]>
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> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
5bd2182d | 81 | #include <linux/audit.h> |
cdc1404a | 82 | #include <linux/security.h> |
e9621e2b | 83 | #include <linux/xattr.h> |
2b188cc1 | 84 | |
c826bd7a DD |
85 | #define CREATE_TRACE_POINTS |
86 | #include <trace/events/io_uring.h> | |
87 | ||
2b188cc1 JA |
88 | #include <uapi/linux/io_uring.h> |
89 | ||
90 | #include "internal.h" | |
561fb04a | 91 | #include "io-wq.h" |
2b188cc1 | 92 | |
5277deaa | 93 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 94 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
4ce8ad95 | 95 | #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8 |
65e19f54 | 96 | |
187f08c1 | 97 | /* only define max */ |
09893e15 | 98 | #define IORING_MAX_FIXED_FILES (1U << 20) |
21b55dbc SG |
99 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
100 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 101 | |
187f08c1 | 102 | #define IO_RSRC_TAG_TABLE_SHIFT (PAGE_SHIFT - 3) |
2d091d62 PB |
103 | #define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT) |
104 | #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1) | |
105 | ||
489809e2 PB |
106 | #define IORING_MAX_REG_BUFFERS (1U << 14) |
107 | ||
68fe256a PB |
108 | #define SQE_COMMON_FLAGS (IOSQE_FIXED_FILE | IOSQE_IO_LINK | \ |
109 | IOSQE_IO_HARDLINK | IOSQE_ASYNC) | |
110 | ||
5562a8d7 PB |
111 | #define SQE_VALID_FLAGS (SQE_COMMON_FLAGS | IOSQE_BUFFER_SELECT | \ |
112 | IOSQE_IO_DRAIN | IOSQE_CQE_SKIP_SUCCESS) | |
68fe256a | 113 | |
c854357b | 114 | #define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \ |
9cae36a0 JA |
115 | REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS | \ |
116 | REQ_F_ASYNC_DATA) | |
b16fed66 | 117 | |
a538be5b PB |
118 | #define IO_REQ_CLEAN_SLOW_FLAGS (REQ_F_REFCOUNT | REQ_F_LINK | REQ_F_HARDLINK |\ |
119 | IO_REQ_CLEAN_FLAGS) | |
120 | ||
227685eb HX |
121 | #define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED) |
122 | ||
09899b19 PB |
123 | #define IO_TCTX_REFS_CACHE_NR (1U << 10) |
124 | ||
2b188cc1 JA |
125 | struct io_uring { |
126 | u32 head ____cacheline_aligned_in_smp; | |
127 | u32 tail ____cacheline_aligned_in_smp; | |
128 | }; | |
129 | ||
1e84b97b | 130 | /* |
75b28aff HV |
131 | * This data is shared with the application through the mmap at offsets |
132 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
133 | * |
134 | * The offsets to the member fields are published through struct | |
135 | * io_sqring_offsets when calling io_uring_setup. | |
136 | */ | |
75b28aff | 137 | struct io_rings { |
1e84b97b SB |
138 | /* |
139 | * Head and tail offsets into the ring; the offsets need to be | |
140 | * masked to get valid indices. | |
141 | * | |
75b28aff HV |
142 | * The kernel controls head of the sq ring and the tail of the cq ring, |
143 | * and the application controls tail of the sq ring and the head of the | |
144 | * cq ring. | |
1e84b97b | 145 | */ |
75b28aff | 146 | struct io_uring sq, cq; |
1e84b97b | 147 | /* |
75b28aff | 148 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
149 | * ring_entries - 1) |
150 | */ | |
75b28aff HV |
151 | u32 sq_ring_mask, cq_ring_mask; |
152 | /* Ring sizes (constant, power of 2) */ | |
153 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
154 | /* |
155 | * Number of invalid entries dropped by the kernel due to | |
156 | * invalid index stored in array | |
157 | * | |
158 | * Written by the kernel, shouldn't be modified by the | |
159 | * application (i.e. get number of "new events" by comparing to | |
160 | * cached value). | |
161 | * | |
162 | * After a new SQ head value was read by the application this | |
163 | * counter includes all submissions that were dropped reaching | |
164 | * the new SQ head (and possibly more). | |
165 | */ | |
75b28aff | 166 | u32 sq_dropped; |
1e84b97b | 167 | /* |
0d9b5b3a | 168 | * Runtime SQ flags |
1e84b97b SB |
169 | * |
170 | * Written by the kernel, shouldn't be modified by the | |
171 | * application. | |
172 | * | |
173 | * The application needs a full memory barrier before checking | |
174 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
175 | */ | |
3a4b89a2 | 176 | atomic_t sq_flags; |
0d9b5b3a SG |
177 | /* |
178 | * Runtime CQ flags | |
179 | * | |
180 | * Written by the application, shouldn't be modified by the | |
181 | * kernel. | |
182 | */ | |
fe7e3257 | 183 | u32 cq_flags; |
1e84b97b SB |
184 | /* |
185 | * Number of completion events lost because the queue was full; | |
186 | * this should be avoided by the application by making sure | |
0b4295b5 | 187 | * there are not more requests pending than there is space in |
1e84b97b SB |
188 | * the completion queue. |
189 | * | |
190 | * Written by the kernel, shouldn't be modified by the | |
191 | * application (i.e. get number of "new events" by comparing to | |
192 | * cached value). | |
193 | * | |
194 | * As completion events come in out of order this counter is not | |
195 | * ordered with any other data. | |
196 | */ | |
75b28aff | 197 | u32 cq_overflow; |
1e84b97b SB |
198 | /* |
199 | * Ring buffer of completion events. | |
200 | * | |
201 | * The kernel writes completion events fresh every time they are | |
202 | * produced, so the application is allowed to modify pending | |
203 | * entries. | |
204 | */ | |
75b28aff | 205 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
206 | }; |
207 | ||
edafccee JA |
208 | struct io_mapped_ubuf { |
209 | u64 ubuf; | |
4751f53d | 210 | u64 ubuf_end; |
edafccee | 211 | unsigned int nr_bvecs; |
de293938 | 212 | unsigned long acct_pages; |
41edf1a5 | 213 | struct bio_vec bvec[]; |
edafccee JA |
214 | }; |
215 | ||
50238531 BM |
216 | struct io_ring_ctx; |
217 | ||
6c2450ae | 218 | struct io_overflow_cqe { |
6c2450ae | 219 | struct list_head list; |
e45a3e05 | 220 | struct io_uring_cqe cqe; |
6c2450ae PB |
221 | }; |
222 | ||
5e45690a JA |
223 | /* |
224 | * FFS_SCM is only available on 64-bit archs, for 32-bit we just define it as 0 | |
225 | * and define IO_URING_SCM_ALL. For this case, we use SCM for all files as we | |
226 | * can't safely always dereference the file when the task has exited and ring | |
227 | * cleanup is done. If a file is tracked and part of SCM, then unix gc on | |
228 | * process exit may reap it before __io_sqe_files_unregister() is run. | |
229 | */ | |
230 | #define FFS_NOWAIT 0x1UL | |
231 | #define FFS_ISREG 0x2UL | |
232 | #if defined(CONFIG_64BIT) | |
233 | #define FFS_SCM 0x4UL | |
234 | #else | |
235 | #define IO_URING_SCM_ALL | |
236 | #define FFS_SCM 0x0UL | |
237 | #endif | |
238 | #define FFS_MASK ~(FFS_NOWAIT|FFS_ISREG|FFS_SCM) | |
239 | ||
a04b0ac0 PB |
240 | struct io_fixed_file { |
241 | /* file * with additional FFS_* flags */ | |
242 | unsigned long file_ptr; | |
243 | }; | |
244 | ||
269bbe5f BM |
245 | struct io_rsrc_put { |
246 | struct list_head list; | |
b60c8dce | 247 | u64 tag; |
50238531 BM |
248 | union { |
249 | void *rsrc; | |
250 | struct file *file; | |
bd54b6fe | 251 | struct io_mapped_ubuf *buf; |
50238531 | 252 | }; |
269bbe5f BM |
253 | }; |
254 | ||
aeca241b | 255 | struct io_file_table { |
042b0d85 | 256 | struct io_fixed_file *files; |
d78bd8ad | 257 | unsigned long *bitmap; |
b70b8e33 | 258 | unsigned int alloc_hint; |
31b51510 JA |
259 | }; |
260 | ||
b895c9a6 | 261 | struct io_rsrc_node { |
05589553 XW |
262 | struct percpu_ref refs; |
263 | struct list_head node; | |
269bbe5f | 264 | struct list_head rsrc_list; |
b895c9a6 | 265 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 266 | struct llist_node llist; |
e297822b | 267 | bool done; |
05589553 XW |
268 | }; |
269 | ||
40ae0ff7 PB |
270 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
271 | ||
b895c9a6 | 272 | struct io_rsrc_data { |
05f3fb3c JA |
273 | struct io_ring_ctx *ctx; |
274 | ||
2d091d62 PB |
275 | u64 **tags; |
276 | unsigned int nr; | |
40ae0ff7 | 277 | rsrc_put_fn *do_put; |
3e942498 | 278 | atomic_t refs; |
05f3fb3c | 279 | struct completion done; |
8bad28d8 | 280 | bool quiesce; |
05f3fb3c JA |
281 | }; |
282 | ||
c7fb1942 | 283 | #define IO_BUFFER_LIST_BUF_PER_PAGE (PAGE_SIZE / sizeof(struct io_uring_buf)) |
dbc7d452 | 284 | struct io_buffer_list { |
c7fb1942 JA |
285 | /* |
286 | * If ->buf_nr_pages is set, then buf_pages/buf_ring are used. If not, | |
287 | * then these are classic provided buffers and ->buf_list is used. | |
288 | */ | |
289 | union { | |
290 | struct list_head buf_list; | |
291 | struct { | |
292 | struct page **buf_pages; | |
293 | struct io_uring_buf_ring *buf_ring; | |
294 | }; | |
295 | }; | |
dbc7d452 | 296 | __u16 bgid; |
c7fb1942 JA |
297 | |
298 | /* below is for ring provided buffers */ | |
299 | __u16 buf_nr_pages; | |
300 | __u16 nr_entries; | |
c6e9fa5c DY |
301 | __u16 head; |
302 | __u16 mask; | |
dbc7d452 JA |
303 | }; |
304 | ||
5a2e745d JA |
305 | struct io_buffer { |
306 | struct list_head list; | |
307 | __u64 addr; | |
d1f82808 | 308 | __u32 len; |
5a2e745d | 309 | __u16 bid; |
b1c62645 | 310 | __u16 bgid; |
5a2e745d JA |
311 | }; |
312 | ||
21b55dbc SG |
313 | struct io_restriction { |
314 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
315 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
316 | u8 sqe_flags_allowed; | |
317 | u8 sqe_flags_required; | |
7e84e1c7 | 318 | bool registered; |
21b55dbc SG |
319 | }; |
320 | ||
37d1e2e3 JA |
321 | enum { |
322 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
323 | IO_SQ_THREAD_SHOULD_PARK, | |
324 | }; | |
325 | ||
534ca6d6 JA |
326 | struct io_sq_data { |
327 | refcount_t refs; | |
9e138a48 | 328 | atomic_t park_pending; |
09a6f4ef | 329 | struct mutex lock; |
69fb2131 JA |
330 | |
331 | /* ctx's that are using this sqd */ | |
332 | struct list_head ctx_list; | |
69fb2131 | 333 | |
534ca6d6 JA |
334 | struct task_struct *thread; |
335 | struct wait_queue_head wait; | |
08369246 XW |
336 | |
337 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
338 | int sq_cpu; |
339 | pid_t task_pid; | |
5c2469e0 | 340 | pid_t task_tgid; |
37d1e2e3 JA |
341 | |
342 | unsigned long state; | |
37d1e2e3 | 343 | struct completion exited; |
534ca6d6 JA |
344 | }; |
345 | ||
6dd0be1e | 346 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 347 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 348 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 | 349 | |
a1ab7b35 PB |
350 | struct io_submit_link { |
351 | struct io_kiocb *head; | |
352 | struct io_kiocb *last; | |
353 | }; | |
354 | ||
258b29a9 | 355 | struct io_submit_state { |
5a158c6b PB |
356 | /* inline/task_work completion list, under ->uring_lock */ |
357 | struct io_wq_work_node free_list; | |
358 | /* batch completion logic */ | |
359 | struct io_wq_work_list compl_reqs; | |
a1ab7b35 | 360 | struct io_submit_link link; |
258b29a9 | 361 | |
258b29a9 | 362 | bool plug_started; |
4b628aeb | 363 | bool need_plug; |
3d4aeb9f | 364 | bool flush_cqes; |
5ca7a8b3 | 365 | unsigned short submit_nr; |
5a158c6b | 366 | struct blk_plug plug; |
258b29a9 PB |
367 | }; |
368 | ||
77bc59b4 UA |
369 | struct io_ev_fd { |
370 | struct eventfd_ctx *cq_ev_fd; | |
c75312dd | 371 | unsigned int eventfd_async: 1; |
77bc59b4 UA |
372 | struct rcu_head rcu; |
373 | }; | |
374 | ||
9cfc7e94 | 375 | #define BGID_ARRAY 64 |
dbc7d452 | 376 | |
2b188cc1 | 377 | struct io_ring_ctx { |
b52ecf8c | 378 | /* const or read-mostly hot data */ |
2b188cc1 JA |
379 | struct { |
380 | struct percpu_ref refs; | |
2b188cc1 | 381 | |
b52ecf8c | 382 | struct io_rings *rings; |
2b188cc1 | 383 | unsigned int flags; |
9f010507 | 384 | enum task_work_notify_mode notify_method; |
e1d85334 | 385 | unsigned int compat: 1; |
e1d85334 | 386 | unsigned int drain_next: 1; |
21b55dbc | 387 | unsigned int restricted: 1; |
f18ee4cf | 388 | unsigned int off_timeout_used: 1; |
10c66904 | 389 | unsigned int drain_active: 1; |
5562a8d7 | 390 | unsigned int drain_disabled: 1; |
9aa8dfde | 391 | unsigned int has_evfd: 1; |
773697b6 | 392 | unsigned int syscall_iopoll: 1; |
b52ecf8c | 393 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 394 | |
7f1129d2 | 395 | /* submission data */ |
b52ecf8c | 396 | struct { |
0499e582 PB |
397 | struct mutex uring_lock; |
398 | ||
75b28aff HV |
399 | /* |
400 | * Ring buffer of indices into array of io_uring_sqe, which is | |
401 | * mmapped by the application using the IORING_OFF_SQES offset. | |
402 | * | |
403 | * This indirection could e.g. be used to assign fixed | |
404 | * io_uring_sqe entries to operations and only submit them to | |
405 | * the queue when needed. | |
406 | * | |
407 | * The kernel modifies neither the indices array nor the entries | |
408 | * array. | |
409 | */ | |
410 | u32 *sq_array; | |
c7af47cf | 411 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
412 | unsigned cached_sq_head; |
413 | unsigned sq_entries; | |
de0617e4 | 414 | struct list_head defer_list; |
7f1129d2 PB |
415 | |
416 | /* | |
417 | * Fixed resources fast path, should be accessed only under | |
418 | * uring_lock, and updated through io_uring_register(2) | |
419 | */ | |
420 | struct io_rsrc_node *rsrc_node; | |
ab409402 | 421 | int rsrc_cached_refs; |
8e29da69 | 422 | atomic_t cancel_seq; |
7f1129d2 PB |
423 | struct io_file_table file_table; |
424 | unsigned nr_user_files; | |
425 | unsigned nr_user_bufs; | |
426 | struct io_mapped_ubuf **user_bufs; | |
427 | ||
428 | struct io_submit_state submit_state; | |
660cbfa2 | 429 | |
9cfc7e94 JA |
430 | struct io_buffer_list *io_bl; |
431 | struct xarray io_bl_xa; | |
cc3cec83 | 432 | struct list_head io_buffers_cache; |
660cbfa2 | 433 | |
5262f567 | 434 | struct list_head timeout_list; |
ef9dd637 | 435 | struct list_head ltimeout_list; |
1d7bb1d5 | 436 | struct list_head cq_overflow_list; |
4d9237e3 | 437 | struct list_head apoll_cache; |
7f1129d2 PB |
438 | struct xarray personalities; |
439 | u32 pers_next; | |
440 | unsigned sq_thread_idle; | |
2b188cc1 JA |
441 | } ____cacheline_aligned_in_smp; |
442 | ||
d0acdee2 | 443 | /* IRQ completion list, under ->completion_lock */ |
c2b6c6bc | 444 | struct io_wq_work_list locked_free_list; |
d0acdee2 | 445 | unsigned int locked_free_nr; |
3c1a2ead | 446 | |
7c30f36a | 447 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
448 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
449 | ||
90554200 | 450 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 451 | struct list_head sqd_list; |
75b28aff | 452 | |
10988a0a | 453 | unsigned long check_cq; |
5ed7a37d | 454 | |
206aefde | 455 | struct { |
d8da428b PB |
456 | /* |
457 | * We cache a range of free CQEs we can use, once exhausted it | |
458 | * should go through a slower range setup, see __io_get_cqe() | |
459 | */ | |
460 | struct io_uring_cqe *cqe_cached; | |
461 | struct io_uring_cqe *cqe_sentinel; | |
462 | ||
206aefde JA |
463 | unsigned cached_cq_tail; |
464 | unsigned cq_entries; | |
77bc59b4 | 465 | struct io_ev_fd __rcu *io_ev_fd; |
0499e582 PB |
466 | struct wait_queue_head cq_wait; |
467 | unsigned cq_extra; | |
468 | atomic_t cq_timeouts; | |
0499e582 | 469 | unsigned cq_last_tm_flush; |
206aefde | 470 | } ____cacheline_aligned_in_smp; |
2b188cc1 | 471 | |
2b188cc1 JA |
472 | struct { |
473 | spinlock_t completion_lock; | |
e94f141b | 474 | |
89850fce JA |
475 | spinlock_t timeout_lock; |
476 | ||
def596e9 | 477 | /* |
540e32a0 | 478 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
479 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
480 | * For SQPOLL, only the single threaded io_sq_thread() will | |
481 | * manipulate the list, hence no extra locking is needed there. | |
482 | */ | |
5eef4e87 | 483 | struct io_wq_work_list iopoll_list; |
78076bb6 JA |
484 | struct hlist_head *cancel_hash; |
485 | unsigned cancel_hash_bits; | |
915b3dde | 486 | bool poll_multi_queue; |
cc3cec83 JA |
487 | |
488 | struct list_head io_buffers_comp; | |
2b188cc1 | 489 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 490 | |
21b55dbc | 491 | struct io_restriction restrictions; |
3c1a2ead | 492 | |
b13a8918 PB |
493 | /* slow path rsrc auxilary data, used by update/register */ |
494 | struct { | |
495 | struct io_rsrc_node *rsrc_backup_node; | |
496 | struct io_mapped_ubuf *dummy_ubuf; | |
497 | struct io_rsrc_data *file_data; | |
498 | struct io_rsrc_data *buf_data; | |
499 | ||
500 | struct delayed_work rsrc_put_work; | |
501 | struct llist_head rsrc_put_llist; | |
502 | struct list_head rsrc_ref_list; | |
503 | spinlock_t rsrc_ref_lock; | |
cc3cec83 JA |
504 | |
505 | struct list_head io_buffers_pages; | |
b13a8918 PB |
506 | }; |
507 | ||
3c1a2ead | 508 | /* Keep this last, we don't need it for the fast path */ |
b986af7e PB |
509 | struct { |
510 | #if defined(CONFIG_UNIX) | |
511 | struct socket *ring_sock; | |
512 | #endif | |
513 | /* hashed buffered write serialization */ | |
514 | struct io_wq_hash *hash_map; | |
515 | ||
516 | /* Only used for accounting purposes */ | |
517 | struct user_struct *user; | |
518 | struct mm_struct *mm_account; | |
519 | ||
520 | /* ctx exit and cancelation */ | |
9011bf9a PB |
521 | struct llist_head fallback_llist; |
522 | struct delayed_work fallback_work; | |
b986af7e PB |
523 | struct work_struct exit_work; |
524 | struct list_head tctx_list; | |
525 | struct completion ref_comp; | |
e139a1ec PB |
526 | u32 iowq_limits[2]; |
527 | bool iowq_limits_set; | |
b986af7e | 528 | }; |
2b188cc1 JA |
529 | }; |
530 | ||
e7a6c00d JA |
531 | /* |
532 | * Arbitrary limit, can be raised if need be | |
533 | */ | |
534 | #define IO_RINGFD_REG_MAX 16 | |
535 | ||
53e043b2 SM |
536 | struct io_uring_task { |
537 | /* submission side */ | |
09899b19 | 538 | int cached_refs; |
53e043b2 SM |
539 | struct xarray xa; |
540 | struct wait_queue_head wait; | |
ee53fb2b SM |
541 | const struct io_ring_ctx *last; |
542 | struct io_wq *io_wq; | |
53e043b2 | 543 | struct percpu_counter inflight; |
9cae36a0 | 544 | atomic_t inflight_tracked; |
53e043b2 | 545 | atomic_t in_idle; |
53e043b2 SM |
546 | |
547 | spinlock_t task_lock; | |
548 | struct io_wq_work_list task_list; | |
3fe07bcd | 549 | struct io_wq_work_list prio_task_list; |
53e043b2 | 550 | struct callback_head task_work; |
e7a6c00d | 551 | struct file **registered_rings; |
6294f368 | 552 | bool task_running; |
53e043b2 SM |
553 | }; |
554 | ||
09bb8394 JA |
555 | /* |
556 | * First field must be the file pointer in all the | |
557 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
558 | */ | |
221c5eb2 JA |
559 | struct io_poll_iocb { |
560 | struct file *file; | |
018043be | 561 | struct wait_queue_head *head; |
221c5eb2 | 562 | __poll_t events; |
392edb45 | 563 | struct wait_queue_entry wait; |
221c5eb2 JA |
564 | }; |
565 | ||
9d805892 | 566 | struct io_poll_update { |
018043be | 567 | struct file *file; |
9d805892 PB |
568 | u64 old_user_data; |
569 | u64 new_user_data; | |
570 | __poll_t events; | |
b69de288 JA |
571 | bool update_events; |
572 | bool update_user_data; | |
018043be PB |
573 | }; |
574 | ||
b5dba59e JA |
575 | struct io_close { |
576 | struct file *file; | |
b5dba59e | 577 | int fd; |
7df778be | 578 | u32 file_slot; |
b5dba59e JA |
579 | }; |
580 | ||
ad8a48ac JA |
581 | struct io_timeout_data { |
582 | struct io_kiocb *req; | |
583 | struct hrtimer timer; | |
584 | struct timespec64 ts; | |
585 | enum hrtimer_mode mode; | |
50c1df2b | 586 | u32 flags; |
ad8a48ac JA |
587 | }; |
588 | ||
8ed8d3c3 JA |
589 | struct io_accept { |
590 | struct file *file; | |
591 | struct sockaddr __user *addr; | |
592 | int __user *addr_len; | |
593 | int flags; | |
aaa4db12 | 594 | u32 file_slot; |
09952e3e | 595 | unsigned long nofile; |
8ed8d3c3 JA |
596 | }; |
597 | ||
1374e08e JA |
598 | struct io_socket { |
599 | struct file *file; | |
600 | int domain; | |
601 | int type; | |
602 | int protocol; | |
603 | int flags; | |
604 | u32 file_slot; | |
605 | unsigned long nofile; | |
606 | }; | |
607 | ||
8ed8d3c3 JA |
608 | struct io_sync { |
609 | struct file *file; | |
610 | loff_t len; | |
611 | loff_t off; | |
612 | int flags; | |
d63d1b5e | 613 | int mode; |
8ed8d3c3 JA |
614 | }; |
615 | ||
fbf23849 JA |
616 | struct io_cancel { |
617 | struct file *file; | |
618 | u64 addr; | |
8e29da69 | 619 | u32 flags; |
4bf94615 | 620 | s32 fd; |
fbf23849 JA |
621 | }; |
622 | ||
b29472ee JA |
623 | struct io_timeout { |
624 | struct file *file; | |
bfe68a22 PB |
625 | u32 off; |
626 | u32 target_seq; | |
135fcde8 | 627 | struct list_head list; |
90cd7e42 PB |
628 | /* head of the link, used by linked timeouts only */ |
629 | struct io_kiocb *head; | |
89b263f6 JA |
630 | /* for linked completions */ |
631 | struct io_kiocb *prev; | |
b29472ee JA |
632 | }; |
633 | ||
0bdf7a2d PB |
634 | struct io_timeout_rem { |
635 | struct file *file; | |
636 | u64 addr; | |
9c8e11b3 PB |
637 | |
638 | /* timeout update */ | |
639 | struct timespec64 ts; | |
640 | u32 flags; | |
f1042b6c | 641 | bool ltimeout; |
0bdf7a2d PB |
642 | }; |
643 | ||
9adbd45d JA |
644 | struct io_rw { |
645 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
646 | struct kiocb kiocb; | |
647 | u64 addr; | |
584b0180 | 648 | u32 len; |
20cbd21d | 649 | rwf_t flags; |
9adbd45d JA |
650 | }; |
651 | ||
3fbb51c1 JA |
652 | struct io_connect { |
653 | struct file *file; | |
654 | struct sockaddr __user *addr; | |
655 | int addr_len; | |
656 | }; | |
657 | ||
e47293fd JA |
658 | struct io_sr_msg { |
659 | struct file *file; | |
fddaface | 660 | union { |
4af3417a PB |
661 | struct compat_msghdr __user *umsg_compat; |
662 | struct user_msghdr __user *umsg; | |
663 | void __user *buf; | |
fddaface | 664 | }; |
e47293fd | 665 | int msg_flags; |
fddaface | 666 | size_t len; |
7ba89d2a | 667 | size_t done_io; |
0455d4cc | 668 | unsigned int flags; |
e47293fd JA |
669 | }; |
670 | ||
15b71abe JA |
671 | struct io_open { |
672 | struct file *file; | |
673 | int dfd; | |
b9445598 | 674 | u32 file_slot; |
15b71abe | 675 | struct filename *filename; |
c12cedf2 | 676 | struct open_how how; |
4022e7af | 677 | unsigned long nofile; |
15b71abe JA |
678 | }; |
679 | ||
269bbe5f | 680 | struct io_rsrc_update { |
05f3fb3c JA |
681 | struct file *file; |
682 | u64 arg; | |
683 | u32 nr_args; | |
684 | u32 offset; | |
685 | }; | |
686 | ||
4840e418 JA |
687 | struct io_fadvise { |
688 | struct file *file; | |
689 | u64 offset; | |
690 | u32 len; | |
691 | u32 advice; | |
692 | }; | |
693 | ||
c1ca757b JA |
694 | struct io_madvise { |
695 | struct file *file; | |
696 | u64 addr; | |
697 | u32 len; | |
698 | u32 advice; | |
699 | }; | |
700 | ||
3e4827b0 JA |
701 | struct io_epoll { |
702 | struct file *file; | |
703 | int epfd; | |
704 | int op; | |
705 | int fd; | |
706 | struct epoll_event event; | |
e47293fd JA |
707 | }; |
708 | ||
7d67af2c PB |
709 | struct io_splice { |
710 | struct file *file_out; | |
7d67af2c PB |
711 | loff_t off_out; |
712 | loff_t off_in; | |
713 | u64 len; | |
a3e4bc23 | 714 | int splice_fd_in; |
7d67af2c PB |
715 | unsigned int flags; |
716 | }; | |
717 | ||
ddf0322d JA |
718 | struct io_provide_buf { |
719 | struct file *file; | |
720 | __u64 addr; | |
38134ada | 721 | __u32 len; |
ddf0322d JA |
722 | __u32 bgid; |
723 | __u16 nbufs; | |
724 | __u16 bid; | |
725 | }; | |
726 | ||
1d9e1288 BM |
727 | struct io_statx { |
728 | struct file *file; | |
729 | int dfd; | |
730 | unsigned int mask; | |
731 | unsigned int flags; | |
1b6fe6e0 | 732 | struct filename *filename; |
1d9e1288 BM |
733 | struct statx __user *buffer; |
734 | }; | |
735 | ||
36f4fa68 JA |
736 | struct io_shutdown { |
737 | struct file *file; | |
738 | int how; | |
739 | }; | |
740 | ||
80a261fd JA |
741 | struct io_rename { |
742 | struct file *file; | |
743 | int old_dfd; | |
744 | int new_dfd; | |
745 | struct filename *oldpath; | |
746 | struct filename *newpath; | |
747 | int flags; | |
748 | }; | |
749 | ||
14a1143b JA |
750 | struct io_unlink { |
751 | struct file *file; | |
752 | int dfd; | |
753 | int flags; | |
754 | struct filename *filename; | |
755 | }; | |
756 | ||
e34a02dc DK |
757 | struct io_mkdir { |
758 | struct file *file; | |
759 | int dfd; | |
760 | umode_t mode; | |
761 | struct filename *filename; | |
762 | }; | |
763 | ||
7a8721f8 DK |
764 | struct io_symlink { |
765 | struct file *file; | |
766 | int new_dfd; | |
767 | struct filename *oldpath; | |
768 | struct filename *newpath; | |
769 | }; | |
770 | ||
cf30da90 DK |
771 | struct io_hardlink { |
772 | struct file *file; | |
773 | int old_dfd; | |
774 | int new_dfd; | |
775 | struct filename *oldpath; | |
776 | struct filename *newpath; | |
777 | int flags; | |
778 | }; | |
779 | ||
4f57f06c JA |
780 | struct io_msg { |
781 | struct file *file; | |
782 | u64 user_data; | |
783 | u32 len; | |
784 | }; | |
785 | ||
f499a021 JA |
786 | struct io_async_connect { |
787 | struct sockaddr_storage address; | |
788 | }; | |
789 | ||
03b1230c JA |
790 | struct io_async_msghdr { |
791 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
792 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
793 | struct iovec *free_iov; | |
03b1230c JA |
794 | struct sockaddr __user *uaddr; |
795 | struct msghdr msg; | |
b537916c | 796 | struct sockaddr_storage addr; |
03b1230c JA |
797 | }; |
798 | ||
538941e2 | 799 | struct io_rw_state { |
ff6165b2 | 800 | struct iov_iter iter; |
cd658695 | 801 | struct iov_iter_state iter_state; |
c88598a9 | 802 | struct iovec fast_iov[UIO_FASTIOV]; |
538941e2 PB |
803 | }; |
804 | ||
805 | struct io_async_rw { | |
806 | struct io_rw_state s; | |
807 | const struct iovec *free_iovec; | |
227c0c96 | 808 | size_t bytes_done; |
bcf5a063 | 809 | struct wait_page_queue wpq; |
f67676d1 JA |
810 | }; |
811 | ||
e9621e2b SR |
812 | struct io_xattr { |
813 | struct file *file; | |
814 | struct xattr_ctx ctx; | |
815 | struct filename *filename; | |
816 | }; | |
817 | ||
6b47ee6e PB |
818 | enum { |
819 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
820 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
821 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
822 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
823 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 824 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
04c76b41 | 825 | REQ_F_CQE_SKIP_BIT = IOSQE_CQE_SKIP_SUCCESS_BIT, |
6b47ee6e | 826 | |
dddca226 | 827 | /* first byte is taken by user flags, shift it to not overlap */ |
93d2bcd2 | 828 | REQ_F_FAIL_BIT = 8, |
6b47ee6e PB |
829 | REQ_F_INFLIGHT_BIT, |
830 | REQ_F_CUR_POS_BIT, | |
831 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 832 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 833 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 834 | REQ_F_POLLED_BIT, |
