submitted = 0; |
6c271ce2 | 7846 | |
51d48dab | 7847 | if (unlikely(!entries)) |
69629809 | 7848 | return 0; |
ee7d46d9 | 7849 | /* make sure SQ entry isn't read before tail */ |
69629809 | 7850 | nr = min3(nr, ctx->sq_entries, entries); |
9a10867a | 7851 | io_get_task_refs(nr); |
6c271ce2 | 7852 | |
ba88ff11 | 7853 | io_submit_state_start(&ctx->submit_state, nr); |
69629809 | 7854 | do { |
3529d8c2 | 7855 | const struct io_uring_sqe *sqe; |
196be95c | 7856 | struct io_kiocb *req; |
fb5ccc98 | 7857 | |
a33ae9ce | 7858 | if (unlikely(!io_alloc_req_refill(ctx))) { |
196be95c PB |
7859 | if (!submitted) |
7860 | submitted = -EAGAIN; | |
fb5ccc98 | 7861 | break; |
196be95c | 7862 | } |
a33ae9ce | 7863 | req = io_alloc_req(ctx); |
4fccfcbb PB |
7864 | sqe = io_get_sqe(ctx); |
7865 | if (unlikely(!sqe)) { | |
c2b6c6bc | 7866 | wq_stack_add_head(&req->comp_list, &ctx->submit_state.free_list); |
4fccfcbb PB |
7867 | break; |
7868 | } | |
d3656344 JA |
7869 | /* will complete beyond this point, count as submitted */ |
7870 | submitted++; | |
bcbb7bf6 JA |
7871 | if (io_submit_sqe(ctx, req, sqe)) { |
7872 | /* | |
7873 | * Continue submitting even for sqe failure if the | |
7874 | * ring was setup with IORING_SETUP_SUBMIT_ALL | |
7875 | */ | |
7876 | if (!(ctx->flags & IORING_SETUP_SUBMIT_ALL)) | |
7877 | break; | |
7878 | } | |
69629809 | 7879 | } while (submitted < nr); |
6c271ce2 | 7880 | |
9466f437 PB |
7881 | if (unlikely(submitted != nr)) { |
7882 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 | 7883 | int unused = nr - ref_used; |
9466f437 | 7884 | |
09899b19 | 7885 | current->io_uring->cached_refs += unused; |
9466f437 | 7886 | } |
6c271ce2 | 7887 | |
553deffd | 7888 | io_submit_state_end(ctx); |
ae9428ca PB |
7889 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
7890 | io_commit_sqring(ctx); | |
7891 | ||
6c271ce2 JA |
7892 | return submitted; |
7893 | } | |
7894 | ||
e4b6d902 PB |
7895 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
7896 | { | |
7897 | return READ_ONCE(sqd->state); | |
7898 | } | |
7899 | ||
23b3628e XW |
7900 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
7901 | { | |
7902 | /* Tell userspace we may need a wakeup call */ | |
79ebeaee | 7903 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7904 | WRITE_ONCE(ctx->rings->sq_flags, |
7905 | ctx->rings->sq_flags | IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7906 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7907 | } |
7908 | ||
7909 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
7910 | { | |
79ebeaee | 7911 | spin_lock(&ctx->completion_lock); |
20c0b380 NA |
7912 | WRITE_ONCE(ctx->rings->sq_flags, |
7913 | ctx->rings->sq_flags & ~IORING_SQ_NEED_WAKEUP); | |
79ebeaee | 7914 | spin_unlock(&ctx->completion_lock); |
23b3628e XW |
7915 | } |
7916 | ||
08369246 | 7917 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 7918 | { |
c8d1ba58 | 7919 | unsigned int to_submit; |
bdcd3eab | 7920 | int ret = 0; |
6c271ce2 | 7921 | |
c8d1ba58 | 7922 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 7923 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
7924 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
7925 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 7926 | |
5eef4e87 | 7927 | if (!wq_list_empty(&ctx->iopoll_list) || to_submit) { |
948e1947 PB |
7928 | const struct cred *creds = NULL; |
7929 | ||
7930 | if (ctx->sq_creds != current_cred()) | |
7931 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 7932 | |
c8d1ba58 | 7933 | mutex_lock(&ctx->uring_lock); |
5eef4e87 | 7934 | if (!wq_list_empty(&ctx->iopoll_list)) |
5ba3c874 | 7935 | io_do_iopoll(ctx, true); |
906a3c6f | 7936 | |
3b763ba1 PB |
7937 | /* |
7938 | * Don't submit if refs are dying, good for io_uring_register(), | |
7939 | * but also it is relied upon by io_ring_exit_work() | |
7940 | */ | |
0298ef96 PB |
7941 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
7942 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 7943 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 7944 | mutex_unlock(&ctx->uring_lock); |
adc8682e OL |
7945 | #ifdef CONFIG_NET_RX_BUSY_POLL |
7946 | spin_lock(&ctx->napi_lock); | |
7947 | if (!list_empty(&ctx->napi_list) && | |
7948 | io_napi_busy_loop(&ctx->napi_list)) | |
7949 | ++ret; | |
7950 | spin_unlock(&ctx->napi_lock); | |
7951 | #endif | |
acfb381d PB |
7952 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
7953 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
7954 | if (creds) |
7955 | revert_creds(creds); | |
acfb381d | 7956 | } |
6c271ce2 | 7957 | |
08369246 XW |
7958 | return ret; |
7959 | } | |
6c271ce2 | 7960 | |
c072481d | 7961 | static __cold void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
08369246 XW |
7962 | { |
7963 | struct io_ring_ctx *ctx; | |
7964 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7965 | |
c9dca27d PB |
7966 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7967 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 7968 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7969 | } |
6c271ce2 | 7970 | |
e4b6d902 PB |
7971 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
7972 | { | |
7973 | bool did_sig = false; | |
7974 | struct ksignal ksig; | |
7975 | ||
7976 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
7977 | signal_pending(current)) { | |
7978 | mutex_unlock(&sqd->lock); | |
7979 | if (signal_pending(current)) | |
7980 | did_sig = get_signal(&ksig); | |
7981 | cond_resched(); | |
7982 | mutex_lock(&sqd->lock); | |
7983 | } | |
e4b6d902 PB |
7984 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
7985 | } | |
7986 | ||
c8d1ba58 JA |
7987 | static int io_sq_thread(void *data) |
7988 | { | |
69fb2131 JA |
7989 | struct io_sq_data *sqd = data; |
7990 | struct io_ring_ctx *ctx; | |
a0d9205f | 7991 | unsigned long timeout = 0; |
37d1e2e3 | 7992 | char buf[TASK_COMM_LEN]; |
08369246 | 7993 | DEFINE_WAIT(wait); |
6c271ce2 | 7994 | |
696ee88a | 7995 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 7996 | set_task_comm(current, buf); |
37d1e2e3 JA |
7997 | |
7998 | if (sqd->sq_cpu != -1) | |
7999 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
8000 | else | |
8001 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
8002 | current->flags |= PF_NO_SETAFFINITY; | |
8003 | ||
5bd2182d PM |
8004 | audit_alloc_kernel(current); |
8005 | ||
09a6f4ef | 8006 | mutex_lock(&sqd->lock); |
e4b6d902 | 8007 | while (1) { |
1a924a80 | 8008 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 8009 | |
e4b6d902 PB |
8010 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
8011 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 8012 | break; |
08369246 XW |
8013 | timeout = jiffies + sqd->sq_thread_idle; |
8014 | } | |
e4b6d902 | 8015 | |
e95eee2d | 8016 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 8017 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 8018 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 8019 | |
5eef4e87 | 8020 | if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list))) |
08369246 | 8021 | sqt_spin = true; |
69fb2131 | 8022 | } |
dd432ea5 PB |
8023 | if (io_run_task_work()) |
8024 | sqt_spin = true; | |
6c271ce2 | 8025 | |
08369246 | 8026 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 8027 | cond_resched(); |
08369246 XW |
8028 | if (sqt_spin) |
8029 | timeout = jiffies + sqd->sq_thread_idle; | |
8030 | continue; | |
8031 | } | |
8032 | ||
08369246 | 8033 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
dd432ea5 | 8034 | if (!io_sqd_events_pending(sqd) && !current->task_works) { |
1a924a80 PB |
8035 | bool needs_sched = true; |
8036 | ||
724cb4f9 | 8037 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
aaa9f0f4 PB |
8038 | io_ring_set_wakeup_flag(ctx); |
8039 | ||
724cb4f9 | 8040 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
5eef4e87 | 8041 | !wq_list_empty(&ctx->iopoll_list)) { |
724cb4f9 HX |
8042 | needs_sched = false; |
8043 | break; | |
8044 | } | |
8045 | if (io_sqring_entries(ctx)) { | |
8046 | needs_sched = false; | |
8047 | break; | |
8048 | } | |
8049 | } | |
8050 | ||
8051 | if (needs_sched) { | |
8052 | mutex_unlock(&sqd->lock); | |
8053 | schedule(); | |
8054 | mutex_lock(&sqd->lock); | |
8055 | } | |
69fb2131 JA |
8056 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
8057 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 8058 | } |
08369246 XW |
8059 | |
8060 | finish_wait(&sqd->wait, &wait); | |
8061 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 8062 | } |
28cea78a | 8063 | |
78cc687b | 8064 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 8065 | sqd->thread = NULL; |
05962f95 | 8066 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 8067 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 8068 | io_run_task_work(); |
734551df PB |
8069 | mutex_unlock(&sqd->lock); |
8070 | ||
5bd2182d PM |
8071 | audit_free(current); |
8072 | ||
37d1e2e3 JA |
8073 | complete(&sqd->exited); |
8074 | do_exit(0); | |
6c271ce2 JA |
8075 | } |
8076 | ||
bda52162 JA |
8077 | struct io_wait_queue { |
8078 | struct wait_queue_entry wq; | |
8079 | struct io_ring_ctx *ctx; | |
5fd46178 | 8080 | unsigned cq_tail; |
bda52162 | 8081 | unsigned nr_timeouts; |
adc8682e OL |
8082 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8083 | unsigned busy_poll_to; | |
8084 | #endif | |
bda52162 JA |
8085 | }; |
8086 | ||
6c503150 | 8087 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
8088 | { |
8089 | struct io_ring_ctx *ctx = iowq->ctx; | |
5fd46178 | 8090 | int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; |
bda52162 JA |
8091 | |
8092 | /* | |
d195a66e | 8093 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
8094 | * started waiting. For timeouts, we always want to return to userspace, |
8095 | * regardless of event count. | |
8096 | */ | |
5fd46178 | 8097 | return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
bda52162 JA |
8098 | } |
8099 | ||
8100 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
8101 | int wake_flags, void *key) | |
8102 | { | |
8103 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
8104 | wq); | |
8105 | ||
6c503150 PB |
8106 | /* |
8107 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
8108 | * the task, and the next invocation will do it. | |
8109 | */ | |
5ed7a37d | 8110 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) |
6c503150 PB |
8111 | return autoremove_wake_function(curr, mode, wake_flags, key); |
8112 | return -1; | |
bda52162 JA |
8113 | } |
8114 | ||
af9c1a44 JA |
8115 | static int io_run_task_work_sig(void) |
8116 | { | |
8117 | if (io_run_task_work()) | |
8118 | return 1; | |
0b8cfa97 | 8119 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 8120 | return -ERESTARTSYS; |
c5020bc8 OL |
8121 | if (task_sigpending(current)) |
8122 | return -EINTR; | |
8123 | return 0; | |
af9c1a44 JA |
8124 | } |
8125 | ||
eeb60b9a PB |
8126 | /* when returns >0, the caller should retry */ |
8127 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
8128 | struct io_wait_queue *iowq, | |
22833966 | 8129 | ktime_t timeout) |
eeb60b9a PB |
8130 | { |
8131 | int ret; | |
8132 | ||
8133 | /* make sure we run task_work before checking for signals */ | |
8134 | ret = io_run_task_work_sig(); | |
8135 | if (ret || io_should_wake(iowq)) | |
8136 | return ret; | |
8137 | /* let the caller flush overflows, retry */ | |
5ed7a37d | 8138 | if (test_bit(0, &ctx->check_cq_overflow)) |
eeb60b9a PB |
8139 | return 1; |
8140 | ||
22833966 JA |
8141 | if (!schedule_hrtimeout(&timeout, HRTIMER_MODE_ABS)) |
8142 | return -ETIME; | |
8143 | return 1; | |
eeb60b9a PB |
8144 | } |
8145 | ||
adc8682e OL |
8146 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8147 | static void io_adjust_busy_loop_timeout(struct timespec64 *ts, | |
8148 | struct io_wait_queue *iowq) | |
8149 | { | |
8150 | unsigned busy_poll_to = READ_ONCE(sysctl_net_busy_poll); | |
8151 | struct timespec64 pollto = ns_to_timespec64(1000 * (s64)busy_poll_to); | |
8152 | ||
8153 | if (timespec64_compare(ts, &pollto) > 0) { | |
8154 | *ts = timespec64_sub(*ts, pollto); | |
8155 | iowq->busy_poll_to = busy_poll_to; | |
8156 | } else { | |
8157 | u64 to = timespec64_to_ns(ts); | |
8158 | ||
8159 | do_div(to, 1000); | |
8160 | iowq->busy_poll_to = to; | |
8161 | ts->tv_sec = 0; | |
8162 | ts->tv_nsec = 0; | |
8163 | } | |
8164 | } | |
8165 | ||
8166 | static inline bool io_busy_loop_timeout(unsigned long start_time, | |
8167 | unsigned long bp_usec) | |
8168 | { | |
8169 | if (bp_usec) { | |
8170 | unsigned long end_time = start_time + bp_usec; | |
8171 | unsigned long now = busy_loop_current_time(); | |
8172 | ||
8173 | return time_after(now, end_time); | |
8174 | } | |
8175 | return true; | |
8176 | } | |
8177 | ||
8178 | static bool io_busy_loop_end(void *p, unsigned long start_time) | |
8179 | { | |
8180 | struct io_wait_queue *iowq = p; | |
8181 | ||
8182 | return signal_pending(current) || | |
8183 | io_should_wake(iowq) || | |
8184 | io_busy_loop_timeout(start_time, iowq->busy_poll_to); | |
8185 | } | |
8186 | ||
8187 | static void io_blocking_napi_busy_loop(struct list_head *napi_list, | |
8188 | struct io_wait_queue *iowq) | |
8189 | { | |
8190 | unsigned long start_time = | |
8191 | list_is_singular(napi_list) ? 0 : | |
8192 | busy_loop_current_time(); | |
8193 | ||
8194 | do { | |
8195 | if (list_is_singular(napi_list)) { | |
8196 | struct napi_entry *ne = | |
8197 | list_first_entry(napi_list, | |
8198 | struct napi_entry, list); | |
8199 | ||
8200 | napi_busy_loop(ne->napi_id, io_busy_loop_end, iowq, | |
8201 | true, BUSY_POLL_BUDGET); | |
8202 | io_check_napi_entry_timeout(ne); | |
8203 | break; | |
8204 | } | |
8205 | } while (io_napi_busy_loop(napi_list) && | |
8206 | !io_busy_loop_end(iowq, start_time)); | |
8207 | } | |
8208 | ||
8209 | static void io_putback_napi_list(struct io_ring_ctx *ctx, | |
8210 | struct list_head *napi_list) | |
8211 | { | |
8212 | struct napi_entry *cne, *lne; | |
8213 | ||
8214 | spin_lock(&ctx->napi_lock); | |
8215 | list_for_each_entry(cne, &ctx->napi_list, list) | |
8216 | list_for_each_entry(lne, napi_list, list) | |
8217 | if (cne->napi_id == lne->napi_id) { | |
8218 | list_del(&lne->list); | |
8219 | kfree(lne); | |
8220 | break; | |
8221 | } | |
8222 | list_splice(napi_list, &ctx->napi_list); | |
8223 | spin_unlock(&ctx->napi_lock); | |
8224 | } | |
8225 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
8226 | ||
2b188cc1 JA |
8227 | /* |
8228 | * Wait until events become available, if we don't already have some. The | |
8229 | * application must reap them itself, as they reside on the shared cq ring. | |
8230 | */ | |
8231 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
8232 | const sigset_t __user *sig, size_t sigsz, |
8233 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 8234 | { |
90291099 | 8235 | struct io_wait_queue iowq; |
75b28aff | 8236 | struct io_rings *rings = ctx->rings; |
22833966 | 8237 | ktime_t timeout = KTIME_MAX; |
c1d5a224 | 8238 | int ret; |
adc8682e OL |
8239 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8240 | LIST_HEAD(local_napi_list); | |
8241 | #endif | |
2b188cc1 | 8242 | |
b41e9852 | 8243 | do { |
90f67366 | 8244 | io_cqring_overflow_flush(ctx); |
6c503150 | 8245 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 8246 | return 0; |
4c6e277c | 8247 | if (!io_run_task_work()) |
b41e9852 | 8248 | break; |
b41e9852 | 8249 | } while (1); |
2b188cc1 JA |
8250 | |
8251 | if (sig) { | |
9e75ad5d AB |
8252 | #ifdef CONFIG_COMPAT |
8253 | if (in_compat_syscall()) | |
8254 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 8255 | sigsz); |
9e75ad5d AB |
8256 | else |
8257 | #endif | |
b772434b | 8258 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 8259 | |
2b188cc1 JA |
8260 | if (ret) |
8261 | return ret; | |
8262 | } | |
8263 | ||
adc8682e OL |
8264 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8265 | iowq.busy_poll_to = 0; | |
8266 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) { | |
8267 | spin_lock(&ctx->napi_lock); | |
8268 | list_splice_init(&ctx->napi_list, &local_napi_list); | |
8269 | spin_unlock(&ctx->napi_lock); | |
8270 | } | |
8271 | #endif | |
950e79dd OL |
8272 | if (uts) { |
8273 | struct timespec64 ts; | |
8274 | ||
8275 | if (get_timespec64(&ts, uts)) | |
8276 | return -EFAULT; | |
adc8682e OL |
8277 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8278 | if (!list_empty(&local_napi_list)) | |
8279 | io_adjust_busy_loop_timeout(&ts, &iowq); | |
8280 | #endif | |
950e79dd OL |
8281 | timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns()); |
8282 | } | |
adc8682e OL |
8283 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8284 | else if (!list_empty(&local_napi_list)) | |
8285 | iowq.busy_poll_to = READ_ONCE(sysctl_net_busy_poll); | |
8286 | #endif | |
950e79dd | 8287 | |
90291099 PB |
8288 | init_waitqueue_func_entry(&iowq.wq, io_wake_function); |
8289 | iowq.wq.private = current; | |
8290 | INIT_LIST_HEAD(&iowq.wq.entry); | |
8291 | iowq.ctx = ctx; | |
bda52162 | 8292 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
5fd46178 | 8293 | iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; |
90291099 | 8294 | |
c826bd7a | 8295 | trace_io_uring_cqring_wait(ctx, min_events); |
adc8682e OL |
8296 | #ifdef CONFIG_NET_RX_BUSY_POLL |
8297 | if (iowq.busy_poll_to) | |
8298 | io_blocking_napi_busy_loop(&local_napi_list, &iowq); | |
8299 | if (!list_empty(&local_napi_list)) | |
8300 | io_putback_napi_list(ctx, &local_napi_list); | |
8301 | #endif | |
bda52162 | 8302 | do { |
ca0a2651 | 8303 | /* if we can't even flush overflow, don't wait for more */ |
90f67366 | 8304 | if (!io_cqring_overflow_flush(ctx)) { |
ca0a2651 JA |
8305 | ret = -EBUSY; |
8306 | break; | |
8307 | } | |
311997b3 | 8308 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 8309 | TASK_INTERRUPTIBLE); |
22833966 | 8310 | ret = io_cqring_wait_schedule(ctx, &iowq, timeout); |
311997b3 | 8311 | finish_wait(&ctx->cq_wait, &iowq.wq); |
ca0a2651 | 8312 | cond_resched(); |
eeb60b9a | 8313 | } while (ret > 0); |
bda52162 | 8314 | |
b7db41c9 | 8315 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 8316 | |
75b28aff | 8317 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
8318 | } |
8319 | ||
9123c8ff | 8320 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 8321 | { |
9123c8ff | 8322 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 8323 | |
846a4ef2 | 8324 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
8325 | kfree(table[i]); |
8326 | kfree(table); | |
8327 | } | |
8328 | ||
c072481d | 8329 | static __cold void **io_alloc_page_table(size_t size) |
9123c8ff PB |
8330 | { |
8331 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
8332 | size_t init_size = size; | |
8333 | void **table; | |
8334 | ||
0bea96f5 | 8335 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
8336 | if (!table) |
8337 | return NULL; | |
8338 | ||
8339 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 8340 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff | 8341 | |