bcda7baa | 835 | REQ_F_BUFFER_SELECTED_BIT, |
c7fb1942 | 836 | REQ_F_BUFFER_RING_BIT, |
e342c807 | 837 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 838 | REQ_F_REISSUE_BIT, |
b8e64b53 | 839 | REQ_F_CREDS_BIT, |
20e60a38 | 840 | REQ_F_REFCOUNT_BIT, |
4d13d1a4 | 841 | REQ_F_ARM_LTIMEOUT_BIT, |
d886e185 | 842 | REQ_F_ASYNC_DATA_BIT, |
04c76b41 | 843 | REQ_F_SKIP_LINK_CQES_BIT, |
91eac1c6 JA |
844 | REQ_F_SINGLE_POLL_BIT, |
845 | REQ_F_DOUBLE_POLL_BIT, | |
8a3e8ee5 | 846 | REQ_F_PARTIAL_IO_BIT, |
2caf9822 | 847 | REQ_F_CQE32_INIT_BIT, |
227685eb | 848 | REQ_F_APOLL_MULTISHOT_BIT, |
7b29f92d | 849 | /* keep async read/write and isreg together and in order */ |
35645ac3 | 850 | REQ_F_SUPPORT_NOWAIT_BIT, |
7b29f92d | 851 | REQ_F_ISREG_BIT, |
84557871 JA |
852 | |
853 | /* not a real bit, just to check we're not overflowing the space */ | |
854 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
855 | }; |
856 | ||
857 | enum { | |
858 | /* ctx owns file */ | |
859 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
860 | /* drain existing IO first */ | |
861 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
862 | /* linked sqes */ | |
863 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
864 | /* doesn't sever on completion < 0 */ | |
865 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
866 | /* IOSQE_ASYNC */ | |
867 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
868 | /* IOSQE_BUFFER_SELECT */ |
869 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
04c76b41 PB |
870 | /* IOSQE_CQE_SKIP_SUCCESS */ |
871 | REQ_F_CQE_SKIP = BIT(REQ_F_CQE_SKIP_BIT), | |
6b47ee6e | 872 | |
6b47ee6e | 873 | /* fail rest of links */ |
93d2bcd2 | 874 | REQ_F_FAIL = BIT(REQ_F_FAIL_BIT), |
b05a1bcd | 875 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
876 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
877 | /* read/write uses file position */ | |
878 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
879 | /* must not punt to workers */ | |
880 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 881 | /* has or had linked timeout */ |
6b47ee6e | 882 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
883 | /* needs cleanup */ |
884 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
885 | /* already went through poll handler */ |
886 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
887 | /* buffer already selected */ |
888 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
c7fb1942 JA |
889 | /* buffer selected from ring, needs commit */ |
890 | REQ_F_BUFFER_RING = BIT(REQ_F_BUFFER_RING_BIT), | |
e342c807 PB |
891 | /* completion is deferred through io_comp_state */ |
892 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
893 | /* caller should reissue async */ |
894 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
35645ac3 PB |
895 | /* supports async reads/writes */ |
896 | REQ_F_SUPPORT_NOWAIT = BIT(REQ_F_SUPPORT_NOWAIT_BIT), | |
7b29f92d JA |
897 | /* regular file */ |
898 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
b8e64b53 PB |
899 | /* has creds assigned */ |
900 | REQ_F_CREDS = BIT(REQ_F_CREDS_BIT), | |
20e60a38 PB |
901 | /* skip refcounting if not set */ |
902 | REQ_F_REFCOUNT = BIT(REQ_F_REFCOUNT_BIT), | |
4d13d1a4 PB |
903 | /* there is a linked timeout that has to be armed */ |
904 | REQ_F_ARM_LTIMEOUT = BIT(REQ_F_ARM_LTIMEOUT_BIT), | |
d886e185 PB |
905 | /* ->async_data allocated */ |
906 | REQ_F_ASYNC_DATA = BIT(REQ_F_ASYNC_DATA_BIT), | |
04c76b41 PB |
907 | /* don't post CQEs while failing linked requests */ |
908 | REQ_F_SKIP_LINK_CQES = BIT(REQ_F_SKIP_LINK_CQES_BIT), | |
91eac1c6 JA |
909 | /* single poll may be active */ |
910 | REQ_F_SINGLE_POLL = BIT(REQ_F_SINGLE_POLL_BIT), | |
911 | /* double poll may active */ | |
912 | REQ_F_DOUBLE_POLL = BIT(REQ_F_DOUBLE_POLL_BIT), | |
8a3e8ee5 JA |
913 | /* request has already done partial IO */ |
914 | REQ_F_PARTIAL_IO = BIT(REQ_F_PARTIAL_IO_BIT), | |
227685eb HX |
915 | /* fast poll multishot mode */ |
916 | REQ_F_APOLL_MULTISHOT = BIT(REQ_F_APOLL_MULTISHOT_BIT), | |
2caf9822 PB |
917 | /* ->extra1 and ->extra2 are initialised */ |
918 | REQ_F_CQE32_INIT = BIT(REQ_F_CQE32_INIT_BIT), | |
d7718a9d JA |
919 | }; |
920 | ||
921 | struct async_poll { | |
922 | struct io_poll_iocb poll; | |
807abcb0 | 923 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
924 | }; |
925 | ||
f237c30a | 926 | typedef void (*io_req_tw_func_t)(struct io_kiocb *req, bool *locked); |
5b0a6acc | 927 | |
7cbf1722 | 928 | struct io_task_work { |
5b0a6acc PB |
929 | union { |
930 | struct io_wq_work_node node; | |
931 | struct llist_node fallback_node; | |
932 | }; | |
933 | io_req_tw_func_t func; | |
7cbf1722 JA |
934 | }; |
935 | ||
992da01a PB |
936 | enum { |
937 | IORING_RSRC_FILE = 0, | |
938 | IORING_RSRC_BUFFER = 1, | |
939 | }; | |
940 | ||
cef216fc PB |
941 | struct io_cqe { |
942 | __u64 user_data; | |
943 | __s32 res; | |
944 | /* fd initially, then cflags for completion */ | |
945 | union { | |
946 | __u32 flags; | |
947 | int fd; | |
948 | }; | |
949 | }; | |
950 | ||
10988a0a DY |
951 | enum { |
952 | IO_CHECK_CQ_OVERFLOW_BIT, | |
155bc950 | 953 | IO_CHECK_CQ_DROPPED_BIT, |
10988a0a DY |
954 | }; |
955 | ||
09bb8394 JA |
956 | /* |
957 | * NOTE! Each of the iocb union members has the file pointer | |
958 | * as the first entry in their struct definition. So you can | |
959 | * access the file pointer through any of the sub-structs, | |
63c36549 | 960 | * or directly as just 'file' in this struct. |
09bb8394 | 961 | */ |
2b188cc1 | 962 | struct io_kiocb { |
221c5eb2 | 963 | union { |
09bb8394 | 964 | struct file *file; |
9adbd45d | 965 | struct io_rw rw; |
221c5eb2 | 966 | struct io_poll_iocb poll; |
9d805892 | 967 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
968 | struct io_accept accept; |
969 | struct io_sync sync; | |
fbf23849 | 970 | struct io_cancel cancel; |
b29472ee | 971 | struct io_timeout timeout; |
0bdf7a2d | 972 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 973 | struct io_connect connect; |
e47293fd | 974 | struct io_sr_msg sr_msg; |
15b71abe | 975 | struct io_open open; |
b5dba59e | 976 | struct io_close close; |
269bbe5f | 977 | struct io_rsrc_update rsrc_update; |
4840e418 | 978 | struct io_fadvise fadvise; |
c1ca757b | 979 | struct io_madvise madvise; |
3e4827b0 | 980 | struct io_epoll epoll; |
7d67af2c | 981 | struct io_splice splice; |
ddf0322d | 982 | struct io_provide_buf pbuf; |
1d9e1288 | 983 | struct io_statx statx; |
36f4fa68 | 984 | struct io_shutdown shutdown; |
80a261fd | 985 | struct io_rename rename; |
14a1143b | 986 | struct io_unlink unlink; |
e34a02dc | 987 | struct io_mkdir mkdir; |
7a8721f8 | 988 | struct io_symlink symlink; |
cf30da90 | 989 | struct io_hardlink hardlink; |
4f57f06c | 990 | struct io_msg msg; |
e9621e2b | 991 | struct io_xattr xattr; |
1374e08e | 992 | struct io_socket sock; |
ee692a21 | 993 | struct io_uring_cmd uring_cmd; |
221c5eb2 | 994 | }; |
2b188cc1 | 995 | |
d625c6ee | 996 | u8 opcode; |
65a6543d XW |
997 | /* polled IO has completed */ |
998 | u8 iopoll_completed; | |
1dbd023e JA |
999 | /* |
1000 | * Can be either a fixed buffer index, or used with provided buffers. | |
1001 | * For the latter, before issue it points to the buffer group ID, | |
1002 | * and after selection it points to the buffer ID itself. | |
1003 | */ | |
4f4eeba8 | 1004 | u16 buf_index; |
d17e56eb PB |
1005 | unsigned int flags; |
1006 | ||
cef216fc | 1007 | struct io_cqe cqe; |
4f4eeba8 | 1008 | |
010e8e6b | 1009 | struct io_ring_ctx *ctx; |
010e8e6b | 1010 | struct task_struct *task; |
d7718a9d | 1011 | |
c1bdf8ed | 1012 | struct io_rsrc_node *rsrc_node; |
a4f8d94c JA |
1013 | |
1014 | union { | |
1015 | /* store used ubuf, so we can prevent reloading */ | |
1016 | struct io_mapped_ubuf *imu; | |
1017 | ||
1018 | /* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */ | |
1019 | struct io_buffer *kbuf; | |
c7fb1942 JA |
1020 | |
1021 | /* | |
1022 | * stores buffer ID for ring provided buffers, valid IFF | |
1023 | * REQ_F_BUFFER_RING is set. | |
1024 | */ | |
1025 | struct io_buffer_list *buf_list; | |
a4f8d94c | 1026 | }; |
fcb323cc | 1027 | |
2804ecd8 JA |
1028 | union { |
1029 | /* used by request caches, completion batching and iopoll */ | |
1030 | struct io_wq_work_node comp_list; | |
1031 | /* cache ->apoll->events */ | |
58f5c8d3 | 1032 | __poll_t apoll_events; |
2804ecd8 | 1033 | }; |
d17e56eb | 1034 | atomic_t refs; |
521d61fc | 1035 | atomic_t poll_refs; |
5b0a6acc | 1036 | struct io_task_work io_task_work; |
010e8e6b | 1037 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
4e5bc0a9 SR |
1038 | union { |
1039 | struct hlist_node hash_node; | |
1040 | struct { | |
1041 | u64 extra1; | |
1042 | u64 extra2; | |
1043 | }; | |
1044 | }; | |
7e3709d5 | 1045 | /* internal polling, see IORING_FEAT_FAST_POLL */ |
010e8e6b | 1046 | struct async_poll *apoll; |
d886e185 PB |
1047 | /* opcode allocated if it needs to store data for async defer */ |
1048 | void *async_data; | |
41cdcc22 | 1049 | /* linked requests, IFF REQ_F_HARDLINK or REQ_F_LINK are set */ |
34d2bfe7 | 1050 | struct io_kiocb *link; |
41cdcc22 | 1051 | /* custom credentials, valid IFF REQ_F_CREDS is set */ |
521d61fc JA |
1052 | const struct cred *creds; |
1053 | struct io_wq_work work; | |
2b188cc1 | 1054 | }; |
05589553 | 1055 | |
13bf43f5 PB |
1056 | struct io_tctx_node { |
1057 | struct list_head ctx_node; | |
1058 | struct task_struct *task; | |
13bf43f5 PB |
1059 | struct io_ring_ctx *ctx; |
1060 | }; | |
1061 | ||
27dc8338 PB |
1062 | struct io_defer_entry { |
1063 | struct list_head list; | |
1064 | struct io_kiocb *req; | |
9cf7c104 | 1065 | u32 seq; |
2b188cc1 JA |
1066 | }; |
1067 | ||
b21432b4 JA |
1068 | struct io_cancel_data { |
1069 | struct io_ring_ctx *ctx; | |
4bf94615 JA |
1070 | union { |
1071 | u64 data; | |
1072 | struct file *file; | |
1073 | }; | |
8e29da69 JA |
1074 | u32 flags; |
1075 | int seq; | |
b21432b4 JA |
1076 | }; |
1077 | ||
ee692a21 JA |
1078 | /* |
1079 | * The URING_CMD payload starts at 'cmd' in the first sqe, and continues into | |
1080 | * the following sqe if SQE128 is used. | |
1081 | */ | |
1082 | #define uring_cmd_pdu_size(is_sqe128) \ | |
1083 | ((1 + !!(is_sqe128)) * sizeof(struct io_uring_sqe) - \ | |
1084 | offsetof(struct io_uring_sqe, cmd)) | |
1085 | ||
d3656344 | 1086 | struct io_op_def { |
d3656344 JA |
1087 | /* needs req->file assigned */ |
1088 | unsigned needs_file : 1; | |
6d63416d PB |
1089 | /* should block plug */ |
1090 | unsigned plug : 1; | |
d3656344 JA |
1091 | /* hash wq insertion if file is a regular file */ |
1092 | unsigned hash_reg_file : 1; | |
1093 | /* unbound wq insertion if file is a non-regular file */ | |
1094 | unsigned unbound_nonreg_file : 1; | |
8a72758c JA |
1095 | /* set if opcode supports polled "wait" */ |
1096 | unsigned pollin : 1; | |
1097 | unsigned pollout : 1; | |
52dd8640 | 1098 | unsigned poll_exclusive : 1; |
bcda7baa JA |
1099 | /* op supports buffer selection */ |
1100 | unsigned buffer_select : 1; | |
26f0505a PB |
1101 | /* do prep async if is going to be punted */ |
1102 | unsigned needs_async_setup : 1; | |
6d63416d PB |
1103 | /* opcode is not supported by this kernel */ |
1104 | unsigned not_supported : 1; | |
5bd2182d PM |
1105 | /* skip auditing */ |
1106 | unsigned audit_skip : 1; | |
73911426 JA |
1107 | /* supports ioprio */ |
1108 | unsigned ioprio : 1; | |
1109 | /* supports iopoll */ | |
1110 | unsigned iopoll : 1; | |
e8c2bc1f JA |
1111 | /* size of async data needed, if any */ |
1112 | unsigned short async_size; | |
d3656344 JA |
1113 | }; |
1114 | ||
0918682b | 1115 | static const struct io_op_def io_op_defs[] = { |
73911426 JA |
1116 | [IORING_OP_NOP] = { |
1117 | .audit_skip = 1, | |
1118 | .iopoll = 1, | |
1119 | }, | |
0463b6c5 | 1120 | [IORING_OP_READV] = { |
d3656344 JA |
1121 | .needs_file = 1, |
1122 | .unbound_nonreg_file = 1, | |
8a72758c | 1123 | .pollin = 1, |
4d954c25 | 1124 | .buffer_select = 1, |
26f0505a | 1125 | .needs_async_setup = 1, |
27926b68 | 1126 | .plug = 1, |
5bd2182d | 1127 | .audit_skip = 1, |
73911426 JA |
1128 | .ioprio = 1, |
1129 | .iopoll = 1, | |
e8c2bc1f | 1130 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 1131 | }, |
0463b6c5 | 1132 | [IORING_OP_WRITEV] = { |
d3656344 JA |
1133 | .needs_file = 1, |
1134 | .hash_reg_file = 1, | |
1135 | .unbound_nonreg_file = 1, | |
8a72758c | 1136 | .pollout = 1, |
26f0505a | 1137 | .needs_async_setup = 1, |
27926b68 | 1138 | .plug = 1, |
5bd2182d | 1139 | .audit_skip = 1, |
73911426 JA |
1140 | .ioprio = 1, |
1141 | .iopoll = 1, | |
e8c2bc1f | 1142 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 1143 | }, |
0463b6c5 | 1144 | [IORING_OP_FSYNC] = { |
d3656344 | 1145 | .needs_file = 1, |
5bd2182d | 1146 | .audit_skip = 1, |
d3656344 | 1147 | }, |
0463b6c5 | 1148 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
1149 | .needs_file = 1, |
1150 | .unbound_nonreg_file = 1, | |
8a72758c | 1151 | .pollin = 1, |
27926b68 | 1152 | .plug = 1, |
5bd2182d | 1153 | .audit_skip = 1, |
73911426 JA |
1154 | .ioprio = 1, |
1155 | .iopoll = 1, | |
e8c2bc1f | 1156 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 1157 | }, |
0463b6c5 | 1158 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
1159 | .needs_file = 1, |
1160 | .hash_reg_file = 1, | |
1161 | .unbound_nonreg_file = 1, | |
8a72758c | 1162 | .pollout = 1, |
27926b68 | 1163 | .plug = 1, |
5bd2182d | 1164 | .audit_skip = 1, |
73911426 JA |
1165 | .ioprio = 1, |
1166 | .iopoll = 1, | |
e8c2bc1f | 1167 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 1168 | }, |
0463b6c5 | 1169 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
1170 | .needs_file = 1, |
1171 | .unbound_nonreg_file = 1, | |
5bd2182d PM |
1172 | .audit_skip = 1, |
1173 | }, | |
1174 | [IORING_OP_POLL_REMOVE] = { | |
1175 | .audit_skip = 1, | |
d3656344 | 1176 | }, |
0463b6c5 | 1177 | [IORING_OP_SYNC_FILE_RANGE] = { |
d3656344 | 1178 | .needs_file = 1, |
5bd2182d | 1179 | .audit_skip = 1, |
d3656344 | 1180 | }, |
0463b6c5 | 1181 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
1182 | .needs_file = 1, |
1183 | .unbound_nonreg_file = 1, | |
8a72758c | 1184 | .pollout = 1, |
26f0505a | 1185 | .needs_async_setup = 1, |
e8c2bc1f | 1186 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 1187 | }, |
0463b6c5 | 1188 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
1189 | .needs_file = 1, |
1190 | .unbound_nonreg_file = 1, | |
8a72758c | 1191 | .pollin = 1, |
52de1fe1 | 1192 | .buffer_select = 1, |
26f0505a | 1193 | .needs_async_setup = 1, |
e8c2bc1f | 1194 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 1195 | }, |
0463b6c5 | 1196 | [IORING_OP_TIMEOUT] = { |
5bd2182d | 1197 | .audit_skip = 1, |
e8c2bc1f | 1198 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 1199 | }, |
9c8e11b3 PB |
1200 | [IORING_OP_TIMEOUT_REMOVE] = { |
1201 | /* used by timeout updates' prep() */ | |
5bd2182d | 1202 | .audit_skip = 1, |
9c8e11b3 | 1203 | }, |
0463b6c5 | 1204 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
1205 | .needs_file = 1, |
1206 | .unbound_nonreg_file = 1, | |
8a72758c | 1207 | .pollin = 1, |
52dd8640 | 1208 | .poll_exclusive = 1, |
4e86a2c9 | 1209 | .ioprio = 1, /* used for flags */ |
d3656344 | 1210 | }, |
5bd2182d PM |
1211 | [IORING_OP_ASYNC_CANCEL] = { |
1212 | .audit_skip = 1, | |
1213 | }, | |
0463b6c5 | 1214 | [IORING_OP_LINK_TIMEOUT] = { |
5bd2182d | 1215 | .audit_skip = 1, |
e8c2bc1f | 1216 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 1217 | }, |
0463b6c5 | 1218 | [IORING_OP_CONNECT] = { |
d3656344 JA |
1219 | .needs_file = 1, |
1220 | .unbound_nonreg_file = 1, | |
8a72758c | 1221 | .pollout = 1, |
26f0505a | 1222 | .needs_async_setup = 1, |
e8c2bc1f | 1223 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 1224 | }, |
0463b6c5 | 1225 | [IORING_OP_FALLOCATE] = { |
d3656344 | 1226 | .needs_file = 1, |
d3656344 | 1227 | }, |
44526bed JA |
1228 | [IORING_OP_OPENAT] = {}, |
1229 | [IORING_OP_CLOSE] = {}, | |
5bd2182d PM |
1230 | [IORING_OP_FILES_UPDATE] = { |
1231 | .audit_skip = 1, | |
73911426 | 1232 | .iopoll = 1, |
5bd2182d PM |
1233 | }, |
1234 | [IORING_OP_STATX] = { | |
1235 | .audit_skip = 1, | |
1236 | }, | |
0463b6c5 | 1237 | [IORING_OP_READ] = { |
3a6820f2 JA |
1238 | .needs_file = 1, |
1239 | .unbound_nonreg_file = 1, | |
8a72758c | 1240 | .pollin = 1, |
bcda7baa | 1241 | .buffer_select = 1, |
27926b68 | 1242 | .plug = 1, |
5bd2182d | 1243 | .audit_skip = 1, |
73911426 JA |
1244 | .ioprio = 1, |
1245 | .iopoll = 1, | |
e8c2bc1f | 1246 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1247 | }, |
0463b6c5 | 1248 | [IORING_OP_WRITE] = { |
3a6820f2 | 1249 | .needs_file = 1, |
7b3188e7 | 1250 | .hash_reg_file = 1, |
3a6820f2 | 1251 | .unbound_nonreg_file = 1, |
8a72758c | 1252 | .pollout = 1, |
27926b68 | 1253 | .plug = 1, |
5bd2182d | 1254 | .audit_skip = 1, |
73911426 JA |
1255 | .ioprio = 1, |
1256 | .iopoll = 1, | |
e8c2bc1f | 1257 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 1258 | }, |
0463b6c5 | 1259 | [IORING_OP_FADVISE] = { |
4840e418 | 1260 | .needs_file = 1, |
5bd2182d | 1261 | .audit_skip = 1, |
c1ca757b | 1262 | }, |
44526bed | 1263 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 1264 | [IORING_OP_SEND] = { |
fddaface JA |
1265 | .needs_file = 1, |
1266 | .unbound_nonreg_file = 1, | |
8a72758c | 1267 | .pollout = 1, |
5bd2182d | 1268 | .audit_skip = 1, |
fddaface | 1269 | }, |
0463b6c5 | 1270 | [IORING_OP_RECV] = { |
fddaface JA |
1271 | .needs_file = 1, |
1272 | .unbound_nonreg_file = 1, | |
8a72758c | 1273 | .pollin = 1, |
bcda7baa | 1274 | .buffer_select = 1, |
5bd2182d | 1275 | .audit_skip = 1, |
fddaface | 1276 | }, |
0463b6c5 | 1277 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1278 | }, |
3e4827b0 JA |
1279 | [IORING_OP_EPOLL_CTL] = { |
1280 | .unbound_nonreg_file = 1, | |
5bd2182d | 1281 | .audit_skip = 1, |
3e4827b0 | 1282 | }, |
7d67af2c PB |
1283 | [IORING_OP_SPLICE] = { |
1284 | .needs_file = 1, | |
1285 | .hash_reg_file = 1, | |
1286 | .unbound_nonreg_file = 1, | |
5bd2182d PM |
1287 | .audit_skip = 1, |
1288 | }, | |
1289 | [IORING_OP_PROVIDE_BUFFERS] = { | |
1290 | .audit_skip = 1, | |
73911426 | 1291 | .iopoll = 1, |
5bd2182d PM |
1292 | }, |
1293 | [IORING_OP_REMOVE_BUFFERS] = { | |
1294 | .audit_skip = 1, | |
73911426 | 1295 | .iopoll = 1, |
ddf0322d | 1296 | }, |
f2a8d5c7 PB |
1297 | [IORING_OP_TEE] = { |
1298 | .needs_file = 1, | |
1299 | .hash_reg_file = 1, | |
1300 | .unbound_nonreg_file = 1, | |
5bd2182d | 1301 | .audit_skip = 1, |
f2a8d5c7 | 1302 | }, |
36f4fa68 JA |
1303 | [IORING_OP_SHUTDOWN] = { |
1304 | .needs_file = 1, | |
1305 | }, | |
44526bed JA |
1306 | [IORING_OP_RENAMEAT] = {}, |
1307 | [IORING_OP_UNLINKAT] = {}, | |
e34a02dc | 1308 | [IORING_OP_MKDIRAT] = {}, |
7a8721f8 | 1309 | [IORING_OP_SYMLINKAT] = {}, |
cf30da90 | 1310 | [IORING_OP_LINKAT] = {}, |
4f57f06c JA |
1311 | [IORING_OP_MSG_RING] = { |
1312 | .needs_file = 1, | |
73911426 | 1313 | .iopoll = 1, |
4f57f06c | 1314 | }, |
e9621e2b SR |
1315 | [IORING_OP_FSETXATTR] = { |
1316 | .needs_file = 1 | |
1317 | }, | |
1318 | [IORING_OP_SETXATTR] = {}, | |
a56834e0 SR |
1319 | [IORING_OP_FGETXATTR] = { |
1320 | .needs_file = 1 | |
1321 | }, | |
1322 | [IORING_OP_GETXATTR] = {}, | |
1374e08e JA |
1323 | [IORING_OP_SOCKET] = { |
1324 | .audit_skip = 1, | |
1325 | }, | |
ee692a21 JA |
1326 | [IORING_OP_URING_CMD] = { |
1327 | .needs_file = 1, | |
1328 | .plug = 1, | |
1329 | .needs_async_setup = 1, | |
1330 | .async_size = uring_cmd_pdu_size(1), | |
1331 | }, | |
d3656344 JA |
1332 | }; |
1333 | ||
0756a869 PB |
1334 | /* requests with any of those set should undergo io_disarm_next() */ |
1335 | #define IO_DISARM_MASK (REQ_F_ARM_LTIMEOUT | REQ_F_LINK_TIMEOUT | REQ_F_FAIL) | |
da1a08c5 | 1336 | #define IO_REQ_LINK_FLAGS (REQ_F_LINK | REQ_F_HARDLINK) |
0756a869 | 1337 | |
7a612350 | 1338 | static bool io_disarm_next(struct io_kiocb *req); |
eef51daa | 1339 | static void io_uring_del_tctx_node(unsigned long index); |
9936c7c2 PB |
1340 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1341 | struct task_struct *task, | |
3dd0c97a | 1342 | bool cancel_all); |
78cc687b | 1343 | static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); |
1ffc5422 | 1344 | |
4e118cd9 | 1345 | static void __io_req_complete_post(struct io_kiocb *req, s32 res, u32 cflags); |
c7dae4ba | 1346 | static void io_dismantle_req(struct io_kiocb *req); |
94ae5e77 | 1347 | static void io_queue_linked_timeout(struct io_kiocb *req); |
fdecb662 | 1348 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1349 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 1350 | unsigned nr_args); |
68fb8979 | 1351 | static void io_clean_op(struct io_kiocb *req); |
5106dd6e JA |
1352 | static inline struct file *io_file_get_fixed(struct io_kiocb *req, int fd, |
1353 | unsigned issue_flags); | |
ee67ba3b | 1354 | static struct file *io_file_get_normal(struct io_kiocb *req, int fd); |
cbc2e203 | 1355 | static void io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1356 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1357 | |
907d1df3 | 1358 | static void io_req_task_queue(struct io_kiocb *req); |
c450178d | 1359 | static void __io_submit_flush_completions(struct io_ring_ctx *ctx); |
179ae0d1 | 1360 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1361 | |
b9445598 PB |
1362 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
1363 | unsigned int issue_flags, u32 slot_index); | |
a7c41b46 XW |
1364 | static int __io_close_fixed(struct io_kiocb *req, unsigned int issue_flags, |
1365 | unsigned int offset); | |
1366 | static inline int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags); | |
7df778be | 1367 | |
f1042b6c | 1368 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); |
9aa8dfde | 1369 | static void io_eventfd_signal(struct io_ring_ctx *ctx); |
4e118cd9 | 1370 | static void io_req_tw_post_queue(struct io_kiocb *req, s32 res, u32 cflags); |
b9445598 | 1371 | |
2b188cc1 JA |
1372 | static struct kmem_cache *req_cachep; |
1373 | ||
0918682b | 1374 | static const struct file_operations io_uring_fops; |
2b188cc1 | 1375 | |
33337d03 DY |
1376 | const char *io_uring_get_opcode(u8 opcode) |
1377 | { | |
1378 | switch ((enum io_uring_op)opcode) { | |
1379 | case IORING_OP_NOP: | |
1380 | return "NOP"; | |
1381 | case IORING_OP_READV: | |
1382 | return "READV"; | |
1383 | case IORING_OP_WRITEV: | |
1384 | return "WRITEV"; | |
1385 | case IORING_OP_FSYNC: | |
1386 | return "FSYNC"; | |
1387 | case IORING_OP_READ_FIXED: | |
1388 | return "READ_FIXED"; | |
1389 | case IORING_OP_WRITE_FIXED: | |
1390 | return "WRITE_FIXED"; | |
1391 | case IORING_OP_POLL_ADD: | |
1392 | return "POLL_ADD"; | |
1393 | case IORING_OP_POLL_REMOVE: | |
1394 | return "POLL_REMOVE"; | |
1395 | case IORING_OP_SYNC_FILE_RANGE: | |
1396 | return "SYNC_FILE_RANGE"; | |
1397 | case IORING_OP_SENDMSG: | |
1398 | return "SENDMSG"; | |
1399 | case IORING_OP_RECVMSG: | |
1400 | return "RECVMSG"; | |
1401 | case IORING_OP_TIMEOUT: | |
1402 | return "TIMEOUT"; | |
1403 | case IORING_OP_TIMEOUT_REMOVE: | |
1404 | return "TIMEOUT_REMOVE"; | |
1405 | case IORING_OP_ACCEPT: | |
1406 | return "ACCEPT"; | |
1407 | case IORING_OP_ASYNC_CANCEL: | |
1408 | return "ASYNC_CANCEL"; | |
1409 | case IORING_OP_LINK_TIMEOUT: | |
1410 | return "LINK_TIMEOUT"; | |
1411 | case IORING_OP_CONNECT: | |
1412 | return "CONNECT"; | |
1413 | case IORING_OP_FALLOCATE: | |
1414 | return "FALLOCATE"; | |
1415 | case IORING_OP_OPENAT: | |
1416 | return "OPENAT"; | |
1417 | case IORING_OP_CLOSE: | |
1418 | return "CLOSE"; | |
1419 | case IORING_OP_FILES_UPDATE: | |
1420 | return "FILES_UPDATE"; | |
1421 | case IORING_OP_STATX: | |
1422 | return "STATX"; | |
1423 | case IORING_OP_READ: | |
1424 | return "READ"; | |
1425 | case IORING_OP_WRITE: | |
1426 | return "WRITE"; | |
1427 | case IORING_OP_FADVISE: | |
1428 | return "FADVISE"; | |
1429 | case IORING_OP_MADVISE: | |
1430 | return "MADVISE"; | |
1431 | case IORING_OP_SEND: | |
1432 | return "SEND"; | |
1433 | case IORING_OP_RECV: | |
1434 | return "RECV"; | |
1435 | case IORING_OP_OPENAT2: | |
1436 | return "OPENAT2"; | |
1437 | case IORING_OP_EPOLL_CTL: | |
1438 | return "EPOLL_CTL"; | |
1439 | case IORING_OP_SPLICE: | |
1440 | return "SPLICE"; | |
1441 | case IORING_OP_PROVIDE_BUFFERS: | |
1442 | return "PROVIDE_BUFFERS"; | |
1443 | case IORING_OP_REMOVE_BUFFERS: | |
1444 | return "REMOVE_BUFFERS"; | |
1445 | case IORING_OP_TEE: | |
1446 | return "TEE"; | |
1447 | case IORING_OP_SHUTDOWN: | |
1448 | return "SHUTDOWN"; | |
1449 | case IORING_OP_RENAMEAT: | |
1450 | return "RENAMEAT"; | |
1451 | case IORING_OP_UNLINKAT: | |
1452 | return "UNLINKAT"; | |
1453 | case IORING_OP_MKDIRAT: | |
1454 | return "MKDIRAT"; | |
1455 | case IORING_OP_SYMLINKAT: | |
1456 | return "SYMLINKAT"; | |
1457 | case IORING_OP_LINKAT: | |
1458 | return "LINKAT"; | |
1459 | case IORING_OP_MSG_RING: | |
1460 | return "MSG_RING"; | |
1461 | case IORING_OP_FSETXATTR: | |
1462 | return "FSETXATTR"; | |
1463 | case IORING_OP_SETXATTR: | |
1464 | return "SETXATTR"; | |
1465 | case IORING_OP_FGETXATTR: | |
1466 | return "FGETXATTR"; | |
1467 | case IORING_OP_GETXATTR: | |
1468 | return "GETXATTR"; | |
1469 | case IORING_OP_SOCKET: | |
1470 | return "SOCKET"; | |
ee692a21 JA |
1471 | case IORING_OP_URING_CMD: |
1472 | return "URING_CMD"; | |
33337d03 DY |
1473 | case IORING_OP_LAST: |
1474 | return "INVALID"; | |
1475 | } | |
1476 | return "INVALID"; | |
1477 | } | |
1478 | ||
2b188cc1 JA |
1479 | struct sock *io_uring_get_socket(struct file *file) |
1480 | { | |
1481 | #if defined(CONFIG_UNIX) | |
1482 | if (file->f_op == &io_uring_fops) { | |
1483 | struct io_ring_ctx *ctx = file->private_data; | |
1484 | ||
1485 | return ctx->ring_sock->sk; | |
1486 | } | |
1487 | #endif | |
1488 | return NULL; | |
1489 | } | |
1490 | EXPORT_SYMBOL(io_uring_get_socket); | |
1491 | ||
1f59bc0f PB |
1492 | #if defined(CONFIG_UNIX) |
1493 | static inline bool io_file_need_scm(struct file *filp) | |
1494 | { | |
5e45690a JA |
1495 | #if defined(IO_URING_SCM_ALL) |
1496 | return true; | |
1497 | #else | |
1f59bc0f | 1498 | return !!unix_get_socket(filp); |
5e45690a | 1499 | #endif |
1f59bc0f PB |
1500 | } |
1501 | #else | |
1502 | static inline bool io_file_need_scm(struct file *filp) | |
1503 | { | |
5e45690a | 1504 | return false; |
1f59bc0f PB |
1505 | } |
1506 | #endif | |
1507 | ||
f8929630 PB |
1508 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, unsigned issue_flags) |