0bea96f5 | 8342 | table[i] = kzalloc(this_size, GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
8343 | if (!table[i]) { |
8344 | io_free_page_table(table, init_size); | |
8345 | return NULL; | |
8346 | } | |
8347 | size -= this_size; | |
8348 | } | |
8349 | return table; | |
05f3fb3c JA |
8350 | } |
8351 | ||
28a9fe25 | 8352 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 8353 | { |
28a9fe25 PB |
8354 | percpu_ref_exit(&ref_node->refs); |
8355 | kfree(ref_node); | |
1642b445 PB |
8356 | } |
8357 | ||
c072481d | 8358 | static __cold void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
b9bd2bea PB |
8359 | { |
8360 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); | |
8361 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; | |
8362 | unsigned long flags; | |
8363 | bool first_add = false; | |
b36a2050 | 8364 | unsigned long delay = HZ; |
b9bd2bea PB |
8365 | |
8366 | spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); | |
8367 | node->done = true; | |
8368 | ||
b36a2050 DY |
8369 | /* if we are mid-quiesce then do not delay */ |
8370 | if (node->rsrc_data->quiesce) | |
8371 | delay = 0; | |
8372 | ||
b9bd2bea PB |
8373 | while (!list_empty(&ctx->rsrc_ref_list)) { |
8374 | node = list_first_entry(&ctx->rsrc_ref_list, | |
8375 | struct io_rsrc_node, node); | |
8376 | /* recycle ref nodes in order */ | |
8377 | if (!node->done) | |
8378 | break; | |
8379 | list_del(&node->node); | |
8380 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
8381 | } | |
8382 | spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); | |
8383 | ||
8384 | if (first_add) | |
b36a2050 | 8385 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
b9bd2bea PB |
8386 | } |
8387 | ||
f6133fbd | 8388 | static struct io_rsrc_node *io_rsrc_node_alloc(void) |
b9bd2bea PB |
8389 | { |
8390 | struct io_rsrc_node *ref_node; | |
8391 | ||
8392 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); | |
8393 | if (!ref_node) | |
8394 | return NULL; | |
8395 | ||
8396 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, | |
8397 | 0, GFP_KERNEL)) { | |
8398 | kfree(ref_node); | |
8399 | return NULL; | |
8400 | } | |
8401 | INIT_LIST_HEAD(&ref_node->node); | |
8402 | INIT_LIST_HEAD(&ref_node->rsrc_list); | |
8403 | ref_node->done = false; | |
8404 | return ref_node; | |
8405 | } | |
8406 | ||
a7f0ed5a PB |
8407 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
8408 | struct io_rsrc_data *data_to_kill) | |
ab409402 | 8409 | __must_hold(&ctx->uring_lock) |
6b06314c | 8410 | { |
a7f0ed5a PB |
8411 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
8412 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 8413 | |
ab409402 PB |
8414 | io_rsrc_refs_drop(ctx); |
8415 | ||
a7f0ed5a PB |
8416 | if (data_to_kill) { |
8417 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 8418 | |
a7f0ed5a | 8419 | rsrc_node->rsrc_data = data_to_kill; |
4956b9ea | 8420 | spin_lock_irq(&ctx->rsrc_ref_lock); |
a7f0ed5a | 8421 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); |
4956b9ea | 8422 | spin_unlock_irq(&ctx->rsrc_ref_lock); |
82fbcfa9 | 8423 | |
3e942498 | 8424 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
8425 | percpu_ref_kill(&rsrc_node->refs); |
8426 | ctx->rsrc_node = NULL; | |
8427 | } | |
6b06314c | 8428 | |
a7f0ed5a PB |
8429 | if (!ctx->rsrc_node) { |
8430 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
8431 | ctx->rsrc_backup_node = NULL; | |
8432 | } | |
8bad28d8 HX |
8433 | } |
8434 | ||
a7f0ed5a | 8435 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
8436 | { |
8437 | if (ctx->rsrc_backup_node) | |
8438 | return 0; | |
f6133fbd | 8439 | ctx->rsrc_backup_node = io_rsrc_node_alloc(); |
8dd03afe | 8440 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
8441 | } |
8442 | ||
c072481d PB |
8443 | static __cold int io_rsrc_ref_quiesce(struct io_rsrc_data *data, |
8444 | struct io_ring_ctx *ctx) | |
8bad28d8 HX |
8445 | { |
8446 | int ret; | |
05589553 | 8447 | |
215c3902 | 8448 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
8449 | if (data->quiesce) |
8450 | return -ENXIO; | |
05589553 | 8451 | |
8bad28d8 | 8452 | data->quiesce = true; |
1ffc5422 | 8453 | do { |
a7f0ed5a | 8454 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 8455 | if (ret) |
f2303b1f | 8456 | break; |
a7f0ed5a | 8457 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 8458 | |
3e942498 PB |
8459 | /* kill initial ref, already quiesced if zero */ |
8460 | if (atomic_dec_and_test(&data->refs)) | |
8461 | break; | |
c018db4a | 8462 | mutex_unlock(&ctx->uring_lock); |
8bad28d8 | 8463 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 | 8464 | ret = wait_for_completion_interruptible(&data->done); |
c018db4a JA |
8465 | if (!ret) { |
8466 | mutex_lock(&ctx->uring_lock); | |
80912cef DY |
8467 | if (atomic_read(&data->refs) > 0) { |
8468 | /* | |
8469 | * it has been revived by another thread while | |
8470 | * we were unlocked | |
8471 | */ | |
8472 | mutex_unlock(&ctx->uring_lock); | |
8473 | } else { | |
8474 | break; | |
8475 | } | |
c018db4a | 8476 | } |
8bad28d8 | 8477 | |
3e942498 PB |
8478 | atomic_inc(&data->refs); |
8479 | /* wait for all works potentially completing data->done */ | |
8480 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 8481 | reinit_completion(&data->done); |
8dd03afe | 8482 | |
1ffc5422 | 8483 | ret = io_run_task_work_sig(); |
8bad28d8 | 8484 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 8485 | } while (ret >= 0); |
8bad28d8 | 8486 | data->quiesce = false; |
05f3fb3c | 8487 | |
8bad28d8 | 8488 | return ret; |
d7954b2b BM |
8489 | } |
8490 | ||
2d091d62 PB |
8491 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
8492 | { | |
8493 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
8494 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
8495 | ||
8496 | return &data->tags[table_idx][off]; | |
8497 | } | |
8498 | ||
44b31f2f | 8499 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 8500 | { |
2d091d62 PB |
8501 | size_t size = data->nr * sizeof(data->tags[0][0]); |
8502 | ||
8503 | if (data->tags) | |
8504 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
8505 | kfree(data); |
8506 | } | |
8507 | ||
c072481d PB |
8508 | static __cold int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
8509 | u64 __user *utags, unsigned nr, | |
8510 | struct io_rsrc_data **pdata) | |
1ad555c6 | 8511 | { |
b895c9a6 | 8512 | struct io_rsrc_data *data; |
2d091d62 | 8513 | int ret = -ENOMEM; |
d878c816 | 8514 | unsigned i; |
1ad555c6 BM |
8515 | |
8516 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
8517 | if (!data) | |
d878c816 | 8518 | return -ENOMEM; |
2d091d62 | 8519 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 8520 | if (!data->tags) { |
1ad555c6 | 8521 | kfree(data); |
d878c816 PB |
8522 | return -ENOMEM; |
8523 | } | |
2d091d62 PB |
8524 | |
8525 | data->nr = nr; | |
8526 | data->ctx = ctx; | |
8527 | data->do_put = do_put; | |
d878c816 | 8528 | if (utags) { |
2d091d62 | 8529 | ret = -EFAULT; |
d878c816 | 8530 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
8531 | u64 *tag_slot = io_get_tag_slot(data, i); |
8532 | ||
8533 | if (copy_from_user(tag_slot, &utags[i], | |
8534 | sizeof(*tag_slot))) | |
2d091d62 | 8535 | goto fail; |
d878c816 | 8536 | } |
1ad555c6 | 8537 | } |
b60c8dce | 8538 | |
3e942498 | 8539 | atomic_set(&data->refs, 1); |
1ad555c6 | 8540 | init_completion(&data->done); |
d878c816 PB |
8541 | *pdata = data; |
8542 | return 0; | |
2d091d62 PB |
8543 | fail: |
8544 | io_rsrc_data_free(data); | |
8545 | return ret; | |
1ad555c6 BM |
8546 | } |
8547 | ||
9123c8ff PB |
8548 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
8549 | { | |
0bea96f5 PB |
8550 | table->files = kvcalloc(nr_files, sizeof(table->files[0]), |
8551 | GFP_KERNEL_ACCOUNT); | |
9123c8ff PB |
8552 | return !!table->files; |
8553 | } | |
8554 | ||
042b0d85 | 8555 | static void io_free_file_tables(struct io_file_table *table) |
9123c8ff | 8556 | { |
042b0d85 | 8557 | kvfree(table->files); |
9123c8ff PB |
8558 | table->files = NULL; |
8559 | } | |
8560 | ||
fff4db76 | 8561 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 8562 | { |
fff4db76 PB |
8563 | #if defined(CONFIG_UNIX) |
8564 | if (ctx->ring_sock) { | |
8565 | struct sock *sock = ctx->ring_sock->sk; | |
8566 | struct sk_buff *skb; | |
8567 | ||
8568 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
8569 | kfree_skb(skb); | |
8570 | } | |
8571 | #else | |
8572 | int i; | |
8573 | ||
8574 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8575 | struct file *file; | |
8576 | ||
8577 | file = io_file_from_index(ctx, i); | |
8578 | if (file) | |
8579 | fput(file); | |
8580 | } | |
8581 | #endif | |
042b0d85 | 8582 | io_free_file_tables(&ctx->file_table); |
44b31f2f | 8583 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
8584 | ctx->file_data = NULL; |
8585 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
8586 | } |
8587 | ||
d7954b2b BM |
8588 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
8589 | { | |
d7954b2b BM |
8590 | int ret; |
8591 | ||
08480400 | 8592 | if (!ctx->file_data) |
d7954b2b | 8593 | return -ENXIO; |
08480400 PB |
8594 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
8595 | if (!ret) | |
8596 | __io_sqe_files_unregister(ctx); | |
8597 | return ret; | |
6b06314c JA |
8598 | } |
8599 | ||
37d1e2e3 | 8600 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 8601 | __releases(&sqd->lock) |
37d1e2e3 | 8602 | { |
521d6a73 PB |
8603 | WARN_ON_ONCE(sqd->thread == current); |
8604 | ||
9e138a48 PB |
8605 | /* |
8606 | * Do the dance but not conditional clear_bit() because it'd race with | |
8607 | * other threads incrementing park_pending and setting the bit. | |
8608 | */ | |
37d1e2e3 | 8609 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
8610 | if (atomic_dec_return(&sqd->park_pending)) |
8611 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 8612 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
8613 | } |
8614 | ||
86e0d676 | 8615 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 8616 | __acquires(&sqd->lock) |
37d1e2e3 | 8617 | { |
521d6a73 PB |
8618 | WARN_ON_ONCE(sqd->thread == current); |
8619 | ||
9e138a48 | 8620 | atomic_inc(&sqd->park_pending); |
86e0d676 | 8621 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 8622 | mutex_lock(&sqd->lock); |
05962f95 | 8623 | if (sqd->thread) |
86e0d676 | 8624 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
8625 | } |
8626 | ||
8627 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
8628 | { | |
521d6a73 | 8629 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 8630 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 8631 | |
05962f95 | 8632 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 8633 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
8634 | if (sqd->thread) |
8635 | wake_up_process(sqd->thread); | |
09a6f4ef | 8636 | mutex_unlock(&sqd->lock); |
05962f95 | 8637 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
8638 | } |
8639 | ||
534ca6d6 | 8640 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 8641 | { |
534ca6d6 | 8642 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
8643 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
8644 | ||
37d1e2e3 JA |
8645 | io_sq_thread_stop(sqd); |
8646 | kfree(sqd); | |
8647 | } | |
8648 | } | |
8649 | ||
8650 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
8651 | { | |
8652 | struct io_sq_data *sqd = ctx->sq_data; | |
8653 | ||
8654 | if (sqd) { | |
05962f95 | 8655 | io_sq_thread_park(sqd); |
521d6a73 | 8656 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 8657 | io_sqd_update_thread_idle(sqd); |
05962f95 | 8658 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
8659 | |
8660 | io_put_sq_data(sqd); | |
8661 | ctx->sq_data = NULL; | |
534ca6d6 JA |
8662 | } |
8663 | } | |
8664 | ||
aa06165d JA |
8665 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
8666 | { | |
8667 | struct io_ring_ctx *ctx_attach; | |
8668 | struct io_sq_data *sqd; | |
8669 | struct fd f; | |
8670 | ||
8671 | f = fdget(p->wq_fd); | |
8672 | if (!f.file) | |
8673 | return ERR_PTR(-ENXIO); | |
8674 | if (f.file->f_op != &io_uring_fops) { | |
8675 | fdput(f); | |
8676 | return ERR_PTR(-EINVAL); | |
8677 | } | |
8678 | ||
8679 | ctx_attach = f.file->private_data; | |
8680 | sqd = ctx_attach->sq_data; | |
8681 | if (!sqd) { | |
8682 | fdput(f); | |
8683 | return ERR_PTR(-EINVAL); | |
8684 | } | |
5c2469e0 JA |
8685 | if (sqd->task_tgid != current->tgid) { |
8686 | fdput(f); | |
8687 | return ERR_PTR(-EPERM); | |
8688 | } | |
aa06165d JA |
8689 | |
8690 | refcount_inc(&sqd->refs); | |
8691 | fdput(f); | |
8692 | return sqd; | |
8693 | } | |
8694 | ||
26984fbf PB |
8695 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
8696 | bool *attached) | |
534ca6d6 JA |
8697 | { |
8698 | struct io_sq_data *sqd; | |
8699 | ||
26984fbf | 8700 | *attached = false; |
5c2469e0 JA |
8701 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
8702 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
8703 | if (!IS_ERR(sqd)) { |
8704 | *attached = true; | |
5c2469e0 | 8705 | return sqd; |
26984fbf | 8706 | } |
5c2469e0 JA |
8707 | /* fall through for EPERM case, setup new sqd/task */ |
8708 | if (PTR_ERR(sqd) != -EPERM) | |
8709 | return sqd; | |
8710 | } | |
aa06165d | 8711 | |
534ca6d6 JA |
8712 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
8713 | if (!sqd) | |
8714 | return ERR_PTR(-ENOMEM); | |
8715 | ||
9e138a48 | 8716 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 8717 | refcount_set(&sqd->refs, 1); |
69fb2131 | 8718 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 8719 | mutex_init(&sqd->lock); |
534ca6d6 | 8720 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 8721 | init_completion(&sqd->exited); |
534ca6d6 JA |
8722 | return sqd; |
8723 | } | |
8724 | ||
6b06314c | 8725 | #if defined(CONFIG_UNIX) |
6b06314c JA |
8726 | /* |
8727 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
8728 | * the io_uring can be safely unregistered on process exit, even if we have | |
8729 | * loops in the file referencing. | |
8730 | */ | |
8731 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
8732 | { | |
8733 | struct sock *sk = ctx->ring_sock->sk; | |
8734 | struct scm_fp_list *fpl; | |
8735 | struct sk_buff *skb; | |
08a45173 | 8736 | int i, nr_files; |
6b06314c | 8737 | |
6b06314c JA |
8738 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
8739 | if (!fpl) | |
8740 | return -ENOMEM; | |
8741 | ||
8742 | skb = alloc_skb(0, GFP_KERNEL); | |
8743 | if (!skb) { | |
8744 | kfree(fpl); | |
8745 | return -ENOMEM; | |
8746 | } | |
8747 | ||
8748 | skb->sk = sk; | |
6b06314c | 8749 | |
08a45173 | 8750 | nr_files = 0; |
62e398be | 8751 | fpl->user = get_uid(current_user()); |
6b06314c | 8752 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
8753 | struct file *file = io_file_from_index(ctx, i + offset); |
8754 | ||
8755 | if (!file) | |
08a45173 | 8756 | continue; |
65e19f54 | 8757 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
8758 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
8759 | nr_files++; | |
6b06314c JA |
8760 | } |
8761 | ||
08a45173 JA |
8762 | if (nr_files) { |
8763 | fpl->max = SCM_MAX_FD; | |
8764 | fpl->count = nr_files; | |
8765 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 8766 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
8767 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
8768 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 8769 | |
08a45173 JA |
8770 | for (i = 0; i < nr_files; i++) |
8771 | fput(fpl->fp[i]); | |
8772 | } else { | |
8773 | kfree_skb(skb); | |
8774 | kfree(fpl); | |
8775 | } | |
6b06314c JA |
8776 | |
8777 | return 0; | |
8778 | } | |
8779 | ||
8780 | /* | |
8781 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
8782 | * causes regular reference counting to break down. We rely on the UNIX | |
8783 | * garbage collection to take care of this problem for us. | |
8784 | */ | |
8785 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8786 | { | |
8787 | unsigned left, total; | |
8788 | int ret = 0; | |
8789 | ||
8790 | total = 0; | |
8791 | left = ctx->nr_user_files; | |
8792 | while (left) { | |
8793 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
8794 | |
8795 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
8796 | if (ret) | |
8797 | break; | |
8798 | left -= this_files; | |
8799 | total += this_files; | |
8800 | } | |
8801 | ||
8802 | if (!ret) | |
8803 | return 0; | |
8804 | ||
8805 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
8806 | struct file *file = io_file_from_index(ctx, total); |
8807 | ||
8808 | if (file) | |
8809 | fput(file); | |
6b06314c JA |
8810 | total++; |
8811 | } | |
8812 | ||
8813 | return ret; | |
8814 | } | |
8815 | #else | |
8816 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
8817 | { | |
8818 | return 0; | |
8819 | } | |
8820 | #endif | |
8821 | ||
47e90392 | 8822 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 8823 | { |
50238531 | 8824 | struct file *file = prsrc->file; |
05f3fb3c JA |
8825 | #if defined(CONFIG_UNIX) |
8826 | struct sock *sock = ctx->ring_sock->sk; | |
8827 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
8828 | struct sk_buff *skb; | |
8829 | int i; | |
8830 | ||
8831 | __skb_queue_head_init(&list); | |
8832 | ||
8833 | /* | |
8834 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
8835 | * remove this entry and rearrange the file array. | |
8836 | */ | |
8837 | skb = skb_dequeue(head); | |
8838 | while (skb) { | |
8839 | struct scm_fp_list *fp; | |
8840 | ||
8841 | fp = UNIXCB(skb).fp; | |
8842 | for (i = 0; i < fp->count; i++) { | |
8843 | int left; | |
8844 | ||
8845 | if (fp->fp[i] != file) | |
8846 | continue; | |
8847 | ||
8848 | unix_notinflight(fp->user, fp->fp[i]); | |
8849 | left = fp->count - 1 - i; | |
8850 | if (left) { | |
8851 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
8852 | left * sizeof(struct file *)); | |
8853 | } | |
8854 | fp->count--; | |
8855 | if (!fp->count) { | |
8856 | kfree_skb(skb); | |
8857 | skb = NULL; | |
8858 | } else { | |
8859 | __skb_queue_tail(&list, skb); | |
8860 | } | |
8861 | fput(file); | |
8862 | file = NULL; | |
8863 | break; | |
8864 | } | |
8865 | ||
8866 | if (!file) | |
8867 | break; | |
8868 | ||
8869 | __skb_queue_tail(&list, skb); | |
8870 | ||
8871 | skb = skb_dequeue(head); | |
8872 | } | |
8873 | ||
8874 | if (skb_peek(&list)) { | |
8875 | spin_lock_irq(&head->lock); | |
8876 | while ((skb = __skb_dequeue(&list)) != NULL) | |
8877 | __skb_queue_tail(head, skb); | |
8878 | spin_unlock_irq(&head->lock); | |
8879 | } | |
8880 | #else | |
8881 | fput(file); | |
8882 | #endif | |
8883 | } | |
8884 | ||
b895c9a6 | 8885 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 8886 | { |
b895c9a6 | 8887 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
8888 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
8889 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 8890 | |
269bbe5f BM |
8891 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