1509 | { | |
1510 | lockdep_assert_held(&ctx->uring_lock); | |
1511 | if (issue_flags & IO_URING_F_UNLOCKED) | |
1512 | mutex_unlock(&ctx->uring_lock); | |
1513 | } | |
1514 | ||
1515 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, unsigned issue_flags) | |
1516 | { | |
1517 | /* | |
1518 | * "Normal" inline submissions always hold the uring_lock, since we | |
1519 | * grab it from the system call. Same is true for the SQPOLL offload. | |
1520 | * The only exception is when we've detached the request and issue it | |
1521 | * from an async worker thread, grab the lock for that case. | |
1522 | */ | |
1523 | if (issue_flags & IO_URING_F_UNLOCKED) | |
1524 | mutex_lock(&ctx->uring_lock); | |
1525 | lockdep_assert_held(&ctx->uring_lock); | |
1526 | } | |
1527 | ||
f237c30a PB |
1528 | static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) |
1529 | { | |
1530 | if (!*locked) { | |
1531 | mutex_lock(&ctx->uring_lock); | |
1532 | *locked = true; | |
1533 | } | |
1534 | } | |
1535 | ||
f2f87370 PB |
1536 | #define io_for_each_link(pos, head) \ |
1537 | for (pos = (head); pos; pos = pos->link) | |
1538 | ||
21c843d5 PB |
1539 | /* |
1540 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1541 | * see commit f958d7b528b1 for details. | |
1542 | */ | |
1543 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1544 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1545 | ||
1546 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) | |
1547 | { | |
20e60a38 | 1548 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1549 | return atomic_inc_not_zero(&req->refs); |
1550 | } | |
1551 | ||
21c843d5 PB |
1552 | static inline bool req_ref_put_and_test(struct io_kiocb *req) |
1553 | { | |
20e60a38 PB |
1554 | if (likely(!(req->flags & REQ_F_REFCOUNT))) |
1555 | return true; | |
1556 | ||
21c843d5 PB |
1557 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1558 | return atomic_dec_and_test(&req->refs); | |
1559 | } | |
1560 | ||
21c843d5 PB |
1561 | static inline void req_ref_get(struct io_kiocb *req) |
1562 | { | |
20e60a38 | 1563 | WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); |
21c843d5 PB |
1564 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1565 | atomic_inc(&req->refs); | |
1566 | } | |
1567 | ||
c450178d PB |
1568 | static inline void io_submit_flush_completions(struct io_ring_ctx *ctx) |
1569 | { | |
6f33b0bc | 1570 | if (!wq_list_empty(&ctx->submit_state.compl_reqs)) |
c450178d PB |
1571 | __io_submit_flush_completions(ctx); |
1572 | } | |
1573 | ||
48dcd38d | 1574 | static inline void __io_req_set_refcount(struct io_kiocb *req, int nr) |
20e60a38 PB |
1575 | { |
1576 | if (!(req->flags & REQ_F_REFCOUNT)) { | |
1577 | req->flags |= REQ_F_REFCOUNT; | |
48dcd38d | 1578 | atomic_set(&req->refs, nr); |
20e60a38 PB |
1579 | } |
1580 | } | |
1581 | ||
48dcd38d PB |
1582 | static inline void io_req_set_refcount(struct io_kiocb *req) |
1583 | { | |
1584 | __io_req_set_refcount(req, 1); | |
1585 | } | |
1586 | ||
ab409402 PB |
1587 | #define IO_RSRC_REF_BATCH 100 |
1588 | ||
25a15d3c PB |
1589 | static void io_rsrc_put_node(struct io_rsrc_node *node, int nr) |
1590 | { | |
1591 | percpu_ref_put_many(&node->refs, nr); | |
1592 | } | |
1593 | ||
ab409402 PB |
1594 | static inline void io_req_put_rsrc_locked(struct io_kiocb *req, |
1595 | struct io_ring_ctx *ctx) | |
1596 | __must_hold(&ctx->uring_lock) | |
36f72fe2 | 1597 | { |
c1bdf8ed | 1598 | struct io_rsrc_node *node = req->rsrc_node; |
ab409402 | 1599 | |
c1bdf8ed PB |
1600 | if (node) { |
1601 | if (node == ctx->rsrc_node) | |
ab409402 PB |
1602 | ctx->rsrc_cached_refs++; |
1603 | else | |
25a15d3c | 1604 | io_rsrc_put_node(node, 1); |
ab409402 PB |
1605 | } |
1606 | } | |
1607 | ||
7ac1edc4 | 1608 | static inline void io_req_put_rsrc(struct io_kiocb *req) |
ab409402 | 1609 | { |
c1bdf8ed | 1610 | if (req->rsrc_node) |
25a15d3c | 1611 | io_rsrc_put_node(req->rsrc_node, 1); |
ab409402 PB |
1612 | } |
1613 | ||
1614 | static __cold void io_rsrc_refs_drop(struct io_ring_ctx *ctx) | |
1615 | __must_hold(&ctx->uring_lock) | |
1616 | { | |
1617 | if (ctx->rsrc_cached_refs) { | |
25a15d3c | 1618 | io_rsrc_put_node(ctx->rsrc_node, ctx->rsrc_cached_refs); |
ab409402 PB |
1619 | ctx->rsrc_cached_refs = 0; |
1620 | } | |
1621 | } | |
1622 | ||
1623 | static void io_rsrc_refs_refill(struct io_ring_ctx *ctx) | |
1624 | __must_hold(&ctx->uring_lock) | |
1625 | { | |
1626 | ctx->rsrc_cached_refs += IO_RSRC_REF_BATCH; | |
1627 | percpu_ref_get_many(&ctx->rsrc_node->refs, IO_RSRC_REF_BATCH); | |
1628 | } | |
36f72fe2 | 1629 | |
a46be971 | 1630 | static inline void io_req_set_rsrc_node(struct io_kiocb *req, |
5106dd6e JA |
1631 | struct io_ring_ctx *ctx, |
1632 | unsigned int issue_flags) | |
36f72fe2 | 1633 | { |
c1bdf8ed PB |
1634 | if (!req->rsrc_node) { |
1635 | req->rsrc_node = ctx->rsrc_node; | |
5106dd6e JA |
1636 | |
1637 | if (!(issue_flags & IO_URING_F_UNLOCKED)) { | |
1638 | lockdep_assert_held(&ctx->uring_lock); | |
1639 | ctx->rsrc_cached_refs--; | |
1640 | if (unlikely(ctx->rsrc_cached_refs < 0)) | |
1641 | io_rsrc_refs_refill(ctx); | |
1642 | } else { | |
c1bdf8ed | 1643 | percpu_ref_get(&req->rsrc_node->refs); |
5106dd6e | 1644 | } |
36f72fe2 PB |
1645 | } |
1646 | } | |
1647 | ||
cc3cec83 | 1648 | static unsigned int __io_put_kbuf(struct io_kiocb *req, struct list_head *list) |
3648e526 | 1649 | { |
c7fb1942 JA |
1650 | if (req->flags & REQ_F_BUFFER_RING) { |
1651 | if (req->buf_list) | |
1652 | req->buf_list->head++; | |
1653 | req->flags &= ~REQ_F_BUFFER_RING; | |
1654 | } else { | |
1655 | list_add(&req->kbuf->list, list); | |
1656 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
1657 | } | |
3648e526 | 1658 | |
1dbd023e | 1659 | return IORING_CQE_F_BUFFER | (req->buf_index << IORING_CQE_BUFFER_SHIFT); |
3648e526 HX |
1660 | } |
1661 | ||
cc3cec83 | 1662 | static inline unsigned int io_put_kbuf_comp(struct io_kiocb *req) |
3648e526 | 1663 | { |
8197b053 PB |
1664 | lockdep_assert_held(&req->ctx->completion_lock); |
1665 | ||
c7fb1942 | 1666 | if (!(req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING))) |
3648e526 | 1667 | return 0; |
cc3cec83 JA |
1668 | return __io_put_kbuf(req, &req->ctx->io_buffers_comp); |
1669 | } | |
1670 | ||
1671 | static inline unsigned int io_put_kbuf(struct io_kiocb *req, | |
1672 | unsigned issue_flags) | |
1673 | { | |
1674 | unsigned int cflags; | |
1675 | ||
c7fb1942 | 1676 | if (!(req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING))) |
cc3cec83 JA |
1677 | return 0; |
1678 | ||
1679 | /* | |
1680 | * We can add this buffer back to two lists: | |
1681 | * | |
1682 | * 1) The io_buffers_cache list. This one is protected by the | |
1683 | * ctx->uring_lock. If we already hold this lock, add back to this | |
1684 | * list as we can grab it from issue as well. | |
1685 | * 2) The io_buffers_comp list. This one is protected by the | |
1686 | * ctx->completion_lock. | |
1687 | * | |
1688 | * We migrate buffers from the comp_list to the issue cache list | |
1689 | * when we need one. | |
1690 | */ | |
c7fb1942 JA |
1691 | if (req->flags & REQ_F_BUFFER_RING) { |
1692 | /* no buffers to recycle for this case */ | |
1693 | cflags = __io_put_kbuf(req, NULL); | |
1694 | } else if (issue_flags & IO_URING_F_UNLOCKED) { | |
cc3cec83 JA |
1695 | struct io_ring_ctx *ctx = req->ctx; |
1696 | ||
1697 | spin_lock(&ctx->completion_lock); | |
1698 | cflags = __io_put_kbuf(req, &ctx->io_buffers_comp); | |
1699 | spin_unlock(&ctx->completion_lock); | |
1700 | } else { | |
ab0ac095 PB |
1701 | lockdep_assert_held(&req->ctx->uring_lock); |
1702 | ||
cc3cec83 JA |
1703 | cflags = __io_put_kbuf(req, &req->ctx->io_buffers_cache); |
1704 | } | |
1705 | ||
1706 | return cflags; | |
3648e526 HX |
1707 | } |
1708 | ||
dbc7d452 JA |
1709 | static struct io_buffer_list *io_buffer_get_list(struct io_ring_ctx *ctx, |
1710 | unsigned int bgid) | |
1711 | { | |
9cfc7e94 JA |
1712 | if (ctx->io_bl && bgid < BGID_ARRAY) |
1713 | return &ctx->io_bl[bgid]; | |
dbc7d452 | 1714 | |
9cfc7e94 | 1715 | return xa_load(&ctx->io_bl_xa, bgid); |
dbc7d452 JA |
1716 | } |
1717 | ||
4d55f238 | 1718 | static void io_kbuf_recycle(struct io_kiocb *req, unsigned issue_flags) |
b1c62645 JA |
1719 | { |
1720 | struct io_ring_ctx *ctx = req->ctx; | |
dbc7d452 JA |
1721 | struct io_buffer_list *bl; |
1722 | struct io_buffer *buf; | |
b1c62645 | 1723 | |
c7fb1942 | 1724 | if (!(req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING))) |
b1c62645 | 1725 | return; |
42db0c00 HX |
1726 | /* |
1727 | * For legacy provided buffer mode, don't recycle if we already did | |
1728 | * IO to this buffer. For ring-mapped provided buffer mode, we should | |
1729 | * increment ring->head to explicitly monopolize the buffer to avoid | |
1730 | * multiple use. | |
1731 | */ | |
1732 | if ((req->flags & REQ_F_BUFFER_SELECTED) && | |
1733 | (req->flags & REQ_F_PARTIAL_IO)) | |
8a3e8ee5 | 1734 | return; |
42db0c00 | 1735 | |
c7fb1942 JA |
1736 | /* |
1737 | * We don't need to recycle for REQ_F_BUFFER_RING, we can just clear | |
1738 | * the flag and hence ensure that bl->head doesn't get incremented. | |
1739 | * If the tail has already been incremented, hang on to it. | |
1740 | */ | |
1741 | if (req->flags & REQ_F_BUFFER_RING) { | |
1742 | if (req->buf_list) { | |
42db0c00 HX |
1743 | if (req->flags & REQ_F_PARTIAL_IO) { |
1744 | req->buf_list->head++; | |
1745 | req->buf_list = NULL; | |
1746 | } else { | |
1747 | req->buf_index = req->buf_list->bgid; | |
1748 | req->flags &= ~REQ_F_BUFFER_RING; | |
1749 | } | |
c7fb1942 JA |
1750 | } |
1751 | return; | |
1752 | } | |
b1c62645 | 1753 | |
f8929630 | 1754 | io_ring_submit_lock(ctx, issue_flags); |
b1c62645 JA |
1755 | |
1756 | buf = req->kbuf; | |
dbc7d452 JA |
1757 | bl = io_buffer_get_list(ctx, buf->bgid); |
1758 | list_add(&buf->list, &bl->buf_list); | |
b1c62645 | 1759 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
1dbd023e | 1760 | req->buf_index = buf->bgid; |
4d55f238 | 1761 | |
f8929630 | 1762 | io_ring_submit_unlock(ctx, issue_flags); |
b1c62645 JA |
1763 | } |
1764 | ||
3dd0c97a PB |
1765 | static bool io_match_task(struct io_kiocb *head, struct task_struct *task, |
1766 | bool cancel_all) | |
6af3f48b | 1767 | __must_hold(&req->ctx->timeout_lock) |
08d23634 | 1768 | { |
9cae36a0 JA |
1769 | struct io_kiocb *req; |
1770 | ||
68207680 | 1771 | if (task && head->task != task) |
08d23634 | 1772 | return false; |
9cae36a0 JA |
1773 | if (cancel_all) |
1774 | return true; | |
1775 | ||
1776 | io_for_each_link(req, head) { | |
1777 | if (req->flags & REQ_F_INFLIGHT) | |
1778 | return true; | |
1779 | } | |
1780 | return false; | |
1781 | } | |
1782 | ||
1783 | static bool io_match_linked(struct io_kiocb *head) | |
1784 | { | |
1785 | struct io_kiocb *req; | |
1786 | ||
1787 | io_for_each_link(req, head) { | |
1788 | if (req->flags & REQ_F_INFLIGHT) | |
1789 | return true; | |
1790 | } | |
1791 | return false; | |
6af3f48b PB |
1792 | } |
1793 | ||
1794 | /* | |
1795 | * As io_match_task() but protected against racing with linked timeouts. | |
1796 | * User must not hold timeout_lock. | |
1797 | */ | |
1798 | static bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, | |
1799 | bool cancel_all) | |
1800 | { | |
9cae36a0 JA |
1801 | bool matched; |
1802 | ||
6af3f48b PB |
1803 | if (task && head->task != task) |
1804 | return false; | |
9cae36a0 JA |
1805 | if (cancel_all) |
1806 | return true; | |
1807 | ||
1808 | if (head->flags & REQ_F_LINK_TIMEOUT) { | |
1809 | struct io_ring_ctx *ctx = head->ctx; | |
1810 | ||
1811 | /* protect against races with linked timeouts */ | |
1812 | spin_lock_irq(&ctx->timeout_lock); | |
1813 | matched = io_match_linked(head); | |
1814 | spin_unlock_irq(&ctx->timeout_lock); | |
1815 | } else { | |
1816 | matched = io_match_linked(head); | |
1817 | } | |
1818 | return matched; | |
6af3f48b PB |
1819 | } |
1820 | ||
d886e185 PB |
1821 | static inline bool req_has_async_data(struct io_kiocb *req) |
1822 | { | |
1823 | return req->flags & REQ_F_ASYNC_DATA; | |
1824 | } | |
1825 | ||
93d2bcd2 | 1826 | static inline void req_set_fail(struct io_kiocb *req) |
c40f6379 | 1827 | { |
93d2bcd2 | 1828 | req->flags |= REQ_F_FAIL; |
04c76b41 PB |
1829 | if (req->flags & REQ_F_CQE_SKIP) { |
1830 | req->flags &= ~REQ_F_CQE_SKIP; | |
1831 | req->flags |= REQ_F_SKIP_LINK_CQES; | |
1832 | } | |
c40f6379 | 1833 | } |
4a38aed2 | 1834 | |
a8295b98 HX |
1835 | static inline void req_fail_link_node(struct io_kiocb *req, int res) |
1836 | { | |
1837 | req_set_fail(req); | |
cef216fc | 1838 | req->cqe.res = res; |
a8295b98 HX |
1839 | } |
1840 | ||
fa05457a PB |
1841 | static inline void io_req_add_to_cache(struct io_kiocb *req, struct io_ring_ctx *ctx) |
1842 | { | |
1843 | wq_stack_add_head(&req->comp_list, &ctx->submit_state.free_list); | |
a8295b98 HX |
1844 | } |
1845 | ||
c072481d | 1846 | static __cold void io_ring_ctx_ref_free(struct percpu_ref *ref) |
2b188cc1 JA |
1847 | { |
1848 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1849 | ||
0f158b4c | 1850 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1851 | } |
1852 | ||
8eb7e2d0 PB |
1853 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1854 | { | |
1855 | return !req->timeout.off; | |
1856 | } | |
1857 | ||
c072481d | 1858 | static __cold void io_fallback_req_func(struct work_struct *work) |
f56165e6 PB |
1859 | { |
1860 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, | |
1861 | fallback_work.work); | |
1862 | struct llist_node *node = llist_del_all(&ctx->fallback_llist); | |
1863 | struct io_kiocb *req, *tmp; | |
f237c30a | 1864 | bool locked = false; |
f56165e6 PB |
1865 | |
1866 | percpu_ref_get(&ctx->refs); | |
1867 | llist_for_each_entry_safe(req, tmp, node, io_task_work.fallback_node) | |
f237c30a | 1868 | req->io_task_work.func(req, &locked); |
5636c00d | 1869 | |
f237c30a | 1870 | if (locked) { |
c450178d | 1871 | io_submit_flush_completions(ctx); |
f237c30a PB |
1872 | mutex_unlock(&ctx->uring_lock); |
1873 | } | |
f56165e6 PB |
1874 | percpu_ref_put(&ctx->refs); |
1875 | } | |
1876 | ||
c072481d | 1877 | static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
2b188cc1 JA |
1878 | { |
1879 | struct io_ring_ctx *ctx; | |
9cfc7e94 | 1880 | int hash_bits; |
2b188cc1 JA |
1881 | |
1882 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1883 | if (!ctx) | |
1884 | return NULL; | |
1885 | ||
9cfc7e94 JA |
1886 | xa_init(&ctx->io_bl_xa); |
1887 | ||
78076bb6 JA |
1888 | /* |
1889 | * Use 5 bits less than the max cq entries, that should give us around | |
1890 | * 32 entries per hash list if totally full and uniformly spread. | |
1891 | */ | |
1892 | hash_bits = ilog2(p->cq_entries); | |
1893 | hash_bits -= 5; | |
1894 | if (hash_bits <= 0) | |
1895 | hash_bits = 1; | |
1896 | ctx->cancel_hash_bits = hash_bits; | |
1897 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1898 | GFP_KERNEL); | |
1899 | if (!ctx->cancel_hash) | |
1900 | goto err; | |
1901 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1902 | ||
6224843d PB |
1903 | ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); |
1904 | if (!ctx->dummy_ubuf) | |
1905 | goto err; | |
1906 | /* set invalid range, so io_import_fixed() fails meeting it */ | |
1907 | ctx->dummy_ubuf->ubuf = -1UL; | |
1908 | ||
21482896 | 1909 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1910 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1911 | goto err; | |
2b188cc1 JA |
1912 | |
1913 | ctx->flags = p->flags; | |
90554200 | 1914 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1915 | INIT_LIST_HEAD(&ctx->sqd_list); |
1d7bb1d5 | 1916 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
cc3cec83 | 1917 | INIT_LIST_HEAD(&ctx->io_buffers_cache); |
4d9237e3 | 1918 | INIT_LIST_HEAD(&ctx->apoll_cache); |
0f158b4c | 1919 | init_completion(&ctx->ref_comp); |
61cf9370 | 1920 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 | 1921 | mutex_init(&ctx->uring_lock); |
311997b3 | 1922 | init_waitqueue_head(&ctx->cq_wait); |
2b188cc1 | 1923 | spin_lock_init(&ctx->completion_lock); |
89850fce | 1924 | spin_lock_init(&ctx->timeout_lock); |
5eef4e87 | 1925 | INIT_WQ_LIST(&ctx->iopoll_list); |
cc3cec83 JA |
1926 | INIT_LIST_HEAD(&ctx->io_buffers_pages); |
1927 | INIT_LIST_HEAD(&ctx->io_buffers_comp); | |
de0617e4 | 1928 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1929 | INIT_LIST_HEAD(&ctx->timeout_list); |
ef9dd637 | 1930 | INIT_LIST_HEAD(&ctx->ltimeout_list); |
d67d2263 BM |
1931 | spin_lock_init(&ctx->rsrc_ref_lock); |
1932 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1933 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1934 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1935 | INIT_LIST_HEAD(&ctx->tctx_list); |
c2b6c6bc PB |
1936 | ctx->submit_state.free_list.next = NULL; |
1937 | INIT_WQ_LIST(&ctx->locked_free_list); | |
9011bf9a | 1938 | INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); |
6f33b0bc | 1939 | INIT_WQ_LIST(&ctx->submit_state.compl_reqs); |
2b188cc1 | 1940 | return ctx; |
206aefde | 1941 | err: |
6224843d | 1942 | kfree(ctx->dummy_ubuf); |
78076bb6 | 1943 | kfree(ctx->cancel_hash); |
9cfc7e94 JA |
1944 | kfree(ctx->io_bl); |
1945 | xa_destroy(&ctx->io_bl_xa); | |
206aefde JA |
1946 | kfree(ctx); |
1947 | return NULL; | |
2b188cc1 JA |
1948 | } |
1949 | ||
8f6ed49a PB |
1950 | static void io_account_cq_overflow(struct io_ring_ctx *ctx) |
1951 | { | |
1952 | struct io_rings *r = ctx->rings; | |
1953 | ||
1954 | WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); | |
1955 | ctx->cq_extra--; | |
1956 | } | |
1957 | ||
9cf7c104 | 1958 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1959 | { |
2bc9930e JA |
1960 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1961 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1962 | |
8f6ed49a | 1963 | return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; |
2bc9930e | 1964 | } |
de0617e4 | 1965 | |
9d858b21 | 1966 | return false; |
de0617e4 JA |
1967 | } |
1968 | ||
c97d8a0f PB |
1969 | static inline bool io_req_ffs_set(struct io_kiocb *req) |
1970 | { | |
35645ac3 | 1971 | return req->flags & REQ_F_FIXED_FILE; |
c97d8a0f PB |
1972 | } |
1973 | ||
9cae36a0 JA |
1974 | static inline void io_req_track_inflight(struct io_kiocb *req) |
1975 | { | |
1976 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
1977 | req->flags |= REQ_F_INFLIGHT; | |
1978 | atomic_inc(¤t->io_uring->inflight_tracked); | |
1979 | } | |
1980 | } | |
1981 | ||
fd08e530 PB |
1982 | static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req) |
1983 | { | |
906c6caa PB |
1984 | if (WARN_ON_ONCE(!req->link)) |
1985 | return NULL; | |
1986 | ||
4d13d1a4 PB |
1987 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
1988 | req->flags |= REQ_F_LINK_TIMEOUT; | |
fd08e530 PB |
1989 | |
1990 | /* linked timeouts should have two refs once prep'ed */ | |
48dcd38d | 1991 | io_req_set_refcount(req); |
4d13d1a4 PB |
1992 | __io_req_set_refcount(req->link, 2); |
1993 | return req->link; | |
fd08e530 PB |
1994 | } |
1995 | ||
1996 | static inline struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) | |
1997 | { | |
4d13d1a4 | 1998 | if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) |
fd08e530 PB |
1999 | return NULL; |
2000 | return __io_prep_linked_timeout(req); | |
2001 | } | |
2002 | ||
cb2d344c PB |
2003 | static noinline void __io_arm_ltimeout(struct io_kiocb *req) |
2004 | { | |
2005 | io_queue_linked_timeout(__io_prep_linked_timeout(req)); | |
2006 | } | |
2007 | ||
2008 | static inline void io_arm_ltimeout(struct io_kiocb *req) | |
2009 | { | |
2010 | if (unlikely(req->flags & REQ_F_ARM_LTIMEOUT)) | |
2011 | __io_arm_ltimeout(req); | |
2012 | } | |
2013 | ||
1e6fa521 JA |
2014 | static void io_prep_async_work(struct io_kiocb *req) |
2015 | { | |
2016 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
2017 | struct io_ring_ctx *ctx = req->ctx; |
2018 | ||
b8e64b53 PB |
2019 | if (!(req->flags & REQ_F_CREDS)) { |
2020 | req->flags |= REQ_F_CREDS; | |
c10d1f98 | 2021 | req->creds = get_current_cred(); |
b8e64b53 | 2022 | } |
003e8dcc | 2023 | |
e1d675df PB |
2024 | req->work.list.next = NULL; |
2025 | req->work.flags = 0; | |
8e29da69 | 2026 | req->work.cancel_seq = atomic_read(&ctx->cancel_seq); |
feaadc4f PB |
2027 | if (req->flags & REQ_F_FORCE_ASYNC) |
2028 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
2029 | ||
1e6fa521 JA |
2030 | if (req->flags & REQ_F_ISREG) { |
2031 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
2032 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 2033 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
2034 | if (def->unbound_nonreg_file) |
2035 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2036 | } | |
561fb04a | 2037 | } |
cccf0ee8 | 2038 | |
cbdcb435 | 2039 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 2040 | { |
cbdcb435 | 2041 | struct io_kiocb *cur; |
54a91f3b | 2042 | |
44eff40a PB |
2043 | if (req->flags & REQ_F_LINK_TIMEOUT) { |
2044 | struct io_ring_ctx *ctx = req->ctx; | |
2045 | ||
674ee8e1 | 2046 | spin_lock_irq(&ctx->timeout_lock); |
44eff40a PB |
2047 | io_for_each_link(cur, req) |
2048 | io_prep_async_work(cur); | |
674ee8e1 | 2049 | spin_unlock_irq(&ctx->timeout_lock); |
44eff40a PB |
2050 | } else { |
2051 | io_for_each_link(cur, req) | |
2052 | io_prep_async_work(cur); | |
2053 | } | |
561fb04a JA |
2054 | } |
2055 | ||
fff4e40e PB |
2056 | static inline void io_req_add_compl_list(struct io_kiocb *req) |
2057 | { | |
775a1f2f | 2058 | struct io_submit_state *state = &req->ctx->submit_state; |
fff4e40e | 2059 | |
3d4aeb9f | 2060 | if (!(req->flags & REQ_F_CQE_SKIP)) |
775a1f2f | 2061 | state->flush_cqes = true; |
fff4e40e PB |
2062 | wq_list_add_tail(&req->comp_list, &state->compl_reqs); |
2063 | } | |
2064 | ||
77955efb | 2065 | static void io_queue_iowq(struct io_kiocb *req, bool *dont_use) |
561fb04a | 2066 | { |
cbdcb435 | 2067 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 2068 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 2069 | |
3bfe6106 JA |
2070 | BUG_ON(!tctx); |
2071 | BUG_ON(!tctx->io_wq); | |
561fb04a | 2072 | |
cbdcb435 PB |
2073 | /* init ->work of the whole link before punting */ |
2074 | io_prep_async_link(req); | |
991468dc JA |
2075 | |
2076 | /* | |
2077 | * Not expected to happen, but if we do have a bug where this _can_ | |
2078 | * happen, catch it here and ensure the request is marked as | |
2079 | * canceled. That will make io-wq go through the usual work cancel | |
2080 | * procedure rather than attempt to run this request (or create a new | |
2081 | * worker for it). | |
2082 | */ | |
2083 | if (WARN_ON_ONCE(!same_thread_group(req->task, current))) | |
2084 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
2085 | ||
971cf9c1 PB |
2086 | trace_io_uring_queue_async_work(req->ctx, req, req->cqe.user_data, |
2087 | req->opcode, req->flags, &req->work, | |
2088 | io_wq_is_hashed(&req->work)); | |
ebf93667 | 2089 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
2090 | if (link) |
2091 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
2092 | } |
2093 | ||
1ee4160c | 2094 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 2095 | __must_hold(&req->ctx->completion_lock) |
89850fce | 2096 | __must_hold(&req->ctx->timeout_lock) |
5262f567 | 2097 | { |
e8c2bc1f | 2098 | struct io_timeout_data *io = req->async_data; |
5262f567 | 2099 | |
fd9c7bc5 | 2100 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
2ae2eb9d PB |
2101 | if (status) |
2102 | req_set_fail(req); | |
01cec8c1 PB |
2103 | atomic_set(&req->ctx->cq_timeouts, |
2104 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 2105 | list_del_init(&req->timeout.list); |
4e118cd9 | 2106 | io_req_tw_post_queue(req, status, 0); |
5262f567 JA |
2107 | } |
2108 | } | |
2109 | ||
c072481d | 2110 | static __cold void io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 2111 | { |
441b8a78 | 2112 | while (!list_empty(&ctx->defer_list)) { |
27dc8338 PB |
2113 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
2114 | struct io_defer_entry, list); | |
de0617e4 | 2115 | |
9cf7c104 | 2116 | if (req_need_defer(de->req, de->seq)) |
04518945 | 2117 | break; |
27dc8338 | 2118 | list_del_init(&de->list); |
907d1df3 | 2119 | io_req_task_queue(de->req); |
27dc8338 | 2120 | kfree(de); |
441b8a78 | 2121 | } |
04518945 PB |
2122 | } |
2123 | ||
c072481d | 2124 | static __cold void io_flush_timeouts(struct io_ring_ctx *ctx) |
89850fce | 2125 | __must_hold(&ctx->completion_lock) |
de0617e4 | 2126 | { |
441b8a78 | 2127 | u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
e677edbc | 2128 | struct io_kiocb *req, *tmp; |
f010505b | 2129 | |
79ebeaee | 2130 | spin_lock_irq(&ctx->timeout_lock); |
e677edbc | 2131 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
f010505b | 2132 | u32 events_needed, events_got; |
de0617e4 | 2133 | |
8eb7e2d0 | 2134 | if (io_is_timeout_noseq(req)) |
360428f8 | 2135 | break; |
f010505b MDG |
2136 | |
2137 | /* | |
2138 | * Since seq can easily wrap around over time, subtract | |
2139 | * the last seq at which timeouts were flushed before comparing. | |
2140 | * Assuming not more than 2^31-1 events have happened since, | |
2141 | * these subtractions won't have wrapped, so we can check if | |
2142 | * target is in [last_seq, current_seq] by comparing the two. | |
2143 | */ | |
2144 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
2145 | events_got = seq - ctx->cq_last_tm_flush; | |
2146 | if (events_got < events_needed) | |
360428f8 | 2147 | break; |
bfe68a22 | 2148 | |
1ee4160c | 2149 | io_kill_timeout(req, 0); |
f18ee4cf | 2150 | } |
f010505b | 2151 | ctx->cq_last_tm_flush = seq; |
79ebeaee | 2152 | spin_unlock_irq(&ctx->timeout_lock); |
360428f8 | 2153 | } |
5262f567 | 2154 | |
9333f6b4 PB |
2155 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) |
2156 | { | |
2157 | /* order cqe stores with ring update */ | |
2158 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
2159 | } | |
2160 | ||
9aa8dfde | 2161 | static void __io_commit_cqring_flush(struct io_ring_ctx *ctx) |
360428f8 | 2162 | { |
9aa8dfde PB |
2163 | if (ctx->off_timeout_used || ctx->drain_active) { |
2164 | spin_lock(&ctx->completion_lock); | |
2165 | if (ctx->off_timeout_used) | |
2166 | io_flush_timeouts(ctx); | |
2167 | if (ctx->drain_active) | |
2168 | io_queue_deferred(ctx); | |
2169 | io_commit_cqring(ctx); | |
2170 | spin_unlock(&ctx->completion_lock); | |
2171 | } | |
2172 | if (ctx->has_evfd) | |
2173 | io_eventfd_signal(ctx); | |
de0617e4 JA |
2174 | } |
2175 | ||
90554200 JA |