8892 | list_del(&prsrc->list); | |
b60c8dce PB |
8893 | |
8894 | if (prsrc->tag) { | |
8895 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
b60c8dce PB |
8896 | |
8897 | io_ring_submit_lock(ctx, lock_ring); | |
79ebeaee | 8898 | spin_lock(&ctx->completion_lock); |
913a571a | 8899 | io_fill_cqe_aux(ctx, prsrc->tag, 0, 0); |
b60c8dce | 8900 | io_commit_cqring(ctx); |
79ebeaee | 8901 | spin_unlock(&ctx->completion_lock); |
b60c8dce PB |
8902 | io_cqring_ev_posted(ctx); |
8903 | io_ring_submit_unlock(ctx, lock_ring); | |
8904 | } | |
8905 | ||
40ae0ff7 | 8906 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 8907 | kfree(prsrc); |
65e19f54 | 8908 | } |
05589553 | 8909 | |
28a9fe25 | 8910 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
8911 | if (atomic_dec_and_test(&rsrc_data->refs)) |
8912 | complete(&rsrc_data->done); | |
2faf852d | 8913 | } |
65e19f54 | 8914 | |
269bbe5f | 8915 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
8916 | { |
8917 | struct io_ring_ctx *ctx; | |
8918 | struct llist_node *node; | |
8919 | ||
269bbe5f BM |
8920 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
8921 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
8922 | |
8923 | while (node) { | |
b895c9a6 | 8924 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
8925 | struct llist_node *next = node->next; |
8926 | ||
b895c9a6 | 8927 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 8928 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
8929 | node = next; |
8930 | } | |
8931 | } | |
8932 | ||
6b06314c | 8933 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 8934 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
8935 | { |
8936 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 8937 | struct file *file; |
f3baed39 | 8938 | int fd, ret; |
846a4ef2 | 8939 | unsigned i; |
6b06314c | 8940 | |
05f3fb3c | 8941 | if (ctx->file_data) |
6b06314c JA |
8942 | return -EBUSY; |
8943 | if (!nr_args) | |
8944 | return -EINVAL; | |
8945 | if (nr_args > IORING_MAX_FIXED_FILES) | |
8946 | return -EMFILE; | |
3a1b8a4e PB |
8947 | if (nr_args > rlimit(RLIMIT_NOFILE)) |
8948 | return -EMFILE; | |
a7f0ed5a | 8949 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
8950 | if (ret) |
8951 | return ret; | |
d878c816 PB |
8952 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
8953 | &ctx->file_data); | |
8954 | if (ret) | |
8955 | return ret; | |
6b06314c | 8956 | |
f3baed39 | 8957 | ret = -ENOMEM; |
aeca241b | 8958 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 8959 | goto out_free; |
65e19f54 | 8960 | |
08a45173 | 8961 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
d878c816 | 8962 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 PB |
8963 | ret = -EFAULT; |
8964 | goto out_fput; | |
8965 | } | |
08a45173 | 8966 | /* allow sparse sets */ |
792e3582 PB |
8967 | if (fd == -1) { |
8968 | ret = -EINVAL; | |
2d091d62 | 8969 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
792e3582 | 8970 | goto out_fput; |
08a45173 | 8971 | continue; |
792e3582 | 8972 | } |
6b06314c | 8973 | |
05f3fb3c | 8974 | file = fget(fd); |
6b06314c | 8975 | ret = -EBADF; |
792e3582 | 8976 | if (unlikely(!file)) |
600cf3f8 | 8977 | goto out_fput; |
05f3fb3c | 8978 | |
6b06314c JA |
8979 | /* |
8980 | * Don't allow io_uring instances to be registered. If UNIX | |
8981 | * isn't enabled, then this causes a reference cycle and this | |
8982 | * instance can never get freed. If UNIX is enabled we'll | |
8983 | * handle it just fine, but there's still no point in allowing | |
8984 | * a ring fd as it doesn't support regular read/write anyway. | |
8985 | */ | |
05f3fb3c JA |
8986 | if (file->f_op == &io_uring_fops) { |
8987 | fput(file); | |
600cf3f8 | 8988 | goto out_fput; |
6b06314c | 8989 | } |
aeca241b | 8990 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
8991 | } |
8992 | ||
6b06314c | 8993 | ret = io_sqe_files_scm(ctx); |
05589553 | 8994 | if (ret) { |
08480400 | 8995 | __io_sqe_files_unregister(ctx); |
05589553 XW |
8996 | return ret; |
8997 | } | |
6b06314c | 8998 | |
a7f0ed5a | 8999 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 9000 | return ret; |
600cf3f8 PB |
9001 | out_fput: |
9002 | for (i = 0; i < ctx->nr_user_files; i++) { | |
9003 | file = io_file_from_index(ctx, i); | |
9004 | if (file) | |
9005 | fput(file); | |
9006 | } | |
042b0d85 | 9007 | io_free_file_tables(&ctx->file_table); |
600cf3f8 | 9008 | ctx->nr_user_files = 0; |
600cf3f8 | 9009 | out_free: |
44b31f2f | 9010 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 9011 | ctx->file_data = NULL; |
6b06314c JA |
9012 | return ret; |
9013 | } | |
9014 | ||
c3a31e60 JA |
9015 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
9016 | int index) | |
9017 | { | |
9018 | #if defined(CONFIG_UNIX) | |
9019 | struct sock *sock = ctx->ring_sock->sk; | |
9020 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
9021 | struct sk_buff *skb; | |
9022 | ||
9023 | /* | |
9024 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
9025 | * file set. If there's no room, fall back to allocating a new skb | |
9026 | * and filling it in. | |
9027 | */ | |
9028 | spin_lock_irq(&head->lock); | |
9029 | skb = skb_peek(head); | |
9030 | if (skb) { | |
9031 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
9032 | ||
9033 | if (fpl->count < SCM_MAX_FD) { | |
9034 | __skb_unlink(skb, head); | |
9035 | spin_unlock_irq(&head->lock); | |
9036 | fpl->fp[fpl->count] = get_file(file); | |
9037 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
9038 | fpl->count++; | |
9039 | spin_lock_irq(&head->lock); | |
9040 | __skb_queue_head(head, skb); | |
9041 | } else { | |
9042 | skb = NULL; | |
9043 | } | |
9044 | } | |
9045 | spin_unlock_irq(&head->lock); | |
9046 | ||
9047 | if (skb) { | |
9048 | fput(file); | |
9049 | return 0; | |
9050 | } | |
9051 | ||
9052 | return __io_sqe_files_scm(ctx, 1, index); | |
9053 | #else | |
9054 | return 0; | |
9055 | #endif | |
9056 | } | |
9057 | ||
9c7b0ba8 PB |
9058 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
9059 | struct io_rsrc_node *node, void *rsrc) | |
9060 | { | |
9061 | struct io_rsrc_put *prsrc; | |
9062 | ||
9063 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); | |
9064 | if (!prsrc) | |
9065 | return -ENOMEM; | |
9066 | ||
9067 | prsrc->tag = *io_get_tag_slot(data, idx); | |
9068 | prsrc->rsrc = rsrc; | |
9069 | list_add(&prsrc->list, &node->rsrc_list); | |
9070 | return 0; | |
9071 | } | |
9072 | ||
b9445598 PB |
9073 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
9074 | unsigned int issue_flags, u32 slot_index) | |
9075 | { | |
9076 | struct io_ring_ctx *ctx = req->ctx; | |
3b44b371 | 9077 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
9c7b0ba8 | 9078 | bool needs_switch = false; |
b9445598 PB |
9079 | struct io_fixed_file *file_slot; |
9080 | int ret = -EBADF; | |
9081 | ||
3b44b371 | 9082 | io_ring_submit_lock(ctx, needs_lock); |
b9445598 PB |
9083 | if (file->f_op == &io_uring_fops) |
9084 | goto err; | |
9085 | ret = -ENXIO; | |
9086 | if (!ctx->file_data) | |
9087 | goto err; | |
9088 | ret = -EINVAL; | |
9089 | if (slot_index >= ctx->nr_user_files) | |
9090 | goto err; | |
9091 | ||
9092 | slot_index = array_index_nospec(slot_index, ctx->nr_user_files); | |
9093 | file_slot = io_fixed_file_slot(&ctx->file_table, slot_index); | |
9c7b0ba8 PB |
9094 | |
9095 | if (file_slot->file_ptr) { | |
9096 | struct file *old_file; | |
9097 | ||
9098 | ret = io_rsrc_node_switch_start(ctx); | |
9099 | if (ret) | |
9100 | goto err; | |
9101 | ||
9102 | old_file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
9103 | ret = io_queue_rsrc_removal(ctx->file_data, slot_index, | |
9104 | ctx->rsrc_node, old_file); | |
9105 | if (ret) | |
9106 | goto err; | |
9107 | file_slot->file_ptr = 0; | |
9108 | needs_switch = true; | |
9109 | } | |
b9445598 PB |
9110 | |
9111 | *io_get_tag_slot(ctx->file_data, slot_index) = 0; | |
9112 | io_fixed_file_set(file_slot, file); | |
9113 | ret = io_sqe_file_register(ctx, file, slot_index); | |
9114 | if (ret) { | |
9115 | file_slot->file_ptr = 0; | |
9116 | goto err; | |
9117 | } | |
9118 | ||
9119 | ret = 0; | |
9120 | err: | |
9c7b0ba8 PB |
9121 | if (needs_switch) |
9122 | io_rsrc_node_switch(ctx, ctx->file_data); | |
3b44b371 | 9123 | io_ring_submit_unlock(ctx, needs_lock); |
b9445598 PB |
9124 | if (ret) |
9125 | fput(file); | |
9126 | return ret; | |
9127 | } | |
9128 | ||
7df778be PB |
9129 | static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) |
9130 | { | |
9131 | unsigned int offset = req->close.file_slot - 1; | |
9132 | struct io_ring_ctx *ctx = req->ctx; | |
3b44b371 | 9133 | bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
7df778be PB |
9134 | struct io_fixed_file *file_slot; |
9135 | struct file *file; | |
9136 | int ret, i; | |
9137 | ||
3b44b371 | 9138 | io_ring_submit_lock(ctx, needs_lock); |
7df778be PB |
9139 | ret = -ENXIO; |
9140 | if (unlikely(!ctx->file_data)) | |
9141 | goto out; | |
9142 | ret = -EINVAL; | |
9143 | if (offset >= ctx->nr_user_files) | |
9144 | goto out; | |
9145 | ret = io_rsrc_node_switch_start(ctx); | |
9146 | if (ret) | |
9147 | goto out; | |
9148 | ||
9149 | i = array_index_nospec(offset, ctx->nr_user_files); | |
9150 | file_slot = io_fixed_file_slot(&ctx->file_table, i); | |
9151 | ret = -EBADF; | |
9152 | if (!file_slot->file_ptr) | |
9153 | goto out; | |
9154 | ||
9155 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
9156 | ret = io_queue_rsrc_removal(ctx->file_data, offset, ctx->rsrc_node, file); | |
9157 | if (ret) | |
9158 | goto out; | |
9159 | ||
9160 | file_slot->file_ptr = 0; | |
9161 | io_rsrc_node_switch(ctx, ctx->file_data); | |
9162 | ret = 0; | |
9163 | out: | |
3b44b371 | 9164 | io_ring_submit_unlock(ctx, needs_lock); |
7df778be PB |
9165 | return ret; |
9166 | } | |
9167 | ||
05f3fb3c | 9168 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
c3bdad02 | 9169 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
9170 | unsigned nr_args) |
9171 | { | |
c3bdad02 | 9172 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 9173 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 9174 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
9175 | struct io_fixed_file *file_slot; |
9176 | struct file *file; | |
98f0b3b4 PB |
9177 | int fd, i, err = 0; |
9178 | unsigned int done; | |
05589553 | 9179 | bool needs_switch = false; |
c3a31e60 | 9180 | |
98f0b3b4 PB |
9181 | if (!ctx->file_data) |
9182 | return -ENXIO; | |
9183 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
9184 | return -EINVAL; |
9185 | ||
67973b93 | 9186 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
9187 | u64 tag = 0; |
9188 | ||
9189 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
9190 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
9191 | err = -EFAULT; |
9192 | break; | |
9193 | } | |
c3bdad02 PB |
9194 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
9195 | err = -EINVAL; | |
9196 | break; | |
9197 | } | |
4e0377a1 | 9198 | if (fd == IORING_REGISTER_FILES_SKIP) |
9199 | continue; | |
9200 | ||
67973b93 | 9201 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 9202 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 9203 | |
a04b0ac0 PB |
9204 | if (file_slot->file_ptr) { |
9205 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
9206 | err = io_queue_rsrc_removal(data, up->offset + done, |
9207 | ctx->rsrc_node, file); | |
a5318d3c HD |
9208 | if (err) |
9209 | break; | |
a04b0ac0 | 9210 | file_slot->file_ptr = 0; |
05589553 | 9211 | needs_switch = true; |
c3a31e60 JA |
9212 | } |
9213 | if (fd != -1) { | |
c3a31e60 JA |
9214 | file = fget(fd); |
9215 | if (!file) { | |
9216 | err = -EBADF; | |
9217 | break; | |
9218 | } | |
9219 | /* | |
9220 | * Don't allow io_uring instances to be registered. If | |
9221 | * UNIX isn't enabled, then this causes a reference | |
9222 | * cycle and this instance can never get freed. If UNIX | |
9223 | * is enabled we'll handle it just fine, but there's | |
9224 | * still no point in allowing a ring fd as it doesn't | |
9225 | * support regular read/write anyway. | |
9226 | */ | |
9227 | if (file->f_op == &io_uring_fops) { | |
9228 | fput(file); | |
9229 | err = -EBADF; | |
9230 | break; | |
9231 | } | |
2d091d62 | 9232 | *io_get_tag_slot(data, up->offset + done) = tag; |
9a321c98 | 9233 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 9234 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 9235 | if (err) { |
a04b0ac0 | 9236 | file_slot->file_ptr = 0; |
f3bd9dae | 9237 | fput(file); |
c3a31e60 | 9238 | break; |
f3bd9dae | 9239 | } |
c3a31e60 | 9240 | } |
05f3fb3c JA |
9241 | } |
9242 | ||
a7f0ed5a PB |
9243 | if (needs_switch) |
9244 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
9245 | return done ? done : err; |
9246 | } | |
05589553 | 9247 | |
685fe7fe JA |
9248 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
9249 | struct task_struct *task) | |
24369c2e | 9250 | { |
e941894e | 9251 | struct io_wq_hash *hash; |
24369c2e | 9252 | struct io_wq_data data; |
24369c2e | 9253 | unsigned int concurrency; |
24369c2e | 9254 | |
362a9e65 | 9255 | mutex_lock(&ctx->uring_lock); |
e941894e JA |
9256 | hash = ctx->hash_map; |
9257 | if (!hash) { | |
9258 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
362a9e65 YY |
9259 | if (!hash) { |
9260 | mutex_unlock(&ctx->uring_lock); | |
e941894e | 9261 | return ERR_PTR(-ENOMEM); |
362a9e65 | 9262 | } |
e941894e JA |
9263 | refcount_set(&hash->refs, 1); |
9264 | init_waitqueue_head(&hash->wait); | |
9265 | ctx->hash_map = hash; | |
24369c2e | 9266 | } |
362a9e65 | 9267 | mutex_unlock(&ctx->uring_lock); |
24369c2e | 9268 | |
e941894e | 9269 | data.hash = hash; |
685fe7fe | 9270 | data.task = task; |
ebc11b6c | 9271 | data.free_work = io_wq_free_work; |
f5fa38c5 | 9272 | data.do_work = io_wq_submit_work; |
24369c2e | 9273 | |
d25e3a3d JA |
9274 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
9275 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 9276 | |
5aa75ed5 | 9277 | return io_wq_create(concurrency, &data); |
24369c2e PB |
9278 | } |
9279 | ||
c072481d PB |
9280 | static __cold int io_uring_alloc_task_context(struct task_struct *task, |
9281 | struct io_ring_ctx *ctx) | |
0f212204 JA |
9282 | { |
9283 | struct io_uring_task *tctx; | |
d8a6df10 | 9284 | int ret; |
0f212204 | 9285 | |
09899b19 | 9286 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
9287 | if (unlikely(!tctx)) |
9288 | return -ENOMEM; | |
9289 | ||
e7a6c00d JA |
9290 | tctx->registered_rings = kcalloc(IO_RINGFD_REG_MAX, |
9291 | sizeof(struct file *), GFP_KERNEL); | |
9292 | if (unlikely(!tctx->registered_rings)) { | |
9293 | kfree(tctx); | |
9294 | return -ENOMEM; | |
9295 | } | |
9296 | ||
d8a6df10 JA |
9297 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
9298 | if (unlikely(ret)) { | |
e7a6c00d | 9299 | kfree(tctx->registered_rings); |
d8a6df10 JA |
9300 | kfree(tctx); |
9301 | return ret; | |
9302 | } | |
9303 | ||
685fe7fe | 9304 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
9305 | if (IS_ERR(tctx->io_wq)) { |
9306 | ret = PTR_ERR(tctx->io_wq); | |
9307 | percpu_counter_destroy(&tctx->inflight); | |
e7a6c00d | 9308 | kfree(tctx->registered_rings); |
5aa75ed5 JA |
9309 | kfree(tctx); |
9310 | return ret; | |
9311 | } | |
9312 | ||
0f212204 JA |
9313 | xa_init(&tctx->xa); |
9314 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 9315 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 9316 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 9317 | task->io_uring = tctx; |
7cbf1722 JA |
9318 | spin_lock_init(&tctx->task_lock); |
9319 | INIT_WQ_LIST(&tctx->task_list); | |
4813c377 | 9320 | INIT_WQ_LIST(&tctx->prior_task_list); |
7cbf1722 | 9321 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
9322 | return 0; |
9323 | } | |
9324 | ||
9325 | void __io_uring_free(struct task_struct *tsk) | |
9326 | { | |
9327 | struct io_uring_task *tctx = tsk->io_uring; | |
9328 | ||
9329 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 9330 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 9331 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 9332 | |
e7a6c00d | 9333 | kfree(tctx->registered_rings); |
d8a6df10 | 9334 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
9335 | kfree(tctx); |
9336 | tsk->io_uring = NULL; | |
9337 | } | |
9338 | ||
c072481d PB |
9339 | static __cold int io_sq_offload_create(struct io_ring_ctx *ctx, |
9340 | struct io_uring_params *p) | |
2b188cc1 JA |
9341 | { |
9342 | int ret; | |
9343 | ||
d25e3a3d JA |
9344 | /* Retain compatibility with failing for an invalid attach attempt */ |
9345 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
9346 | IORING_SETUP_ATTACH_WQ) { | |
9347 | struct fd f; | |
9348 | ||
9349 | f = fdget(p->wq_fd); | |
9350 | if (!f.file) | |
9351 | return -ENXIO; | |
0cc936f7 JA |
9352 | if (f.file->f_op != &io_uring_fops) { |
9353 | fdput(f); | |
f2a48dd0 | 9354 | return -EINVAL; |
0cc936f7 JA |
9355 | } |
9356 | fdput(f); | |
d25e3a3d | 9357 | } |
6c271ce2 | 9358 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 9359 | struct task_struct *tsk; |
534ca6d6 | 9360 | struct io_sq_data *sqd; |
26984fbf | 9361 | bool attached; |
534ca6d6 | 9362 | |
cdc1404a PM |
9363 | ret = security_uring_sqpoll(); |
9364 | if (ret) | |
9365 | return ret; | |
9366 | ||
26984fbf | 9367 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
9368 | if (IS_ERR(sqd)) { |
9369 | ret = PTR_ERR(sqd); | |
9370 | goto err; | |
9371 | } | |
69fb2131 | 9372 | |
7c30f36a | 9373 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 9374 | ctx->sq_data = sqd; |
917257da JA |
9375 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
9376 | if (!ctx->sq_thread_idle) | |
9377 | ctx->sq_thread_idle = HZ; | |
9378 | ||
78d7f6ba | 9379 | io_sq_thread_park(sqd); |
de75a3d3 PB |
9380 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
9381 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 9382 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 9383 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
9384 | io_sq_thread_unpark(sqd); |
9385 | ||
de75a3d3 PB |
9386 | if (ret < 0) |
9387 | goto err; | |
9388 | if (attached) | |
5aa75ed5 | 9389 | return 0; |
aa06165d | 9390 | |
6c271ce2 | 9391 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 9392 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 9393 | |
917257da | 9394 | ret = -EINVAL; |
f2a48dd0 | 9395 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 9396 | goto err_sqpoll; |
37d1e2e3 | 9397 | sqd->sq_cpu = cpu; |
6c271ce2 | 9398 | } else { |
37d1e2e3 | 9399 | sqd->sq_cpu = -1; |
6c271ce2 | 9400 | } |
37d1e2e3 JA |
9401 | |
9402 | sqd->task_pid = current->pid; | |
5c2469e0 | 9403 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
9404 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
9405 | if (IS_ERR(tsk)) { | |
9406 | ret = PTR_ERR(tsk); | |
e8f98f24 | 9407 | goto err_sqpoll; |
6c271ce2 | 9408 | } |
97a73a0f | 9409 | |
46fe18b1 | 9410 | sqd->thread = tsk; |
97a73a0f | 9411 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 9412 | wake_up_new_task(tsk); |
0f212204 JA |
9413 | if (ret) |
9414 | goto err; | |
6c271ce2 JA |
9415 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
9416 | /* Can't have SQ_AFF without SQPOLL */ | |
9417 | ret = -EINVAL; | |
9418 | goto err; | |
9419 | } | |
9420 | ||
2b188cc1 | 9421 | return 0; |
f2a48dd0 PB |
9422 | err_sqpoll: |
9423 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 9424 | err: |
37d1e2e3 | 9425 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
9426 | return ret; |
9427 | } | |
9428 | ||
a087e2b5 BM |
9429 | static inline void __io_unaccount_mem(struct user_struct *user, |
9430 | unsigned long nr_pages) | |
2b188cc1 JA |
9431 | { |
9432 | atomic_long_sub(nr_pages, &user->locked_vm); | |
9433 | } | |
9434 | ||
a087e2b5 BM |
9435 | static inline int __io_account_mem(struct user_struct *user, |
9436 | unsigned long nr_pages) | |
2b188cc1 JA |
9437 | { |
9438 | unsigned long page_limit, cur_pages, new_pages; | |
9439 | ||
9440 | /* Don't allow more pages than we can safely lock */ | |
9441 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
9442 | ||
9443 | do { | |
9444 | cur_pages = atomic_long_read(&user->locked_vm); | |
9445 | new_pages = cur_pages + nr_pages; | |
9446 | if (new_pages > page_limit) | |
9447 | return -ENOMEM; | |
9448 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
9449 | new_pages) != cur_pages); | |
9450 | ||
9451 | return 0; | |
9452 | } | |
9453 | ||
26bfa89e | 9454 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 9455 | { |
62e398be | 9456 | if (ctx->user) |
a087e2b5 | 9457 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 9458 | |
26bfa89e JA |
9459 | if (ctx->mm_account) |
9460 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
9461 | } |
9462 | ||
26bfa89e | 9463 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 9464 | { |
30975825 BM |
9465 | int ret; |
9466 | ||
62e398be | 9467 | if (ctx->user) { |
30975825 BM |
9468 | ret = __io_account_mem(ctx->user, nr_pages); |
9469 | if (ret) | |
9470 | return ret; | |
9471 | } | |
9472 | ||
26bfa89e JA |
9473 | if (ctx->mm_account) |
9474 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
9475 | |
9476 | return 0; | |
9477 | } | |
9478 | ||
2b188cc1 JA |
9479 | static void io_mem_free(void *ptr) |
9480 | { | |
52e04ef4 MR |
9481 | struct page *page; |
9482 | ||
9483 | if (!ptr) | |
9484 | return; | |
2b188cc1 | 9485 | |
52e04ef4 | 9486 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
9487 | if (put_page_testzero(page)) |
9488 | free_compound_page(page); | |
9489 | } | |
9490 | ||
9491 | static void *io_mem_alloc(size_t size) | |
9492 | { | |
0a3f1e0b | 9493 | gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP; |
2b188cc1 | 9494 | |
0a3f1e0b | 9495 | return (void *) __get_free_pages(gfp, get_order(size)); |
2b188cc1 JA |
9496 | } |
9497 | ||
75b28aff HV |
9498 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
9499 | size_t *sq_offset) | |
9500 | { | |
9501 | struct io_rings *rings; | |
9502 | size_t off, sq_array_size; | |
9503 | ||
9504 | off = struct_size(rings, cqes, cq_entries); | |
9505 | if (off == SIZE_MAX) | |
9506 | return SIZE_MAX; | |
9507 | ||
9508 | #ifdef CONFIG_SMP | |
9509 | off = ALIGN(off, SMP_CACHE_BYTES); | |
9510 | if (off == 0) | |
9511 | return SIZE_MAX; | |
9512 | #endif | |
9513 | ||
b36200f5 DV |
9514 | if (sq_offset) |
9515 | *sq_offset = off; | |
9516 | ||
75b28aff HV |
9517 | sq_array_size = array_size(sizeof(u32), sq_entries); |
9518 | if (sq_array_size == SIZE_MAX) | |
9519 | return SIZE_MAX; | |
9520 | ||
9521 | if (check_add_overflow(off, sq_array_size, &off)) | |
9522 | return SIZE_MAX; | |
9523 | ||
75b28aff HV |
9524 | return off; |
9525 | } | |
9526 | ||
41edf1a5 | 9527 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 9528 | { |
41edf1a5 | 9529 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
9530 | unsigned int i; |
9531 | ||
6224843d PB |
9532 | if (imu != ctx->dummy_ubuf) { |
9533 | for (i = 0; i < imu->nr_bvecs; i++) | |
9534 | unpin_user_page(imu->bvec[i].bv_page); | |
9535 | if (imu->acct_pages) | |
9536 | io_unaccount_mem(ctx, imu->acct_pages); | |
9537 | kvfree(imu); | |
9538 | } | |
41edf1a5 | 9539 | *slot = NULL; |
7f61a1e9 PB |
9540 | } |
9541 | ||
bd54b6fe | 9542 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 9543 | { |
634d00df PB |
9544 | io_buffer_unmap(ctx, &prsrc->buf); |
9545 | prsrc->buf = NULL; | |
bd54b6fe | 9546 | } |
edafccee | 9547 | |
bd54b6fe BM |
9548 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
9549 | { | |
9550 | unsigned int i; | |
edafccee | 9551 | |
7f61a1e9 PB |
9552 | for (i = 0; i < ctx->nr_user_bufs; i++) |
9553 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 9554 | kfree(ctx->user_bufs); |
bb6659cc | 9555 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 9556 | ctx->user_bufs = NULL; |
bd54b6fe | 9557 | ctx->buf_data = NULL; |
edafccee | 9558 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
9559 | } |
9560 | ||
0a96bbe4 | 9561 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 9562 | { |
bd54b6fe | 9563 | int ret; |
edafccee | 9564 | |
bd54b6fe | 9565 | if (!ctx->buf_data) |
edafccee JA |
9566 | return -ENXIO; |
9567 | ||
bd54b6fe BM |
9568 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
9569 | if (!ret) | |
9570 | __io_sqe_buffers_unregister(ctx); | |
9571 | return ret; | |
edafccee JA |
9572 | } |
9573 | ||
9574 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
9575 | void __user *arg, unsigned index) | |
9576 | { | |
9577 | struct iovec __user *src; | |
9578 | ||
9579 | #ifdef CONFIG_COMPAT | |
9580 | if (ctx->compat) { | |
9581 | struct compat_iovec __user *ciovs; | |
9582 | struct compat_iovec ciov; | |
9583 | ||
9584 | ciovs = (struct compat_iovec __user *) arg; | |
9585 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
9586 | return -EFAULT; | |
9587 | ||
d55e5f5b | 9588 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
9589 | dst->iov_len = ciov.iov_len; |
9590 | return 0; | |
9591 | } | |
9592 | #endif | |
9593 | src = (struct iovec __user *) arg; | |
9594 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
9595 | return -EFAULT; | |
9596 | return 0; | |
9597 | } | |
9598 | ||
de293938 JA |
9599 | /* |
9600 | * Not super efficient, but this is just a registration time. And we do cache | |
9601 | * the last compound head, so generally we'll only do a full search if we don't | |
9602 | * match that one. | |
9603 | * | |
9604 | * We check if the given compound head page has already been accounted, to | |
9605 | * avoid double accounting it. This allows us to account the full size of the | |
9606 | * page, not just the constituent pages of a huge page. | |
9607 | */ | |
9608 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
9609 | int nr_pages, struct page *hpage) | |
9610 | { | |
9611 | int i, j; | |
9612 | ||
9613 | /* check current page array */ | |
9614 | for (i = 0; i < nr_pages; i++) { | |
9615 | if (!PageCompound(pages[i])) | |
9616 | continue; | |
9617 | if (compound_head(pages[i]) == hpage) | |
9618 | return true; | |
9619 | } | |
9620 | ||
9621 | /* check previously registered pages */ | |
9622 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 9623 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
9624 | |
9625 | for (j = 0; j < imu->nr_bvecs; j++) { | |
9626 | if (!PageCompound(imu->bvec[j].bv_page)) | |
9627 | continue; | |
9628 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
9629 | return true; | |
9630 | } | |
9631 | } | |
9632 | ||
9633 | return false; | |
9634 | } | |
9635 | ||
9636 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
9637 | int nr_pages, struct io_mapped_ubuf *imu, | |
9638 | struct page **last_hpage) | |
9639 | { | |
9640 | int i, ret; | |
9641 | ||
216e5835 | 9642 | imu->acct_pages = 0; |
de293938 JA |
9643 | for (i = 0; i < nr_pages; i++) { |
9644 | if (!PageCompound(pages[i])) { | |
9645 | imu->acct_pages++; | |
9646 | } else { | |
9647 | struct page *hpage; | |
9648 | ||
9649 | hpage = compound_head(pages[i]); | |
9650 | if (hpage == *last_hpage) | |
9651 | continue; | |
9652 | *last_hpage = hpage; | |
9653 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
9654 | continue; | |
9655 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
9656 | } | |
9657 | } | |
9658 | ||
9659 | if (!imu->acct_pages) | |
9660 | return 0; | |
9661 | ||
26bfa89e | 9662 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
9663 | if (ret) |
9664 | imu->acct_pages = 0; | |
9665 | return ret; | |
9666 | } | |
9667 | ||
0a96bbe4 | 9668 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 9669 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 9670 | struct page **last_hpage) |
edafccee | 9671 | { |
41edf1a5 | 9672 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
9673 | struct vm_area_struct **vmas = NULL; |
9674 | struct page **pages = NULL; | |
0a96bbe4 BM |
9675 | unsigned long off, start, end, ubuf; |
9676 | size_t size; | |
9677 | int ret, pret, nr_pages, i; | |
9678 | ||
6224843d PB |
9679 | if (!iov->iov_base) { |
9680 | *pimu = ctx->dummy_ubuf; | |
9681 | return 0; | |
9682 | } | |
9683 | ||
0a96bbe4 BM |
9684 | ubuf = (unsigned long) iov->iov_base; |
9685 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
9686 | start = ubuf >> PAGE_SHIFT; | |
9687 | nr_pages = end - start; | |
9688 | ||
41edf1a5 | 9689 | *pimu = NULL; |
0a96bbe4 BM |
9690 | ret = -ENOMEM; |
9691 | ||
9692 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
9693 | if (!pages) | |
9694 | goto done; | |
9695 | ||
9696 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
9697 | GFP_KERNEL); | |
9698 | if (!vmas) | |
9699 | goto done; | |
edafccee | 9700 | |
41edf1a5 | 9701 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
a2b4198c | 9702 | if (!imu) |
0a96bbe4 BM |
9703 | goto done; |
9704 | ||
9705 | ret = 0; | |
9706 | mmap_read_lock(current->mm); | |
9707 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
9708 | pages, vmas); | |
9709 | if (pret == nr_pages) { | |
9710 | /* don't support file backed memory */ | |
9711 | for (i = 0; i < nr_pages; i++) { | |
9712 | struct vm_area_struct *vma = vmas[i]; | |
9713 | ||
40dad765 PB |
9714 | if (vma_is_shmem(vma)) |
9715 | continue; | |
0a96bbe4 BM |
9716 | if (vma->vm_file && |
9717 | !is_file_hugepages(vma->vm_file)) { | |
9718 | ret = -EOPNOTSUPP; | |
9719 | break; | |
9720 | } | |
9721 | } | |
9722 | } else { | |
9723 | ret = pret < 0 ? pret : -EFAULT; | |
9724 | } | |
9725 | mmap_read_unlock(current->mm); | |
9726 | if (ret) { | |
9727 | /* | |
9728 | * if we did partial map, or found file backed vmas, | |
9729 | * release any pages we did get | |
9730 | */ | |
9731 | if (pret > 0) | |
9732 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
9733 | goto done; |
9734 | } | |
9735 | ||
9736 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
9737 | if (ret) { | |
9738 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
9739 | goto done; |
9740 | } | |
9741 | ||
9742 | off = ubuf & ~PAGE_MASK; | |
9743 | size = iov->iov_len; | |
9744 | for (i = 0; i < nr_pages; i++) { | |
9745 | size_t vec_len; | |
9746 | ||
9747 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
9748 | imu->bvec[i].bv_page = pages[i]; | |
9749 | imu->bvec[i].bv_len = vec_len; | |
9750 | imu->bvec[i].bv_offset = off; | |
9751 | off = 0; | |
9752 | size -= vec_len; | |
9753 | } | |
9754 | /* store original address for later verification */ | |
9755 | imu->ubuf = ubuf; | |
4751f53d | 9756 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 9757 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 9758 | *pimu = imu; |
0a96bbe4 BM |
9759 | ret = 0; |
9760 | done: | |
41edf1a5 PB |
9761 | if (ret) |
9762 | kvfree(imu); | |
0a96bbe4 BM |
9763 | kvfree(pages); |
9764 | kvfree(vmas); | |
9765 | return ret; | |
9766 | } | |
9767 | ||
2b358604 | 9768 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 9769 | { |
87094465 PB |
9770 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
9771 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 9772 | } |
edafccee | 9773 | |
2b358604 BM |
9774 | static int io_buffer_validate(struct iovec *iov) |
9775 | { | |
50e96989 PB |
9776 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
9777 | ||
2b358604 BM |
9778 | /* |
9779 | * Don't impose further limits on the size and buffer | |
9780 | * constraints here, we'll -EINVAL later when IO is | |
9781 | * submitted if they are wrong. | |
9782 | */ | |
6224843d PB |
9783 | if (!iov->iov_base) |
9784 | return iov->iov_len ? -EFAULT : 0; | |
9785 | if (!iov->iov_len) | |
2b358604 | 9786 | return -EFAULT; |
edafccee | 9787 | |
2b358604 BM |
9788 | /* arbitrary limit, but we need something */ |
9789 | if (iov->iov_len > SZ_1G) | |
9790 | return -EFAULT; | |
edafccee | 9791 | |
50e96989 PB |
9792 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
9793 | return -EOVERFLOW; | |
9794 | ||
2b358604 BM |
9795 | return 0; |
9796 | } | |
edafccee | 9797 | |
2b358604 | 9798 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 9799 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 9800 | { |
bd54b6fe BM |
9801 | struct page *last_hpage = NULL; |
9802 | struct io_rsrc_data *data; | |
2b358604 BM |
9803 | int i, ret; |
9804 | struct iovec iov; | |
edafccee | 9805 | |
87094465 PB |
9806 | if (ctx->user_bufs) |
9807 | return -EBUSY; | |
489809e2 | 9808 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 9809 | return -EINVAL; |
bd54b6fe | 9810 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
9811 | if (ret) |
9812 | return ret; | |
d878c816 PB |
9813 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
9814 | if (ret) | |
9815 | return ret; | |
bd54b6fe BM |
9816 | ret = io_buffers_map_alloc(ctx, nr_args); |
9817 | if (ret) { | |
bb6659cc | 9818 | io_rsrc_data_free(data); |
bd54b6fe BM |
9819 | return ret; |
9820 | } | |
edafccee | 9821 | |
87094465 | 9822 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
9823 | ret = io_copy_iov(ctx, &iov, arg, i); |
9824 | if (ret) | |
0a96bbe4 | 9825 | break; |
2b358604 BM |
9826 | ret = io_buffer_validate(&iov); |
9827 | if (ret) | |
0a96bbe4 | 9828 | break; |
2d091d62 | 9829 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
9830 | ret = -EINVAL; |
9831 | break; | |
9832 | } | |
edafccee | 9833 | |
41edf1a5 PB |
9834 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
9835 | &last_hpage); | |
0a96bbe4 BM |
9836 | if (ret) |
9837 | break; | |
edafccee | 9838 | } |
0a96bbe4 | 9839 | |
bd54b6fe | 9840 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 9841 | |
bd54b6fe BM |
9842 | ctx->buf_data = data; |
9843 | if (ret) | |
9844 | __io_sqe_buffers_unregister(ctx); | |
9845 | else | |
9846 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
9847 | return ret; |
9848 | } | |
9849 | ||
634d00df PB |
9850 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
9851 | struct io_uring_rsrc_update2 *up, | |
9852 | unsigned int nr_args) | |
9853 | { | |
9854 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
9855 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
9856 | struct page *last_hpage = NULL; |
9857 | bool needs_switch = false; | |
9858 | __u32 done; | |
9859 | int i, err; | |
9860 | ||
9861 | if (!ctx->buf_data) | |
9862 | return -ENXIO; | |
9863 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
9864 | return -EINVAL; | |
9865 | ||
9866 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
9867 | struct io_mapped_ubuf *imu; |
9868 | int offset = up->offset + done; | |
634d00df PB |
9869 | u64 tag = 0; |
9870 | ||
9871 | err = io_copy_iov(ctx, &iov, iovs, done); | |
9872 | if (err) | |
9873 | break; | |
9874 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
9875 | err = -EFAULT; | |
9876 | break; | |
9877 | } | |
0b8c0e7c PB |
9878 | err = io_buffer_validate(&iov); |
9879 | if (err) | |
9880 | break; | |
cf3770e7 CIK |
9881 | if (!iov.iov_base && tag) { |
9882 | err = -EINVAL; | |
9883 | break; | |
9884 | } | |
0b8c0e7c PB |
9885 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
9886 | if (err) | |
9887 | break; | |
634d00df | 9888 | |
0b8c0e7c | 9889 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 9890 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
0b8c0e7c PB |
9891 | err = io_queue_rsrc_removal(ctx->buf_data, offset, |
9892 | ctx->rsrc_node, ctx->user_bufs[i]); | |
9893 | if (unlikely(err)) { | |
9894 | io_buffer_unmap(ctx, &imu); | |
634d00df | 9895 | break; |
0b8c0e7c | 9896 | } |
634d00df PB |
9897 | ctx->user_bufs[i] = NULL; |
9898 | needs_switch = true; | |
9899 | } | |
9900 | ||
0b8c0e7c | 9901 | ctx->user_bufs[i] = imu; |
2d091d62 | 9902 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
9903 | } |
9904 | ||
9905 | if (needs_switch) | |
9906 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
9907 | return done ? done : err; | |
9908 | } | |
9909 | ||
c75312dd UA |
9910 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg, |
9911 | unsigned int eventfd_async) | |
9b402849 | 9912 | { |
77bc59b4 | 9913 | struct io_ev_fd *ev_fd; |
9b402849 | 9914 | __s32 __user *fds = arg; |
f0a4e62b | 9915 | int fd; |
9b402849 | 9916 | |
77bc59b4 UA |
9917 | ev_fd = rcu_dereference_protected(ctx->io_ev_fd, |
9918 | lockdep_is_held(&ctx->uring_lock)); | |
9919 | if (ev_fd) | |
9b402849 JA |
9920 | return -EBUSY; |
9921 | ||
9922 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
9923 | return -EFAULT; | |
9924 | ||
77bc59b4 UA |
9925 | ev_fd = kmalloc(sizeof(*ev_fd), GFP_KERNEL); |
9926 | if (!ev_fd) | |
9927 | return -ENOMEM; | |
fe7e3257 | 9928 | |
77bc59b4 UA |
9929 | ev_fd->cq_ev_fd = eventfd_ctx_fdget(fd); |
9930 | if (IS_ERR(ev_fd->cq_ev_fd)) { | |
f0a4e62b | 9931 | int ret = PTR_ERR(ev_fd->cq_ev_fd); |
77bc59b4 | 9932 | kfree(ev_fd); |
9b402849 JA |
9933 | return ret; |
9934 | } | |
c75312dd | 9935 | ev_fd->eventfd_async = eventfd_async; |
9aa8dfde | 9936 | ctx->has_evfd = true; |
77bc59b4 | 9937 | rcu_assign_pointer(ctx->io_ev_fd, ev_fd); |
f0a4e62b | 9938 | return 0; |
77bc59b4 UA |
9939 | } |
9940 | ||
9941 | static void io_eventfd_put(struct rcu_head *rcu) | |
9942 | { | |
9943 | struct io_ev_fd *ev_fd = container_of(rcu, struct io_ev_fd, rcu); | |
9944 | ||
9945 | eventfd_ctx_put(ev_fd->cq_ev_fd); | |
9946 | kfree(ev_fd); | |
9b402849 JA |
9947 | } |
9948 | ||
9949 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
9950 | { | |
77bc59b4 UA |
9951 | struct io_ev_fd *ev_fd; |
9952 | ||
9953 | ev_fd = rcu_dereference_protected(ctx->io_ev_fd, | |
9954 | lockdep_is_held(&ctx->uring_lock)); | |
9955 | if (ev_fd) { | |
9aa8dfde | 9956 | ctx->has_evfd = false; |
77bc59b4 UA |
9957 | rcu_assign_pointer(ctx->io_ev_fd, NULL); |
9958 | call_rcu(&ev_fd->rcu, io_eventfd_put); | |
9b402849 JA |
9959 | return 0; |
9960 | } | |
9961 | ||
9962 | return -ENXIO; | |
9963 | } | |
9964 | ||
5a2e745d JA |
9965 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
9966 | { | |
dbc7d452 JA |
9967 | int i; |
9968 | ||
9969 | for (i = 0; i < (1U << IO_BUFFERS_HASH_BITS); i++) { | |
9970 | struct list_head *list = &ctx->io_buffers[i]; | |