2176 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
2177 | { | |
2178 | struct io_rings *r = ctx->rings; | |
2179 | ||
a566c556 | 2180 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; |
90554200 JA |
2181 | } |
2182 | ||
888aae2e PB |
2183 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
2184 | { | |
2185 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
2186 | } | |
2187 | ||
d8da428b PB |
2188 | /* |
2189 | * writes to the cq entry need to come after reading head; the | |
2190 | * control dependency is enough as we're using WRITE_ONCE to | |
2191 | * fill the cq entry | |
2192 | */ | |
2193 | static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx) | |
2b188cc1 | 2194 | { |
75b28aff | 2195 | struct io_rings *rings = ctx->rings; |
d8da428b | 2196 | unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1); |
2fee6bc6 | 2197 | unsigned int shift = 0; |
d8da428b PB |
2198 | unsigned int free, queued, len; |
2199 | ||
2fee6bc6 SR |
2200 | if (ctx->flags & IORING_SETUP_CQE32) |
2201 | shift = 1; | |
2202 | ||
d8da428b PB |
2203 | /* userspace may cheat modifying the tail, be safe and do min */ |
2204 | queued = min(__io_cqring_events(ctx), ctx->cq_entries); | |
2205 | free = ctx->cq_entries - queued; | |
2206 | /* we need a contiguous range, limit based on the current array offset */ | |
2207 | len = min(free, ctx->cq_entries - off); | |
2208 | if (!len) | |
2b188cc1 JA |
2209 | return NULL; |
2210 | ||
d8da428b PB |
2211 | ctx->cached_cq_tail++; |
2212 | ctx->cqe_cached = &rings->cqes[off]; | |
2213 | ctx->cqe_sentinel = ctx->cqe_cached + len; | |
2fee6bc6 SR |
2214 | ctx->cqe_cached++; |
2215 | return &rings->cqes[off << shift]; | |
d8da428b PB |
2216 | } |
2217 | ||
2218 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) | |
2219 | { | |
2220 | if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) { | |
2fee6bc6 SR |
2221 | struct io_uring_cqe *cqe = ctx->cqe_cached; |
2222 | ||
2223 | if (ctx->flags & IORING_SETUP_CQE32) { | |
2224 | unsigned int off = ctx->cqe_cached - ctx->rings->cqes; | |
2225 | ||
2226 | cqe += off; | |
2227 | } | |
2228 | ||
d8da428b | 2229 | ctx->cached_cq_tail++; |
2fee6bc6 SR |
2230 | ctx->cqe_cached++; |
2231 | return cqe; | |
d8da428b | 2232 | } |
2fee6bc6 | 2233 | |
d8da428b | 2234 | return __io_get_cqe(ctx); |
2b188cc1 JA |
2235 | } |
2236 | ||
77bc59b4 | 2237 | static void io_eventfd_signal(struct io_ring_ctx *ctx) |
f2842ab5 | 2238 | { |
77bc59b4 UA |
2239 | struct io_ev_fd *ev_fd; |
2240 | ||
77bc59b4 UA |
2241 | rcu_read_lock(); |
2242 | /* | |
2243 | * rcu_dereference ctx->io_ev_fd once and use it for both for checking | |
2244 | * and eventfd_signal | |
2245 | */ | |
2246 | ev_fd = rcu_dereference(ctx->io_ev_fd); | |
2247 | ||
2248 | /* | |
2249 | * Check again if ev_fd exists incase an io_eventfd_unregister call | |
2250 | * completed between the NULL check of ctx->io_ev_fd at the start of | |
2251 | * the function and rcu_read_lock. | |
2252 | */ | |
2253 | if (unlikely(!ev_fd)) | |
2254 | goto out; | |
7e55a19c | 2255 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
77bc59b4 UA |
2256 | goto out; |
2257 | ||
c75312dd | 2258 | if (!ev_fd->eventfd_async || io_wq_current_is_worker()) |
77bc59b4 | 2259 | eventfd_signal(ev_fd->cq_ev_fd, 1); |
77bc59b4 UA |
2260 | out: |
2261 | rcu_read_unlock(); | |
f2842ab5 JA |
2262 | } |
2263 | ||
9aa8dfde PB |
2264 | static inline void io_cqring_wake(struct io_ring_ctx *ctx) |
2265 | { | |
2266 | /* | |
2267 | * wake_up_all() may seem excessive, but io_wake_function() and | |
2268 | * io_should_wake() handle the termination of the loop and only | |
2269 | * wake as many waiters as we need to. | |
2270 | */ | |
2271 | if (wq_has_sleeper(&ctx->cq_wait)) | |
2272 | wake_up_all(&ctx->cq_wait); | |
2273 | } | |
2274 | ||
2c5d763c JA |
2275 | /* |
2276 | * This should only get called when at least one event has been posted. | |
2277 | * Some applications rely on the eventfd notification count only changing | |
2278 | * IFF a new CQE has been added to the CQ ring. There's no depedency on | |
2279 | * 1:1 relationship between how many times this function is called (and | |
2280 | * hence the eventfd count) and number of CQEs posted to the CQ ring. | |
2281 | */ | |
66fc25ca | 2282 | static inline void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 2283 | { |
9aa8dfde PB |
2284 | if (unlikely(ctx->off_timeout_used || ctx->drain_active || |
2285 | ctx->has_evfd)) | |
9333f6b4 PB |
2286 | __io_commit_cqring_flush(ctx); |
2287 | ||
9aa8dfde | 2288 | io_cqring_wake(ctx); |
1d7bb1d5 JA |
2289 | } |
2290 | ||
80c18e4a PB |
2291 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
2292 | { | |
9aa8dfde PB |
2293 | if (unlikely(ctx->off_timeout_used || ctx->drain_active || |
2294 | ctx->has_evfd)) | |
9333f6b4 PB |
2295 | __io_commit_cqring_flush(ctx); |
2296 | ||
9aa8dfde PB |
2297 | if (ctx->flags & IORING_SETUP_SQPOLL) |
2298 | io_cqring_wake(ctx); | |
80c18e4a PB |
2299 | } |
2300 | ||
c4a2ed72 | 2301 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 2302 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 | 2303 | { |
b18032bb | 2304 | bool all_flushed, posted; |
e45a3e05 | 2305 | size_t cqe_size = sizeof(struct io_uring_cqe); |
1d7bb1d5 | 2306 | |
a566c556 | 2307 | if (!force && __io_cqring_events(ctx) == ctx->cq_entries) |
e23de15f | 2308 | return false; |
1d7bb1d5 | 2309 | |
e45a3e05 SR |
2310 | if (ctx->flags & IORING_SETUP_CQE32) |
2311 | cqe_size <<= 1; | |
2312 | ||
b18032bb | 2313 | posted = false; |
79ebeaee | 2314 | spin_lock(&ctx->completion_lock); |
6c2450ae | 2315 | while (!list_empty(&ctx->cq_overflow_list)) { |
d068b506 | 2316 | struct io_uring_cqe *cqe = io_get_cqe(ctx); |
6c2450ae | 2317 | struct io_overflow_cqe *ocqe; |
e6c8aa9a | 2318 | |
1d7bb1d5 JA |
2319 | if (!cqe && !force) |
2320 | break; | |
6c2450ae PB |
2321 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
2322 | struct io_overflow_cqe, list); | |
2323 | if (cqe) | |
e45a3e05 | 2324 | memcpy(cqe, &ocqe->cqe, cqe_size); |
6c2450ae | 2325 | else |
8f6ed49a PB |
2326 | io_account_cq_overflow(ctx); |
2327 | ||
b18032bb | 2328 | posted = true; |
6c2450ae PB |
2329 | list_del(&ocqe->list); |
2330 | kfree(ocqe); | |
1d7bb1d5 JA |
2331 | } |
2332 | ||
09e88404 PB |
2333 | all_flushed = list_empty(&ctx->cq_overflow_list); |
2334 | if (all_flushed) { | |
10988a0a | 2335 | clear_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq); |
3a4b89a2 | 2336 | atomic_andnot(IORING_SQ_CQ_OVERFLOW, &ctx->rings->sq_flags); |
09e88404 | 2337 | } |
46930143 | 2338 | |
60053be8 | 2339 | io_commit_cqring(ctx); |
79ebeaee | 2340 | spin_unlock(&ctx->completion_lock); |
b18032bb JA |
2341 | if (posted) |
2342 | io_cqring_ev_posted(ctx); | |
09e88404 | 2343 | return all_flushed; |
1d7bb1d5 JA |
2344 | } |
2345 | ||
90f67366 | 2346 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx) |
6c503150 | 2347 | { |
ca0a2651 JA |
2348 | bool ret = true; |
2349 | ||
10988a0a | 2350 | if (test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq)) { |
6c503150 PB |
2351 | /* iopoll syncs against uring_lock, not completion_lock */ |
2352 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
2353 | mutex_lock(&ctx->uring_lock); | |
90f67366 | 2354 | ret = __io_cqring_overflow_flush(ctx, false); |
6c503150 PB |
2355 | if (ctx->flags & IORING_SETUP_IOPOLL) |
2356 | mutex_unlock(&ctx->uring_lock); | |
2357 | } | |
ca0a2651 JA |
2358 | |
2359 | return ret; | |
6c503150 PB |
2360 | } |
2361 | ||
9d170164 | 2362 | static void __io_put_task(struct task_struct *task, int nr) |
6a290a14 PB |
2363 | { |
2364 | struct io_uring_task *tctx = task->io_uring; | |
2365 | ||
9d170164 PB |
2366 | percpu_counter_sub(&tctx->inflight, nr); |
2367 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2368 | wake_up(&tctx->wait); | |
2369 | put_task_struct_many(task, nr); | |
2370 | } | |
2371 | ||
2372 | /* must to be called somewhat shortly after putting a request */ | |
2373 | static inline void io_put_task(struct task_struct *task, int nr) | |
2374 | { | |
2375 | if (likely(task == current)) | |
2376 | task->io_uring->cached_refs += nr; | |
2377 | else | |
2378 | __io_put_task(task, nr); | |
6a290a14 PB |
2379 | } |
2380 | ||
9a10867a PB |
2381 | static void io_task_refs_refill(struct io_uring_task *tctx) |
2382 | { | |
2383 | unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; | |
2384 | ||
2385 | percpu_counter_add(&tctx->inflight, refill); | |
2386 | refcount_add(refill, ¤t->usage); | |
2387 | tctx->cached_refs += refill; | |
2388 | } | |
2389 | ||
2390 | static inline void io_get_task_refs(int nr) | |
2391 | { | |
2392 | struct io_uring_task *tctx = current->io_uring; | |
2393 | ||
2394 | tctx->cached_refs -= nr; | |
2395 | if (unlikely(tctx->cached_refs < 0)) | |
2396 | io_task_refs_refill(tctx); | |
2397 | } | |
2398 | ||
3cc7fdb9 PB |
2399 | static __cold void io_uring_drop_tctx_refs(struct task_struct *task) |
2400 | { | |
2401 | struct io_uring_task *tctx = task->io_uring; | |
2402 | unsigned int refs = tctx->cached_refs; | |
2403 | ||
2404 | if (refs) { | |
2405 | tctx->cached_refs = 0; | |
2406 | percpu_counter_sub(&tctx->inflight, refs); | |
2407 | put_task_struct_many(task, refs); | |
2408 | } | |
2409 | } | |
2410 | ||
d4d19c19 | 2411 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
e45a3e05 SR |
2412 | s32 res, u32 cflags, u64 extra1, |
2413 | u64 extra2) | |
2b188cc1 | 2414 | { |
cce4b8b0 | 2415 | struct io_overflow_cqe *ocqe; |
e45a3e05 SR |
2416 | size_t ocq_size = sizeof(struct io_overflow_cqe); |
2417 | bool is_cqe32 = (ctx->flags & IORING_SETUP_CQE32); | |
2b188cc1 | 2418 | |
e45a3e05 SR |
2419 | if (is_cqe32) |
2420 | ocq_size += sizeof(struct io_uring_cqe); | |
2b188cc1 | 2421 | |
e45a3e05 | 2422 | ocqe = kmalloc(ocq_size, GFP_ATOMIC | __GFP_ACCOUNT); |
08dcd028 | 2423 | trace_io_uring_cqe_overflow(ctx, user_data, res, cflags, ocqe); |
cce4b8b0 PB |
2424 | if (!ocqe) { |
2425 | /* | |
2426 | * If we're in ring overflow flush mode, or in task cancel mode, | |
2427 | * or cannot allocate an overflow entry, then we need to drop it | |
2428 | * on the floor. | |
2429 | */ | |
8f6ed49a | 2430 | io_account_cq_overflow(ctx); |
155bc950 | 2431 | set_bit(IO_CHECK_CQ_DROPPED_BIT, &ctx->check_cq); |
cce4b8b0 | 2432 | return false; |
2b188cc1 | 2433 | } |
cce4b8b0 | 2434 | if (list_empty(&ctx->cq_overflow_list)) { |
10988a0a | 2435 | set_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq); |
3a4b89a2 | 2436 | atomic_or(IORING_SQ_CQ_OVERFLOW, &ctx->rings->sq_flags); |
20c0b380 | 2437 | |
cce4b8b0 | 2438 | } |
d4d19c19 | 2439 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
2440 | ocqe->cqe.res = res; |
2441 | ocqe->cqe.flags = cflags; | |
e45a3e05 SR |
2442 | if (is_cqe32) { |
2443 | ocqe->cqe.big_cqe[0] = extra1; | |
2444 | ocqe->cqe.big_cqe[1] = extra2; | |
2445 | } | |
cce4b8b0 PB |
2446 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); |
2447 | return true; | |
2b188cc1 JA |
2448 | } |
2449 | ||
91ef75a7 PB |
2450 | static inline bool __io_fill_cqe_req(struct io_ring_ctx *ctx, |
2451 | struct io_kiocb *req) | |
d5ec1dfa | 2452 | { |
90e7c35f PB |
2453 | struct io_uring_cqe *cqe; |
2454 | ||
f43de1f8 PB |
2455 | if (!(ctx->flags & IORING_SETUP_CQE32)) { |
2456 | trace_io_uring_complete(req->ctx, req, req->cqe.user_data, | |
2457 | req->cqe.res, req->cqe.flags, 0, 0); | |
90e7c35f | 2458 | |
f43de1f8 PB |
2459 | /* |
2460 | * If we can't get a cq entry, userspace overflowed the | |
2461 | * submission (by quite a lot). Increment the overflow count in | |
2462 | * the ring. | |
2463 | */ | |
2464 | cqe = io_get_cqe(ctx); | |
2465 | if (likely(cqe)) { | |
2466 | memcpy(cqe, &req->cqe, sizeof(*cqe)); | |
2467 | return true; | |
2468 | } | |
2469 | ||
2470 | return io_cqring_event_overflow(ctx, req->cqe.user_data, | |
2471 | req->cqe.res, req->cqe.flags, | |
2472 | 0, 0); | |
2473 | } else { | |
2caf9822 PB |
2474 | u64 extra1 = 0, extra2 = 0; |
2475 | ||
2476 | if (req->flags & REQ_F_CQE32_INIT) { | |
2477 | extra1 = req->extra1; | |
2478 | extra2 = req->extra2; | |
2479 | } | |
f43de1f8 PB |
2480 | |
2481 | trace_io_uring_complete(req->ctx, req, req->cqe.user_data, | |
2482 | req->cqe.res, req->cqe.flags, extra1, extra2); | |
2483 | ||
2484 | /* | |
2485 | * If we can't get a cq entry, userspace overflowed the | |
2486 | * submission (by quite a lot). Increment the overflow count in | |
2487 | * the ring. | |
2488 | */ | |
2489 | cqe = io_get_cqe(ctx); | |
2490 | if (likely(cqe)) { | |
2491 | memcpy(cqe, &req->cqe, sizeof(struct io_uring_cqe)); | |
2492 | WRITE_ONCE(cqe->big_cqe[0], extra1); | |
2493 | WRITE_ONCE(cqe->big_cqe[1], extra2); | |
2494 | return true; | |
2495 | } | |
2496 | ||
2497 | return io_cqring_event_overflow(ctx, req->cqe.user_data, | |
2498 | req->cqe.res, req->cqe.flags, | |
2499 | extra1, extra2); | |
90e7c35f | 2500 | } |
d5ec1dfa SR |
2501 | } |
2502 | ||
913a571a PB |
2503 | static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data, |
2504 | s32 res, u32 cflags) | |
bcda7baa | 2505 | { |
cd94903d PB |
2506 | struct io_uring_cqe *cqe; |
2507 | ||
913a571a | 2508 | ctx->cq_extra++; |
c4bb964f | 2509 | trace_io_uring_complete(ctx, NULL, user_data, res, cflags, 0, 0); |
cd94903d PB |
2510 | |
2511 | /* | |
2512 | * If we can't get a cq entry, userspace overflowed the | |
2513 | * submission (by quite a lot). Increment the overflow count in | |
2514 | * the ring. | |
2515 | */ | |
2516 | cqe = io_get_cqe(ctx); | |
2517 | if (likely(cqe)) { | |
2518 | WRITE_ONCE(cqe->user_data, user_data); | |
2519 | WRITE_ONCE(cqe->res, res); | |
2520 | WRITE_ONCE(cqe->flags, cflags); | |
c5595975 PB |
2521 | |
2522 | if (ctx->flags & IORING_SETUP_CQE32) { | |
2523 | WRITE_ONCE(cqe->big_cqe[0], 0); | |
2524 | WRITE_ONCE(cqe->big_cqe[1], 0); | |
2525 | } | |
cd94903d PB |
2526 | return true; |
2527 | } | |
2528 | return io_cqring_event_overflow(ctx, user_data, res, cflags, 0, 0); | |
bcda7baa JA |
2529 | } |
2530 | ||
effcf8bd | 2531 | static void __io_req_complete_put(struct io_kiocb *req) |
2b188cc1 | 2532 | { |
c7dae4ba JA |
2533 | /* |
2534 | * If we're the last reference to this request, add to our locked | |
2535 | * free_list cache. | |
2536 | */ | |
de9b4cca | 2537 | if (req_ref_put_and_test(req)) { |
effcf8bd SR |
2538 | struct io_ring_ctx *ctx = req->ctx; |
2539 | ||
da1a08c5 | 2540 | if (req->flags & IO_REQ_LINK_FLAGS) { |
0756a869 | 2541 | if (req->flags & IO_DISARM_MASK) |
7a612350 PB |
2542 | io_disarm_next(req); |
2543 | if (req->link) { | |
2544 | io_req_task_queue(req->link); | |
2545 | req->link = NULL; | |
2546 | } | |
2547 | } | |
7ac1edc4 | 2548 | io_req_put_rsrc(req); |
8197b053 PB |
2549 | /* |
2550 | * Selected buffer deallocation in io_clean_op() assumes that | |
2551 | * we don't hold ->completion_lock. Clean them here to avoid | |
2552 | * deadlocks. | |
2553 | */ | |
2554 | io_put_kbuf_comp(req); | |
c7dae4ba JA |
2555 | io_dismantle_req(req); |
2556 | io_put_task(req->task, 1); | |
c2b6c6bc | 2557 | wq_list_add_head(&req->comp_list, &ctx->locked_free_list); |
d0acdee2 | 2558 | ctx->locked_free_nr++; |
180f829f | 2559 | } |
a37fae8a HX |
2560 | } |
2561 | ||
effcf8bd SR |
2562 | static void __io_req_complete_post(struct io_kiocb *req, s32 res, |
2563 | u32 cflags) | |
2564 | { | |
91ef75a7 PB |
2565 | if (!(req->flags & REQ_F_CQE_SKIP)) { |
2566 | req->cqe.res = res; | |
2567 | req->cqe.flags = cflags; | |
2568 | __io_fill_cqe_req(req->ctx, req); | |
2569 | } | |
effcf8bd SR |
2570 | __io_req_complete_put(req); |
2571 | } | |
2572 | ||
effcf8bd | 2573 | static void io_req_complete_post(struct io_kiocb *req, s32 res, u32 cflags) |
a37fae8a HX |
2574 | { |
2575 | struct io_ring_ctx *ctx = req->ctx; | |
2576 | ||
2577 | spin_lock(&ctx->completion_lock); | |
2578 | __io_req_complete_post(req, res, cflags); | |
7a612350 | 2579 | io_commit_cqring(ctx); |
79ebeaee | 2580 | spin_unlock(&ctx->completion_lock); |
a3f34907 | 2581 | io_cqring_ev_posted(ctx); |
4e3d9ff9 JA |
2582 | } |
2583 | ||
54daa9b2 PB |
2584 | static inline void io_req_complete_state(struct io_kiocb *req, s32 res, |
2585 | u32 cflags) | |
229a7b63 | 2586 | { |
cef216fc PB |
2587 | req->cqe.res = res; |
2588 | req->cqe.flags = cflags; | |
e342c807 | 2589 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
2590 | } |
2591 | ||
889fca73 | 2592 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
54daa9b2 | 2593 | s32 res, u32 cflags) |
bcda7baa | 2594 | { |
889fca73 PB |
2595 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
2596 | io_req_complete_state(req, res, cflags); | |
a38d68db | 2597 | else |
c7dae4ba | 2598 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
2599 | } |
2600 | ||
54daa9b2 | 2601 | static inline void io_req_complete(struct io_kiocb *req, s32 res) |
0ddf92e8 | 2602 | { |
4ffaa94b KJ |
2603 | if (res < 0) |
2604 | req_set_fail(req); | |
889fca73 | 2605 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
2606 | } |
2607 | ||
54daa9b2 | 2608 | static void io_req_complete_failed(struct io_kiocb *req, s32 res) |
f41db273 | 2609 | { |
93d2bcd2 | 2610 | req_set_fail(req); |
ab0ac095 | 2611 | io_req_complete_post(req, res, io_put_kbuf(req, IO_URING_F_UNLOCKED)); |
f41db273 PB |
2612 | } |
2613 | ||
864ea921 PB |
2614 | /* |
2615 | * Don't initialise the fields below on every allocation, but do that in | |
2616 | * advance and keep them valid across allocations. | |
2617 | */ | |
2618 | static void io_preinit_req(struct io_kiocb *req, struct io_ring_ctx *ctx) | |
2619 | { | |
2620 | req->ctx = ctx; | |
2621 | req->link = NULL; | |
2622 | req->async_data = NULL; | |
2623 | /* not necessary, but safer to zero */ | |
cef216fc | 2624 | req->cqe.res = 0; |
864ea921 PB |
2625 | } |
2626 | ||
dac7a098 | 2627 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
cd0ca2e0 | 2628 | struct io_submit_state *state) |
dac7a098 | 2629 | { |
79ebeaee | 2630 | spin_lock(&ctx->completion_lock); |
c2b6c6bc | 2631 | wq_list_splice(&ctx->locked_free_list, &state->free_list); |
d0acdee2 | 2632 | ctx->locked_free_nr = 0; |
79ebeaee | 2633 | spin_unlock(&ctx->completion_lock); |
dac7a098 PB |
2634 | } |
2635 | ||
88ab95be | 2636 | static inline bool io_req_cache_empty(struct io_ring_ctx *ctx) |
0ddf92e8 | 2637 | { |
88ab95be | 2638 | return !ctx->submit_state.free_list.next; |
0ddf92e8 JA |
2639 | } |
2640 | ||
5d5901a3 PB |
2641 | /* |
2642 | * A request might get retired back into the request caches even before opcode | |
2643 | * handlers and io_issue_sqe() are done with it, e.g. inline completion path. | |
2644 | * Because of that, io_alloc_req() should be called only under ->uring_lock | |
2645 | * and with extra caution to not get a request that is still worked on. | |
2646 | */ | |
c072481d | 2647 | static __cold bool __io_alloc_req_refill(struct io_ring_ctx *ctx) |
5d5901a3 | 2648 | __must_hold(&ctx->uring_lock) |
2b188cc1 | 2649 | { |
864ea921 | 2650 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
3ab665b7 | 2651 | void *reqs[IO_REQ_ALLOC_BATCH]; |
864ea921 | 2652 | int ret, i; |
e5d1bc0a | 2653 | |
23a5c43b PB |
2654 | /* |
2655 | * If we have more than a batch's worth of requests in our IRQ side | |
2656 | * locked cache, grab the lock and move them over to our submission | |
2657 | * side cache. | |
2658 | */ | |
a6d97a8a | 2659 | if (data_race(ctx->locked_free_nr) > IO_COMPL_BATCH) { |
23a5c43b | 2660 | io_flush_cached_locked_reqs(ctx, &ctx->submit_state); |
88ab95be | 2661 | if (!io_req_cache_empty(ctx)) |
23a5c43b PB |
2662 | return true; |
2663 | } | |
e5d1bc0a | 2664 | |
3ab665b7 | 2665 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, ARRAY_SIZE(reqs), reqs); |
fd6fab2c | 2666 | |
864ea921 PB |
2667 | /* |
2668 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
2669 | * retry single alloc to be on the safe side. | |
2670 | */ | |
2671 | if (unlikely(ret <= 0)) { | |
3ab665b7 PB |
2672 | reqs[0] = kmem_cache_alloc(req_cachep, gfp); |
2673 | if (!reqs[0]) | |
a33ae9ce | 2674 | return false; |
864ea921 | 2675 | ret = 1; |
2b188cc1 | 2676 | } |
864ea921 | 2677 | |
37f0e767 | 2678 | percpu_ref_get_many(&ctx->refs, ret); |
3ab665b7 | 2679 | for (i = 0; i < ret; i++) { |
23a5c43b | 2680 | struct io_kiocb *req = reqs[i]; |
3ab665b7 PB |
2681 | |
2682 | io_preinit_req(req, ctx); | |
fa05457a | 2683 | io_req_add_to_cache(req, ctx); |
3ab665b7 | 2684 | } |
a33ae9ce PB |
2685 | return true; |
2686 | } | |
2687 | ||
2688 | static inline bool io_alloc_req_refill(struct io_ring_ctx *ctx) | |
2689 | { | |
88ab95be | 2690 | if (unlikely(io_req_cache_empty(ctx))) |
a33ae9ce PB |
2691 | return __io_alloc_req_refill(ctx); |
2692 | return true; | |
2693 | } | |
2694 | ||
2695 | static inline struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) | |
2696 | { | |
2697 | struct io_wq_work_node *node; | |
2698 | ||
2699 | node = wq_stack_extract(&ctx->submit_state.free_list); | |
c2b6c6bc | 2700 | return container_of(node, struct io_kiocb, comp_list); |
2b188cc1 JA |
2701 | } |
2702 | ||
e1d767f0 | 2703 | static inline void io_put_file(struct file *file) |
8da11c19 | 2704 | { |
e1d767f0 | 2705 | if (file) |
8da11c19 PB |
2706 | fput(file); |
2707 | } | |
2708 | ||
6b639522 | 2709 | static inline void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2710 | { |
094bae49 | 2711 | unsigned int flags = req->flags; |
929a3af9 | 2712 | |
867f8fa5 | 2713 | if (unlikely(flags & IO_REQ_CLEAN_FLAGS)) |
3a0a6902 | 2714 | io_clean_op(req); |
e1d767f0 PB |
2715 | if (!(flags & REQ_F_FIXED_FILE)) |
2716 | io_put_file(req->file); | |
e65ef56d JA |
2717 | } |
2718 | ||
f5c6cf2a | 2719 | static __cold void io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2720 | { |
51a4cc11 | 2721 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2722 | |
7ac1edc4 | 2723 | io_req_put_rsrc(req); |
216578e5 | 2724 | io_dismantle_req(req); |
7c660731 | 2725 | io_put_task(req->task, 1); |
c6ca97b3 | 2726 | |
79ebeaee | 2727 | spin_lock(&ctx->completion_lock); |
c2b6c6bc | 2728 | wq_list_add_head(&req->comp_list, &ctx->locked_free_list); |
c34b025f | 2729 | ctx->locked_free_nr++; |
79ebeaee | 2730 | spin_unlock(&ctx->completion_lock); |
e65ef56d JA |
2731 | } |
2732 | ||
f2f87370 PB |
2733 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2734 | { | |
2735 | struct io_kiocb *nxt = req->link; | |
2736 | ||
2737 | req->link = nxt->link; | |
2738 | nxt->link = NULL; | |
2739 | } | |
2740 | ||
81ec803b | 2741 | static struct io_kiocb *io_disarm_linked_timeout(struct io_kiocb *req) |
33cc89a9 | 2742 | __must_hold(&req->ctx->completion_lock) |
89b263f6 | 2743 | __must_hold(&req->ctx->timeout_lock) |
2665abfd | 2744 | { |
33cc89a9 | 2745 | struct io_kiocb *link = req->link; |
f2f87370 | 2746 | |
b97e736a | 2747 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
c9abd7ad | 2748 | struct io_timeout_data *io = link->async_data; |
7c86ffee | 2749 | |
f2f87370 | 2750 | io_remove_next_linked(req); |
90cd7e42 | 2751 | link->timeout.head = NULL; |
fd9c7bc5 | 2752 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ef9dd637 | 2753 | list_del(&link->timeout.list); |
81ec803b | 2754 | return link; |
c9abd7ad PB |
2755 | } |
2756 | } | |
81ec803b | 2757 | return NULL; |
7c86ffee PB |
2758 | } |
2759 | ||
d148ca4b | 2760 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 2761 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 2762 | { |
33cc89a9 | 2763 | struct io_kiocb *nxt, *link = req->link; |
04c76b41 | 2764 | bool ignore_cqes = req->flags & REQ_F_SKIP_LINK_CQES; |
9e645e11 | 2765 | |
f2f87370 | 2766 | req->link = NULL; |
f2f87370 | 2767 | while (link) { |
a8295b98 HX |
2768 | long res = -ECANCELED; |
2769 | ||
2770 | if (link->flags & REQ_F_FAIL) | |
cef216fc | 2771 | res = link->cqe.res; |
a8295b98 | 2772 | |
f2f87370 PB |
2773 | nxt = link->link; |
2774 | link->link = NULL; | |
2665abfd | 2775 | |
cef216fc | 2776 | trace_io_uring_fail_link(req->ctx, req, req->cqe.user_data, |
502c87d6 SR |
2777 | req->opcode, link); |
2778 | ||
4e118cd9 PB |
2779 | if (ignore_cqes) |
2780 | link->flags |= REQ_F_CQE_SKIP; | |
2781 | else | |
04c76b41 | 2782 | link->flags &= ~REQ_F_CQE_SKIP; |
4e118cd9 | 2783 | __io_req_complete_post(link, res, 0); |
f2f87370 | 2784 | link = nxt; |
9e645e11 | 2785 | } |
33cc89a9 | 2786 | } |
9e645e11 | 2787 | |
33cc89a9 PB |
2788 | static bool io_disarm_next(struct io_kiocb *req) |
2789 | __must_hold(&req->ctx->completion_lock) | |
2790 | { | |
81ec803b | 2791 | struct io_kiocb *link = NULL; |
33cc89a9 PB |
2792 | bool posted = false; |
2793 | ||
0756a869 | 2794 | if (req->flags & REQ_F_ARM_LTIMEOUT) { |
81ec803b | 2795 | link = req->link; |
906c6caa | 2796 | req->flags &= ~REQ_F_ARM_LTIMEOUT; |
0756a869 PB |
2797 | if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { |
2798 | io_remove_next_linked(req); | |
4e118cd9 | 2799 | io_req_tw_post_queue(link, -ECANCELED, 0); |
0756a869 PB |
2800 | posted = true; |
2801 | } | |
2802 | } else if (req->flags & REQ_F_LINK_TIMEOUT) { | |
89b263f6 JA |
2803 | struct io_ring_ctx *ctx = req->ctx; |
2804 | ||
2805 | spin_lock_irq(&ctx->timeout_lock); | |
81ec803b | 2806 | link = io_disarm_linked_timeout(req); |
89b263f6 | 2807 | spin_unlock_irq(&ctx->timeout_lock); |
81ec803b PB |
2808 | if (link) { |
2809 | posted = true; | |
2810 | io_req_tw_post_queue(link, -ECANCELED, 0); | |
2811 | } | |
89b263f6 | 2812 | } |
93d2bcd2 | 2813 | if (unlikely((req->flags & REQ_F_FAIL) && |
e4335ed3 | 2814 | !(req->flags & REQ_F_HARDLINK))) { |
33cc89a9 PB |
2815 | posted |= (req->link != NULL); |
2816 | io_fail_links(req); | |
2817 | } | |
2818 | return posted; | |
9e645e11 JA |
2819 | } |
2820 | ||
d81499bf PB |
2821 | static void __io_req_find_next_prep(struct io_kiocb *req) |
2822 | { | |
2823 | struct io_ring_ctx *ctx = req->ctx; | |
2824 | bool posted; | |
2825 | ||
2826 | spin_lock(&ctx->completion_lock); | |
2827 | posted = io_disarm_next(req); | |
60053be8 | 2828 | io_commit_cqring(ctx); |
d81499bf PB |
2829 | spin_unlock(&ctx->completion_lock); |
2830 | if (posted) | |
2831 | io_cqring_ev_posted(ctx); | |
2832 | } | |
2833 | ||
2834 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) | |
c69f8dbe | 2835 | { |
33cc89a9 | 2836 | struct io_kiocb *nxt; |
944e58bf | 2837 | |
9e645e11 JA |
2838 | /* |
2839 | * If LINK is set, we have dependent requests in this chain. If we | |
2840 | * didn't fail this request, queue the first one up, moving any other | |
2841 | * dependencies to the next request. In case of failure, fail the rest | |
2842 | * of the chain. | |
2843 | */ | |
d81499bf PB |
2844 | if (unlikely(req->flags & IO_DISARM_MASK)) |