9e15c3a0 | 9971 | |
dbc7d452 JA |
9972 | while (!list_empty(list)) { |
9973 | struct io_buffer_list *bl; | |
9974 | ||
9975 | bl = list_first_entry(list, struct io_buffer_list, list); | |
9976 | __io_remove_buffers(ctx, bl, -1U); | |
9977 | list_del(&bl->list); | |
9978 | kfree(bl); | |
9979 | } | |
9980 | } | |
cc3cec83 JA |
9981 | |
9982 | while (!list_empty(&ctx->io_buffers_pages)) { | |
9983 | struct page *page; | |
9984 | ||
9985 | page = list_first_entry(&ctx->io_buffers_pages, struct page, lru); | |
9986 | list_del_init(&page->lru); | |
9987 | __free_page(page); | |
9988 | } | |
5a2e745d JA |
9989 | } |
9990 | ||
4010fec4 | 9991 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 9992 | { |
cd0ca2e0 | 9993 | struct io_submit_state *state = &ctx->submit_state; |
37f0e767 | 9994 | int nr = 0; |
bf019da7 | 9995 | |
9a4fdbd8 | 9996 | mutex_lock(&ctx->uring_lock); |
cd0ca2e0 | 9997 | io_flush_cached_locked_reqs(ctx, state); |
9a4fdbd8 | 9998 | |
c2b6c6bc PB |
9999 | while (state->free_list.next) { |
10000 | struct io_wq_work_node *node; | |
10001 | struct io_kiocb *req; | |
9a4fdbd8 | 10002 | |
c2b6c6bc PB |
10003 | node = wq_stack_extract(&state->free_list); |
10004 | req = container_of(node, struct io_kiocb, comp_list); | |
10005 | kmem_cache_free(req_cachep, req); | |
37f0e767 | 10006 | nr++; |
c2b6c6bc | 10007 | } |
37f0e767 PB |
10008 | if (nr) |
10009 | percpu_ref_put_many(&ctx->refs, nr); | |
9a4fdbd8 JA |
10010 | mutex_unlock(&ctx->uring_lock); |
10011 | } | |
10012 | ||
43597aac | 10013 | static void io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 10014 | { |
43597aac | 10015 | if (data && !atomic_dec_and_test(&data->refs)) |
bd54b6fe | 10016 | wait_for_completion(&data->done); |
bd54b6fe | 10017 | } |
04fc6c80 | 10018 | |
4d9237e3 JA |
10019 | static void io_flush_apoll_cache(struct io_ring_ctx *ctx) |
10020 | { | |
10021 | struct async_poll *apoll; | |
10022 | ||
10023 | while (!list_empty(&ctx->apoll_cache)) { | |
10024 | apoll = list_first_entry(&ctx->apoll_cache, struct async_poll, | |
10025 | poll.wait.entry); | |
10026 | list_del(&apoll->poll.wait.entry); | |
10027 | kfree(apoll); | |
10028 | } | |
10029 | } | |
10030 | ||
c072481d | 10031 | static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx) |
2b188cc1 | 10032 | { |
37d1e2e3 | 10033 | io_sq_thread_finish(ctx); |
2aede0e4 | 10034 | |
37d1e2e3 | 10035 | if (ctx->mm_account) { |
2aede0e4 JA |
10036 | mmdrop(ctx->mm_account); |
10037 | ctx->mm_account = NULL; | |
30975825 | 10038 | } |
def596e9 | 10039 | |
ab409402 | 10040 | io_rsrc_refs_drop(ctx); |
43597aac PB |
10041 | /* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */ |
10042 | io_wait_rsrc_data(ctx->buf_data); | |
10043 | io_wait_rsrc_data(ctx->file_data); | |
10044 | ||
8bad28d8 | 10045 | mutex_lock(&ctx->uring_lock); |
43597aac | 10046 | if (ctx->buf_data) |
bd54b6fe | 10047 | __io_sqe_buffers_unregister(ctx); |
43597aac | 10048 | if (ctx->file_data) |
08480400 | 10049 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
10050 | if (ctx->rings) |
10051 | __io_cqring_overflow_flush(ctx, true); | |
9b402849 | 10052 | io_eventfd_unregister(ctx); |
4d9237e3 | 10053 | io_flush_apoll_cache(ctx); |
77bc59b4 | 10054 | mutex_unlock(&ctx->uring_lock); |
5a2e745d | 10055 | io_destroy_buffers(ctx); |
07db298a PB |
10056 | if (ctx->sq_creds) |
10057 | put_cred(ctx->sq_creds); | |
def596e9 | 10058 | |
a7f0ed5a PB |
10059 | /* there are no registered resources left, nobody uses it */ |
10060 | if (ctx->rsrc_node) | |
10061 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 10062 | if (ctx->rsrc_backup_node) |
b895c9a6 | 10063 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a | 10064 | flush_delayed_work(&ctx->rsrc_put_work); |
756ab7c0 | 10065 | flush_delayed_work(&ctx->fallback_work); |
a7f0ed5a PB |
10066 | |
10067 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
10068 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 10069 | |
2b188cc1 | 10070 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
10071 | if (ctx->ring_sock) { |
10072 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 10073 | sock_release(ctx->ring_sock); |
355e8d26 | 10074 | } |
2b188cc1 | 10075 | #endif |
ef9dd637 | 10076 | WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list)); |
2b188cc1 | 10077 | |
75b28aff | 10078 | io_mem_free(ctx->rings); |
2b188cc1 | 10079 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
10080 | |
10081 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 10082 | free_uid(ctx->user); |
4010fec4 | 10083 | io_req_caches_free(ctx); |
e941894e JA |
10084 | if (ctx->hash_map) |
10085 | io_wq_put_hash(ctx->hash_map); | |
adc8682e | 10086 | io_free_napi_list(ctx); |
78076bb6 | 10087 | kfree(ctx->cancel_hash); |
6224843d | 10088 | kfree(ctx->dummy_ubuf); |
dbc7d452 | 10089 | kfree(ctx->io_buffers); |
2b188cc1 JA |
10090 | kfree(ctx); |
10091 | } | |
10092 | ||
10093 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
10094 | { | |
10095 | struct io_ring_ctx *ctx = file->private_data; | |
10096 | __poll_t mask = 0; | |
10097 | ||
d60aa65b | 10098 | poll_wait(file, &ctx->cq_wait, wait); |
4f7067c3 SB |
10099 | /* |
10100 | * synchronizes with barrier from wq_has_sleeper call in | |
10101 | * io_commit_cqring | |
10102 | */ | |
2b188cc1 | 10103 | smp_rmb(); |
90554200 | 10104 | if (!io_sqring_full(ctx)) |
2b188cc1 | 10105 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
10106 | |
10107 | /* | |
10108 | * Don't flush cqring overflow list here, just do a simple check. | |
10109 | * Otherwise there could possible be ABBA deadlock: | |
10110 | * CPU0 CPU1 | |
10111 | * ---- ---- | |
10112 | * lock(&ctx->uring_lock); | |
10113 | * lock(&ep->mtx); | |
10114 | * lock(&ctx->uring_lock); | |
10115 | * lock(&ep->mtx); | |
10116 | * | |
10117 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
10118 | * pushs them to do the flush. | |
10119 | */ | |
5ed7a37d | 10120 | if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) |
2b188cc1 JA |
10121 | mask |= EPOLLIN | EPOLLRDNORM; |
10122 | ||
10123 | return mask; | |
10124 | } | |
10125 | ||
0bead8cd | 10126 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 10127 | { |
4379bf8b | 10128 | const struct cred *creds; |
071698e1 | 10129 | |
61cf9370 | 10130 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
10131 | if (creds) { |
10132 | put_cred(creds); | |
0bead8cd | 10133 | return 0; |
1e6fa521 | 10134 | } |
0bead8cd YD |
10135 | |
10136 | return -EINVAL; | |
10137 | } | |
10138 | ||
d56d938b PB |
10139 | struct io_tctx_exit { |
10140 | struct callback_head task_work; | |
10141 | struct completion completion; | |
baf186c4 | 10142 | struct io_ring_ctx *ctx; |
d56d938b PB |
10143 | }; |
10144 | ||
c072481d | 10145 | static __cold void io_tctx_exit_cb(struct callback_head *cb) |
d56d938b PB |
10146 | { |
10147 | struct io_uring_task *tctx = current->io_uring; | |
10148 | struct io_tctx_exit *work; | |
10149 | ||
10150 | work = container_of(cb, struct io_tctx_exit, task_work); | |
10151 | /* | |
10152 | * When @in_idle, we're in cancellation and it's racy to remove the | |
10153 | * node. It'll be removed by the end of cancellation, just ignore it. | |
10154 | */ | |
10155 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 10156 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
10157 | complete(&work->completion); |
10158 | } | |
10159 | ||
c072481d | 10160 | static __cold bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
28090c13 PB |
10161 | { |
10162 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
10163 | ||
10164 | return req->ctx == data; | |
10165 | } | |
10166 | ||
c072481d | 10167 | static __cold void io_ring_exit_work(struct work_struct *work) |
85faa7b8 | 10168 | { |
d56d938b | 10169 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 10170 | unsigned long timeout = jiffies + HZ * 60 * 5; |
58d3be2c | 10171 | unsigned long interval = HZ / 20; |
d56d938b PB |
10172 | struct io_tctx_exit exit; |
10173 | struct io_tctx_node *node; | |
10174 | int ret; | |
85faa7b8 | 10175 | |
56952e91 JA |
10176 | /* |
10177 | * If we're doing polled IO and end up having requests being | |
10178 | * submitted async (out-of-line), then completions can come in while | |
10179 | * we're waiting for refs to drop. We need to reap these manually, | |
10180 | * as nobody else will be looking for them. | |
10181 | */ | |
b2edc0a7 | 10182 | do { |
3dd0c97a | 10183 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
10184 | if (ctx->sq_data) { |
10185 | struct io_sq_data *sqd = ctx->sq_data; | |
10186 | struct task_struct *tsk; | |
10187 | ||
10188 | io_sq_thread_park(sqd); | |
10189 | tsk = sqd->thread; | |
10190 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
10191 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
10192 | io_cancel_ctx_cb, ctx, true); | |
10193 | io_sq_thread_unpark(sqd); | |
10194 | } | |
b5bb3a24 | 10195 | |
37f0e767 PB |
10196 | io_req_caches_free(ctx); |
10197 | ||
58d3be2c PB |
10198 | if (WARN_ON_ONCE(time_after(jiffies, timeout))) { |
10199 | /* there is little hope left, don't run it too often */ | |
10200 | interval = HZ * 60; | |
10201 | } | |
10202 | } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); | |
d56d938b | 10203 | |
7f00651a PB |
10204 | init_completion(&exit.completion); |
10205 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
10206 | exit.ctx = ctx; | |
89b5066e PB |
10207 | /* |
10208 | * Some may use context even when all refs and requests have been put, | |
10209 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 10210 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
10211 | * this lock/unlock section also waits them to finish. |
10212 | */ | |
d56d938b PB |
10213 | mutex_lock(&ctx->uring_lock); |
10214 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
10215 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
10216 | ||
d56d938b PB |
10217 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
10218 | ctx_node); | |
7f00651a PB |
10219 | /* don't spin on a single task if cancellation failed */ |
10220 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
10221 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
10222 | if (WARN_ON_ONCE(ret)) | |
10223 | continue; | |
d56d938b PB |
10224 | |
10225 | mutex_unlock(&ctx->uring_lock); | |
10226 | wait_for_completion(&exit.completion); | |
d56d938b PB |
10227 | mutex_lock(&ctx->uring_lock); |
10228 | } | |
10229 | mutex_unlock(&ctx->uring_lock); | |
79ebeaee JA |
10230 | spin_lock(&ctx->completion_lock); |
10231 | spin_unlock(&ctx->completion_lock); | |
d56d938b | 10232 | |
85faa7b8 JA |
10233 | io_ring_ctx_free(ctx); |
10234 | } | |
10235 | ||
80c4cbdb | 10236 | /* Returns true if we found and killed one or more timeouts */ |
c072481d PB |
10237 | static __cold bool io_kill_timeouts(struct io_ring_ctx *ctx, |
10238 | struct task_struct *tsk, bool cancel_all) | |
80c4cbdb PB |
10239 | { |
10240 | struct io_kiocb *req, *tmp; | |
10241 | int canceled = 0; | |
10242 | ||
79ebeaee JA |
10243 | spin_lock(&ctx->completion_lock); |
10244 | spin_lock_irq(&ctx->timeout_lock); | |
80c4cbdb | 10245 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
3dd0c97a | 10246 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
10247 | io_kill_timeout(req, -ECANCELED); |
10248 | canceled++; | |
10249 | } | |
10250 | } | |
79ebeaee | 10251 | spin_unlock_irq(&ctx->timeout_lock); |
51520426 PB |
10252 | if (canceled != 0) |
10253 | io_commit_cqring(ctx); | |
79ebeaee | 10254 | spin_unlock(&ctx->completion_lock); |
80c4cbdb PB |
10255 | if (canceled != 0) |
10256 | io_cqring_ev_posted(ctx); | |
10257 | return canceled != 0; | |
10258 | } | |
10259 | ||
c072481d | 10260 | static __cold void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
2b188cc1 | 10261 | { |
61cf9370 MWO |
10262 | unsigned long index; |
10263 | struct creds *creds; | |
10264 | ||
2b188cc1 JA |
10265 | mutex_lock(&ctx->uring_lock); |
10266 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 10267 | if (ctx->rings) |
6c2450ae | 10268 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
10269 | xa_for_each(&ctx->personalities, index, creds) |
10270 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
10271 | mutex_unlock(&ctx->uring_lock); |
10272 | ||
3dd0c97a PB |
10273 | io_kill_timeouts(ctx, NULL, true); |
10274 | io_poll_remove_all(ctx, NULL, true); | |
561fb04a | 10275 | |
15dff286 | 10276 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 10277 | io_iopoll_try_reap_events(ctx); |
309fc03a | 10278 | |
85faa7b8 | 10279 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
10280 | /* |
10281 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
10282 | * if we're exiting a ton of rings at the same time. It just adds | |
10283 | * noise and overhead, there's no discernable change in runtime | |
10284 | * over using system_wq. | |
10285 | */ | |
10286 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
10287 | } |
10288 | ||
10289 | static int io_uring_release(struct inode *inode, struct file *file) | |
10290 | { | |
10291 | struct io_ring_ctx *ctx = file->private_data; | |
10292 | ||
10293 | file->private_data = NULL; | |
10294 | io_ring_ctx_wait_and_kill(ctx); | |
10295 | return 0; | |
10296 | } | |
10297 | ||
f6edbabb PB |
10298 | struct io_task_cancel { |
10299 | struct task_struct *task; | |
3dd0c97a | 10300 | bool all; |
f6edbabb | 10301 | }; |
f254ac04 | 10302 | |
f6edbabb | 10303 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 10304 | { |
9a472ef7 | 10305 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 10306 | struct io_task_cancel *cancel = data; |
9a472ef7 | 10307 | |
6af3f48b | 10308 | return io_match_task_safe(req, cancel->task, cancel->all); |
b711d4ea JA |
10309 | } |
10310 | ||
c072481d PB |
10311 | static __cold bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
10312 | struct task_struct *task, | |
10313 | bool cancel_all) | |
b7ddce3c | 10314 | { |
e1915f76 | 10315 | struct io_defer_entry *de; |
b7ddce3c PB |
10316 | LIST_HEAD(list); |
10317 | ||
79ebeaee | 10318 | spin_lock(&ctx->completion_lock); |
b7ddce3c | 10319 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { |
6af3f48b | 10320 | if (io_match_task_safe(de->req, task, cancel_all)) { |
b7ddce3c PB |
10321 | list_cut_position(&list, &ctx->defer_list, &de->list); |
10322 | break; | |
10323 | } | |
10324 | } | |
79ebeaee | 10325 | spin_unlock(&ctx->completion_lock); |
e1915f76 PB |
10326 | if (list_empty(&list)) |
10327 | return false; | |
b7ddce3c PB |
10328 | |
10329 | while (!list_empty(&list)) { | |
10330 | de = list_first_entry(&list, struct io_defer_entry, list); | |
10331 | list_del_init(&de->list); | |
f41db273 | 10332 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
10333 | kfree(de); |
10334 | } | |
e1915f76 | 10335 | return true; |
b7ddce3c PB |
10336 | } |
10337 | ||
c072481d | 10338 | static __cold bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
1b00764f PB |
10339 | { |
10340 | struct io_tctx_node *node; | |
10341 | enum io_wq_cancel cret; | |
10342 | bool ret = false; | |
10343 | ||
10344 | mutex_lock(&ctx->uring_lock); | |
10345 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
10346 | struct io_uring_task *tctx = node->task->io_uring; | |
10347 | ||
10348 | /* | |
10349 | * io_wq will stay alive while we hold uring_lock, because it's | |
10350 | * killed after ctx nodes, which requires to take the lock. | |
10351 | */ | |
10352 | if (!tctx || !tctx->io_wq) | |
10353 | continue; | |
10354 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
10355 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
10356 | } | |
10357 | mutex_unlock(&ctx->uring_lock); | |
10358 | ||
10359 | return ret; | |
10360 | } | |
10361 | ||
c072481d PB |
10362 | static __cold void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
10363 | struct task_struct *task, | |
10364 | bool cancel_all) | |
9936c7c2 | 10365 | { |
3dd0c97a | 10366 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 10367 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
10368 | |
10369 | while (1) { | |
10370 | enum io_wq_cancel cret; | |
10371 | bool ret = false; | |
10372 | ||
1b00764f PB |
10373 | if (!task) { |
10374 | ret |= io_uring_try_cancel_iowq(ctx); | |
10375 | } else if (tctx && tctx->io_wq) { | |
10376 | /* | |
10377 | * Cancels requests of all rings, not only @ctx, but | |
10378 | * it's fine as the task is in exit/exec. | |
10379 | */ | |
5aa75ed5 | 10380 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
10381 | &cancel, true); |
10382 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
10383 | } | |
10384 | ||
10385 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 10386 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 10387 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
5eef4e87 | 10388 | while (!wq_list_empty(&ctx->iopoll_list)) { |
9936c7c2 PB |
10389 | io_iopoll_try_reap_events(ctx); |
10390 | ret = true; | |
10391 | } | |
10392 | } | |
10393 | ||
3dd0c97a PB |
10394 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
10395 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
10396 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
10397 | if (task) |
10398 | ret |= io_run_task_work(); | |
9936c7c2 PB |
10399 | if (!ret) |
10400 | break; | |
10401 | cond_resched(); | |
10402 | } | |
10403 | } | |
10404 | ||
eef51daa | 10405 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 10406 | { |
236434c3 | 10407 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 10408 | struct io_tctx_node *node; |
a528b04e | 10409 | int ret; |
236434c3 MWO |
10410 | |
10411 | if (unlikely(!tctx)) { | |
5aa75ed5 | 10412 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
10413 | if (unlikely(ret)) |
10414 | return ret; | |
e139a1ec | 10415 | |
236434c3 | 10416 | tctx = current->io_uring; |
e139a1ec PB |
10417 | if (ctx->iowq_limits_set) { |
10418 | unsigned int limits[2] = { ctx->iowq_limits[0], | |
10419 | ctx->iowq_limits[1], }; | |
10420 | ||
10421 | ret = io_wq_max_workers(tctx->io_wq, limits); | |
10422 | if (ret) | |
10423 | return ret; | |
10424 | } | |
0f212204 | 10425 | } |
cf27f3b1 PB |
10426 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
10427 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
10428 | if (!node) | |
10429 | return -ENOMEM; | |
10430 | node->ctx = ctx; | |
10431 | node->task = current; | |
13bf43f5 | 10432 | |
cf27f3b1 PB |
10433 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
10434 | node, GFP_KERNEL)); | |
10435 | if (ret) { | |
10436 | kfree(node); | |
10437 | return ret; | |
0f212204 | 10438 | } |
cf27f3b1 PB |
10439 | |
10440 | mutex_lock(&ctx->uring_lock); | |
10441 | list_add(&node->ctx_node, &ctx->tctx_list); | |
10442 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 10443 | } |
cf27f3b1 | 10444 | tctx->last = ctx; |
0f212204 JA |
10445 | return 0; |
10446 | } | |
10447 | ||
cf27f3b1 PB |
10448 | /* |
10449 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
10450 | */ | |
eef51daa | 10451 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
10452 | { |
10453 | struct io_uring_task *tctx = current->io_uring; | |
10454 | ||
10455 | if (likely(tctx && tctx->last == ctx)) | |
10456 | return 0; | |
eef51daa | 10457 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
10458 | } |
10459 | ||
0f212204 JA |
10460 | /* |
10461 | * Remove this io_uring_file -> task mapping. | |
10462 | */ | |
c072481d | 10463 | static __cold void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
10464 | { |
10465 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 10466 | struct io_tctx_node *node; |
2941267b | 10467 | |
eebd2e37 PB |
10468 | if (!tctx) |
10469 | return; | |
13bf43f5 PB |
10470 | node = xa_erase(&tctx->xa, index); |
10471 | if (!node) | |
2941267b | 10472 | return; |
0f212204 | 10473 | |
13bf43f5 PB |
10474 | WARN_ON_ONCE(current != node->task); |
10475 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
10476 | ||
10477 | mutex_lock(&node->ctx->uring_lock); | |
10478 | list_del(&node->ctx_node); | |
10479 | mutex_unlock(&node->ctx->uring_lock); | |
10480 | ||
baf186c4 | 10481 | if (tctx->last == node->ctx) |
0f212204 | 10482 | tctx->last = NULL; |
13bf43f5 | 10483 | kfree(node); |
0f212204 JA |
10484 | } |
10485 | ||
c072481d | 10486 | static __cold void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 10487 | { |
ba5ef6dc | 10488 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 10489 | struct io_tctx_node *node; |
de7f1d9e PB |
10490 | unsigned long index; |
10491 | ||
8bab4c09 | 10492 | xa_for_each(&tctx->xa, index, node) { |
eef51daa | 10493 | io_uring_del_tctx_node(index); |
8bab4c09 JA |
10494 | cond_resched(); |
10495 | } | |
b16ef427 ME |
10496 | if (wq) { |
10497 | /* | |
f6f9b278 | 10498 | * Must be after io_uring_del_tctx_node() (removes nodes under |
b16ef427 ME |
10499 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). |
10500 | */ | |
ba5ef6dc | 10501 | io_wq_put_and_exit(wq); |
dadebc35 | 10502 | tctx->io_wq = NULL; |
b16ef427 | 10503 | } |
de7f1d9e PB |
10504 | } |
10505 | ||
3f48cf18 | 10506 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 10507 | { |
3f48cf18 PB |
10508 | if (tracked) |
10509 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
10510 | return percpu_counter_sum(&tctx->inflight); |
10511 | } | |
10512 | ||
78cc687b PB |
10513 | /* |
10514 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
78a78060 | 10515 | * requests. @sqd should be not-null IFF it's an SQPOLL thread cancellation. |
78cc687b | 10516 | */ |
c072481d PB |
10517 | static __cold void io_uring_cancel_generic(bool cancel_all, |
10518 | struct io_sq_data *sqd) | |
0e9ddb39 | 10519 | { |
521d6a73 | 10520 | struct io_uring_task *tctx = current->io_uring; |
734551df | 10521 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
10522 | s64 inflight; |
10523 | DEFINE_WAIT(wait); | |
fdaf083c | 10524 | |
78cc687b PB |
10525 | WARN_ON_ONCE(sqd && sqd->thread != current); |
10526 | ||
6d042ffb PO |
10527 | if (!current->io_uring) |
10528 | return; | |
17a91051 PB |
10529 | if (tctx->io_wq) |
10530 | io_wq_exit_start(tctx->io_wq); | |
10531 | ||
0e9ddb39 PB |
10532 | atomic_inc(&tctx->in_idle); |
10533 | do { | |
e9dbe221 | 10534 | io_uring_drop_tctx_refs(current); |
0e9ddb39 | 10535 | /* read completions before cancelations */ |
78cc687b | 10536 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
10537 | if (!inflight) |
10538 | break; | |
fdaf083c | 10539 | |
78cc687b PB |
10540 | if (!sqd) { |
10541 | struct io_tctx_node *node; | |
10542 | unsigned long index; | |
0f212204 | 10543 | |
78cc687b PB |
10544 | xa_for_each(&tctx->xa, index, node) { |
10545 | /* sqpoll task will cancel all its requests */ | |
10546 | if (node->ctx->sq_data) | |
10547 | continue; | |
10548 | io_uring_try_cancel_requests(node->ctx, current, | |
10549 | cancel_all); | |
10550 | } | |
10551 | } else { | |
10552 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
10553 | io_uring_try_cancel_requests(ctx, current, | |
10554 | cancel_all); | |
10555 | } | |
17a91051 | 10556 | |
78a78060 JA |
10557 | prepare_to_wait(&tctx->wait, &wait, TASK_INTERRUPTIBLE); |
10558 | io_run_task_work(); | |
e9dbe221 | 10559 | io_uring_drop_tctx_refs(current); |
78a78060 | 10560 | |
0f212204 | 10561 | /* |
a1bb3cd5 PB |
10562 | * If we've seen completions, retry without waiting. This |
10563 | * avoids a race where a completion comes in before we did | |
10564 | * prepare_to_wait(). | |
0f212204 | 10565 | */ |
3dd0c97a | 10566 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 10567 | schedule(); |
f57555ed | 10568 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 10569 | } while (1); |
de7f1d9e | 10570 | |
8452d4a6 | 10571 | io_uring_clean_tctx(tctx); |
3dd0c97a | 10572 | if (cancel_all) { |
3cc7fdb9 PB |
10573 | /* |
10574 | * We shouldn't run task_works after cancel, so just leave | |
10575 | * ->in_idle set for normal exit. | |
10576 | */ | |
10577 | atomic_dec(&tctx->in_idle); | |
3f48cf18 PB |
10578 | /* for exec all current's requests should be gone, kill tctx */ |
10579 | __io_uring_free(current); | |
10580 | } | |
44e728b8 PB |
10581 | } |
10582 | ||
f552a27a | 10583 | void __io_uring_cancel(bool cancel_all) |
78cc687b | 10584 | { |
f552a27a | 10585 | io_uring_cancel_generic(cancel_all, NULL); |
78cc687b PB |
10586 | } |
10587 | ||
e7a6c00d JA |
10588 | void io_uring_unreg_ringfd(void) |
10589 | { | |
10590 | struct io_uring_task *tctx = current->io_uring; | |
10591 | int i; | |
10592 | ||
10593 | for (i = 0; i < IO_RINGFD_REG_MAX; i++) { | |
10594 | if (tctx->registered_rings[i]) { | |
10595 | fput(tctx->registered_rings[i]); | |
10596 | tctx->registered_rings[i] = NULL; | |
10597 | } | |
10598 | } | |
10599 | } | |
10600 | ||
10601 | static int io_ring_add_registered_fd(struct io_uring_task *tctx, int fd, | |
10602 | int start, int end) | |
10603 | { | |
10604 | struct file *file; | |
10605 | int offset; | |
10606 | ||
10607 | for (offset = start; offset < end; offset++) { | |
10608 | offset = array_index_nospec(offset, IO_RINGFD_REG_MAX); | |
10609 | if (tctx->registered_rings[offset]) | |
10610 | continue; | |
10611 | ||
10612 | file = fget(fd); | |
10613 | if (!file) { | |
10614 | return -EBADF; | |
10615 | } else if (file->f_op != &io_uring_fops) { | |
10616 | fput(file); | |
10617 | return -EOPNOTSUPP; | |
10618 | } | |
10619 | tctx->registered_rings[offset] = file; | |
10620 | return offset; | |
10621 | } | |
10622 | ||
10623 | return -EBUSY; | |
10624 | } | |
10625 | ||
10626 | /* | |
10627 | * Register a ring fd to avoid fdget/fdput for each io_uring_enter() | |
10628 | * invocation. User passes in an array of struct io_uring_rsrc_update | |
10629 | * with ->data set to the ring_fd, and ->offset given for the desired | |
10630 | * index. If no index is desired, application may set ->offset == -1U | |
10631 | * and we'll find an available index. Returns number of entries | |
10632 | * successfully processed, or < 0 on error if none were processed. | |
10633 | */ | |
10634 | static int io_ringfd_register(struct io_ring_ctx *ctx, void __user *__arg, | |
10635 | unsigned nr_args) | |
10636 | { | |
10637 | struct io_uring_rsrc_update __user *arg = __arg; | |
10638 | struct io_uring_rsrc_update reg; | |
10639 | struct io_uring_task *tctx; | |
10640 | int ret, i; | |
10641 | ||
10642 | if (!nr_args || nr_args > IO_RINGFD_REG_MAX) | |
10643 | return -EINVAL; | |
10644 | ||
10645 | mutex_unlock(&ctx->uring_lock); | |
10646 | ret = io_uring_add_tctx_node(ctx); | |
10647 | mutex_lock(&ctx->uring_lock); | |
10648 | if (ret) | |
10649 | return ret; | |
10650 | ||
10651 | tctx = current->io_uring; | |
10652 | for (i = 0; i < nr_args; i++) { | |
10653 | int start, end; | |
10654 | ||
10655 | if (copy_from_user(®, &arg[i], sizeof(reg))) { | |
10656 | ret = -EFAULT; | |
10657 | break; | |
10658 | } | |
10659 | ||
10660 | if (reg.offset == -1U) { | |
10661 | start = 0; | |
10662 | end = IO_RINGFD_REG_MAX; | |
10663 | } else { | |
10664 | if (reg.offset >= IO_RINGFD_REG_MAX) { | |
10665 | ret = -EINVAL; | |
10666 | break; | |
10667 | } | |
10668 | start = reg.offset; | |
10669 | end = start + 1; | |
10670 | } | |
10671 | ||
10672 | ret = io_ring_add_registered_fd(tctx, reg.data, start, end); | |
10673 | if (ret < 0) | |
10674 | break; | |
10675 | ||
10676 | reg.offset = ret; | |
10677 | if (copy_to_user(&arg[i], ®, sizeof(reg))) { | |
10678 | fput(tctx->registered_rings[reg.offset]); | |
10679 | tctx->registered_rings[reg.offset] = NULL; | |
10680 | ret = -EFAULT; | |
10681 | break; | |
10682 | } | |
10683 | } | |
10684 | ||
10685 | return i ? i : ret; | |
10686 | } | |
10687 | ||
10688 | static int io_ringfd_unregister(struct io_ring_ctx *ctx, void __user *__arg, | |
10689 | unsigned nr_args) | |
10690 | { | |
10691 | struct io_uring_rsrc_update __user *arg = __arg; | |
10692 | struct io_uring_task *tctx = current->io_uring; | |
10693 | struct io_uring_rsrc_update reg; | |
10694 | int ret = 0, i; | |
10695 | ||
10696 | if (!nr_args || nr_args > IO_RINGFD_REG_MAX) | |
10697 | return -EINVAL; | |
10698 | if (!tctx) | |
10699 | return 0; | |
10700 | ||
10701 | for (i = 0; i < nr_args; i++) { | |
10702 | if (copy_from_user(®, &arg[i], sizeof(reg))) { | |
10703 | ret = -EFAULT; | |
10704 | break; | |
10705 | } | |
10706 | if (reg.offset >= IO_RINGFD_REG_MAX) { | |
10707 | ret = -EINVAL; | |
10708 | break; | |
10709 | } | |
10710 | ||
10711 | reg.offset = array_index_nospec(reg.offset, IO_RINGFD_REG_MAX); | |
10712 | if (tctx->registered_rings[reg.offset]) { | |
10713 | fput(tctx->registered_rings[reg.offset]); | |
10714 | tctx->registered_rings[reg.offset] = NULL; | |
10715 | } | |
10716 | } | |
10717 | ||
10718 | return i ? i : ret; | |
10719 | } | |
10720 | ||
6c5c240e RP |
10721 | static void *io_uring_validate_mmap_request(struct file *file, |
10722 | loff_t pgoff, size_t sz) | |
2b188cc1 | 10723 | { |
2b188cc1 | 10724 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 10725 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
10726 | struct page *page; |
10727 | void *ptr; | |
10728 | ||
10729 | switch (offset) { | |
10730 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
10731 | case IORING_OFF_CQ_RING: |
10732 | ptr = ctx->rings; | |
2b188cc1 JA |
10733 | break; |
10734 | case IORING_OFF_SQES: | |
10735 | ptr = ctx->sq_sqes; | |
10736 | break; | |
2b188cc1 | 10737 | default: |
6c5c240e | 10738 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
10739 | } |
10740 | ||
10741 | page = virt_to_head_page(ptr); | |
a50b854e | 10742 | if (sz > page_size(page)) |
6c5c240e RP |
10743 | return ERR_PTR(-EINVAL); |
10744 | ||
10745 | return ptr; | |
10746 | } | |
10747 | ||
10748 | #ifdef CONFIG_MMU | |
10749 | ||
c072481d | 10750 | static __cold int io_uring_mmap(struct file *file, struct vm_area_struct *vma) |
6c5c240e RP |
10751 | { |
10752 | size_t sz = vma->vm_end - vma->vm_start; | |
10753 | unsigned long pfn; | |
10754 | void *ptr; | |
10755 | ||
10756 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
10757 | if (IS_ERR(ptr)) | |
10758 | return PTR_ERR(ptr); | |
2b188cc1 JA |
10759 | |
10760 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
10761 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
10762 | } | |
10763 | ||
6c5c240e RP |
10764 | #else /* !CONFIG_MMU */ |
10765 | ||
10766 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
10767 | { | |
10768 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
10769 | } | |
10770 | ||
10771 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
10772 | { | |
10773 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
10774 | } | |
10775 | ||
10776 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
10777 | unsigned long addr, unsigned long len, | |
10778 | unsigned long pgoff, unsigned long flags) | |
10779 | { | |
10780 | void *ptr; | |
10781 | ||
10782 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
10783 | if (IS_ERR(ptr)) | |
10784 | return PTR_ERR(ptr); | |
10785 | ||
10786 | return (unsigned long) ptr; | |
10787 | } | |
10788 | ||
10789 | #endif /* !CONFIG_MMU */ | |
10790 | ||
d9d05217 | 10791 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
10792 | { |
10793 | DEFINE_WAIT(wait); | |
10794 | ||
10795 | do { | |
10796 | if (!io_sqring_full(ctx)) | |
10797 | break; | |
90554200 JA |
10798 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
10799 | ||
10800 | if (!io_sqring_full(ctx)) | |
10801 | break; | |
90554200 JA |
10802 | schedule(); |
10803 | } while (!signal_pending(current)); | |
10804 | ||
10805 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 10806 | return 0; |
90554200 JA |
10807 | } |
10808 | ||
c73ebb68 HX |
10809 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
10810 | struct __kernel_timespec __user **ts, | |
10811 | const sigset_t __user **sig) | |
10812 | { | |
10813 | struct io_uring_getevents_arg arg; | |
10814 | ||
10815 | /* | |
10816 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
10817 | * is just a pointer to the sigset_t. | |
10818 | */ | |
10819 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
10820 | *sig = (const sigset_t __user *) argp; | |
10821 | *ts = NULL; | |
10822 | return 0; | |
10823 | } | |
10824 | ||
10825 | /* | |
10826 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
10827 | * timespec and sigset_t pointers if good. | |
10828 | */ | |
10829 | if (*argsz != sizeof(arg)) | |
10830 | return -EINVAL; | |
10831 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
10832 | return -EFAULT; | |
10833 | *sig = u64_to_user_ptr(arg.sigmask); | |
10834 | *argsz = arg.sigmask_sz; | |
10835 | *ts = u64_to_user_ptr(arg.ts); | |
10836 | return 0; | |
10837 | } | |
10838 | ||
2b188cc1 | 10839 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
10840 | u32, min_complete, u32, flags, const void __user *, argp, |
10841 | size_t, argsz) | |
2b188cc1 JA |
10842 | { |
10843 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
10844 | int submitted = 0; |
10845 | struct fd f; | |
33f993da | 10846 | long ret; |
2b188cc1 | 10847 | |
4c6e277c | 10848 | io_run_task_work(); |
b41e9852 | 10849 | |
33f993da | 10850 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
e7a6c00d JA |
10851 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG | |
10852 | IORING_ENTER_REGISTERED_RING))) | |
2b188cc1 JA |
10853 | return -EINVAL; |
10854 | ||
e7a6c00d JA |
10855 | /* |
10856 | * Ring fd has been registered via IORING_REGISTER_RING_FDS, we | |
10857 | * need only dereference our task private array to find it. | |
10858 | */ | |
10859 | if (flags & IORING_ENTER_REGISTERED_RING) { | |
10860 | struct io_uring_task *tctx = current->io_uring; | |
10861 | ||
10862 | if (!tctx || fd >= IO_RINGFD_REG_MAX) | |
10863 | return -EINVAL; | |
10864 | fd = array_index_nospec(fd, IO_RINGFD_REG_MAX); | |
10865 | f.file = tctx->registered_rings[fd]; | |
10866 | if (unlikely(!f.file)) | |
10867 | return -EBADF; | |
10868 | } else { | |
10869 | f = fdget(fd); | |
10870 | if (unlikely(!f.file)) | |
10871 | return -EBADF; | |
10872 | } | |
2b188cc1 JA |
10873 | |
10874 | ret = -EOPNOTSUPP; | |
33f993da | 10875 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
10876 | goto out_fput; |
10877 | ||
10878 | ret = -ENXIO; | |
10879 | ctx = f.file->private_data; | |
33f993da | 10880 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
10881 | goto out_fput; |
10882 | ||
7e84e1c7 | 10883 | ret = -EBADFD; |
33f993da | 10884 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
10885 | goto out; |
10886 | ||
6c271ce2 JA |
10887 | /* |
10888 | * For SQ polling, the thread will do all submissions and completions. | |
10889 | * Just return the requested submit count, and wake the thread if | |
10890 | * we were asked to. | |
10891 | */ | |
b2a9eada | 10892 | ret = 0; |
6c271ce2 | 10893 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
90f67366 | 10894 | io_cqring_overflow_flush(ctx); |
89448c47 | 10895 | |
21f96522 JA |
10896 | if (unlikely(ctx->sq_data->thread == NULL)) { |
10897 | ret = -EOWNERDEAD; | |
04147488 | 10898 | goto out; |
21f96522 | 10899 | } |
6c271ce2 | 10900 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 10901 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
10902 | if (flags & IORING_ENTER_SQ_WAIT) { |
10903 | ret = io_sqpoll_wait_sq(ctx); | |
10904 | if (ret) | |
10905 | goto out; | |
10906 | } | |
6c271ce2 | 10907 | submitted = to_submit; |
b2a9eada | 10908 | } else if (to_submit) { |
eef51daa | 10909 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
10910 | if (unlikely(ret)) |
10911 | goto out; | |
2b188cc1 | 10912 | mutex_lock(&ctx->uring_lock); |
0f212204 | 10913 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 10914 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
10915 | |
10916 | if (submitted != to_submit) | |
10917 | goto out; | |
2b188cc1 JA |
10918 | } |
10919 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
10920 | const sigset_t __user *sig; |
10921 | struct __kernel_timespec __user *ts; | |
10922 | ||
10923 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
10924 | if (unlikely(ret)) | |
10925 | goto out; | |
10926 | ||
2b188cc1 JA |
10927 | min_complete = min(min_complete, ctx->cq_entries); |
10928 | ||
32b2244a XW |
10929 | /* |
10930 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
10931 | * space applications don't need to do io completion events | |
10932 | * polling again, they can rely on io_sq_thread to do polling | |
10933 | * work, which can reduce cpu usage and uring_lock contention. | |
10934 | */ | |
10935 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
10936 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 10937 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 10938 | } else { |
c73ebb68 | 10939 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 10940 | } |
2b188cc1 JA |
10941 | } |
10942 | ||
7c504e65 | 10943 | out: |
6805b32e | 10944 | percpu_ref_put(&ctx->refs); |
2b188cc1 | 10945 | out_fput: |
e7a6c00d JA |
10946 | if (!(flags & IORING_ENTER_REGISTERED_RING)) |
10947 | fdput(f); | |
2b188cc1 JA |
10948 | return submitted ? submitted : ret; |
10949 | } | |
10950 | ||
bebdb65e | 10951 | #ifdef CONFIG_PROC_FS |
c072481d | 10952 | static __cold int io_uring_show_cred(struct seq_file *m, unsigned int id, |
61cf9370 | 10953 | const struct cred *cred) |
87ce955b | 10954 | { |
87ce955b JA |
10955 | struct user_namespace *uns = seq_user_ns(m); |
10956 | struct group_info *gi; | |
10957 | kernel_cap_t cap; | |
10958 | unsigned __capi; | |
10959 | int g; | |
10960 | ||
10961 | seq_printf(m, "%5d\n", id); | |
10962 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
10963 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
10964 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
10965 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
10966 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
10967 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
10968 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
10969 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