2845 | __io_req_find_next_prep(req); | |
33cc89a9 PB |
2846 | nxt = req->link; |
2847 | req->link = NULL; | |
2848 | return nxt; | |
4d7dd462 | 2849 | } |
9e645e11 | 2850 | |
f237c30a | 2851 | static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked) |
2c32395d PB |
2852 | { |
2853 | if (!ctx) | |
2854 | return; | |
ef060ea9 JA |
2855 | if (ctx->flags & IORING_SETUP_TASKRUN_FLAG) |
2856 | atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags); | |
f237c30a | 2857 | if (*locked) { |
c450178d | 2858 | io_submit_flush_completions(ctx); |
2c32395d | 2859 | mutex_unlock(&ctx->uring_lock); |
f237c30a | 2860 | *locked = false; |
2c32395d PB |
2861 | } |
2862 | percpu_ref_put(&ctx->refs); | |
2863 | } | |
2864 | ||
f28c240e HX |
2865 | static inline void ctx_commit_and_unlock(struct io_ring_ctx *ctx) |
2866 | { | |
2867 | io_commit_cqring(ctx); | |
2868 | spin_unlock(&ctx->completion_lock); | |
2869 | io_cqring_ev_posted(ctx); | |
2870 | } | |
2871 | ||
2872 | static void handle_prev_tw_list(struct io_wq_work_node *node, | |
2873 | struct io_ring_ctx **ctx, bool *uring_locked) | |
2874 | { | |
2875 | if (*ctx && !*uring_locked) | |
2876 | spin_lock(&(*ctx)->completion_lock); | |
2877 | ||
2878 | do { | |
2879 | struct io_wq_work_node *next = node->next; | |
2880 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2881 | io_task_work.node); | |
2882 | ||
34d2bfe7 JA |
2883 | prefetch(container_of(next, struct io_kiocb, io_task_work.node)); |
2884 | ||
f28c240e HX |
2885 | if (req->ctx != *ctx) { |
2886 | if (unlikely(!*uring_locked && *ctx)) | |
2887 | ctx_commit_and_unlock(*ctx); | |
2888 | ||
2889 | ctx_flush_and_put(*ctx, uring_locked); | |
2890 | *ctx = req->ctx; | |
2891 | /* if not contended, grab and improve batching */ | |
2892 | *uring_locked = mutex_trylock(&(*ctx)->uring_lock); | |
2893 | percpu_ref_get(&(*ctx)->refs); | |
2894 | if (unlikely(!*uring_locked)) | |
2895 | spin_lock(&(*ctx)->completion_lock); | |
2896 | } | |
2897 | if (likely(*uring_locked)) | |
2898 | req->io_task_work.func(req, uring_locked); | |
2899 | else | |
cef216fc | 2900 | __io_req_complete_post(req, req->cqe.res, |
cc3cec83 | 2901 | io_put_kbuf_comp(req)); |
f28c240e HX |
2902 | node = next; |
2903 | } while (node); | |
2904 | ||
2905 | if (unlikely(!*uring_locked)) | |
2906 | ctx_commit_and_unlock(*ctx); | |
2907 | } | |
2908 | ||
2909 | static void handle_tw_list(struct io_wq_work_node *node, | |
2910 | struct io_ring_ctx **ctx, bool *locked) | |
9f8d032a HX |
2911 | { |
2912 | do { | |
2913 | struct io_wq_work_node *next = node->next; | |
2914 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
2915 | io_task_work.node); | |
2916 | ||
34d2bfe7 JA |
2917 | prefetch(container_of(next, struct io_kiocb, io_task_work.node)); |
2918 | ||
9f8d032a HX |
2919 | if (req->ctx != *ctx) { |
2920 | ctx_flush_and_put(*ctx, locked); | |
2921 | *ctx = req->ctx; | |
2922 | /* if not contended, grab and improve batching */ | |
2923 | *locked = mutex_trylock(&(*ctx)->uring_lock); | |
2924 | percpu_ref_get(&(*ctx)->refs); | |
2925 | } | |
2926 | req->io_task_work.func(req, locked); | |
2927 | node = next; | |
2928 | } while (node); | |
2929 | } | |
2930 | ||
7cbf1722 | 2931 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2932 | { |
f28c240e | 2933 | bool uring_locked = false; |
ebd0df2e | 2934 | struct io_ring_ctx *ctx = NULL; |
3f18407d PB |
2935 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, |
2936 | task_work); | |
c40f6379 | 2937 | |
16f72070 | 2938 | while (1) { |
f28c240e | 2939 | struct io_wq_work_node *node1, *node2; |
3f18407d PB |
2940 | |
2941 | spin_lock_irq(&tctx->task_lock); | |
3fe07bcd | 2942 | node1 = tctx->prio_task_list.first; |
f28c240e | 2943 | node2 = tctx->task_list.first; |
3f18407d | 2944 | INIT_WQ_LIST(&tctx->task_list); |
3fe07bcd | 2945 | INIT_WQ_LIST(&tctx->prio_task_list); |
f28c240e | 2946 | if (!node2 && !node1) |
6294f368 | 2947 | tctx->task_running = false; |
3f18407d | 2948 | spin_unlock_irq(&tctx->task_lock); |
f28c240e | 2949 | if (!node2 && !node1) |
6294f368 | 2950 | break; |
3f18407d | 2951 | |
f28c240e HX |
2952 | if (node1) |
2953 | handle_prev_tw_list(node1, &ctx, &uring_locked); | |
f28c240e HX |
2954 | if (node2) |
2955 | handle_tw_list(node2, &ctx, &uring_locked); | |
7cbf1722 | 2956 | cond_resched(); |
68ca8fc0 | 2957 | |
a6d97a8a | 2958 | if (data_race(!tctx->task_list.first) && |
3fe07bcd | 2959 | data_race(!tctx->prio_task_list.first) && uring_locked) |
68ca8fc0 | 2960 | io_submit_flush_completions(ctx); |
3f18407d | 2961 | } |
ebd0df2e | 2962 | |
f28c240e | 2963 | ctx_flush_and_put(ctx, &uring_locked); |
3cc7fdb9 PB |
2964 | |
2965 | /* relaxed read is enough as only the task itself sets ->in_idle */ | |
2966 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2967 | io_uring_drop_tctx_refs(current); | |
7cbf1722 JA |
2968 | } |
2969 | ||
3fe07bcd JA |
2970 | static void __io_req_task_work_add(struct io_kiocb *req, |
2971 | struct io_uring_task *tctx, | |
2972 | struct io_wq_work_list *list) | |
7cbf1722 | 2973 | { |
9f010507 | 2974 | struct io_ring_ctx *ctx = req->ctx; |
e09ee510 | 2975 | struct io_wq_work_node *node; |
0b81e80c | 2976 | unsigned long flags; |
6294f368 | 2977 | bool running; |
7cbf1722 | 2978 | |
0b81e80c | 2979 | spin_lock_irqsave(&tctx->task_lock, flags); |
3fe07bcd | 2980 | wq_list_add_tail(&req->io_task_work.node, list); |
6294f368 PB |
2981 | running = tctx->task_running; |
2982 | if (!running) | |
2983 | tctx->task_running = true; | |
0b81e80c | 2984 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2985 | |
2986 | /* task_work already pending, we're done */ | |
6294f368 | 2987 | if (running) |
e09ee510 | 2988 | return; |
7cbf1722 | 2989 | |
ef060ea9 JA |
2990 | if (ctx->flags & IORING_SETUP_TASKRUN_FLAG) |
2991 | atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags); | |
2992 | ||
3fe07bcd | 2993 | if (likely(!task_work_add(req->task, &tctx->task_work, ctx->notify_method))) |
e09ee510 | 2994 | return; |
2215bed9 | 2995 | |
0b81e80c | 2996 | spin_lock_irqsave(&tctx->task_lock, flags); |
6294f368 | 2997 | tctx->task_running = false; |
3fe07bcd | 2998 | node = wq_list_merge(&tctx->prio_task_list, &tctx->task_list); |
0b81e80c | 2999 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 | 3000 | |
e09ee510 PB |
3001 | while (node) { |
3002 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
3003 | node = node->next; | |
3004 | if (llist_add(&req->io_task_work.fallback_node, | |
3005 | &req->ctx->fallback_llist)) | |
3006 | schedule_delayed_work(&req->ctx->fallback_work, 1); | |
3007 | } | |
eab30c4d PB |
3008 | } |
3009 | ||
3fe07bcd JA |
3010 | static void io_req_task_work_add(struct io_kiocb *req) |
3011 | { | |
3012 | struct io_uring_task *tctx = req->task->io_uring; | |
3013 | ||
3014 | __io_req_task_work_add(req, tctx, &tctx->task_list); | |
3015 | } | |
3016 | ||
3017 | static void io_req_task_prio_work_add(struct io_kiocb *req) | |
3018 | { | |
3019 | struct io_uring_task *tctx = req->task->io_uring; | |
3020 | ||
3021 | if (req->ctx->flags & IORING_SETUP_SQPOLL) | |
3022 | __io_req_task_work_add(req, tctx, &tctx->prio_task_list); | |
3023 | else | |
3024 | __io_req_task_work_add(req, tctx, &tctx->task_list); | |
3025 | } | |
3026 | ||
4e118cd9 | 3027 | static void io_req_tw_post(struct io_kiocb *req, bool *locked) |
c40f6379 | 3028 | { |
4e118cd9 PB |
3029 | io_req_complete_post(req, req->cqe.res, req->cqe.flags); |
3030 | } | |
c40f6379 | 3031 | |
4e118cd9 PB |
3032 | static void io_req_tw_post_queue(struct io_kiocb *req, s32 res, u32 cflags) |
3033 | { | |
3034 | req->cqe.res = res; | |
3035 | req->cqe.flags = cflags; | |
3036 | req->io_task_work.func = io_req_tw_post; | |
3fe07bcd | 3037 | io_req_task_work_add(req); |
4e118cd9 PB |
3038 | } |
3039 | ||
f237c30a | 3040 | static void io_req_task_cancel(struct io_kiocb *req, bool *locked) |
c40f6379 | 3041 | { |
b18a1a45 | 3042 | /* not needed for normal modes, but SQPOLL depends on it */ |
971cf9c1 | 3043 | io_tw_lock(req->ctx, locked); |
cef216fc | 3044 | io_req_complete_failed(req, req->cqe.res); |
c40f6379 JA |
3045 | } |
3046 | ||
f237c30a | 3047 | static void io_req_task_submit(struct io_kiocb *req, bool *locked) |
c40f6379 | 3048 | { |
971cf9c1 | 3049 | io_tw_lock(req->ctx, locked); |
316319e8 | 3050 | /* req->task == current here, checking PF_EXITING is safe */ |
af066f31 | 3051 | if (likely(!(req->task->flags & PF_EXITING))) |
cbc2e203 | 3052 | io_queue_sqe(req); |
81b6d05c | 3053 | else |
2593553a | 3054 | io_req_complete_failed(req, -EFAULT); |
c40f6379 JA |
3055 | } |
3056 | ||
2c4b8eb6 | 3057 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 3058 | { |
cef216fc | 3059 | req->cqe.res = ret; |
5b0a6acc | 3060 | req->io_task_work.func = io_req_task_cancel; |
3fe07bcd | 3061 | io_req_task_work_add(req); |
c40f6379 JA |
3062 | } |
3063 | ||
2c4b8eb6 | 3064 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 3065 | { |
5b0a6acc | 3066 | req->io_task_work.func = io_req_task_submit; |
3fe07bcd | 3067 | io_req_task_work_add(req); |
a3df7698 PB |
3068 | } |
3069 | ||
773af691 JA |
3070 | static void io_req_task_queue_reissue(struct io_kiocb *req) |
3071 | { | |
77955efb | 3072 | req->io_task_work.func = io_queue_iowq; |
3fe07bcd | 3073 | io_req_task_work_add(req); |
773af691 JA |
3074 | } |
3075 | ||
57859f4d | 3076 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 3077 | { |
9b5f7bd9 | 3078 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
3079 | |
3080 | if (nxt) | |
906a8c3f | 3081 | io_req_task_queue(nxt); |
c69f8dbe JL |
3082 | } |
3083 | ||
3aa83bfb | 3084 | static void io_free_batch_list(struct io_ring_ctx *ctx, |
1cce17ac | 3085 | struct io_wq_work_node *node) |
3aa83bfb | 3086 | __must_hold(&ctx->uring_lock) |
5af1d13e | 3087 | { |
d4b7a5ef | 3088 | struct task_struct *task = NULL; |
37f0e767 | 3089 | int task_refs = 0; |
5af1d13e | 3090 | |
3aa83bfb PB |
3091 | do { |
3092 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
3093 | comp_list); | |
2d6500d4 | 3094 | |
a538be5b PB |
3095 | if (unlikely(req->flags & IO_REQ_CLEAN_SLOW_FLAGS)) { |
3096 | if (req->flags & REQ_F_REFCOUNT) { | |
3097 | node = req->comp_list.next; | |
3098 | if (!req_ref_put_and_test(req)) | |
3099 | continue; | |
3100 | } | |
b605a7fa PB |
3101 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
3102 | struct async_poll *apoll = req->apoll; | |
3103 | ||
3104 | if (apoll->double_poll) | |
3105 | kfree(apoll->double_poll); | |
3106 | list_add(&apoll->poll.wait.entry, | |
3107 | &ctx->apoll_cache); | |
3108 | req->flags &= ~REQ_F_POLLED; | |
3109 | } | |
da1a08c5 | 3110 | if (req->flags & IO_REQ_LINK_FLAGS) |
57859f4d | 3111 | io_queue_next(req); |
a538be5b PB |
3112 | if (unlikely(req->flags & IO_REQ_CLEAN_FLAGS)) |
3113 | io_clean_op(req); | |
c1e53a69 | 3114 | } |
a538be5b PB |
3115 | if (!(req->flags & REQ_F_FIXED_FILE)) |
3116 | io_put_file(req->file); | |
2d6500d4 | 3117 | |
ab409402 | 3118 | io_req_put_rsrc_locked(req, ctx); |
5af1d13e | 3119 | |
d4b7a5ef PB |
3120 | if (req->task != task) { |
3121 | if (task) | |
3122 | io_put_task(task, task_refs); | |
3123 | task = req->task; | |
3124 | task_refs = 0; | |
3125 | } | |
3126 | task_refs++; | |
c1e53a69 | 3127 | node = req->comp_list.next; |
fa05457a | 3128 | io_req_add_to_cache(req, ctx); |
3aa83bfb | 3129 | } while (node); |
d4b7a5ef | 3130 | |
d4b7a5ef PB |
3131 | if (task) |
3132 | io_put_task(task, task_refs); | |
7a743e22 PB |
3133 | } |
3134 | ||
c450178d | 3135 | static void __io_submit_flush_completions(struct io_ring_ctx *ctx) |
a141dd89 | 3136 | __must_hold(&ctx->uring_lock) |
905c172f | 3137 | { |
6f33b0bc | 3138 | struct io_wq_work_node *node, *prev; |
cd0ca2e0 | 3139 | struct io_submit_state *state = &ctx->submit_state; |
905c172f | 3140 | |
3d4aeb9f PB |
3141 | if (state->flush_cqes) { |
3142 | spin_lock(&ctx->completion_lock); | |
3143 | wq_list_for_each(node, prev, &state->compl_reqs) { | |
3144 | struct io_kiocb *req = container_of(node, struct io_kiocb, | |
6f33b0bc | 3145 | comp_list); |
5182ed2e | 3146 | |
f43de1f8 PB |
3147 | if (!(req->flags & REQ_F_CQE_SKIP)) |
3148 | __io_fill_cqe_req(ctx, req); | |
3d4aeb9f PB |
3149 | } |
3150 | ||
3151 | io_commit_cqring(ctx); | |
3152 | spin_unlock(&ctx->completion_lock); | |
3153 | io_cqring_ev_posted(ctx); | |
3154 | state->flush_cqes = false; | |
905c172f | 3155 | } |
5182ed2e | 3156 | |
1cce17ac | 3157 | io_free_batch_list(ctx, state->compl_reqs.first); |
6f33b0bc | 3158 | INIT_WQ_LIST(&state->compl_reqs); |
7a743e22 PB |
3159 | } |
3160 | ||
ba816ad6 JA |
3161 | /* |
3162 | * Drop reference to request, return next in chain (if there is one) if this | |
3163 | * was the last reference to this request. | |
3164 | */ | |
0d85035a | 3165 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 3166 | { |
9b5f7bd9 PB |
3167 | struct io_kiocb *nxt = NULL; |
3168 | ||
de9b4cca | 3169 | if (req_ref_put_and_test(req)) { |
da1a08c5 | 3170 | if (unlikely(req->flags & IO_REQ_LINK_FLAGS)) |
7819a1f6 | 3171 | nxt = io_req_find_next(req); |
f5c6cf2a | 3172 | io_free_req(req); |
2a44f467 | 3173 | } |
9b5f7bd9 | 3174 | return nxt; |
2b188cc1 JA |
3175 | } |
3176 | ||
0d85035a | 3177 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 3178 | { |
f5c6cf2a PB |
3179 | if (req_ref_put_and_test(req)) { |
3180 | io_queue_next(req); | |
e65ef56d | 3181 | io_free_req(req); |
543af3a1 | 3182 | } |
216578e5 PB |
3183 | } |
3184 | ||
6c503150 | 3185 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
3186 | { |
3187 | /* See comment at the top of this file */ | |
3188 | smp_rmb(); | |
e23de15f | 3189 | return __io_cqring_events(ctx); |
a3a0e43f JA |
3190 | } |
3191 | ||
fb5ccc98 PB |
3192 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
3193 | { | |
3194 | struct io_rings *rings = ctx->rings; | |
3195 | ||
3196 | /* make sure SQ entry isn't read before tail */ | |
3197 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
3198 | } | |
3199 | ||
4c6e277c JA |
3200 | static inline bool io_run_task_work(void) |
3201 | { | |
7f62d40d | 3202 | if (test_thread_flag(TIF_NOTIFY_SIGNAL) || task_work_pending(current)) { |
4c6e277c | 3203 | __set_current_state(TASK_RUNNING); |
7c5d8fa6 EB |
3204 | clear_notify_signal(); |
3205 | if (task_work_pending(current)) | |
3206 | task_work_run(); | |
4c6e277c JA |
3207 | return true; |
3208 | } | |
3209 | ||
3210 | return false; | |
bcda7baa JA |
3211 | } |
3212 | ||
5ba3c874 | 3213 | static int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin) |
def596e9 | 3214 | { |
5eef4e87 | 3215 | struct io_wq_work_node *pos, *start, *prev; |
d729cf9a | 3216 | unsigned int poll_flags = BLK_POLL_NOSLEEP; |
b688f11e | 3217 | DEFINE_IO_COMP_BATCH(iob); |
5ba3c874 | 3218 | int nr_events = 0; |
def596e9 JA |
3219 | |
3220 | /* | |
3221 | * Only spin for completions if we don't have multiple devices hanging | |
87a115fb | 3222 | * off our complete list. |
def596e9 | 3223 | */ |
87a115fb | 3224 | if (ctx->poll_multi_queue || force_nonspin) |
ef99b2d3 | 3225 | poll_flags |= BLK_POLL_ONESHOT; |
def596e9 | 3226 | |
5eef4e87 PB |
3227 | wq_list_for_each(pos, start, &ctx->iopoll_list) { |
3228 | struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list); | |
9adbd45d | 3229 | struct kiocb *kiocb = &req->rw.kiocb; |
a2416e1e | 3230 | int ret; |
def596e9 JA |
3231 | |
3232 | /* | |
581f9810 BM |
3233 | * Move completed and retryable entries to our local lists. |
3234 | * If we find a request that requires polling, break out | |
3235 | * and complete those lists first, if we have entries there. | |
def596e9 | 3236 | */ |
e3f721e6 | 3237 | if (READ_ONCE(req->iopoll_completed)) |
def596e9 JA |
3238 | break; |
3239 | ||
b688f11e | 3240 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, &iob, poll_flags); |
a2416e1e PB |
3241 | if (unlikely(ret < 0)) |
3242 | return ret; | |
3243 | else if (ret) | |
ef99b2d3 | 3244 | poll_flags |= BLK_POLL_ONESHOT; |
def596e9 | 3245 | |
3aadc23e | 3246 | /* iopoll may have completed current req */ |
b688f11e JA |
3247 | if (!rq_list_empty(iob.req_list) || |
3248 | READ_ONCE(req->iopoll_completed)) | |
e3f721e6 | 3249 | break; |
def596e9 JA |
3250 | } |
3251 | ||
b688f11e JA |
3252 | if (!rq_list_empty(iob.req_list)) |
3253 | iob.complete(&iob); | |
5eef4e87 PB |
3254 | else if (!pos) |
3255 | return 0; | |
def596e9 | 3256 | |
5eef4e87 PB |
3257 | prev = start; |
3258 | wq_list_for_each_resume(pos, prev) { | |
3259 | struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list); | |
3260 | ||
b3fa03fd PB |
3261 | /* order with io_complete_rw_iopoll(), e.g. ->result updates */ |
3262 | if (!smp_load_acquire(&req->iopoll_completed)) | |
e3f721e6 | 3263 | break; |
c0713540 | 3264 | nr_events++; |
83a13a41 PB |
3265 | if (unlikely(req->flags & REQ_F_CQE_SKIP)) |
3266 | continue; | |
91ef75a7 PB |
3267 | |
3268 | req->cqe.flags = io_put_kbuf(req, 0); | |
3269 | __io_fill_cqe_req(req->ctx, req); | |
e3f721e6 | 3270 | } |
def596e9 | 3271 | |
f5ed3bcd PB |
3272 | if (unlikely(!nr_events)) |
3273 | return 0; | |
3274 | ||
3275 | io_commit_cqring(ctx); | |
3276 | io_cqring_ev_posted_iopoll(ctx); | |
1cce17ac | 3277 | pos = start ? start->next : ctx->iopoll_list.first; |
5eef4e87 | 3278 | wq_list_cut(&ctx->iopoll_list, prev, start); |
1cce17ac | 3279 | io_free_batch_list(ctx, pos); |
5ba3c874 | 3280 | return nr_events; |
def596e9 JA |
3281 | } |
3282 | ||
def596e9 JA |
3283 | /* |
3284 | * We can't just wait for polled events to come to us, we have to actively | |
3285 | * find and complete them. | |
3286 | */ | |
c072481d | 3287 | static __cold void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
3288 | { |
3289 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
3290 | return; | |
3291 | ||
3292 | mutex_lock(&ctx->uring_lock); | |
5eef4e87 | 3293 | while (!wq_list_empty(&ctx->iopoll_list)) { |
b2edc0a7 | 3294 | /* let it sleep and repeat later if can't complete a request */ |
5ba3c874 | 3295 | if (io_do_iopoll(ctx, true) == 0) |
b2edc0a7 | 3296 | break; |
08f5439f JA |
3297 | /* |
3298 | * Ensure we allow local-to-the-cpu processing to take place, | |
3299 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 3300 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 3301 | */ |
3fcee5a6 PB |
3302 | if (need_resched()) { |
3303 | mutex_unlock(&ctx->uring_lock); | |
3304 | cond_resched(); | |
3305 | mutex_lock(&ctx->uring_lock); | |
3306 | } | |
def596e9 JA |
3307 | } |
3308 | mutex_unlock(&ctx->uring_lock); | |
3309 | } | |
3310 | ||
7668b92a | 3311 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 3312 | { |
7668b92a | 3313 | unsigned int nr_events = 0; |
e9979b36 | 3314 | int ret = 0; |
155bc950 | 3315 | unsigned long check_cq; |
500f9fba | 3316 | |
f39c8a5b PB |
3317 | /* |
3318 | * Don't enter poll loop if we already have events pending. | |
3319 | * If we do, we can potentially be spinning for commands that | |
3320 | * already triggered a CQE (eg in error). | |
3321 | */ | |
155bc950 DY |
3322 | check_cq = READ_ONCE(ctx->check_cq); |
3323 | if (check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT)) | |
f39c8a5b PB |
3324 | __io_cqring_overflow_flush(ctx, false); |
3325 | if (io_cqring_events(ctx)) | |
d487b43c | 3326 | return 0; |
155bc950 DY |
3327 | |
3328 | /* | |
3329 | * Similarly do not spin if we have not informed the user of any | |
3330 | * dropped CQE. | |
3331 | */ | |
3332 | if (unlikely(check_cq & BIT(IO_CHECK_CQ_DROPPED_BIT))) | |
3333 | return -EBADR; | |
3334 | ||
def596e9 | 3335 | do { |
500f9fba JA |
3336 | /* |
3337 | * If a submit got punted to a workqueue, we can have the | |
3338 | * application entering polling for a command before it gets | |
3339 | * issued. That app will hold the uring_lock for the duration | |
3340 | * of the poll right here, so we need to take a breather every | |
3341 | * now and then to ensure that the issue has a chance to add | |
3342 | * the poll to the issued list. Otherwise we can spin here | |
3343 | * forever, while the workqueue is stuck trying to acquire the | |
3344 | * very same mutex. | |
3345 | */ | |
5eef4e87 | 3346 | if (wq_list_empty(&ctx->iopoll_list)) { |
8f487ef2 PB |
3347 | u32 tail = ctx->cached_cq_tail; |
3348 | ||
500f9fba | 3349 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 3350 | io_run_task_work(); |
500f9fba | 3351 | mutex_lock(&ctx->uring_lock); |
def596e9 | 3352 | |
8f487ef2 PB |
3353 | /* some requests don't go through iopoll_list */ |
3354 | if (tail != ctx->cached_cq_tail || | |
5eef4e87 | 3355 | wq_list_empty(&ctx->iopoll_list)) |
e9979b36 | 3356 | break; |
500f9fba | 3357 | } |
5ba3c874 PB |
3358 | ret = io_do_iopoll(ctx, !min); |
3359 | if (ret < 0) | |
3360 | break; | |
3361 | nr_events += ret; | |
3362 | ret = 0; | |
3363 | } while (nr_events < min && !need_resched()); | |
d487b43c | 3364 | |
def596e9 JA |
3365 | return ret; |
3366 | } | |
3367 | ||
491381ce | 3368 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 3369 | { |
491381ce JA |
3370 | /* |
3371 | * Tell lockdep we inherited freeze protection from submission | |
3372 | * thread. | |
3373 | */ | |
3374 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 3375 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 3376 | |
1c98679d PB |
3377 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
3378 | sb_end_write(sb); | |
2b188cc1 JA |
3379 | } |
3380 | } | |
3381 | ||
b63534c4 | 3382 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 3383 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 3384 | { |
ab454438 | 3385 | struct io_async_rw *rw = req->async_data; |
b63534c4 | 3386 | |
d886e185 | 3387 | if (!req_has_async_data(req)) |
ab454438 | 3388 | return !io_req_prep_async(req); |
538941e2 | 3389 | iov_iter_restore(&rw->s.iter, &rw->s.iter_state); |
ab454438 | 3390 | return true; |
b63534c4 | 3391 | } |
b63534c4 | 3392 | |
3e6a0d3c | 3393 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 3394 | { |
355afaeb | 3395 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 3396 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 3397 | |
355afaeb JA |
3398 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
3399 | return false; | |
3e6a0d3c JA |
3400 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
3401 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 3402 | return false; |
7c977a58 JA |
3403 | /* |
3404 | * If ref is dying, we might be running poll reap from the exit work. | |
3405 | * Don't attempt to reissue from that path, just let it fail with | |
3406 | * -EAGAIN. | |
3407 | */ | |
3e6a0d3c JA |
3408 | if (percpu_ref_is_dying(&ctx->refs)) |
3409 | return false; | |
ef046888 JA |
3410 | /* |
3411 | * Play it safe and assume not safe to re-import and reissue if we're | |
3412 | * not in the original thread group (or in task context). | |
3413 | */ | |
3414 | if (!same_thread_group(req->task, current) || !in_task()) | |
3415 | return false; | |
3e6a0d3c JA |
3416 | return true; |
3417 | } | |
e82ad485 | 3418 | #else |
a1ff1e3f | 3419 | static bool io_resubmit_prep(struct io_kiocb *req) |
e82ad485 JA |
3420 | { |
3421 | return false; | |
3422 | } | |
e82ad485 | 3423 | static bool io_rw_should_reissue(struct io_kiocb *req) |
3e6a0d3c | 3424 | { |
b63534c4 JA |
3425 | return false; |
3426 | } | |
3e6a0d3c | 3427 | #endif |
b63534c4 | 3428 | |
8ef12efe | 3429 | static bool __io_complete_rw_common(struct io_kiocb *req, long res) |
a1d7c393 | 3430 | { |
f63cf519 | 3431 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) { |
b65c128f | 3432 | kiocb_end_write(req); |
f63cf519 JA |
3433 | fsnotify_modify(req->file); |
3434 | } else { | |
3435 | fsnotify_access(req->file); | |
3436 | } | |
cef216fc | 3437 | if (unlikely(res != req->cqe.res)) { |
9532b99b PB |
3438 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && |
3439 | io_rw_should_reissue(req)) { | |
1bacd264 | 3440 | req->flags |= REQ_F_REISSUE | REQ_F_PARTIAL_IO; |
8ef12efe | 3441 | return true; |
9532b99b | 3442 | } |
93d2bcd2 | 3443 | req_set_fail(req); |
cef216fc | 3444 | req->cqe.res = res; |
9532b99b | 3445 | } |
8ef12efe JA |
3446 | return false; |
3447 | } | |
3448 | ||
cc8e9ba7 | 3449 | static inline void io_req_task_complete(struct io_kiocb *req, bool *locked) |
8ef12efe | 3450 | { |
cef216fc | 3451 | int res = req->cqe.res; |
126180b9 PB |
3452 | |
3453 | if (*locked) { | |
cc3cec83 | 3454 | io_req_complete_state(req, res, io_put_kbuf(req, 0)); |
fff4e40e | 3455 | io_req_add_compl_list(req); |
126180b9 | 3456 | } else { |
cc3cec83 JA |
3457 | io_req_complete_post(req, res, |
3458 | io_put_kbuf(req, IO_URING_F_UNLOCKED)); | |
126180b9 | 3459 | } |
8ef12efe JA |
3460 | } |
3461 | ||
00f6e68b | 3462 | static void __io_complete_rw(struct io_kiocb *req, long res, |
8ef12efe JA |
3463 | unsigned int issue_flags) |
3464 | { | |
3465 | if (__io_complete_rw_common(req, res)) | |
3466 | return; | |
cef216fc | 3467 | __io_req_complete(req, issue_flags, req->cqe.res, |
cc3cec83 | 3468 | io_put_kbuf(req, issue_flags)); |
ba816ad6 JA |
3469 | } |
3470 | ||
6b19b766 | 3471 | static void io_complete_rw(struct kiocb *kiocb, long res) |
ba816ad6 | 3472 | { |
9adbd45d | 3473 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 3474 | |
8ef12efe JA |
3475 | if (__io_complete_rw_common(req, res)) |
3476 | return; | |
cef216fc | 3477 | req->cqe.res = res; |
8ef12efe | 3478 | req->io_task_work.func = io_req_task_complete; |
3fe07bcd | 3479 | io_req_task_prio_work_add(req); |
2b188cc1 JA |
3480 | } |
3481 | ||
6b19b766 | 3482 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res) |
def596e9 | 3483 | { |
9adbd45d | 3484 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 3485 | |
491381ce JA |
3486 | if (kiocb->ki_flags & IOCB_WRITE) |
3487 | kiocb_end_write(req); | |
cef216fc | 3488 | if (unlikely(res != req->cqe.res)) { |
b66ceaf3 | 3489 | if (res == -EAGAIN && io_rw_should_reissue(req)) { |
1bacd264 | 3490 | req->flags |= REQ_F_REISSUE | REQ_F_PARTIAL_IO; |
b66ceaf3 | 3491 | return; |
9532b99b | 3492 | } |
cef216fc | 3493 | req->cqe.res = res; |
8c130827 | 3494 | } |
bbde017a | 3495 | |
b3fa03fd PB |
3496 | /* order with io_iopoll_complete() checking ->iopoll_completed */ |
3497 | smp_store_release(&req->iopoll_completed, 1); | |
def596e9 JA |
3498 | } |
3499 | ||
3500 | /* | |
3501 | * After the iocb has been issued, it's safe to be found on the poll list. | |
3502 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 3503 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
3504 | * accessing the kiocb cookie. |
3505 | */ | |
9882131c | 3506 | static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags) |
def596e9 JA |
3507 | { |
3508 | struct io_ring_ctx *ctx = req->ctx; | |
3b44b371 | 3509 | const bool needs_lock = issue_flags & IO_URING_F_UNLOCKED; |
cb3d8972 PB |
3510 | |
3511 | /* workqueue context doesn't hold uring_lock, grab it now */ | |
3b44b371 | 3512 | if (unlikely(needs_lock)) |
cb3d8972 | 3513 | mutex_lock(&ctx->uring_lock); |
def596e9 JA |
3514 | |
3515 | /* | |
3516 | * Track whether we have multiple files in our lists. This will impact | |
3517 | * how we do polling eventually, not spinning if we're on potentially | |
3518 | * different devices. | |
3519 | */ | |
5eef4e87 | 3520 | if (wq_list_empty(&ctx->iopoll_list)) { |
915b3dde HX |
3521 | ctx->poll_multi_queue = false; |
3522 | } else if (!ctx->poll_multi_queue) { | |
def596e9 JA |
3523 | struct io_kiocb *list_req; |
3524 | ||
5eef4e87 PB |
3525 | list_req = container_of(ctx->iopoll_list.first, struct io_kiocb, |
3526 | comp_list); | |
30da1b45 | 3527 | if (list_req->file != req->file) |
915b3dde | 3528 | ctx->poll_multi_queue = true; |
def596e9 JA |
3529 | } |
3530 | ||
3531 | /* | |
3532 | * For fast devices, IO may have already completed. If it has, add | |
3533 | * it to the front so we find it first. | |
3534 | */ | |
65a6543d | 3535 | if (READ_ONCE(req->iopoll_completed)) |
5eef4e87 | 3536 | wq_list_add_head(&req->comp_list, &ctx->iopoll_list); |
def596e9 | 3537 | else |
5eef4e87 | 3538 | wq_list_add_tail(&req->comp_list, &ctx->iopoll_list); |
bdcd3eab | 3539 | |
3b44b371 | 3540 | if (unlikely(needs_lock)) { |
cb3d8972 PB |
3541 | /* |
3542 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handle | |
3543 | * in sq thread task context or in io worker task context. If | |
3544 | * current task context is sq thread, we don't need to check | |
3545 | * whether should wake up sq thread. | |
3546 | */ | |
3547 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
3548 | wq_has_sleeper(&ctx->sq_data->wait)) | |
3549 | wake_up(&ctx->sq_data->wait); | |
3550 | ||
3551 | mutex_unlock(&ctx->uring_lock); | |
3552 | } | |
def596e9 JA |
3553 | } |
3554 | ||
4503b767 JA |
3555 | static bool io_bdev_nowait(struct block_device *bdev) |
3556 | { | |
9ba0d0c8 | 3557 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
3558 | } |
3559 | ||
2b188cc1 JA |
3560 | /* |
3561 | * If we tracked the file through the SCM inflight mechanism, we could support | |
3562 | * any file. For now, just ensure that anything potentially problematic is done | |
3563 | * inline. | |
3564 | */ | |
88459b50 | 3565 | static bool __io_file_supports_nowait(struct file *file, umode_t mode) |
2b188cc1 | 3566 | { |
4503b767 | 3567 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
3568 | if (IS_ENABLED(CONFIG_BLOCK) && |
3569 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
3570 | return true; |
3571 | return false; | |
3572 | } | |
976517f1 | 3573 | if (S_ISSOCK(mode)) |
2b188cc1 | 3574 | return true; |
4503b767 | 3575 | if (S_ISREG(mode)) { |
4e7b5671 CH |
3576 | if (IS_ENABLED(CONFIG_BLOCK) && |
3577 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
3578 | file->f_op != &io_uring_fops) |
3579 | return true; | |
3580 | return false; | |
3581 | } | |
2b188cc1 | 3582 | |
c5b85625 JA |
3583 | /* any ->read/write should understand O_NONBLOCK */ |
3584 | if (file->f_flags & O_NONBLOCK) | |
3585 | return true; | |
35645ac3 | 3586 | return file->f_mode & FMODE_NOWAIT; |
2b188cc1 | 3587 | } |
c5b85625 | 3588 | |
88459b50 PB |
3589 | /* |
3590 | * If we tracked the file through the SCM inflight mechanism, we could support | |
3591 | * any file. For now, just ensure that anything potentially problematic is done | |
3592 | * inline. | |
3593 | */ | |
3594 | static unsigned int io_file_get_flags(struct file *file) | |
3595 | { | |
3596 | umode_t mode = file_inode(file)->i_mode; | |
3597 | unsigned int res = 0; | |
af197f50 | 3598 | |
88459b50 PB |
3599 | if (S_ISREG(mode)) |
3600 | res |= FFS_ISREG; | |
3601 | if (__io_file_supports_nowait(file, mode)) | |
3602 | res |= FFS_NOWAIT; | |
5e45690a JA |
3603 | if (io_file_need_scm(file)) |
3604 | res |= FFS_SCM; | |
88459b50 | 3605 | return res; |
2b188cc1 JA |
3606 | } |
3607 | ||
35645ac3 | 3608 | static inline bool io_file_supports_nowait(struct io_kiocb *req) |
7b29f92d | 3609 | { |
88459b50 | 3610 | return req->flags & REQ_F_SUPPORT_NOWAIT; |
7b29f92d JA |
3611 | } |
3612 | ||
b9a6b8f9 | 3613 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 3614 | { |
9adbd45d | 3615 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
3616 | unsigned ioprio; |
3617 | int ret; | |
2b188cc1 | 3618 | |
2b188cc1 | 3619 | kiocb->ki_pos = READ_ONCE(sqe->off); |
05b538c1 PB |
3620 | /* used for fixed read/write too - just read unconditionally */ |
3621 | req->buf_index = READ_ONCE(sqe->buf_index); | |
3622 | ||
3623 | if (req->opcode == IORING_OP_READ_FIXED || | |
3624 | req->opcode == IORING_OP_WRITE_FIXED) { | |
3625 | struct io_ring_ctx *ctx = req->ctx; | |
3626 | u16 index; | |
3627 | ||
3628 | if (unlikely(req->buf_index >= ctx->nr_user_bufs)) | |
3629 | return -EFAULT; | |
3630 | index = array_index_nospec(req->buf_index, ctx->nr_user_bufs); | |
3631 | req->imu = ctx->user_bufs[index]; | |
3632 | io_req_set_rsrc_node(req, ctx, 0); | |
3633 | } | |
9adbd45d | 3634 | |
fb27274a PB |
3635 | ioprio = READ_ONCE(sqe->ioprio); |
3636 | if (ioprio) { | |
3637 | ret = ioprio_check_cap(ioprio); | |
3638 | if (ret) | |
3639 | return ret; | |
3640 | ||
3641 | kiocb->ki_ioprio = ioprio; | |
3642 | } else { | |
3643 | kiocb->ki_ioprio = get_current_ioprio(); | |
eae071c9 PB |
3644 | } |
3645 | ||
3529d8c2 JA |
3646 | req->rw.addr = READ_ONCE(sqe->addr); |
3647 | req->rw.len = READ_ONCE(sqe->len); | |
584b0180 | 3648 | req->rw.flags = READ_ONCE(sqe->rw_flags); |
2b188cc1 | 3649 | return 0; |
2b188cc1 JA |
3650 | } |
3651 | ||
3652 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
3653 | { | |
3654 | switch (ret) { | |
3655 | case -EIOCBQUEUED: | |
3656 | break; | |
3657 | case -ERESTARTSYS: | |
3658 | case -ERESTARTNOINTR: | |
3659 | case -ERESTARTNOHAND: | |
3660 | case -ERESTART_RESTARTBLOCK: | |
3661 | /* | |
3662 | * We can't just restart the syscall, since previously | |
3663 | * submitted sqes may already be in progress. Just fail this | |
3664 | * IO with EINTR. | |
3665 | */ | |
3666 | ret = -EINTR; | |
df561f66 | 3667 | fallthrough; |
2b188cc1 | 3668 | default: |
6b19b766 | 3669 | kiocb->ki_complete(kiocb, ret); |
2b188cc1 JA |
3670 | } |
3671 | } | |
3672 | ||
b4aec400 | 3673 | static inline loff_t *io_kiocb_update_pos(struct io_kiocb *req) |
d34e1e5b DY |
3674 | { |
3675 | struct kiocb *kiocb = &req->rw.kiocb; | |
3676 | ||
6f83ab22 JA |
3677 | if (kiocb->ki_pos != -1) |
3678 | return &kiocb->ki_pos; | |
3679 | ||
3680 | if (!(req->file->f_mode & FMODE_STREAM)) { | |
3681 | req->flags |= REQ_F_CUR_POS; | |
3682 | kiocb->ki_pos = req->file->f_pos; | |
3683 | return &kiocb->ki_pos; | |
d34e1e5b | 3684 | } |
6f83ab22 JA |
3685 | |
3686 | kiocb->ki_pos = 0; | |
3687 | return NULL; | |
d34e1e5b DY |
3688 | } |
3689 | ||
2ea537ca | 3690 | static void kiocb_done(struct io_kiocb *req, ssize_t ret, |
889fca73 | 3691 | unsigned int issue_flags) |
ba816ad6 | 3692 | { |
e8c2bc1f | 3693 | struct io_async_rw *io = req->async_data; |
ba04291e | 3694 | |
227c0c96 | 3695 | /* add previously done IO, if any */ |
d886e185 | 3696 | if (req_has_async_data(req) && io->bytes_done > 0) { |
227c0c96 | 3697 | if (ret < 0) |
e8c2bc1f | 3698 | ret = io->bytes_done; |
227c0c96 | 3699 | else |
e8c2bc1f | 3700 | ret += io->bytes_done; |
227c0c96 JA |
3701 | } |
3702 | ||
ba04291e | 3703 | if (req->flags & REQ_F_CUR_POS) |
2ea537ca PB |
3704 | req->file->f_pos = req->rw.kiocb.ki_pos; |
3705 | if (ret >= 0 && (req->rw.kiocb.ki_complete == io_complete_rw)) | |
00f6e68b | 3706 | __io_complete_rw(req, ret, issue_flags); |
ba816ad6 | 3707 | else |
2ea537ca | 3708 | io_rw_done(&req->rw.kiocb, ret); |
97284637 | 3709 | |
b66ceaf3 | 3710 | if (req->flags & REQ_F_REISSUE) { |
97284637 | 3711 | req->flags &= ~REQ_F_REISSUE; |
b91ef187 | 3712 | if (io_resubmit_prep(req)) |
773af691 | 3713 | io_req_task_queue_reissue(req); |
b91ef187 PB |
3714 | else |
3715 | io_req_task_queue_fail(req, ret); | |
97284637 | 3716 | } |
ba816ad6 JA |
3717 | } |
3718 | ||
eae071c9 PB |
3719 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
3720 | struct io_mapped_ubuf *imu) | |
edafccee | 3721 | { |
9adbd45d | 3722 | size_t len = req->rw.len; |
75769e3f | 3723 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 3724 | size_t offset; |
edafccee | 3725 | |
75769e3f | 3726 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
3727 | return -EFAULT; |
3728 | /* not inside the mapped region */ | |
4751f53d | 3729 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
3730 | return -EFAULT; |
3731 | ||
3732 | /* | |
3733 | * May not be a start of buffer, set size appropriately | |
3734 | * and advance us to the beginning. | |
3735 | */ | |
3736 | offset = buf_addr - imu->ubuf; | |
3737 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3738 | |
3739 | if (offset) { | |
3740 | /* | |
3741 | * Don't use iov_iter_advance() here, as it's really slow for | |
3742 | * using the latter parts of a big fixed buffer - it iterates | |
3743 | * over each segment manually. We can cheat a bit here, because | |
3744 | * we know that: | |
3745 | * | |
3746 | * 1) it's a BVEC iter, we set it up | |
3747 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3748 | * first and last bvec | |
3749 | * | |
3750 | * So just find our index, and adjust the iterator afterwards. | |
3751 | * If the offset is within the first bvec (or the whole first | |
3752 | * bvec, just use iov_iter_advance(). This makes it easier | |
3753 | * since we can just skip the first segment, which may not | |
3754 | * be PAGE_SIZE aligned. | |
3755 | */ | |
3756 | const struct bio_vec *bvec = imu->bvec; | |
3757 | ||
3758 | if (offset <= bvec->bv_len) { | |
3759 | iov_iter_advance(iter, offset); | |
3760 | } else { | |
3761 | unsigned long seg_skip; | |
3762 | ||
3763 | /* skip first vec */ | |
3764 | offset -= bvec->bv_len; | |
3765 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3766 | ||
3767 | iter->bvec = bvec + seg_skip; | |
3768 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3769 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3770 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3771 | } |
3772 | } | |
3773 | ||
847595de | 3774 | return 0; |
edafccee JA |
3775 | } |
3776 | ||
5106dd6e JA |
3777 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
3778 | unsigned int issue_flags) | |
eae071c9 | 3779 | { |
05b538c1 PB |
3780 | if (WARN_ON_ONCE(!req->imu)) |
3781 | return -EFAULT; | |
3782 | return __io_import_fixed(req, rw, iter, req->imu); | |
eae071c9 PB |
3783 | } |
3784 | ||
9cfc7e94 JA |
3785 | static int io_buffer_add_list(struct io_ring_ctx *ctx, |
3786 | struct io_buffer_list *bl, unsigned int bgid) | |
bcda7baa | 3787 | { |
dbc7d452 | 3788 | bl->bgid = bgid; |
9cfc7e94 JA |
3789 | if (bgid < BGID_ARRAY) |
3790 | return 0; | |
3791 | ||
3792 | return xa_err(xa_store(&ctx->io_bl_xa, bgid, bl, GFP_KERNEL)); | |
dbc7d452 JA |
3793 | } |
3794 | ||
149c69b0 | 3795 | static void __user *io_provided_buffer_select(struct io_kiocb *req, size_t *len, |
984824db | 3796 | struct io_buffer_list *bl) |
bcda7baa | 3797 | { |
e7637a49 JA |
3798 | if (!list_empty(&bl->buf_list)) { |
3799 | struct io_buffer *kbuf; | |
bcda7baa | 3800 | |
dbc7d452 JA |
3801 | kbuf = list_first_entry(&bl->buf_list, struct io_buffer, list); |
3802 | list_del(&kbuf->list); | |
bcda7baa JA |
3803 | if (*len > kbuf->len) |
3804 | *len = kbuf->len; | |
30d51dd4 PB |
3805 | req->flags |= REQ_F_BUFFER_SELECTED; |
3806 | req->kbuf = kbuf; | |
e7637a49 | 3807 | req->buf_index = kbuf->bid; |
984824db | 3808 | return u64_to_user_ptr(kbuf->addr); |
e7637a49 | 3809 | } |
984824db | 3810 | return NULL; |
149c69b0 JA |
3811 | } |
3812 | ||
c7fb1942 JA |
3813 | static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len, |
3814 | struct io_buffer_list *bl, | |
3815 | unsigned int issue_flags) | |
3816 | { | |
3817 | struct io_uring_buf_ring *br = bl->buf_ring; | |
3818 | struct io_uring_buf *buf; | |
c6e9fa5c | 3819 | __u16 head = bl->head; |
c7fb1942 | 3820 | |
fc9375e3 | 3821 | if (unlikely(smp_load_acquire(&br->tail) == head)) |
984824db | 3822 | return NULL; |
c7fb1942 JA |
3823 | |
3824 | head &= bl->mask; | |
3825 | if (head < IO_BUFFER_LIST_BUF_PER_PAGE) { | |
3826 | buf = &br->bufs[head]; | |
bcda7baa | 3827 | } else { |
c7fb1942 | 3828 | int off = head & (IO_BUFFER_LIST_BUF_PER_PAGE - 1); |
97da4a53 | 3829 | int index = head / IO_BUFFER_LIST_BUF_PER_PAGE; |
c7fb1942 JA |
3830 | buf = page_address(bl->buf_pages[index]); |
3831 | buf += off; | |
bcda7baa | 3832 | } |
c7fb1942 JA |
3833 | if (*len > buf->len) |
3834 | *len = buf->len; | |
3835 | req->flags |= REQ_F_BUFFER_RING; | |
3836 | req->buf_list = bl; | |
3837 | req->buf_index = buf->bid; | |
bcda7baa | 3838 | |
a76c0b31 | 3839 | if (issue_flags & IO_URING_F_UNLOCKED || !file_can_poll(req->file)) { |
984824db CH |
3840 | /* |
3841 | * If we came in unlocked, we have no choice but to consume the | |
3842 | * buffer here. This does mean it'll be pinned until the IO | |
3843 | * completes. But coming in unlocked means we're in io-wq | |
3844 | * context, hence there should be no further retry. For the | |
3845 | * locked case, the caller must ensure to call the commit when | |
3846 | * the transfer completes (or if we get -EAGAIN and must poll | |
3847 | * or retry). | |
3848 | */ | |
3849 | req->buf_list = NULL; | |
3850 | bl->head++; | |
3851 | } | |
c7fb1942 | 3852 | return u64_to_user_ptr(buf->addr); |
bcda7baa JA |
3853 | } |
3854 | ||
c54d52c2 | 3855 | static void __user *io_buffer_select(struct io_kiocb *req, size_t *len, |
4e906702 | 3856 | unsigned int issue_flags) |
4d954c25 | 3857 | { |
dbc7d452 JA |
3858 | struct io_ring_ctx *ctx = req->ctx; |
3859 | struct io_buffer_list *bl; | |
984824db | 3860 | void __user *ret = NULL; |
bcda7baa | 3861 | |
f8929630 | 3862 | io_ring_submit_lock(req->ctx, issue_flags); |
4d954c25 | 3863 | |
4e906702 | 3864 | bl = io_buffer_get_list(ctx, req->buf_index); |
984824db CH |
3865 | if (likely(bl)) { |
3866 | if (bl->buf_nr_pages) | |
3867 | ret = io_ring_buffer_select(req, len, bl, issue_flags); | |
3868 | else | |
3869 | ret = io_provided_buffer_select(req, len, bl); | |
bcda7baa | 3870 | } |
984824db CH |
3871 | io_ring_submit_unlock(req->ctx, issue_flags); |
3872 | return ret; | |
4d954c25 JA |
3873 | } |
3874 | ||
3875 | #ifdef CONFIG_COMPAT | |
3876 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
51aac424 | 3877 | unsigned int issue_flags) |
4d954c25 JA |
3878 | { |
3879 | struct compat_iovec __user *uiov; | |
3880 | compat_ssize_t clen; | |
3881 | void __user *buf; | |
e5b00349 | 3882 | size_t len; |
4d954c25 JA |
3883 | |
3884 | uiov = u64_to_user_ptr(req->rw.addr); | |
3885 | if (!access_ok(uiov, sizeof(*uiov))) | |
3886 | return -EFAULT; | |
3887 | if (__get_user(clen, &uiov->iov_len)) | |
3888 | return -EFAULT; | |
3889 | if (clen < 0) | |
3890 | return -EINVAL; | |
3891 | ||
3892 | len = clen; | |
4e906702 | 3893 | buf = io_buffer_select(req, &len, issue_flags); |
984824db CH |
3894 | if (!buf) |
3895 | return -ENOBUFS; | |
b66e65f4 | 3896 | req->rw.addr = (unsigned long) buf; |
4d954c25 | 3897 | iov[0].iov_base = buf; |
b66e65f4 | 3898 | req->rw.len = iov[0].iov_len = (compat_size_t) len; |
4d954c25 JA |
3899 | return 0; |
3900 | } | |
3901 | #endif | |
3902 | ||
3903 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
51aac424 | 3904 | unsigned int issue_flags) |
4d954c25 JA |
3905 | { |
3906 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3907 | void __user *buf; | |
3908 | ssize_t len; | |
3909 | ||
3910 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3911 | return -EFAULT; | |
3912 | ||
3913 | len = iov[0].iov_len; | |
3914 | if (len < 0) | |
3915 | return -EINVAL; | |
4e906702 | 3916 | buf = io_buffer_select(req, &len, issue_flags); |
984824db CH |
3917 | if (!buf) |
3918 | return -ENOBUFS; | |
b66e65f4 | 3919 | req->rw.addr = (unsigned long) buf; |
4d954c25 | 3920 | iov[0].iov_base = buf; |
b66e65f4 | 3921 | req->rw.len = iov[0].iov_len = len; |
4d954c25 JA |
3922 | return 0; |
3923 | } | |
3924 | ||
3925 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
51aac424 | 3926 | unsigned int issue_flags) |
4d954c25 | 3927 | { |
c7fb1942 | 3928 | if (req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING)) { |
b66e65f4 JA |
3929 | iov[0].iov_base = u64_to_user_ptr(req->rw.addr); |
3930 | iov[0].iov_len = req->rw.len; | |
4d954c25 | 3931 | return 0; |
dddb3e26 | 3932 | } |
dd201662 | 3933 | if (req->rw.len != 1) |
4d954c25 JA |
3934 | return -EINVAL; |
3935 | ||
3936 | #ifdef CONFIG_COMPAT | |
3937 | if (req->ctx->compat) | |
51aac424 | 3938 | return io_compat_import(req, iov, issue_flags); |
4d954c25 JA |
3939 | #endif |
3940 | ||
51aac424 | 3941 | return __io_iov_buffer_select(req, iov, issue_flags); |
4d954c25 JA |
3942 | } |
3943 | ||
b66e65f4 JA |
3944 | static inline bool io_do_buffer_select(struct io_kiocb *req) |
3945 | { | |
3946 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3947 | return false; | |
c7fb1942 | 3948 | return !(req->flags & (REQ_F_BUFFER_SELECTED|REQ_F_BUFFER_RING)); |
b66e65f4 JA |
3949 | } |
3950 | ||
caa8fe6e PB |
3951 | static struct iovec *__io_import_iovec(int rw, struct io_kiocb *req, |
3952 | struct io_rw_state *s, | |
3953 | unsigned int issue_flags) | |
2b188cc1 | 3954 | { |
5e49c973 | 3955 | struct iov_iter *iter = &s->iter; |
847595de | 3956 | u8 opcode = req->opcode; |
caa8fe6e | 3957 | struct iovec *iovec; |
d1d681b0 PB |
3958 | void __user *buf; |
3959 | size_t sqe_len; | |
4d954c25 | 3960 | ssize_t ret; |
edafccee | 3961 | |
f3251183 | 3962 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
5106dd6e | 3963 | ret = io_import_fixed(req, rw, iter, issue_flags); |
f3251183 PB |
3964 | if (ret) |
3965 | return ERR_PTR(ret); | |
3966 | return NULL; | |
3967 | } | |
2b188cc1 | 3968 | |
d1d681b0 PB |
3969 | buf = u64_to_user_ptr(req->rw.addr); |
3970 | sqe_len = req->rw.len; | |
9adbd45d | 3971 | |
3a6820f2 | 3972 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
b66e65f4 | 3973 | if (io_do_buffer_select(req)) { |
4e906702 | 3974 | buf = io_buffer_select(req, &sqe_len, issue_flags); |
984824db CH |
3975 | if (!buf) |
3976 | return ERR_PTR(-ENOBUFS); | |
b66e65f4 | 3977 | req->rw.addr = (unsigned long) buf; |
3f9d6441 | 3978 | req->rw.len = sqe_len; |
bcda7baa JA |
3979 | } |
3980 | ||
5e49c973 | 3981 | ret = import_single_range(rw, buf, sqe_len, s->fast_iov, iter); |
f3251183 PB |
3982 | if (ret) |
3983 | return ERR_PTR(ret); | |
3984 | return NULL; | |
3a6820f2 JA |
3985 | } |
3986 | ||
caa8fe6e | 3987 | iovec = s->fast_iov; |
4d954c25 | 3988 | if (req->flags & REQ_F_BUFFER_SELECT) { |
caa8fe6e | 3989 | ret = io_iov_buffer_select(req, iovec, issue_flags); |
f3251183 PB |
3990 | if (ret) |
3991 | return ERR_PTR(ret); | |
3992 | iov_iter_init(iter, rw, iovec, 1, iovec->iov_len); | |
3993 | return NULL; | |
4d954c25 JA |
3994 | } |
3995 | ||
caa8fe6e | 3996 | ret = __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, &iovec, iter, |
89cd35c5 | 3997 | req->ctx->compat); |
caa8fe6e PB |
3998 | if (unlikely(ret < 0)) |
3999 | return ERR_PTR(ret); | |
4000 | return iovec; | |
2b188cc1 JA |
4001 | } |
4002 | ||
5e49c973 PB |
4003 | static inline int io_import_iovec(int rw, struct io_kiocb *req, |
4004 | struct iovec **iovec, struct io_rw_state *s, | |
4005 | unsigned int issue_flags) | |
4006 | { | |
caa8fe6e PB |
4007 | *iovec = __io_import_iovec(rw, req, s, issue_flags); |
4008 | if (unlikely(IS_ERR(*iovec))) | |
4009 | return PTR_ERR(*iovec); | |
5e49c973 | 4010 | |
5e49c973 | 4011 | iov_iter_save_state(&s->iter, &s->iter_state); |
caa8fe6e | 4012 | return 0; |
2b188cc1 JA |
4013 | } |
4014 | ||
0fef9483 JA |
4015 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
4016 | { | |
5b09e37e | 4017 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
4018 | } |
4019 | ||
31b51510 | 4020 | /* |
32960613 JA |
4021 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
4022 | * by looping over ->read() or ->write() manually. | |
31b51510 | 4023 | */ |
4017eb91 | 4024 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 4025 | { |
4017eb91 JA |
4026 | struct kiocb *kiocb = &req->rw.kiocb; |
4027 | struct file *file = req->file; | |
32960613 | 4028 | ssize_t ret = 0; |
af9c45ec | 4029 | loff_t *ppos; |
32960613 JA |
4030 | |
4031 | /* | |
4032 | * Don't support polled IO through this interface, and we can't | |
4033 | * support non-blocking either. For the latter, this just causes | |
4034 | * the kiocb to be handled from an async context. | |
4035 | */ | |
4036 | if (kiocb->ki_flags & IOCB_HIPRI) | |
4037 | return -EOPNOTSUPP; | |
35645ac3 PB |
4038 | if ((kiocb->ki_flags & IOCB_NOWAIT) && |
4039 | !(kiocb->ki_filp->f_flags & O_NONBLOCK)) | |
32960613 JA |
4040 | return -EAGAIN; |
4041 | ||
af9c45ec DY |
4042 | ppos = io_kiocb_ppos(kiocb); |
4043 | ||
32960613 | 4044 | while (iov_iter_count(iter)) { |
311ae9e1 | 4045 | struct iovec iovec; |
32960613 JA |
4046 | ssize_t nr; |
4047 | ||
311ae9e1 PB |
4048 | if (!iov_iter_is_bvec(iter)) { |
4049 | iovec = iov_iter_iovec(iter); | |
4050 | } else { | |
4017eb91 JA |
4051 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
4052 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
4053 | } |
4054 | ||
32960613 JA |
4055 | if (rw == READ) { |
4056 | nr = file->f_op->read(file, iovec.iov_base, | |
af9c45ec | 4057 | iovec.iov_len, ppos); |
32960613 JA |
4058 | } else { |
4059 | nr = file->f_op->write(file, iovec.iov_base, | |
af9c45ec | 4060 | iovec.iov_len, ppos); |
32960613 JA |
4061 | } |
4062 | ||
4063 | if (nr < 0) { | |
4064 | if (!ret) | |
4065 | ret = nr; | |
4066 | break; | |
4067 | } | |
5e929367 | 4068 | ret += nr; |
16c8d2df JA |
4069 | if (!iov_iter_is_bvec(iter)) { |
4070 | iov_iter_advance(iter, nr); | |
4071 | } else { | |
16c8d2df | 4072 | req->rw.addr += nr; |
5e929367 JA |
4073 | req->rw.len -= nr; |
4074 | if (!req->rw.len) | |
4075 | break; | |
16c8d2df | 4076 | } |
32960613 JA |
4077 | if (nr != iovec.iov_len) |
4078 | break; | |
32960613 JA |
4079 | } |
4080 | ||
4081 | return ret; | |
4082 | } | |
4083 | ||
ff6165b2 JA |
4084 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
4085 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 4086 | { |
e8c2bc1f | 4087 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 4088 | |
538941e2 | 4089 | memcpy(&rw->s.iter, iter, sizeof(*iter)); |
afb87658 | 4090 | rw->free_iovec = iovec; |
227c0c96 | 4091 | rw->bytes_done = 0; |
ff6165b2 | 4092 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 4093 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 4094 | return; |
b64e3444 | 4095 | if (!iovec) { |
ff6165b2 JA |
4096 | unsigned iov_off = 0; |
4097 | ||
538941e2 | 4098 | rw->s.iter.iov = rw->s.fast_iov; |
ff6165b2 JA |
4099 | if (iter->iov != fast_iov) { |
4100 | iov_off = iter->iov - fast_iov; | |
538941e2 | 4101 | rw->s.iter.iov += iov_off; |
ff6165b2 | 4102 | } |
538941e2 PB |
4103 | if (rw->s.fast_iov != fast_iov) |
4104 | memcpy(rw->s.fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 4105 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
4106 | } else { |
4107 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
4108 | } |
4109 | } | |
4110 | ||
8d4af685 | 4111 | static inline bool io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 4112 | { |
e8c2bc1f JA |
4113 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
4114 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
d886e185 PB |
4115 | if (req->async_data) { |
4116 | req->flags |= REQ_F_ASYNC_DATA; | |
4117 | return false; | |
4118 | } | |
4119 | return true; | |
3d9932a8 XW |
4120 | } |
4121 | ||
ff6165b2 | 4122 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
c88598a9 | 4123 | struct io_rw_state *s, bool force) |
b7bb4f7d | 4124 | { |
26f0505a | 4125 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 4126 | return 0; |
d886e185 | 4127 | if (!req_has_async_data(req)) { |
cd658695 JA |
4128 | struct io_async_rw *iorw; |
4129 | ||
6cb78689 | 4130 | if (io_alloc_async_data(req)) { |
6bf985dc | 4131 | kfree(iovec); |
5d204bcf | 4132 | return -ENOMEM; |
6bf985dc | 4133 | } |
b7bb4f7d | 4134 | |
c88598a9 | 4135 | io_req_map_rw(req, iovec, s->fast_iov, &s->iter); |
cd658695 JA |
4136 | iorw = req->async_data; |
4137 | /* we've copied and mapped the iter, ensure state is saved */ | |
538941e2 | 4138 | iov_iter_save_state(&iorw->s.iter, &iorw->s.iter_state); |
5d204bcf | 4139 | } |
b7bb4f7d | 4140 | return 0; |
f67676d1 JA |
4141 | } |
4142 | ||
73debe68 | 4143 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 4144 | { |
e8c2bc1f | 4145 | struct io_async_rw *iorw = req->async_data; |
5e49c973 | 4146 | struct iovec *iov; |
847595de | 4147 | int ret; |
c3e330a4 | 4148 | |
51aac424 | 4149 | /* submission path, ->uring_lock should already be taken */ |
3b44b371 | 4150 | ret = io_import_iovec(rw, req, &iov, &iorw->s, 0); |
c3e330a4 PB |
4151 | if (unlikely(ret < 0)) |
4152 | return ret; | |
4153 | ||
ab0b196c PB |
4154 | iorw->bytes_done = 0; |
4155 | iorw->free_iovec = iov; | |
4156 | if (iov) | |
4157 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
4158 | return 0; |
4159 | } | |
4160 | ||
157dc813 JA |
4161 | static int io_readv_prep_async(struct io_kiocb *req) |
4162 | { | |
4163 | return io_rw_prep_async(req, READ); | |
4164 | } | |
4165 | ||
4166 | static int io_writev_prep_async(struct io_kiocb *req) | |
4167 | { | |
4168 | return io_rw_prep_async(req, WRITE); | |
4169 | } | |
4170 | ||
c1dd91d1 | 4171 | /* |
ffdc8dab | 4172 | * This is our waitqueue callback handler, registered through __folio_lock_async() |
c1dd91d1 JA |
4173 | * when we initially tried to do the IO with the iocb armed our waitqueue. |
4174 | * This gets called when the page is unlocked, and we generally expect that to | |
4175 | * happen when the page IO is completed and the page is now uptodate. This will | |
4176 | * queue a task_work based retry of the operation, attempting to copy the data | |
4177 | * again. If the latter fails because the page was NOT uptodate, then we will | |
4178 | * do a thread based blocking retry of the operation. That's the unexpected | |
4179 | * slow path. | |
4180 | */ | |
bcf5a063 JA |
4181 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
4182 | int sync, void *arg) | |
4183 | { | |