10970 | seq_puts(m, "\n\tGroups:\t"); | |
10971 | gi = cred->group_info; | |
10972 | for (g = 0; g < gi->ngroups; g++) { | |
10973 | seq_put_decimal_ull(m, g ? " " : "", | |
10974 | from_kgid_munged(uns, gi->gid[g])); | |
10975 | } | |
10976 | seq_puts(m, "\n\tCapEff:\t"); | |
10977 | cap = cred->cap_effective; | |
10978 | CAP_FOR_EACH_U32(__capi) | |
10979 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
10980 | seq_putc(m, '\n'); | |
10981 | return 0; | |
10982 | } | |
10983 | ||
c072481d PB |
10984 | static __cold void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, |
10985 | struct seq_file *m) | |
87ce955b | 10986 | { |
dbbe9c64 | 10987 | struct io_sq_data *sq = NULL; |
83f84356 HX |
10988 | struct io_overflow_cqe *ocqe; |
10989 | struct io_rings *r = ctx->rings; | |
10990 | unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1; | |
83f84356 HX |
10991 | unsigned int sq_head = READ_ONCE(r->sq.head); |
10992 | unsigned int sq_tail = READ_ONCE(r->sq.tail); | |
10993 | unsigned int cq_head = READ_ONCE(r->cq.head); | |
10994 | unsigned int cq_tail = READ_ONCE(r->cq.tail); | |
f75d1183 | 10995 | unsigned int sq_entries, cq_entries; |
fad8e0de | 10996 | bool has_lock; |
83f84356 HX |
10997 | unsigned int i; |
10998 | ||
10999 | /* | |
11000 | * we may get imprecise sqe and cqe info if uring is actively running | |
11001 | * since we get cached_sq_head and cached_cq_tail without uring_lock | |
11002 | * and sq_tail and cq_head are changed by userspace. But it's ok since | |
11003 | * we usually use these info when it is stuck. | |
11004 | */ | |
c0235652 | 11005 | seq_printf(m, "SqMask:\t0x%x\n", sq_mask); |
f75d1183 JA |
11006 | seq_printf(m, "SqHead:\t%u\n", sq_head); |
11007 | seq_printf(m, "SqTail:\t%u\n", sq_tail); | |
11008 | seq_printf(m, "CachedSqHead:\t%u\n", ctx->cached_sq_head); | |
11009 | seq_printf(m, "CqMask:\t0x%x\n", cq_mask); | |
11010 | seq_printf(m, "CqHead:\t%u\n", cq_head); | |
11011 | seq_printf(m, "CqTail:\t%u\n", cq_tail); | |
11012 | seq_printf(m, "CachedCqTail:\t%u\n", ctx->cached_cq_tail); | |
11013 | seq_printf(m, "SQEs:\t%u\n", sq_tail - ctx->cached_sq_head); | |
11014 | sq_entries = min(sq_tail - sq_head, ctx->sq_entries); | |
11015 | for (i = 0; i < sq_entries; i++) { | |
11016 | unsigned int entry = i + sq_head; | |
11017 | unsigned int sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]); | |
a1957780 | 11018 | struct io_uring_sqe *sqe; |
f75d1183 JA |
11019 | |
11020 | if (sq_idx > sq_mask) | |
11021 | continue; | |
11022 | sqe = &ctx->sq_sqes[sq_idx]; | |
11023 | seq_printf(m, "%5u: opcode:%d, fd:%d, flags:%x, user_data:%llu\n", | |
11024 | sq_idx, sqe->opcode, sqe->fd, sqe->flags, | |
11025 | sqe->user_data); | |
83f84356 | 11026 | } |
f75d1183 JA |
11027 | seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head); |
11028 | cq_entries = min(cq_tail - cq_head, ctx->cq_entries); | |
11029 | for (i = 0; i < cq_entries; i++) { | |
11030 | unsigned int entry = i + cq_head; | |
11031 | struct io_uring_cqe *cqe = &r->cqes[entry & cq_mask]; | |
83f84356 HX |
11032 | |
11033 | seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x\n", | |
f75d1183 JA |
11034 | entry & cq_mask, cqe->user_data, cqe->res, |
11035 | cqe->flags); | |
83f84356 | 11036 | } |
87ce955b | 11037 | |
fad8e0de JA |
11038 | /* |
11039 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
11040 | * since fdinfo case grabs it in the opposite direction of normal use | |
11041 | * cases. If we fail to get the lock, we just don't iterate any | |
11042 | * structures that could be going away outside the io_uring mutex. | |
11043 | */ | |
11044 | has_lock = mutex_trylock(&ctx->uring_lock); | |
11045 | ||
5f3f26f9 | 11046 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 11047 | sq = ctx->sq_data; |
5f3f26f9 JA |
11048 | if (!sq->thread) |
11049 | sq = NULL; | |
11050 | } | |
dbbe9c64 JQ |
11051 | |
11052 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
11053 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 11054 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 11055 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 11056 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 11057 | |
87ce955b JA |
11058 | if (f) |
11059 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
11060 | else | |
11061 | seq_printf(m, "%5u: <none>\n", i); | |
11062 | } | |
11063 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 11064 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 11065 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 11066 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 11067 | |
4751f53d | 11068 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 11069 | } |
61cf9370 MWO |
11070 | if (has_lock && !xa_empty(&ctx->personalities)) { |
11071 | unsigned long index; | |
11072 | const struct cred *cred; | |
11073 | ||
87ce955b | 11074 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
11075 | xa_for_each(&ctx->personalities, index, cred) |
11076 | io_uring_show_cred(m, index, cred); | |
87ce955b | 11077 | } |
83f84356 HX |
11078 | if (has_lock) |
11079 | mutex_unlock(&ctx->uring_lock); | |
11080 | ||
11081 | seq_puts(m, "PollList:\n"); | |
79ebeaee | 11082 | spin_lock(&ctx->completion_lock); |
d7718a9d JA |
11083 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
11084 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
11085 | struct io_kiocb *req; | |
11086 | ||
11087 | hlist_for_each_entry(req, list, hash_node) | |
11088 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
11089 | req->task->task_works != NULL); | |
11090 | } | |
83f84356 HX |
11091 | |
11092 | seq_puts(m, "CqOverflowList:\n"); | |
11093 | list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) { | |
11094 | struct io_uring_cqe *cqe = &ocqe->cqe; | |
11095 | ||
11096 | seq_printf(m, " user_data=%llu, res=%d, flags=%x\n", | |
11097 | cqe->user_data, cqe->res, cqe->flags); | |
11098 | ||
11099 | } | |
11100 | ||
79ebeaee | 11101 | spin_unlock(&ctx->completion_lock); |
87ce955b JA |
11102 | } |
11103 | ||
c072481d | 11104 | static __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *f) |
87ce955b JA |
11105 | { |
11106 | struct io_ring_ctx *ctx = f->private_data; | |
11107 | ||
11108 | if (percpu_ref_tryget(&ctx->refs)) { | |
11109 | __io_uring_show_fdinfo(ctx, m); | |
11110 | percpu_ref_put(&ctx->refs); | |
11111 | } | |
11112 | } | |
bebdb65e | 11113 | #endif |
87ce955b | 11114 | |
2b188cc1 JA |
11115 | static const struct file_operations io_uring_fops = { |
11116 | .release = io_uring_release, | |
11117 | .mmap = io_uring_mmap, | |
6c5c240e RP |
11118 | #ifndef CONFIG_MMU |
11119 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
11120 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
11121 | #endif | |
2b188cc1 | 11122 | .poll = io_uring_poll, |
bebdb65e | 11123 | #ifdef CONFIG_PROC_FS |
87ce955b | 11124 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 11125 | #endif |
2b188cc1 JA |
11126 | }; |
11127 | ||
c072481d PB |
11128 | static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx, |
11129 | struct io_uring_params *p) | |
2b188cc1 | 11130 | { |
75b28aff HV |
11131 | struct io_rings *rings; |
11132 | size_t size, sq_array_offset; | |
2b188cc1 | 11133 | |
bd740481 JA |
11134 | /* make sure these are sane, as we already accounted them */ |
11135 | ctx->sq_entries = p->sq_entries; | |
11136 | ctx->cq_entries = p->cq_entries; | |
11137 | ||
75b28aff HV |
11138 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
11139 | if (size == SIZE_MAX) | |
11140 | return -EOVERFLOW; | |
11141 | ||
11142 | rings = io_mem_alloc(size); | |
11143 | if (!rings) | |
2b188cc1 JA |
11144 | return -ENOMEM; |
11145 | ||
75b28aff HV |
11146 | ctx->rings = rings; |
11147 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
11148 | rings->sq_ring_mask = p->sq_entries - 1; | |
11149 | rings->cq_ring_mask = p->cq_entries - 1; | |
11150 | rings->sq_ring_entries = p->sq_entries; | |
11151 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 JA |
11152 | |
11153 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
11154 | if (size == SIZE_MAX) { |
11155 | io_mem_free(ctx->rings); | |
11156 | ctx->rings = NULL; | |
2b188cc1 | 11157 | return -EOVERFLOW; |
eb065d30 | 11158 | } |
2b188cc1 JA |
11159 | |
11160 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
11161 | if (!ctx->sq_sqes) { |
11162 | io_mem_free(ctx->rings); | |
11163 | ctx->rings = NULL; | |
2b188cc1 | 11164 | return -ENOMEM; |
eb065d30 | 11165 | } |
2b188cc1 | 11166 | |
2b188cc1 JA |
11167 | return 0; |
11168 | } | |
11169 | ||
9faadcc8 PB |
11170 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
11171 | { | |
11172 | int ret, fd; | |
11173 | ||
11174 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
11175 | if (fd < 0) | |
11176 | return fd; | |
11177 | ||
eef51daa | 11178 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
11179 | if (ret) { |
11180 | put_unused_fd(fd); | |
11181 | return ret; | |
11182 | } | |
11183 | fd_install(fd, file); | |
11184 | return fd; | |
11185 | } | |
11186 | ||
2b188cc1 JA |
11187 | /* |
11188 | * Allocate an anonymous fd, this is what constitutes the application | |
11189 | * visible backing of an io_uring instance. The application mmaps this | |
11190 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
11191 | * we have to tie this fd to a socket for file garbage collection purposes. | |
11192 | */ | |
9faadcc8 | 11193 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
11194 | { |
11195 | struct file *file; | |
9faadcc8 | 11196 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
11197 | int ret; |
11198 | ||
2b188cc1 JA |
11199 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
11200 | &ctx->ring_sock); | |
11201 | if (ret) | |
9faadcc8 | 11202 | return ERR_PTR(ret); |
2b188cc1 JA |
11203 | #endif |
11204 | ||
91a9ab7c PM |
11205 | file = anon_inode_getfile_secure("[io_uring]", &io_uring_fops, ctx, |
11206 | O_RDWR | O_CLOEXEC, NULL); | |
2b188cc1 | 11207 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
11208 | if (IS_ERR(file)) { |
11209 | sock_release(ctx->ring_sock); | |
11210 | ctx->ring_sock = NULL; | |
11211 | } else { | |
11212 | ctx->ring_sock->file = file; | |
0f212204 | 11213 | } |
2b188cc1 | 11214 | #endif |
9faadcc8 | 11215 | return file; |
2b188cc1 JA |
11216 | } |
11217 | ||
c072481d PB |
11218 | static __cold int io_uring_create(unsigned entries, struct io_uring_params *p, |
11219 | struct io_uring_params __user *params) | |
2b188cc1 | 11220 | { |
2b188cc1 | 11221 | struct io_ring_ctx *ctx; |
9faadcc8 | 11222 | struct file *file; |
2b188cc1 JA |
11223 | int ret; |
11224 | ||
8110c1a6 | 11225 | if (!entries) |
2b188cc1 | 11226 | return -EINVAL; |
8110c1a6 JA |
11227 | if (entries > IORING_MAX_ENTRIES) { |
11228 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
11229 | return -EINVAL; | |
11230 | entries = IORING_MAX_ENTRIES; | |
11231 | } | |
2b188cc1 JA |
11232 | |
11233 | /* | |
11234 | * Use twice as many entries for the CQ ring. It's possible for the | |
11235 | * application to drive a higher depth than the size of the SQ ring, | |
11236 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
11237 | * some flexibility in overcommitting a bit. If the application has |
11238 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
11239 | * of CQ ring entries manually. | |
2b188cc1 JA |
11240 | */ |
11241 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
11242 | if (p->flags & IORING_SETUP_CQSIZE) { |
11243 | /* | |
11244 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
11245 | * to a power-of-two, if it isn't already. We do NOT impose | |
11246 | * any cq vs sq ring sizing. | |
11247 | */ | |
eb2667b3 | 11248 | if (!p->cq_entries) |
33a107f0 | 11249 | return -EINVAL; |
8110c1a6 JA |
11250 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
11251 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
11252 | return -EINVAL; | |
11253 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
11254 | } | |
eb2667b3 JQ |
11255 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
11256 | if (p->cq_entries < p->sq_entries) | |
11257 | return -EINVAL; | |
33a107f0 JA |
11258 | } else { |
11259 | p->cq_entries = 2 * p->sq_entries; | |
11260 | } | |
2b188cc1 | 11261 | |
2b188cc1 | 11262 | ctx = io_ring_ctx_alloc(p); |
62e398be | 11263 | if (!ctx) |
2b188cc1 | 11264 | return -ENOMEM; |
2b188cc1 | 11265 | ctx->compat = in_compat_syscall(); |
62e398be JA |
11266 | if (!capable(CAP_IPC_LOCK)) |
11267 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
11268 | |
11269 | /* | |
11270 | * This is just grabbed for accounting purposes. When a process exits, | |
11271 | * the mm is exited and dropped before the files, hence we need to hang | |
11272 | * on to this mm purely for the purposes of being able to unaccount | |
11273 | * memory (locked/pinned vm). It's not used for anything else. | |
11274 | */ | |
6b7898eb | 11275 | mmgrab(current->mm); |
2aede0e4 | 11276 | ctx->mm_account = current->mm; |
6b7898eb | 11277 | |
2b188cc1 JA |
11278 | ret = io_allocate_scq_urings(ctx, p); |
11279 | if (ret) | |
11280 | goto err; | |
11281 | ||
7e84e1c7 | 11282 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
11283 | if (ret) |
11284 | goto err; | |
eae071c9 | 11285 | /* always set a rsrc node */ |
47b228ce PB |
11286 | ret = io_rsrc_node_switch_start(ctx); |
11287 | if (ret) | |
11288 | goto err; | |
eae071c9 | 11289 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 11290 | |
2b188cc1 | 11291 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
11292 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
11293 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
11294 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
11295 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
11296 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
11297 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
11298 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
11299 | |
11300 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
11301 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
11302 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
11303 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
11304 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
11305 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
11306 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 11307 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 11308 | |
7f13657d XW |
11309 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
11310 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 11311 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 11312 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e | 11313 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
04c76b41 | 11314 | IORING_FEAT_RSRC_TAGS | IORING_FEAT_CQE_SKIP; |
7f13657d XW |
11315 | |
11316 | if (copy_to_user(params, p, sizeof(*p))) { | |
11317 | ret = -EFAULT; | |
11318 | goto err; | |
11319 | } | |
d1719f70 | 11320 | |
9faadcc8 PB |
11321 | file = io_uring_get_file(ctx); |
11322 | if (IS_ERR(file)) { | |
11323 | ret = PTR_ERR(file); | |
11324 | goto err; | |
11325 | } | |
11326 | ||
044c1ab3 JA |
11327 | /* |
11328 | * Install ring fd as the very last thing, so we don't risk someone | |
11329 | * having closed it before we finish setup | |
11330 | */ | |
9faadcc8 PB |
11331 | ret = io_uring_install_fd(ctx, file); |
11332 | if (ret < 0) { | |
11333 | /* fput will clean it up */ | |
11334 | fput(file); | |
11335 | return ret; | |
11336 | } | |
044c1ab3 | 11337 | |
c826bd7a | 11338 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
11339 | return ret; |
11340 | err: | |
11341 | io_ring_ctx_wait_and_kill(ctx); | |
11342 | return ret; | |
11343 | } | |
11344 | ||
11345 | /* | |
11346 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
11347 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
11348 | * params structure passed in. | |
11349 | */ | |
11350 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
11351 | { | |
11352 | struct io_uring_params p; | |
2b188cc1 JA |
11353 | int i; |
11354 | ||
11355 | if (copy_from_user(&p, params, sizeof(p))) | |
11356 | return -EFAULT; | |
11357 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
11358 | if (p.resv[i]) | |
11359 | return -EINVAL; | |
11360 | } | |
11361 | ||
6c271ce2 | 11362 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 11363 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 | 11364 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
bcbb7bf6 | 11365 | IORING_SETUP_R_DISABLED | IORING_SETUP_SUBMIT_ALL)) |
2b188cc1 JA |
11366 | return -EINVAL; |
11367 | ||
7f13657d | 11368 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
11369 | } |
11370 | ||
11371 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
11372 | struct io_uring_params __user *, params) | |
11373 | { | |
11374 | return io_uring_setup(entries, params); | |
11375 | } | |
11376 | ||
c072481d PB |
11377 | static __cold int io_probe(struct io_ring_ctx *ctx, void __user *arg, |
11378 | unsigned nr_args) | |
66f4af93 JA |
11379 | { |
11380 | struct io_uring_probe *p; | |
11381 | size_t size; | |
11382 | int i, ret; | |
11383 | ||
11384 | size = struct_size(p, ops, nr_args); | |
11385 | if (size == SIZE_MAX) | |
11386 | return -EOVERFLOW; | |
11387 | p = kzalloc(size, GFP_KERNEL); | |
11388 | if (!p) | |
11389 | return -ENOMEM; | |
11390 | ||
11391 | ret = -EFAULT; | |
11392 | if (copy_from_user(p, arg, size)) | |
11393 | goto out; | |
11394 | ret = -EINVAL; | |
11395 | if (memchr_inv(p, 0, size)) | |
11396 | goto out; | |
11397 | ||
11398 | p->last_op = IORING_OP_LAST - 1; | |
11399 | if (nr_args > IORING_OP_LAST) | |
11400 | nr_args = IORING_OP_LAST; | |
11401 | ||
11402 | for (i = 0; i < nr_args; i++) { | |
11403 | p->ops[i].op = i; | |
11404 | if (!io_op_defs[i].not_supported) | |
11405 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
11406 | } | |
11407 | p->ops_len = i; | |
11408 | ||
11409 | ret = 0; | |
11410 | if (copy_to_user(arg, p, size)) | |
11411 | ret = -EFAULT; | |
11412 | out: | |
11413 | kfree(p); | |
11414 | return ret; | |
11415 | } | |
11416 | ||
071698e1 JA |
11417 | static int io_register_personality(struct io_ring_ctx *ctx) |
11418 | { | |
4379bf8b | 11419 | const struct cred *creds; |
61cf9370 | 11420 | u32 id; |
1e6fa521 | 11421 | int ret; |
071698e1 | 11422 | |
4379bf8b | 11423 | creds = get_current_cred(); |
1e6fa521 | 11424 | |
61cf9370 MWO |
11425 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
11426 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
a30f895a JA |
11427 | if (ret < 0) { |
11428 | put_cred(creds); | |
11429 | return ret; | |
11430 | } | |
11431 | return id; | |
071698e1 JA |
11432 | } |
11433 | ||
c072481d PB |
11434 | static __cold int io_register_restrictions(struct io_ring_ctx *ctx, |
11435 | void __user *arg, unsigned int nr_args) | |
21b55dbc SG |
11436 | { |
11437 | struct io_uring_restriction *res; | |
11438 | size_t size; | |