4184 | struct wait_page_queue *wpq; | |
4185 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 4186 | struct wait_page_key *key = arg; |
bcf5a063 JA |
4187 | |
4188 | wpq = container_of(wait, struct wait_page_queue, wait); | |
4189 | ||
cdc8fcb4 LT |
4190 | if (!wake_page_match(wpq, key)) |
4191 | return 0; | |
4192 | ||
c8d317aa | 4193 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 | 4194 | list_del_init(&wait->entry); |
921b9054 | 4195 | io_req_task_queue(req); |
bcf5a063 JA |
4196 | return 1; |
4197 | } | |
4198 | ||
c1dd91d1 JA |
4199 | /* |
4200 | * This controls whether a given IO request should be armed for async page | |
4201 | * based retry. If we return false here, the request is handed to the async | |
4202 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
4203 | * we prepare a private wait_page_queue entry and retry the operation. This | |
4204 | * will either succeed because the page is now uptodate and unlocked, or it | |
4205 | * will register a callback when the page is unlocked at IO completion. Through | |
4206 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
4207 | * That retry will attempt the buffered read again. The retry will generally | |
4208 | * succeed, or in rare cases where it fails, we then fall back to using the | |
4209 | * async worker threads for a blocking retry. | |
4210 | */ | |
227c0c96 | 4211 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 4212 | { |
e8c2bc1f JA |
4213 | struct io_async_rw *rw = req->async_data; |
4214 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 4215 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 4216 | |
bcf5a063 JA |
4217 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
4218 | if (req->flags & REQ_F_NOWAIT) | |
4219 | return false; | |
f67676d1 | 4220 | |
227c0c96 | 4221 | /* Only for buffered IO */ |
3b2a4439 | 4222 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 4223 | return false; |
3b2a4439 | 4224 | |
bcf5a063 JA |
4225 | /* |
4226 | * just use poll if we can, and don't attempt if the fs doesn't | |
4227 | * support callback based unlocks | |
4228 | */ | |
4229 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
4230 | return false; | |
f67676d1 | 4231 | |
3b2a4439 JA |
4232 | wait->wait.func = io_async_buf_func; |
4233 | wait->wait.private = req; | |
4234 | wait->wait.flags = 0; | |
4235 | INIT_LIST_HEAD(&wait->wait.entry); | |
4236 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 4237 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 4238 | kiocb->ki_waitq = wait; |
3b2a4439 | 4239 | return true; |
bcf5a063 JA |
4240 | } |
4241 | ||
aeab9506 | 4242 | static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) |
bcf5a063 | 4243 | { |
607b6fb8 | 4244 | if (likely(req->file->f_op->read_iter)) |
bcf5a063 | 4245 | return call_read_iter(req->file, &req->rw.kiocb, iter); |
2dd2111d | 4246 | else if (req->file->f_op->read) |
4017eb91 | 4247 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
4248 | else |
4249 | return -EINVAL; | |
f67676d1 JA |
4250 | } |
4251 | ||
7db30437 ML |
4252 | static bool need_read_all(struct io_kiocb *req) |
4253 | { | |
4254 | return req->flags & REQ_F_ISREG || | |
4255 | S_ISBLK(file_inode(req->file)->i_mode); | |
4256 | } | |
4257 | ||
584b0180 JA |
4258 | static int io_rw_init_file(struct io_kiocb *req, fmode_t mode) |
4259 | { | |
4260 | struct kiocb *kiocb = &req->rw.kiocb; | |
4261 | struct io_ring_ctx *ctx = req->ctx; | |
4262 | struct file *file = req->file; | |
4263 | int ret; | |
4264 | ||
4265 | if (unlikely(!file || !(file->f_mode & mode))) | |
4266 | return -EBADF; | |
4267 | ||
4268 | if (!io_req_ffs_set(req)) | |
4269 | req->flags |= io_file_get_flags(file) << REQ_F_SUPPORT_NOWAIT_BIT; | |
4270 | ||
4271 | kiocb->ki_flags = iocb_flags(file); | |
4272 | ret = kiocb_set_rw_flags(kiocb, req->rw.flags); | |
4273 | if (unlikely(ret)) | |
4274 | return ret; | |
4275 | ||
4276 | /* | |
4277 | * If the file is marked O_NONBLOCK, still allow retry for it if it | |
4278 | * supports async. Otherwise it's impossible to use O_NONBLOCK files | |
4279 | * reliably. If not, or it IOCB_NOWAIT is set, don't retry. | |
4280 | */ | |
4281 | if ((kiocb->ki_flags & IOCB_NOWAIT) || | |
4282 | ((file->f_flags & O_NONBLOCK) && !io_file_supports_nowait(req))) | |
4283 | req->flags |= REQ_F_NOWAIT; | |
4284 | ||
4285 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
4286 | if (!(kiocb->ki_flags & IOCB_DIRECT) || !file->f_op->iopoll) | |
4287 | return -EOPNOTSUPP; | |
4288 | ||
32452a3e | 4289 | kiocb->private = NULL; |
584b0180 JA |
4290 | kiocb->ki_flags |= IOCB_HIPRI | IOCB_ALLOC_CACHE; |
4291 | kiocb->ki_complete = io_complete_rw_iopoll; | |
4292 | req->iopoll_completed = 0; | |
4293 | } else { | |
4294 | if (kiocb->ki_flags & IOCB_HIPRI) | |
4295 | return -EINVAL; | |
4296 | kiocb->ki_complete = io_complete_rw; | |
4297 | } | |
4298 | ||
4299 | return 0; | |
4300 | } | |
4301 | ||
889fca73 | 4302 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 4303 | { |
607b6fb8 | 4304 | struct io_rw_state __s, *s = &__s; |
c88598a9 | 4305 | struct iovec *iovec; |
9adbd45d | 4306 | struct kiocb *kiocb = &req->rw.kiocb; |
45d189c6 | 4307 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
d886e185 | 4308 | struct io_async_rw *rw; |
cd658695 | 4309 | ssize_t ret, ret2; |
b4aec400 | 4310 | loff_t *ppos; |
ff6165b2 | 4311 | |
607b6fb8 PB |
4312 | if (!req_has_async_data(req)) { |
4313 | ret = io_import_iovec(READ, req, &iovec, s, issue_flags); | |
4314 | if (unlikely(ret < 0)) | |
4315 | return ret; | |
4316 | } else { | |
2be2eb02 JA |
4317 | /* |
4318 | * Safe and required to re-import if we're using provided | |
4319 | * buffers, as we dropped the selected one before retry. | |
4320 | */ | |
4321 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4322 | ret = io_import_iovec(READ, req, &iovec, s, issue_flags); | |
4323 | if (unlikely(ret < 0)) | |
4324 | return ret; | |
4325 | } | |
4326 | ||
d886e185 | 4327 | rw = req->async_data; |
c88598a9 | 4328 | s = &rw->s; |
cd658695 JA |
4329 | /* |
4330 | * We come here from an earlier attempt, restore our state to | |
4331 | * match in case it doesn't. It's cheap enough that we don't | |
4332 | * need to make this conditional. | |
4333 | */ | |
c88598a9 | 4334 | iov_iter_restore(&s->iter, &s->iter_state); |
2846c481 | 4335 | iovec = NULL; |
2846c481 | 4336 | } |
584b0180 | 4337 | ret = io_rw_init_file(req, FMODE_READ); |
323b190b JA |
4338 | if (unlikely(ret)) { |
4339 | kfree(iovec); | |
584b0180 | 4340 | return ret; |
323b190b | 4341 | } |
cef216fc | 4342 | req->cqe.res = iov_iter_count(&s->iter); |
2b188cc1 | 4343 | |
607b6fb8 PB |
4344 | if (force_nonblock) { |
4345 | /* If the file doesn't support async, just async punt */ | |
35645ac3 | 4346 | if (unlikely(!io_file_supports_nowait(req))) { |
607b6fb8 PB |
4347 | ret = io_setup_async_rw(req, iovec, s, true); |
4348 | return ret ?: -EAGAIN; | |
4349 | } | |
a88fc400 | 4350 | kiocb->ki_flags |= IOCB_NOWAIT; |
607b6fb8 PB |
4351 | } else { |
4352 | /* Ensure we clear previously set non-block flag */ | |
4353 | kiocb->ki_flags &= ~IOCB_NOWAIT; | |
6713e7a6 | 4354 | } |
9e645e11 | 4355 | |
b4aec400 | 4356 | ppos = io_kiocb_update_pos(req); |
d34e1e5b | 4357 | |
cef216fc | 4358 | ret = rw_verify_area(READ, req->file, ppos, req->cqe.res); |
5ea5dd45 PB |
4359 | if (unlikely(ret)) { |
4360 | kfree(iovec); | |
4361 | return ret; | |
4362 | } | |
2b188cc1 | 4363 | |
c88598a9 | 4364 | ret = io_iter_do_read(req, &s->iter); |
32960613 | 4365 | |
230d50d4 | 4366 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 4367 | req->flags &= ~REQ_F_REISSUE; |
9af177ee JA |
4368 | /* if we can poll, just do that */ |
4369 | if (req->opcode == IORING_OP_READ && file_can_poll(req->file)) | |
4370 | return -EAGAIN; | |
eefdf30f JA |
4371 | /* IOPOLL retry should happen for io-wq threads */ |
4372 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 4373 | goto done; |
75c668cd PB |
4374 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
4375 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 4376 | goto done; |
f38c7e3a | 4377 | ret = 0; |
230d50d4 JA |
4378 | } else if (ret == -EIOCBQUEUED) { |
4379 | goto out_free; | |
cef216fc | 4380 | } else if (ret == req->cqe.res || ret <= 0 || !force_nonblock || |
7db30437 | 4381 | (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { |
7335e3bf | 4382 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 4383 | goto done; |
227c0c96 JA |
4384 | } |
4385 | ||
cd658695 JA |
4386 | /* |
4387 | * Don't depend on the iter state matching what was consumed, or being | |
4388 | * untouched in case of error. Restore it and we'll advance it | |
4389 | * manually if we need to. | |
4390 | */ | |
c88598a9 | 4391 | iov_iter_restore(&s->iter, &s->iter_state); |
cd658695 | 4392 | |
c88598a9 | 4393 | ret2 = io_setup_async_rw(req, iovec, s, true); |
6bf985dc PB |
4394 | if (ret2) |
4395 | return ret2; | |
4396 | ||
fe1cdd55 | 4397 | iovec = NULL; |
e8c2bc1f | 4398 | rw = req->async_data; |
c88598a9 | 4399 | s = &rw->s; |
cd658695 JA |
4400 | /* |
4401 | * Now use our persistent iterator and state, if we aren't already. | |
4402 | * We've restored and mapped the iter to match. | |
4403 | */ | |
227c0c96 | 4404 | |
b23df91b | 4405 | do { |
cd658695 JA |
4406 | /* |
4407 | * We end up here because of a partial read, either from | |
4408 | * above or inside this loop. Advance the iter by the bytes | |
4409 | * that were consumed. | |
4410 | */ | |
c88598a9 PB |
4411 | iov_iter_advance(&s->iter, ret); |
4412 | if (!iov_iter_count(&s->iter)) | |
cd658695 | 4413 | break; |
b23df91b | 4414 | rw->bytes_done += ret; |
c88598a9 | 4415 | iov_iter_save_state(&s->iter, &s->iter_state); |
cd658695 | 4416 | |
b23df91b PB |
4417 | /* if we can retry, do so with the callbacks armed */ |
4418 | if (!io_rw_should_retry(req)) { | |
4419 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
4420 | return -EAGAIN; | |
4421 | } | |
4422 | ||
4423 | /* | |
4424 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
4425 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
4426 | * desired page gets unlocked. We can also get a partial read | |
4427 | * here, and if we do, then just retry at the new offset. | |
4428 | */ | |
c88598a9 | 4429 | ret = io_iter_do_read(req, &s->iter); |
b23df91b PB |
4430 | if (ret == -EIOCBQUEUED) |
4431 | return 0; | |
227c0c96 | 4432 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 4433 | kiocb->ki_flags &= ~IOCB_WAITQ; |
c88598a9 | 4434 | iov_iter_restore(&s->iter, &s->iter_state); |
cd658695 | 4435 | } while (ret > 0); |
227c0c96 | 4436 | done: |
2ea537ca | 4437 | kiocb_done(req, ret, issue_flags); |
fe1cdd55 PB |
4438 | out_free: |
4439 | /* it's faster to check here then delegate to kfree */ | |
4440 | if (iovec) | |
4441 | kfree(iovec); | |
5ea5dd45 | 4442 | return 0; |
2b188cc1 JA |
4443 | } |
4444 | ||
889fca73 | 4445 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 4446 | { |
607b6fb8 | 4447 | struct io_rw_state __s, *s = &__s; |
c88598a9 | 4448 | struct iovec *iovec; |
9adbd45d | 4449 | struct kiocb *kiocb = &req->rw.kiocb; |
45d189c6 | 4450 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
cd658695 | 4451 | ssize_t ret, ret2; |
b4aec400 | 4452 | loff_t *ppos; |
2b188cc1 | 4453 | |
607b6fb8 | 4454 | if (!req_has_async_data(req)) { |
5e49c973 PB |
4455 | ret = io_import_iovec(WRITE, req, &iovec, s, issue_flags); |
4456 | if (unlikely(ret < 0)) | |
2846c481 | 4457 | return ret; |
607b6fb8 PB |
4458 | } else { |
4459 | struct io_async_rw *rw = req->async_data; | |
4460 | ||
4461 | s = &rw->s; | |
4462 | iov_iter_restore(&s->iter, &s->iter_state); | |
2846c481 | 4463 | iovec = NULL; |
2846c481 | 4464 | } |
584b0180 | 4465 | ret = io_rw_init_file(req, FMODE_WRITE); |
323b190b JA |
4466 | if (unlikely(ret)) { |
4467 | kfree(iovec); | |
584b0180 | 4468 | return ret; |
323b190b | 4469 | } |
cef216fc | 4470 | req->cqe.res = iov_iter_count(&s->iter); |
2b188cc1 | 4471 | |
607b6fb8 PB |
4472 | if (force_nonblock) { |
4473 | /* If the file doesn't support async, just async punt */ | |
35645ac3 | 4474 | if (unlikely(!io_file_supports_nowait(req))) |
607b6fb8 | 4475 | goto copy_iov; |
fd6c2e4c | 4476 | |
607b6fb8 PB |
4477 | /* file path doesn't support NOWAIT for non-direct_IO */ |
4478 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
4479 | (req->flags & REQ_F_ISREG)) | |
4480 | goto copy_iov; | |
31b51510 | 4481 | |
607b6fb8 PB |
4482 | kiocb->ki_flags |= IOCB_NOWAIT; |
4483 | } else { | |
4484 | /* Ensure we clear previously set non-block flag */ | |
4485 | kiocb->ki_flags &= ~IOCB_NOWAIT; | |
4486 | } | |
31b51510 | 4487 | |
b4aec400 | 4488 | ppos = io_kiocb_update_pos(req); |
d34e1e5b | 4489 | |
cef216fc | 4490 | ret = rw_verify_area(WRITE, req->file, ppos, req->cqe.res); |
fa15bafb PB |
4491 | if (unlikely(ret)) |
4492 | goto out_free; | |
4ed734b0 | 4493 | |
fa15bafb PB |
4494 | /* |
4495 | * Open-code file_start_write here to grab freeze protection, | |
4496 | * which will be released by another thread in | |
4497 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
4498 | * released so that it doesn't complain about the held lock when | |
4499 | * we return to userspace. | |
4500 | */ | |
4501 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 4502 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
4503 | __sb_writers_release(file_inode(req->file)->i_sb, |
4504 | SB_FREEZE_WRITE); | |
4505 | } | |
4506 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 4507 | |
35645ac3 | 4508 | if (likely(req->file->f_op->write_iter)) |
c88598a9 | 4509 | ret2 = call_write_iter(req->file, kiocb, &s->iter); |
2dd2111d | 4510 | else if (req->file->f_op->write) |
c88598a9 | 4511 | ret2 = loop_rw_iter(WRITE, req, &s->iter); |
2dd2111d GH |
4512 | else |
4513 | ret2 = -EINVAL; | |
4ed734b0 | 4514 | |
6ad7f233 PB |
4515 | if (req->flags & REQ_F_REISSUE) { |
4516 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 4517 | ret2 = -EAGAIN; |
6ad7f233 | 4518 | } |
230d50d4 | 4519 | |
fa15bafb PB |
4520 | /* |
4521 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
4522 | * retry them without IOCB_NOWAIT. | |
4523 | */ | |
4524 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
4525 | ret2 = -EAGAIN; | |
75c668cd PB |
4526 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
4527 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 4528 | goto done; |
fa15bafb | 4529 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f | 4530 | /* IOPOLL retry should happen for io-wq threads */ |
b10841c9 | 4531 | if (ret2 == -EAGAIN && (req->ctx->flags & IORING_SETUP_IOPOLL)) |
eefdf30f | 4532 | goto copy_iov; |
355afaeb | 4533 | done: |
2ea537ca | 4534 | kiocb_done(req, ret2, issue_flags); |
fa15bafb | 4535 | } else { |
f67676d1 | 4536 | copy_iov: |
c88598a9 PB |
4537 | iov_iter_restore(&s->iter, &s->iter_state); |
4538 | ret = io_setup_async_rw(req, iovec, s, false); | |
6bf985dc | 4539 | return ret ?: -EAGAIN; |
2b188cc1 | 4540 | } |
31b51510 | 4541 | out_free: |
f261c168 | 4542 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 4543 | if (iovec) |
6f2cc166 | 4544 | kfree(iovec); |
2b188cc1 JA |
4545 | return ret; |
4546 | } | |
4547 | ||
80a261fd JA |
4548 | static int io_renameat_prep(struct io_kiocb *req, |
4549 | const struct io_uring_sqe *sqe) | |
4550 | { | |
4551 | struct io_rename *ren = &req->rename; | |
4552 | const char __user *oldf, *newf; | |
4553 | ||
73911426 | 4554 | if (sqe->buf_index || sqe->splice_fd_in) |
ed7eb259 | 4555 | return -EINVAL; |
80a261fd JA |
4556 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
4557 | return -EBADF; | |
4558 | ||
4559 | ren->old_dfd = READ_ONCE(sqe->fd); | |
4560 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4561 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4562 | ren->new_dfd = READ_ONCE(sqe->len); | |
4563 | ren->flags = READ_ONCE(sqe->rename_flags); | |
4564 | ||
4565 | ren->oldpath = getname(oldf); | |
4566 | if (IS_ERR(ren->oldpath)) | |
4567 | return PTR_ERR(ren->oldpath); | |
4568 | ||
4569 | ren->newpath = getname(newf); | |
4570 | if (IS_ERR(ren->newpath)) { | |
4571 | putname(ren->oldpath); | |
4572 | return PTR_ERR(ren->newpath); | |
4573 | } | |
4574 | ||
4575 | req->flags |= REQ_F_NEED_CLEANUP; | |
4576 | return 0; | |
4577 | } | |
4578 | ||
45d189c6 | 4579 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
4580 | { |
4581 | struct io_rename *ren = &req->rename; | |
4582 | int ret; | |
4583 | ||
45d189c6 | 4584 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
4585 | return -EAGAIN; |
4586 | ||
4587 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
4588 | ren->newpath, ren->flags); | |
4589 | ||
4590 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
80a261fd JA |
4591 | io_req_complete(req, ret); |
4592 | return 0; | |
4593 | } | |
4594 | ||
e9621e2b SR |
4595 | static inline void __io_xattr_finish(struct io_kiocb *req) |
4596 | { | |
4597 | struct io_xattr *ix = &req->xattr; | |
4598 | ||
4599 | if (ix->filename) | |
4600 | putname(ix->filename); | |
4601 | ||
4602 | kfree(ix->ctx.kname); | |
4603 | kvfree(ix->ctx.kvalue); | |
4604 | } | |
4605 | ||
4606 | static void io_xattr_finish(struct io_kiocb *req, int ret) | |
4607 | { | |
4608 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4609 | ||
4610 | __io_xattr_finish(req); | |
e9621e2b SR |
4611 | io_req_complete(req, ret); |
4612 | } | |
4613 | ||
a56834e0 SR |
4614 | static int __io_getxattr_prep(struct io_kiocb *req, |
4615 | const struct io_uring_sqe *sqe) | |
4616 | { | |
4617 | struct io_xattr *ix = &req->xattr; | |
4618 | const char __user *name; | |
4619 | int ret; | |
4620 | ||
a56834e0 SR |
4621 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
4622 | return -EBADF; | |
4623 | ||
4624 | ix->filename = NULL; | |
4625 | ix->ctx.kvalue = NULL; | |
4626 | name = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4627 | ix->ctx.cvalue = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4628 | ix->ctx.size = READ_ONCE(sqe->len); | |
4629 | ix->ctx.flags = READ_ONCE(sqe->xattr_flags); | |
4630 | ||
4631 | if (ix->ctx.flags) | |
4632 | return -EINVAL; | |
4633 | ||
4634 | ix->ctx.kname = kmalloc(sizeof(*ix->ctx.kname), GFP_KERNEL); | |
4635 | if (!ix->ctx.kname) | |
4636 | return -ENOMEM; | |
4637 | ||
4638 | ret = strncpy_from_user(ix->ctx.kname->name, name, | |
4639 | sizeof(ix->ctx.kname->name)); | |
4640 | if (!ret || ret == sizeof(ix->ctx.kname->name)) | |
4641 | ret = -ERANGE; | |
4642 | if (ret < 0) { | |
4643 | kfree(ix->ctx.kname); | |
4644 | return ret; | |
4645 | } | |
4646 | ||
4647 | req->flags |= REQ_F_NEED_CLEANUP; | |
4648 | return 0; | |
4649 | } | |
4650 | ||
4651 | static int io_fgetxattr_prep(struct io_kiocb *req, | |
4652 | const struct io_uring_sqe *sqe) | |
4653 | { | |
4654 | return __io_getxattr_prep(req, sqe); | |
4655 | } | |
4656 | ||
4657 | static int io_getxattr_prep(struct io_kiocb *req, | |
4658 | const struct io_uring_sqe *sqe) | |
4659 | { | |
4660 | struct io_xattr *ix = &req->xattr; | |
4661 | const char __user *path; | |
4662 | int ret; | |
4663 | ||
4664 | ret = __io_getxattr_prep(req, sqe); | |
4665 | if (ret) | |
4666 | return ret; | |
4667 | ||
4668 | path = u64_to_user_ptr(READ_ONCE(sqe->addr3)); | |
4669 | ||
4670 | ix->filename = getname_flags(path, LOOKUP_FOLLOW, NULL); | |
4671 | if (IS_ERR(ix->filename)) { | |
4672 | ret = PTR_ERR(ix->filename); | |
4673 | ix->filename = NULL; | |
4674 | } | |
4675 | ||
4676 | return ret; | |
4677 | } | |
4678 | ||
4679 | static int io_fgetxattr(struct io_kiocb *req, unsigned int issue_flags) | |
4680 | { | |
4681 | struct io_xattr *ix = &req->xattr; | |
4682 | int ret; | |
4683 | ||
4684 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4685 | return -EAGAIN; | |
4686 | ||
4687 | ret = do_getxattr(mnt_user_ns(req->file->f_path.mnt), | |
4688 | req->file->f_path.dentry, | |
4689 | &ix->ctx); | |
4690 | ||
4691 | io_xattr_finish(req, ret); | |
4692 | return 0; | |
4693 | } | |
4694 | ||
4695 | static int io_getxattr(struct io_kiocb *req, unsigned int issue_flags) | |
4696 | { | |
4697 | struct io_xattr *ix = &req->xattr; | |
4698 | unsigned int lookup_flags = LOOKUP_FOLLOW; | |
4699 | struct path path; | |
4700 | int ret; | |
4701 | ||
4702 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4703 | return -EAGAIN; | |
4704 | ||
4705 | retry: | |
4706 | ret = filename_lookup(AT_FDCWD, ix->filename, lookup_flags, &path, NULL); | |
4707 | if (!ret) { | |
4708 | ret = do_getxattr(mnt_user_ns(path.mnt), | |
4709 | path.dentry, | |
4710 | &ix->ctx); | |
4711 | ||
4712 | path_put(&path); | |
4713 | if (retry_estale(ret, lookup_flags)) { | |
4714 | lookup_flags |= LOOKUP_REVAL; | |
4715 | goto retry; | |
4716 | } | |
4717 | } | |
4718 | ||
4719 | io_xattr_finish(req, ret); | |
4720 | return 0; | |
4721 | } | |
4722 | ||
e9621e2b SR |
4723 | static int __io_setxattr_prep(struct io_kiocb *req, |
4724 | const struct io_uring_sqe *sqe) | |
4725 | { | |
4726 | struct io_xattr *ix = &req->xattr; | |
4727 | const char __user *name; | |
4728 | int ret; | |
4729 | ||
e9621e2b SR |
4730 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
4731 | return -EBADF; | |
4732 | ||
4733 | ix->filename = NULL; | |
4734 | name = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4735 | ix->ctx.cvalue = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4736 | ix->ctx.kvalue = NULL; | |
4737 | ix->ctx.size = READ_ONCE(sqe->len); | |
4738 | ix->ctx.flags = READ_ONCE(sqe->xattr_flags); | |
4739 | ||
4740 | ix->ctx.kname = kmalloc(sizeof(*ix->ctx.kname), GFP_KERNEL); | |
4741 | if (!ix->ctx.kname) | |
4742 | return -ENOMEM; | |
4743 | ||
4744 | ret = setxattr_copy(name, &ix->ctx); | |
4745 | if (ret) { | |
4746 | kfree(ix->ctx.kname); | |
4747 | return ret; | |
4748 | } | |
4749 | ||
4750 | req->flags |= REQ_F_NEED_CLEANUP; | |
4751 | return 0; | |
4752 | } | |
4753 | ||
4754 | static int io_setxattr_prep(struct io_kiocb *req, | |
4755 | const struct io_uring_sqe *sqe) | |
4756 | { | |
4757 | struct io_xattr *ix = &req->xattr; | |
4758 | const char __user *path; | |
4759 | int ret; | |
4760 | ||
4761 | ret = __io_setxattr_prep(req, sqe); | |
4762 | if (ret) | |
4763 | return ret; | |
4764 | ||
4765 | path = u64_to_user_ptr(READ_ONCE(sqe->addr3)); | |
4766 | ||
4767 | ix->filename = getname_flags(path, LOOKUP_FOLLOW, NULL); | |
4768 | if (IS_ERR(ix->filename)) { | |
4769 | ret = PTR_ERR(ix->filename); | |
4770 | ix->filename = NULL; | |
4771 | } | |
4772 | ||
4773 | return ret; | |
4774 | } | |
4775 | ||
4776 | static int io_fsetxattr_prep(struct io_kiocb *req, | |
4777 | const struct io_uring_sqe *sqe) | |
4778 | { | |
4779 | return __io_setxattr_prep(req, sqe); | |
4780 | } | |
4781 | ||
4782 | static int __io_setxattr(struct io_kiocb *req, unsigned int issue_flags, | |
4783 | struct path *path) | |
4784 | { | |
4785 | struct io_xattr *ix = &req->xattr; | |
4786 | int ret; | |
4787 | ||
4788 | ret = mnt_want_write(path->mnt); | |
4789 | if (!ret) { | |
4790 | ret = do_setxattr(mnt_user_ns(path->mnt), path->dentry, &ix->ctx); | |
4791 | mnt_drop_write(path->mnt); | |
4792 | } | |
4793 | ||
4794 | return ret; | |
4795 | } | |
4796 | ||
4797 | static int io_fsetxattr(struct io_kiocb *req, unsigned int issue_flags) | |
4798 | { | |
4799 | int ret; | |
4800 | ||
4801 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4802 | return -EAGAIN; | |
4803 | ||
4804 | ret = __io_setxattr(req, issue_flags, &req->file->f_path); | |
4805 | io_xattr_finish(req, ret); | |
4806 | ||
4807 | return 0; | |
4808 | } | |
4809 | ||
4810 | static int io_setxattr(struct io_kiocb *req, unsigned int issue_flags) | |
4811 | { | |
4812 | struct io_xattr *ix = &req->xattr; | |
4813 | unsigned int lookup_flags = LOOKUP_FOLLOW; | |
4814 | struct path path; | |
4815 | int ret; | |
4816 | ||
4817 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4818 | return -EAGAIN; | |
4819 | ||
4820 | retry: | |
4821 | ret = filename_lookup(AT_FDCWD, ix->filename, lookup_flags, &path, NULL); | |
4822 | if (!ret) { | |
4823 | ret = __io_setxattr(req, issue_flags, &path); | |
4824 | path_put(&path); | |
4825 | if (retry_estale(ret, lookup_flags)) { | |
4826 | lookup_flags |= LOOKUP_REVAL; | |
4827 | goto retry; | |
4828 | } | |
4829 | } | |
4830 | ||
4831 | io_xattr_finish(req, ret); | |
4832 | return 0; | |
4833 | } | |
4834 | ||
14a1143b JA |
4835 | static int io_unlinkat_prep(struct io_kiocb *req, |
4836 | const struct io_uring_sqe *sqe) | |
4837 | { | |
4838 | struct io_unlink *un = &req->unlink; | |
4839 | const char __user *fname; | |
4840 | ||
73911426 | 4841 | if (sqe->off || sqe->len || sqe->buf_index || sqe->splice_fd_in) |
22634bc5 | 4842 | return -EINVAL; |
14a1143b JA |
4843 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
4844 | return -EBADF; | |
4845 | ||
4846 | un->dfd = READ_ONCE(sqe->fd); | |
4847 | ||
4848 | un->flags = READ_ONCE(sqe->unlink_flags); | |
4849 | if (un->flags & ~AT_REMOVEDIR) | |
4850 | return -EINVAL; | |
4851 | ||
4852 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4853 | un->filename = getname(fname); | |
4854 | if (IS_ERR(un->filename)) | |
4855 | return PTR_ERR(un->filename); | |
4856 | ||
4857 | req->flags |= REQ_F_NEED_CLEANUP; | |
4858 | return 0; | |
4859 | } | |
4860 | ||
45d189c6 | 4861 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
4862 | { |
4863 | struct io_unlink *un = &req->unlink; | |
4864 | int ret; | |
4865 | ||
45d189c6 | 4866 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
4867 | return -EAGAIN; |
4868 | ||
4869 | if (un->flags & AT_REMOVEDIR) | |
4870 | ret = do_rmdir(un->dfd, un->filename); | |
4871 | else | |
4872 | ret = do_unlinkat(un->dfd, un->filename); | |
4873 | ||
4874 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
14a1143b JA |
4875 | io_req_complete(req, ret); |
4876 | return 0; | |
4877 | } | |
4878 | ||
e34a02dc DK |
4879 | static int io_mkdirat_prep(struct io_kiocb *req, |
4880 | const struct io_uring_sqe *sqe) | |
4881 | { | |
4882 | struct io_mkdir *mkd = &req->mkdir; | |
4883 | const char __user *fname; | |
4884 | ||
73911426 | 4885 | if (sqe->off || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) |
e34a02dc DK |
4886 | return -EINVAL; |
4887 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
4888 | return -EBADF; | |
4889 | ||
4890 | mkd->dfd = READ_ONCE(sqe->fd); | |
4891 | mkd->mode = READ_ONCE(sqe->len); | |
4892 | ||
4893 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4894 | mkd->filename = getname(fname); | |
4895 | if (IS_ERR(mkd->filename)) | |
4896 | return PTR_ERR(mkd->filename); | |