11439 | int i, ret; | |
11440 | ||
7e84e1c7 SG |
11441 | /* Restrictions allowed only if rings started disabled */ |
11442 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
11443 | return -EBADFD; | |
11444 | ||
21b55dbc | 11445 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 11446 | if (ctx->restrictions.registered) |
21b55dbc SG |
11447 | return -EBUSY; |
11448 | ||
11449 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
11450 | return -EINVAL; | |
11451 | ||
11452 | size = array_size(nr_args, sizeof(*res)); | |
11453 | if (size == SIZE_MAX) | |
11454 | return -EOVERFLOW; | |
11455 | ||
11456 | res = memdup_user(arg, size); | |
11457 | if (IS_ERR(res)) | |
11458 | return PTR_ERR(res); | |
11459 | ||
11460 | ret = 0; | |
11461 | ||
11462 | for (i = 0; i < nr_args; i++) { | |
11463 | switch (res[i].opcode) { | |
11464 | case IORING_RESTRICTION_REGISTER_OP: | |
11465 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
11466 | ret = -EINVAL; | |
11467 | goto out; | |
11468 | } | |
11469 | ||
11470 | __set_bit(res[i].register_op, | |
11471 | ctx->restrictions.register_op); | |
11472 | break; | |
11473 | case IORING_RESTRICTION_SQE_OP: | |
11474 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
11475 | ret = -EINVAL; | |
11476 | goto out; | |
11477 | } | |
11478 | ||
11479 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
11480 | break; | |
11481 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
11482 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
11483 | break; | |
11484 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
11485 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
11486 | break; | |
11487 | default: | |
11488 | ret = -EINVAL; | |
11489 | goto out; | |
11490 | } | |
11491 | } | |
11492 | ||
11493 | out: | |
11494 | /* Reset all restrictions if an error happened */ | |
11495 | if (ret != 0) | |
11496 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
11497 | else | |
7e84e1c7 | 11498 | ctx->restrictions.registered = true; |
21b55dbc SG |
11499 | |
11500 | kfree(res); | |
11501 | return ret; | |
11502 | } | |
11503 | ||
7e84e1c7 SG |
11504 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
11505 | { | |
11506 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
11507 | return -EBADFD; | |
11508 | ||
11509 | if (ctx->restrictions.registered) | |
11510 | ctx->restricted = 1; | |
11511 | ||
0298ef96 PB |
11512 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
11513 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
11514 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
11515 | return 0; |
11516 | } | |
11517 | ||
fdecb662 | 11518 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 11519 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
11520 | unsigned nr_args) |
11521 | { | |
11522 | __u32 tmp; | |
11523 | int err; | |
11524 | ||
c3bdad02 PB |
11525 | if (up->resv) |
11526 | return -EINVAL; | |
98f0b3b4 PB |
11527 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
11528 | return -EOVERFLOW; | |
11529 | err = io_rsrc_node_switch_start(ctx); | |
11530 | if (err) | |
11531 | return err; | |
11532 | ||
fdecb662 PB |
11533 | switch (type) { |
11534 | case IORING_RSRC_FILE: | |
98f0b3b4 | 11535 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
11536 | case IORING_RSRC_BUFFER: |
11537 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
11538 | } |
11539 | return -EINVAL; | |
11540 | } | |
11541 | ||
c3bdad02 PB |
11542 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
11543 | unsigned nr_args) | |
98f0b3b4 | 11544 | { |
c3bdad02 | 11545 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
11546 | |
11547 | if (!nr_args) | |
11548 | return -EINVAL; | |
c3bdad02 PB |
11549 | memset(&up, 0, sizeof(up)); |
11550 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
11551 | return -EFAULT; | |
11552 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
11553 | } | |
11554 | ||
11555 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 11556 | unsigned size, unsigned type) |
c3bdad02 PB |
11557 | { |
11558 | struct io_uring_rsrc_update2 up; | |
11559 | ||
11560 | if (size != sizeof(up)) | |
11561 | return -EINVAL; | |
98f0b3b4 PB |
11562 | if (copy_from_user(&up, arg, sizeof(up))) |
11563 | return -EFAULT; | |
992da01a | 11564 | if (!up.nr || up.resv) |
98f0b3b4 | 11565 | return -EINVAL; |
992da01a | 11566 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
11567 | } |
11568 | ||
c072481d | 11569 | static __cold int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 11570 | unsigned int size, unsigned int type) |
792e3582 PB |
11571 | { |
11572 | struct io_uring_rsrc_register rr; | |
11573 | ||
11574 | /* keep it extendible */ | |
11575 | if (size != sizeof(rr)) | |
11576 | return -EINVAL; | |
11577 | ||
11578 | memset(&rr, 0, sizeof(rr)); | |
11579 | if (copy_from_user(&rr, arg, size)) | |
11580 | return -EFAULT; | |
992da01a | 11581 | if (!rr.nr || rr.resv || rr.resv2) |
792e3582 PB |
11582 | return -EINVAL; |
11583 | ||
992da01a | 11584 | switch (type) { |
792e3582 PB |
11585 | case IORING_RSRC_FILE: |
11586 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
11587 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df PB |
11588 | case IORING_RSRC_BUFFER: |
11589 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), | |
11590 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
11591 | } |
11592 | return -EINVAL; | |
11593 | } | |
11594 | ||
c072481d PB |
11595 | static __cold int io_register_iowq_aff(struct io_ring_ctx *ctx, |
11596 | void __user *arg, unsigned len) | |
fe76421d JA |
11597 | { |
11598 | struct io_uring_task *tctx = current->io_uring; | |
11599 | cpumask_var_t new_mask; | |
11600 | int ret; | |
11601 | ||
11602 | if (!tctx || !tctx->io_wq) | |
11603 | return -EINVAL; | |
11604 | ||
11605 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
11606 | return -ENOMEM; | |
11607 | ||
11608 | cpumask_clear(new_mask); | |
11609 | if (len > cpumask_size()) | |
11610 | len = cpumask_size(); | |
11611 | ||
11612 | if (copy_from_user(new_mask, arg, len)) { | |
11613 | free_cpumask_var(new_mask); | |
11614 | return -EFAULT; | |
11615 | } | |
11616 | ||
11617 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
11618 | free_cpumask_var(new_mask); | |
11619 | return ret; | |
11620 | } | |
11621 | ||
c072481d | 11622 | static __cold int io_unregister_iowq_aff(struct io_ring_ctx *ctx) |
fe76421d JA |
11623 | { |
11624 | struct io_uring_task *tctx = current->io_uring; | |
11625 | ||
11626 | if (!tctx || !tctx->io_wq) | |
11627 | return -EINVAL; | |
11628 | ||
11629 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
11630 | } | |
11631 | ||
c072481d PB |
11632 | static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, |
11633 | void __user *arg) | |
b22fa62a | 11634 | __must_hold(&ctx->uring_lock) |
2e480058 | 11635 | { |
b22fa62a | 11636 | struct io_tctx_node *node; |
fa84693b JA |
11637 | struct io_uring_task *tctx = NULL; |
11638 | struct io_sq_data *sqd = NULL; | |
2e480058 JA |
11639 | __u32 new_count[2]; |
11640 | int i, ret; | |
11641 | ||
2e480058 JA |
11642 | if (copy_from_user(new_count, arg, sizeof(new_count))) |
11643 | return -EFAULT; | |
11644 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
11645 | if (new_count[i] > INT_MAX) | |
11646 | return -EINVAL; | |
11647 | ||
fa84693b JA |
11648 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
11649 | sqd = ctx->sq_data; | |
11650 | if (sqd) { | |
009ad9f0 JA |
11651 | /* |
11652 | * Observe the correct sqd->lock -> ctx->uring_lock | |
11653 | * ordering. Fine to drop uring_lock here, we hold | |
11654 | * a ref to the ctx. | |
11655 | */ | |
41d3a6bd | 11656 | refcount_inc(&sqd->refs); |
009ad9f0 | 11657 | mutex_unlock(&ctx->uring_lock); |
fa84693b | 11658 | mutex_lock(&sqd->lock); |
009ad9f0 | 11659 | mutex_lock(&ctx->uring_lock); |
41d3a6bd JA |
11660 | if (sqd->thread) |
11661 | tctx = sqd->thread->io_uring; | |
fa84693b JA |
11662 | } |
11663 | } else { | |
11664 | tctx = current->io_uring; | |
11665 | } | |
11666 | ||
e139a1ec | 11667 | BUILD_BUG_ON(sizeof(new_count) != sizeof(ctx->iowq_limits)); |
fa84693b | 11668 | |
bad119b9 PB |
11669 | for (i = 0; i < ARRAY_SIZE(new_count); i++) |
11670 | if (new_count[i]) | |
11671 | ctx->iowq_limits[i] = new_count[i]; | |
e139a1ec PB |
11672 | ctx->iowq_limits_set = true; |
11673 | ||
e139a1ec PB |
11674 | if (tctx && tctx->io_wq) { |
11675 | ret = io_wq_max_workers(tctx->io_wq, new_count); | |
11676 | if (ret) | |
11677 | goto err; | |
11678 | } else { | |
11679 | memset(new_count, 0, sizeof(new_count)); | |
11680 | } | |
fa84693b | 11681 | |
41d3a6bd | 11682 | if (sqd) { |
fa84693b | 11683 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
11684 | io_put_sq_data(sqd); |
11685 | } | |
2e480058 JA |
11686 | |
11687 | if (copy_to_user(arg, new_count, sizeof(new_count))) | |
11688 | return -EFAULT; | |
11689 | ||
b22fa62a PB |
11690 | /* that's it for SQPOLL, only the SQPOLL task creates requests */ |
11691 | if (sqd) | |
11692 | return 0; | |
11693 | ||
11694 | /* now propagate the restriction to all registered users */ | |
11695 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
11696 | struct io_uring_task *tctx = node->task->io_uring; | |
11697 | ||
11698 | if (WARN_ON_ONCE(!tctx->io_wq)) | |
11699 | continue; | |
11700 | ||
11701 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
11702 | new_count[i] = ctx->iowq_limits[i]; | |
11703 | /* ignore errors, it always returns zero anyway */ | |
11704 | (void)io_wq_max_workers(tctx->io_wq, new_count); | |
11705 | } | |
2e480058 | 11706 | return 0; |
fa84693b | 11707 | err: |
41d3a6bd | 11708 | if (sqd) { |
fa84693b | 11709 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
11710 | io_put_sq_data(sqd); |
11711 | } | |
fa84693b | 11712 | return ret; |
2e480058 JA |
11713 | } |
11714 | ||
edafccee JA |
11715 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
11716 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
11717 | __releases(ctx->uring_lock) |
11718 | __acquires(ctx->uring_lock) | |
edafccee JA |
11719 | { |
11720 | int ret; | |
11721 | ||
35fa71a0 JA |
11722 | /* |
11723 | * We're inside the ring mutex, if the ref is already dying, then | |
11724 | * someone else killed the ctx or is already going through | |
11725 | * io_uring_register(). | |
11726 | */ | |
11727 | if (percpu_ref_is_dying(&ctx->refs)) | |
11728 | return -ENXIO; | |
11729 | ||
75c4021a PB |
11730 | if (ctx->restricted) { |
11731 | if (opcode >= IORING_REGISTER_LAST) | |
11732 | return -EINVAL; | |
11733 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
11734 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
11735 | return -EACCES; | |
11736 | } | |
11737 | ||
edafccee JA |
11738 | switch (opcode) { |
11739 | case IORING_REGISTER_BUFFERS: | |
634d00df | 11740 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
11741 | break; |
11742 | case IORING_UNREGISTER_BUFFERS: | |
11743 | ret = -EINVAL; | |
11744 | if (arg || nr_args) | |
11745 | break; | |
0a96bbe4 | 11746 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 11747 | break; |
6b06314c | 11748 | case IORING_REGISTER_FILES: |
792e3582 | 11749 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
11750 | break; |
11751 | case IORING_UNREGISTER_FILES: | |
11752 | ret = -EINVAL; | |
11753 | if (arg || nr_args) | |
11754 | break; | |
11755 | ret = io_sqe_files_unregister(ctx); | |
11756 | break; | |
c3a31e60 | 11757 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 11758 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 11759 | break; |
9b402849 JA |
11760 | case IORING_REGISTER_EVENTFD: |
11761 | ret = -EINVAL; | |
11762 | if (nr_args != 1) | |
11763 | break; | |
c75312dd UA |
11764 | ret = io_eventfd_register(ctx, arg, 0); |
11765 | break; | |
11766 | case IORING_REGISTER_EVENTFD_ASYNC: | |
11767 | ret = -EINVAL; | |
11768 | if (nr_args != 1) | |
f2842ab5 | 11769 | break; |
c75312dd | 11770 | ret = io_eventfd_register(ctx, arg, 1); |
9b402849 JA |
11771 | break; |
11772 | case IORING_UNREGISTER_EVENTFD: | |
11773 | ret = -EINVAL; | |
11774 | if (arg || nr_args) | |
11775 | break; | |
11776 | ret = io_eventfd_unregister(ctx); | |
11777 | break; | |
66f4af93 JA |
11778 | case IORING_REGISTER_PROBE: |
11779 | ret = -EINVAL; | |
11780 | if (!arg || nr_args > 256) | |
11781 | break; | |
11782 | ret = io_probe(ctx, arg, nr_args); | |
11783 | break; | |
071698e1 JA |
11784 | case IORING_REGISTER_PERSONALITY: |
11785 | ret = -EINVAL; | |
11786 | if (arg || nr_args) | |
11787 | break; | |
11788 | ret = io_register_personality(ctx); | |
11789 | break; | |
11790 | case IORING_UNREGISTER_PERSONALITY: | |
11791 | ret = -EINVAL; | |
11792 | if (arg) | |
11793 | break; | |
11794 | ret = io_unregister_personality(ctx, nr_args); | |
11795 | break; | |
7e84e1c7 SG |
11796 | case IORING_REGISTER_ENABLE_RINGS: |
11797 | ret = -EINVAL; | |
11798 | if (arg || nr_args) | |
11799 | break; | |
11800 | ret = io_register_enable_rings(ctx); | |
11801 | break; | |
21b55dbc SG |
11802 | case IORING_REGISTER_RESTRICTIONS: |
11803 | ret = io_register_restrictions(ctx, arg, nr_args); | |
11804 | break; | |
992da01a PB |
11805 | case IORING_REGISTER_FILES2: |
11806 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
11807 | break; | |
11808 | case IORING_REGISTER_FILES_UPDATE2: | |
11809 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
11810 | IORING_RSRC_FILE); | |
11811 | break; | |
11812 | case IORING_REGISTER_BUFFERS2: | |
11813 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 11814 | break; |
992da01a PB |
11815 | case IORING_REGISTER_BUFFERS_UPDATE: |
11816 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
11817 | IORING_RSRC_BUFFER); | |
c3bdad02 | 11818 | break; |
fe76421d JA |
11819 | case IORING_REGISTER_IOWQ_AFF: |
11820 | ret = -EINVAL; | |
11821 | if (!arg || !nr_args) | |
11822 | break; | |
11823 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
11824 | break; | |
11825 | case IORING_UNREGISTER_IOWQ_AFF: | |
11826 | ret = -EINVAL; | |
11827 | if (arg || nr_args) | |
11828 | break; | |
11829 | ret = io_unregister_iowq_aff(ctx); | |
11830 | break; | |
2e480058 JA |
11831 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
11832 | ret = -EINVAL; | |
11833 | if (!arg || nr_args != 2) | |
11834 | break; | |
11835 | ret = io_register_iowq_max_workers(ctx, arg); | |
11836 | break; | |
e7a6c00d JA |
11837 | case IORING_REGISTER_RING_FDS: |
11838 | ret = io_ringfd_register(ctx, arg, nr_args); | |
11839 | break; | |
11840 | case IORING_UNREGISTER_RING_FDS: | |
11841 | ret = io_ringfd_unregister(ctx, arg, nr_args); | |
11842 | break; | |
edafccee JA |
11843 | default: |
11844 | ret = -EINVAL; | |
11845 | break; | |
11846 | } | |
11847 | ||
edafccee JA |
11848 | return ret; |
11849 | } | |
11850 | ||
11851 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
11852 | void __user *, arg, unsigned int, nr_args) | |
11853 | { | |
11854 | struct io_ring_ctx *ctx; | |
11855 | long ret = -EBADF; | |
11856 | struct fd f; | |
11857 | ||
11858 | f = fdget(fd); | |
11859 | if (!f.file) | |
11860 | return -EBADF; | |
11861 | ||
11862 | ret = -EOPNOTSUPP; | |
11863 | if (f.file->f_op != &io_uring_fops) | |
11864 | goto out_fput; | |
11865 | ||
11866 | ctx = f.file->private_data; | |
11867 | ||
b6c23dd5 PB |
11868 | io_run_task_work(); |
11869 | ||
edafccee JA |
11870 | mutex_lock(&ctx->uring_lock); |
11871 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
11872 | mutex_unlock(&ctx->uring_lock); | |
2757be22 | 11873 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, ret); |
edafccee JA |
11874 | out_fput: |
11875 | fdput(f); | |
11876 | return ret; | |
11877 | } | |
11878 | ||
2b188cc1 JA |
11879 | static int __init io_uring_init(void) |
11880 | { | |
d7f62e82 SM |
11881 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
11882 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
11883 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
11884 | } while (0) | |
11885 | ||
11886 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
11887 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
11888 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
11889 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
11890 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
11891 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
11892 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
11893 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
11894 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
11895 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 11896 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
11897 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
11898 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
11899 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
11900 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
11901 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
11902 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
11903 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
11904 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
11905 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
11906 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
11907 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
11908 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
11909 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
11910 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
11911 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 11912 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
11913 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
11914 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 11915 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 11916 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 11917 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
b9445598 | 11918 | BUILD_BUG_SQE_ELEM(44, __u32, file_index); |
d7f62e82 | 11919 | |
b0d658ec PB |
11920 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
11921 | sizeof(struct io_uring_rsrc_update)); | |
11922 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
11923 | sizeof(struct io_uring_rsrc_update2)); | |
90499ad0 PB |
11924 | |
11925 | /* ->buf_index is u16 */ | |
11926 | BUILD_BUG_ON(IORING_MAX_REG_BUFFERS >= (1u << 16)); | |
11927 | ||
b0d658ec PB |
11928 | /* should fit into one byte */ |
11929 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
68fe256a PB |
11930 | BUILD_BUG_ON(SQE_COMMON_FLAGS >= (1 << 8)); |
11931 | BUILD_BUG_ON((SQE_VALID_FLAGS | SQE_COMMON_FLAGS) != SQE_VALID_FLAGS); | |
b0d658ec | 11932 | |
d3656344 | 11933 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
32c2d33e | 11934 | BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int)); |
16340eab | 11935 | |
91f245d5 JA |
11936 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
11937 | SLAB_ACCOUNT); | |
2b188cc1 JA |
11938 | return 0; |
11939 | }; | |
11940 | __initcall(io_uring_init); |