4897 | ||
4898 | req->flags |= REQ_F_NEED_CLEANUP; | |
4899 | return 0; | |
4900 | } | |
4901 | ||
04f34081 | 4902 | static int io_mkdirat(struct io_kiocb *req, unsigned int issue_flags) |
e34a02dc DK |
4903 | { |
4904 | struct io_mkdir *mkd = &req->mkdir; | |
4905 | int ret; | |
4906 | ||
4907 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4908 | return -EAGAIN; | |
4909 | ||
4910 | ret = do_mkdirat(mkd->dfd, mkd->filename, mkd->mode); | |
4911 | ||
4912 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
e34a02dc DK |
4913 | io_req_complete(req, ret); |
4914 | return 0; | |
4915 | } | |
4916 | ||
7a8721f8 DK |
4917 | static int io_symlinkat_prep(struct io_kiocb *req, |
4918 | const struct io_uring_sqe *sqe) | |
4919 | { | |
4920 | struct io_symlink *sl = &req->symlink; | |
4921 | const char __user *oldpath, *newpath; | |
4922 | ||
73911426 | 4923 | if (sqe->len || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) |
7a8721f8 DK |
4924 | return -EINVAL; |
4925 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
4926 | return -EBADF; | |
4927 | ||
4928 | sl->new_dfd = READ_ONCE(sqe->fd); | |
4929 | oldpath = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4930 | newpath = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4931 | ||
4932 | sl->oldpath = getname(oldpath); | |
4933 | if (IS_ERR(sl->oldpath)) | |
4934 | return PTR_ERR(sl->oldpath); | |
4935 | ||
4936 | sl->newpath = getname(newpath); | |
4937 | if (IS_ERR(sl->newpath)) { | |
4938 | putname(sl->oldpath); | |
4939 | return PTR_ERR(sl->newpath); | |
4940 | } | |
4941 | ||
4942 | req->flags |= REQ_F_NEED_CLEANUP; | |
4943 | return 0; | |
4944 | } | |
4945 | ||
04f34081 | 4946 | static int io_symlinkat(struct io_kiocb *req, unsigned int issue_flags) |
7a8721f8 DK |
4947 | { |
4948 | struct io_symlink *sl = &req->symlink; | |
4949 | int ret; | |
4950 | ||
4951 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4952 | return -EAGAIN; | |
4953 | ||
4954 | ret = do_symlinkat(sl->oldpath, sl->new_dfd, sl->newpath); | |
4955 | ||
4956 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
7a8721f8 DK |
4957 | io_req_complete(req, ret); |
4958 | return 0; | |
4959 | } | |
4960 | ||
cf30da90 DK |
4961 | static int io_linkat_prep(struct io_kiocb *req, |
4962 | const struct io_uring_sqe *sqe) | |
4963 | { | |
4964 | struct io_hardlink *lnk = &req->hardlink; | |
4965 | const char __user *oldf, *newf; | |
4966 | ||
73911426 | 4967 | if (sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in) |
cf30da90 DK |
4968 | return -EINVAL; |
4969 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
4970 | return -EBADF; | |
4971 | ||
4972 | lnk->old_dfd = READ_ONCE(sqe->fd); | |
4973 | lnk->new_dfd = READ_ONCE(sqe->len); | |
4974 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4975 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
4976 | lnk->flags = READ_ONCE(sqe->hardlink_flags); | |
4977 | ||
4978 | lnk->oldpath = getname(oldf); | |
4979 | if (IS_ERR(lnk->oldpath)) | |
4980 | return PTR_ERR(lnk->oldpath); | |
4981 | ||
4982 | lnk->newpath = getname(newf); | |
4983 | if (IS_ERR(lnk->newpath)) { | |
4984 | putname(lnk->oldpath); | |
4985 | return PTR_ERR(lnk->newpath); | |
4986 | } | |
4987 | ||
4988 | req->flags |= REQ_F_NEED_CLEANUP; | |
4989 | return 0; | |
4990 | } | |
4991 | ||
04f34081 | 4992 | static int io_linkat(struct io_kiocb *req, unsigned int issue_flags) |
cf30da90 DK |
4993 | { |
4994 | struct io_hardlink *lnk = &req->hardlink; | |
4995 | int ret; | |
4996 | ||
4997 | if (issue_flags & IO_URING_F_NONBLOCK) | |
4998 | return -EAGAIN; | |
4999 | ||
5000 | ret = do_linkat(lnk->old_dfd, lnk->oldpath, lnk->new_dfd, | |
5001 | lnk->newpath, lnk->flags); | |
5002 | ||
5003 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
cf30da90 DK |
5004 | io_req_complete(req, ret); |
5005 | return 0; | |
5006 | } | |
5007 | ||
ee692a21 JA |
5008 | static void io_uring_cmd_work(struct io_kiocb *req, bool *locked) |
5009 | { | |
5010 | req->uring_cmd.task_work_cb(&req->uring_cmd); | |
5011 | } | |
5012 | ||
5013 | void io_uring_cmd_complete_in_task(struct io_uring_cmd *ioucmd, | |
5014 | void (*task_work_cb)(struct io_uring_cmd *)) | |
5015 | { | |
5016 | struct io_kiocb *req = container_of(ioucmd, struct io_kiocb, uring_cmd); | |
5017 | ||
5018 | req->uring_cmd.task_work_cb = task_work_cb; | |
5019 | req->io_task_work.func = io_uring_cmd_work; | |
32fc810b | 5020 | io_req_task_work_add(req); |
ee692a21 JA |
5021 | } |
5022 | EXPORT_SYMBOL_GPL(io_uring_cmd_complete_in_task); | |
5023 | ||
29ede201 PB |
5024 | static inline void io_req_set_cqe32_extra(struct io_kiocb *req, |
5025 | u64 extra1, u64 extra2) | |
5026 | { | |
5027 | req->extra1 = extra1; | |
5028 | req->extra2 = extra2; | |
2caf9822 | 5029 | req->flags |= REQ_F_CQE32_INIT; |
29ede201 PB |
5030 | } |
5031 | ||
ee692a21 JA |
5032 | /* |
5033 | * Called by consumers of io_uring_cmd, if they originally returned | |
5034 | * -EIOCBQUEUED upon receiving the command. | |
5035 | */ | |
5036 | void io_uring_cmd_done(struct io_uring_cmd *ioucmd, ssize_t ret, ssize_t res2) | |
5037 | { | |
5038 | struct io_kiocb *req = container_of(ioucmd, struct io_kiocb, uring_cmd); | |
5039 | ||
5040 | if (ret < 0) | |
5041 | req_set_fail(req); | |
29ede201 | 5042 | |
ee692a21 | 5043 | if (req->ctx->flags & IORING_SETUP_CQE32) |
29ede201 PB |
5044 | io_req_set_cqe32_extra(req, res2, 0); |
5045 | io_req_complete(req, ret); | |
ee692a21 JA |
5046 | } |
5047 | EXPORT_SYMBOL_GPL(io_uring_cmd_done); | |
5048 | ||
5049 | static int io_uring_cmd_prep_async(struct io_kiocb *req) | |
5050 | { | |
5051 | size_t cmd_size; | |
5052 | ||
5053 | cmd_size = uring_cmd_pdu_size(req->ctx->flags & IORING_SETUP_SQE128); | |
5054 | ||
5055 | memcpy(req->async_data, req->uring_cmd.cmd, cmd_size); | |
5056 | return 0; | |
5057 | } | |
5058 | ||
5059 | static int io_uring_cmd_prep(struct io_kiocb *req, | |
5060 | const struct io_uring_sqe *sqe) | |
5061 | { | |
5062 | struct io_uring_cmd *ioucmd = &req->uring_cmd; | |
5063 | ||
5064 | if (sqe->rw_flags) | |
5065 | return -EINVAL; | |
5066 | ioucmd->cmd = sqe->cmd; | |
5067 | ioucmd->cmd_op = READ_ONCE(sqe->cmd_op); | |
5068 | return 0; | |
5069 | } | |
5070 | ||
5071 | static int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags) | |
5072 | { | |
5073 | struct io_uring_cmd *ioucmd = &req->uring_cmd; | |
5074 | struct io_ring_ctx *ctx = req->ctx; | |
5075 | struct file *file = req->file; | |
5076 | int ret; | |
5077 | ||
5078 | if (!req->file->f_op->uring_cmd) | |
5079 | return -EOPNOTSUPP; | |
5080 | ||
5081 | if (ctx->flags & IORING_SETUP_SQE128) | |
5082 | issue_flags |= IO_URING_F_SQE128; | |
5083 | if (ctx->flags & IORING_SETUP_CQE32) | |
5084 | issue_flags |= IO_URING_F_CQE32; | |
5085 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
5086 | issue_flags |= IO_URING_F_IOPOLL; | |
5087 | ||
5088 | if (req_has_async_data(req)) | |
5089 | ioucmd->cmd = req->async_data; | |
5090 | ||
5091 | ret = file->f_op->uring_cmd(ioucmd, issue_flags); | |
5092 | if (ret == -EAGAIN) { | |
5093 | if (!req_has_async_data(req)) { | |
5094 | if (io_alloc_async_data(req)) | |
5095 | return -ENOMEM; | |
5096 | io_uring_cmd_prep_async(req); | |
5097 | } | |
5098 | return -EAGAIN; | |
5099 | } | |
5100 | ||
5101 | if (ret != -EIOCBQUEUED) | |
5102 | io_uring_cmd_done(ioucmd, ret, 0); | |
5103 | return 0; | |
5104 | } | |
5105 | ||
f2a8d5c7 PB |
5106 | static int __io_splice_prep(struct io_kiocb *req, |
5107 | const struct io_uring_sqe *sqe) | |
7d67af2c | 5108 | { |
fe7e3257 | 5109 | struct io_splice *sp = &req->splice; |
7d67af2c | 5110 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; |
7d67af2c | 5111 | |
7d67af2c PB |
5112 | sp->len = READ_ONCE(sqe->len); |
5113 | sp->flags = READ_ONCE(sqe->splice_flags); | |
7d67af2c PB |
5114 | if (unlikely(sp->flags & ~valid_flags)) |
5115 | return -EINVAL; | |
a3e4bc23 | 5116 | sp->splice_fd_in = READ_ONCE(sqe->splice_fd_in); |
7d67af2c PB |
5117 | return 0; |
5118 | } | |
5119 | ||
f2a8d5c7 PB |
5120 | static int io_tee_prep(struct io_kiocb *req, |
5121 | const struct io_uring_sqe *sqe) | |
5122 | { | |
5123 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
5124 | return -EINVAL; | |
5125 | return __io_splice_prep(req, sqe); | |
5126 | } | |
5127 | ||
45d189c6 | 5128 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
5129 | { |
5130 | struct io_splice *sp = &req->splice; | |
f2a8d5c7 PB |
5131 | struct file *out = sp->file_out; |
5132 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
a3e4bc23 | 5133 | struct file *in; |
f2a8d5c7 PB |
5134 | long ret = 0; |
5135 | ||
45d189c6 | 5136 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 | 5137 | return -EAGAIN; |
a3e4bc23 | 5138 | |
5106dd6e | 5139 | if (sp->flags & SPLICE_F_FD_IN_FIXED) |
e9419766 | 5140 | in = io_file_get_fixed(req, sp->splice_fd_in, issue_flags); |
5106dd6e JA |
5141 | else |
5142 | in = io_file_get_normal(req, sp->splice_fd_in); | |
a3e4bc23 JA |
5143 | if (!in) { |
5144 | ret = -EBADF; | |
5145 | goto done; | |
5146 | } | |
5147 | ||
f2a8d5c7 PB |
5148 | if (sp->len) |
5149 | ret = do_tee(in, out, sp->len, flags); | |
5150 | ||
e1d767f0 PB |
5151 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
5152 | io_put_file(in); | |
a3e4bc23 | 5153 | done: |
f2a8d5c7 | 5154 | if (ret != sp->len) |
93d2bcd2 | 5155 | req_set_fail(req); |
4ffaa94b | 5156 | __io_req_complete(req, 0, ret, 0); |
f2a8d5c7 PB |
5157 | return 0; |
5158 | } | |
5159 | ||
5160 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5161 | { | |
fe7e3257 | 5162 | struct io_splice *sp = &req->splice; |
f2a8d5c7 PB |
5163 | |
5164 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
5165 | sp->off_out = READ_ONCE(sqe->off); | |
5166 | return __io_splice_prep(req, sqe); | |
5167 | } | |
5168 | ||
45d189c6 | 5169 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
5170 | { |
5171 | struct io_splice *sp = &req->splice; | |
7d67af2c PB |
5172 | struct file *out = sp->file_out; |
5173 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
5174 | loff_t *poff_in, *poff_out; | |
a3e4bc23 | 5175 | struct file *in; |
c9687426 | 5176 | long ret = 0; |
7d67af2c | 5177 | |
45d189c6 | 5178 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 5179 | return -EAGAIN; |
7d67af2c | 5180 | |
5106dd6e | 5181 | if (sp->flags & SPLICE_F_FD_IN_FIXED) |
e9419766 | 5182 | in = io_file_get_fixed(req, sp->splice_fd_in, issue_flags); |
5106dd6e JA |
5183 | else |
5184 | in = io_file_get_normal(req, sp->splice_fd_in); | |
a3e4bc23 JA |
5185 | if (!in) { |
5186 | ret = -EBADF; | |
5187 | goto done; | |
5188 | } | |
5189 | ||
7d67af2c PB |
5190 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; |
5191 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 5192 | |
948a7749 | 5193 | if (sp->len) |
c9687426 | 5194 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 5195 | |
e1d767f0 PB |
5196 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
5197 | io_put_file(in); | |
a3e4bc23 | 5198 | done: |
7d67af2c | 5199 | if (ret != sp->len) |
93d2bcd2 | 5200 | req_set_fail(req); |
4ffaa94b | 5201 | __io_req_complete(req, 0, ret, 0); |
7d67af2c PB |
5202 | return 0; |
5203 | } | |
5204 | ||
2bb04df7 SR |
5205 | static int io_nop_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5206 | { | |
2bb04df7 SR |
5207 | return 0; |
5208 | } | |
5209 | ||
2b188cc1 JA |
5210 | /* |
5211 | * IORING_OP_NOP just posts a completion event, nothing else. | |
5212 | */ | |
889fca73 | 5213 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5214 | { |
aa165d6d | 5215 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
5216 | return 0; |
5217 | } | |
5218 | ||
4f57f06c JA |
5219 | static int io_msg_ring_prep(struct io_kiocb *req, |
5220 | const struct io_uring_sqe *sqe) | |
5221 | { | |
73911426 JA |
5222 | if (unlikely(sqe->addr || sqe->rw_flags || sqe->splice_fd_in || |
5223 | sqe->buf_index || sqe->personality)) | |
4f57f06c JA |
5224 | return -EINVAL; |
5225 | ||
4f57f06c JA |
5226 | req->msg.user_data = READ_ONCE(sqe->off); |
5227 | req->msg.len = READ_ONCE(sqe->len); | |
5228 | return 0; | |
5229 | } | |
5230 | ||
5231 | static int io_msg_ring(struct io_kiocb *req, unsigned int issue_flags) | |
5232 | { | |
5233 | struct io_ring_ctx *target_ctx; | |
5234 | struct io_msg *msg = &req->msg; | |
4f57f06c | 5235 | bool filled; |
3f1d52ab | 5236 | int ret; |
4f57f06c | 5237 | |
3f1d52ab JA |
5238 | ret = -EBADFD; |
5239 | if (req->file->f_op != &io_uring_fops) | |
5240 | goto done; | |
4f57f06c | 5241 | |
3f1d52ab | 5242 | ret = -EOVERFLOW; |
4f57f06c JA |
5243 | target_ctx = req->file->private_data; |
5244 | ||
5245 | spin_lock(&target_ctx->completion_lock); | |
7ef66d18 | 5246 | filled = io_fill_cqe_aux(target_ctx, msg->user_data, msg->len, 0); |
4f57f06c JA |
5247 | io_commit_cqring(target_ctx); |
5248 | spin_unlock(&target_ctx->completion_lock); | |
5249 | ||
5250 | if (filled) { | |
5251 | io_cqring_ev_posted(target_ctx); | |
5252 | ret = 0; | |
5253 | } | |
5254 | ||
3f1d52ab | 5255 | done: |
9666d420 JA |
5256 | if (ret < 0) |
5257 | req_set_fail(req); | |
4f57f06c | 5258 | __io_req_complete(req, issue_flags, ret, 0); |
aa184e86 JA |
5259 | /* put file to avoid an attempt to IOPOLL the req */ |
5260 | io_put_file(req->file); | |
5261 | req->file = NULL; | |
4f57f06c JA |
5262 | return 0; |
5263 | } | |
5264 | ||
1155c76a | 5265 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 5266 | { |
73911426 | 5267 | if (unlikely(sqe->addr || sqe->buf_index || sqe->splice_fd_in)) |
c992fe29 CH |
5268 | return -EINVAL; |
5269 | ||
8ed8d3c3 JA |
5270 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
5271 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
5272 | return -EINVAL; | |
5273 | ||
5274 | req->sync.off = READ_ONCE(sqe->off); | |
5275 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
5276 | return 0; |
5277 | } | |
5278 | ||
45d189c6 | 5279 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 5280 | { |
8ed8d3c3 | 5281 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
5282 | int ret; |
5283 | ||
ac45abc0 | 5284 | /* fsync always requires a blocking context */ |
45d189c6 | 5285 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
5286 | return -EAGAIN; |
5287 | ||
9adbd45d | 5288 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
5289 | end > 0 ? end : LLONG_MAX, |
5290 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
e1e16097 | 5291 | io_req_complete(req, ret); |
c992fe29 CH |
5292 | return 0; |
5293 | } | |
5294 | ||
d63d1b5e JA |
5295 | static int io_fallocate_prep(struct io_kiocb *req, |
5296 | const struct io_uring_sqe *sqe) | |
5297 | { | |
73911426 | 5298 | if (sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in) |
3232dd02 | 5299 | return -EINVAL; |
d63d1b5e JA |
5300 | |
5301 | req->sync.off = READ_ONCE(sqe->off); | |
5302 | req->sync.len = READ_ONCE(sqe->addr); | |
5303 | req->sync.mode = READ_ONCE(sqe->len); | |
5304 | return 0; | |
5305 | } | |
5306 | ||
45d189c6 | 5307 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 5308 | { |
ac45abc0 PB |
5309 | int ret; |
5310 | ||
d63d1b5e | 5311 | /* fallocate always requiring blocking context */ |
45d189c6 | 5312 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 5313 | return -EAGAIN; |
ac45abc0 PB |
5314 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
5315 | req->sync.len); | |
4ffaa94b | 5316 | if (ret >= 0) |
f63cf519 | 5317 | fsnotify_modify(req->file); |
e1e16097 | 5318 | io_req_complete(req, ret); |
5d17b4a4 JA |
5319 | return 0; |
5320 | } | |
5321 | ||
ec65fea5 | 5322 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 5323 | { |
f8748881 | 5324 | const char __user *fname; |
15b71abe | 5325 | int ret; |
b7bb4f7d | 5326 | |
73911426 | 5327 | if (unlikely(sqe->buf_index)) |
15b71abe | 5328 | return -EINVAL; |
ec65fea5 | 5329 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 5330 | return -EBADF; |
03b1230c | 5331 | |
ec65fea5 PB |
5332 | /* open.how should be already initialised */ |
5333 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 5334 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 5335 | |
25e72d10 PB |
5336 | req->open.dfd = READ_ONCE(sqe->fd); |
5337 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 5338 | req->open.filename = getname(fname); |
15b71abe JA |
5339 | if (IS_ERR(req->open.filename)) { |
5340 | ret = PTR_ERR(req->open.filename); | |
5341 | req->open.filename = NULL; | |
5342 | return ret; | |
5343 | } | |
b9445598 PB |
5344 | |
5345 | req->open.file_slot = READ_ONCE(sqe->file_index); | |
5346 | if (req->open.file_slot && (req->open.how.flags & O_CLOEXEC)) | |
5347 | return -EINVAL; | |
5348 | ||
4022e7af | 5349 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 5350 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 5351 | return 0; |
03b1230c JA |
5352 | } |
5353 | ||
ec65fea5 PB |
5354 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5355 | { | |
d3fddf6d PB |
5356 | u64 mode = READ_ONCE(sqe->len); |
5357 | u64 flags = READ_ONCE(sqe->open_flags); | |
ec65fea5 | 5358 | |
ec65fea5 PB |
5359 | req->open.how = build_open_how(flags, mode); |
5360 | return __io_openat_prep(req, sqe); | |
5361 | } | |
5362 | ||
cebdb986 | 5363 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 5364 | { |
cebdb986 | 5365 | struct open_how __user *how; |
cebdb986 | 5366 | size_t len; |
0fa03c62 JA |
5367 | int ret; |
5368 | ||
cebdb986 JA |
5369 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
5370 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
5371 | if (len < OPEN_HOW_SIZE_VER0) |
5372 | return -EINVAL; | |
3529d8c2 | 5373 | |
cebdb986 JA |
5374 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
5375 | len); | |
5376 | if (ret) | |
5377 | return ret; | |
3529d8c2 | 5378 | |
ec65fea5 | 5379 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
5380 | } |
5381 | ||
1339f24b | 5382 | static int io_file_bitmap_get(struct io_ring_ctx *ctx) |
b70b8e33 JA |
5383 | { |
5384 | struct io_file_table *table = &ctx->file_table; | |
5385 | unsigned long nr = ctx->nr_user_files; | |
5386 | int ret; | |
5387 | ||
b70b8e33 JA |
5388 | do { |
5389 | ret = find_next_zero_bit(table->bitmap, nr, table->alloc_hint); | |
4278a0de | 5390 | if (ret != nr) |
b70b8e33 | 5391 | return ret; |
4278a0de | 5392 | |
b70b8e33 JA |
5393 | if (!table->alloc_hint) |
5394 | break; | |
5395 | ||
5396 | nr = table->alloc_hint; | |
5397 | table->alloc_hint = 0; | |
5398 | } while (1); | |
5399 | ||
5400 | return -ENFILE; | |
5401 | } | |
5402 | ||
8c71fe75 XW |
5403 | /* |
5404 | * Note when io_fixed_fd_install() returns error value, it will ensure | |
5405 | * fput() is called correspondingly. | |
5406 | */ | |
1339f24b JA |
5407 | static int io_fixed_fd_install(struct io_kiocb *req, unsigned int issue_flags, |
5408 | struct file *file, unsigned int file_slot) | |
5409 | { | |
5410 | bool alloc_slot = file_slot == IORING_FILE_INDEX_ALLOC; | |
5411 | struct io_ring_ctx *ctx = req->ctx; | |
5412 | int ret; | |
5413 | ||
61c1b44a PB |
5414 | io_ring_submit_lock(ctx, issue_flags); |
5415 | ||
1339f24b | 5416 | if (alloc_slot) { |
1339f24b | 5417 | ret = io_file_bitmap_get(ctx); |
61c1b44a PB |
5418 | if (unlikely(ret < 0)) |
5419 | goto err; | |
1339f24b JA |
5420 | file_slot = ret; |
5421 | } else { | |
5422 | file_slot--; | |
5423 | } | |
5424 | ||
5425 | ret = io_install_fixed_file(req, file, issue_flags, file_slot); | |
61c1b44a PB |
5426 | if (!ret && alloc_slot) |
5427 | ret = file_slot; | |
5428 | err: | |
5429 | io_ring_submit_unlock(ctx, issue_flags); | |
5430 | if (unlikely(ret < 0)) | |
5431 | fput(file); | |
1339f24b JA |
5432 | return ret; |
5433 | } | |
5434 | ||
45d189c6 | 5435 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
5436 | { |
5437 | struct open_flags op; | |
15b71abe | 5438 | struct file *file; |
b9445598 PB |
5439 | bool resolve_nonblock, nonblock_set; |
5440 | bool fixed = !!req->open.file_slot; | |
15b71abe JA |
5441 | int ret; |
5442 | ||
cebdb986 | 5443 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
5444 | if (ret) |
5445 | goto err; | |
3a81fd02 JA |
5446 | nonblock_set = op.open_flag & O_NONBLOCK; |
5447 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 5448 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
5449 | /* |
5450 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
5451 | * it'll always -EAGAIN | |
5452 | */ | |
5453 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
5454 | return -EAGAIN; | |
5455 | op.lookup_flags |= LOOKUP_CACHED; | |
5456 | op.open_flag |= O_NONBLOCK; | |
5457 | } | |
15b71abe | 5458 | |
b9445598 PB |
5459 | if (!fixed) { |
5460 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); | |
5461 | if (ret < 0) | |
5462 | goto err; | |
5463 | } | |
15b71abe JA |
5464 | |
5465 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
12dcb58a | 5466 | if (IS_ERR(file)) { |
944d1444 | 5467 | /* |
12dcb58a PB |
5468 | * We could hang on to this 'fd' on retrying, but seems like |
5469 | * marginal gain for something that is now known to be a slower | |
5470 | * path. So just put it, and we'll get a new one when we retry. | |
944d1444 | 5471 | */ |
b9445598 PB |
5472 | if (!fixed) |
5473 | put_unused_fd(ret); | |
3a81fd02 | 5474 | |
15b71abe | 5475 | ret = PTR_ERR(file); |
12dcb58a PB |
5476 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
5477 | if (ret == -EAGAIN && | |
5478 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) | |
5479 | return -EAGAIN; | |
5480 | goto err; | |
15b71abe | 5481 | } |
12dcb58a PB |
5482 | |
5483 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) | |
5484 | file->f_flags &= ~O_NONBLOCK; | |
5485 | fsnotify_open(file); | |
b9445598 PB |
5486 | |
5487 | if (!fixed) | |
5488 | fd_install(ret, file); | |
5489 | else | |
1339f24b JA |
5490 | ret = io_fixed_fd_install(req, issue_flags, file, |
5491 | req->open.file_slot); | |
15b71abe JA |
5492 | err: |
5493 | putname(req->open.filename); | |
8fef80bf | 5494 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe | 5495 | if (ret < 0) |
93d2bcd2 | 5496 | req_set_fail(req); |
0bdf3398 | 5497 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
5498 | return 0; |
5499 | } | |
5500 | ||
45d189c6 | 5501 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 5502 | { |
e45cff58 | 5503 | return io_openat2(req, issue_flags); |
cebdb986 JA |
5504 | } |
5505 | ||
067524e9 JA |
5506 | static int io_remove_buffers_prep(struct io_kiocb *req, |
5507 | const struct io_uring_sqe *sqe) | |
5508 | { | |
5509 | struct io_provide_buf *p = &req->pbuf; | |
5510 | u64 tmp; | |
5511 | ||
73911426 | 5512 | if (sqe->rw_flags || sqe->addr || sqe->len || sqe->off || |
26578cda | 5513 | sqe->splice_fd_in) |
067524e9 JA |
5514 | return -EINVAL; |
5515 | ||
5516 | tmp = READ_ONCE(sqe->fd); | |
5517 | if (!tmp || tmp > USHRT_MAX) | |
5518 | return -EINVAL; | |
5519 | ||
5520 | memset(p, 0, sizeof(*p)); | |
5521 | p->nbufs = tmp; | |
5522 | p->bgid = READ_ONCE(sqe->buf_group); | |
5523 | return 0; | |
5524 | } | |
5525 | ||
dbc7d452 JA |
5526 | static int __io_remove_buffers(struct io_ring_ctx *ctx, |
5527 | struct io_buffer_list *bl, unsigned nbufs) | |
067524e9 JA |
5528 | { |
5529 | unsigned i = 0; | |
5530 | ||
5531 | /* shouldn't happen */ | |
5532 | if (!nbufs) | |
5533 | return 0; | |
5534 | ||
c7fb1942 JA |
5535 | if (bl->buf_nr_pages) { |
5536 | int j; | |
5537 | ||
5538 | i = bl->buf_ring->tail - bl->head; | |
5539 | for (j = 0; j < bl->buf_nr_pages; j++) | |
5540 | unpin_user_page(bl->buf_pages[j]); | |
5541 | kvfree(bl->buf_pages); | |
5542 | bl->buf_pages = NULL; | |
5543 | bl->buf_nr_pages = 0; | |
1d0dbbfa JA |
5544 | /* make sure it's seen as empty */ |
5545 | INIT_LIST_HEAD(&bl->buf_list); | |
c7fb1942 JA |
5546 | return i; |
5547 | } | |
5548 | ||
067524e9 | 5549 | /* the head kbuf is the list itself */ |
dbc7d452 | 5550 | while (!list_empty(&bl->buf_list)) { |
067524e9 JA |
5551 | struct io_buffer *nxt; |
5552 | ||
dbc7d452 | 5553 | nxt = list_first_entry(&bl->buf_list, struct io_buffer, list); |
067524e9 | 5554 | list_del(&nxt->list); |
067524e9 JA |
5555 | if (++i == nbufs) |
5556 | return i; | |
1d0254e6 | 5557 | cond_resched(); |
067524e9 JA |
5558 | } |
5559 | i++; | |
067524e9 JA |
5560 | |
5561 | return i; | |
5562 | } | |
5563 | ||
889fca73 | 5564 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
5565 | { |
5566 | struct io_provide_buf *p = &req->pbuf; | |
5567 | struct io_ring_ctx *ctx = req->ctx; | |
dbc7d452 | 5568 | struct io_buffer_list *bl; |
067524e9 | 5569 | int ret = 0; |
067524e9 | 5570 | |
f8929630 | 5571 | io_ring_submit_lock(ctx, issue_flags); |
067524e9 JA |
5572 | |
5573 | ret = -ENOENT; | |
dbc7d452 | 5574 | bl = io_buffer_get_list(ctx, p->bgid); |
c7fb1942 JA |
5575 | if (bl) { |
5576 | ret = -EINVAL; | |
5577 | /* can't use provide/remove buffers command on mapped buffers */ | |
5578 | if (!bl->buf_nr_pages) | |
5579 | ret = __io_remove_buffers(ctx, bl, p->nbufs); | |
5580 | } | |
067524e9 | 5581 | if (ret < 0) |
93d2bcd2 | 5582 | req_set_fail(req); |
067524e9 | 5583 | |
9fb8cb49 PB |
5584 | /* complete before unlock, IOPOLL may need the lock */ |
5585 | __io_req_complete(req, issue_flags, ret, 0); | |
f8929630 | 5586 | io_ring_submit_unlock(ctx, issue_flags); |
067524e9 JA |
5587 | return 0; |
5588 | } | |
5589 | ||
ddf0322d JA |
5590 | static int io_provide_buffers_prep(struct io_kiocb *req, |
5591 | const struct io_uring_sqe *sqe) | |
5592 | { | |
38134ada | 5593 | unsigned long size, tmp_check; |
ddf0322d JA |
5594 | struct io_provide_buf *p = &req->pbuf; |
5595 | u64 tmp; | |
5596 | ||
73911426 | 5597 | if (sqe->rw_flags || sqe->splice_fd_in) |
ddf0322d JA |
5598 | return -EINVAL; |
5599 | ||
5600 | tmp = READ_ONCE(sqe->fd); | |
5601 | if (!tmp || tmp > USHRT_MAX) | |
5602 | return -E2BIG; | |
5603 | p->nbufs = tmp; | |
5604 | p->addr = READ_ONCE(sqe->addr); | |
5605 | p->len = READ_ONCE(sqe->len); | |
5606 | ||
38134ada PB |
5607 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
5608 | &size)) | |
5609 | return -EOVERFLOW; | |
5610 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
5611 | return -EOVERFLOW; | |
5612 | ||
d81269fe PB |
5613 | size = (unsigned long)p->len * p->nbufs; |
5614 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
5615 | return -EFAULT; |
5616 | ||
5617 | p->bgid = READ_ONCE(sqe->buf_group); | |
5618 | tmp = READ_ONCE(sqe->off); | |
5619 | if (tmp > USHRT_MAX) | |
5620 | return -E2BIG; | |
5621 | p->bid = tmp; | |
5622 | return 0; | |
5623 | } | |
5624 | ||
cc3cec83 JA |
5625 | static int io_refill_buffer_cache(struct io_ring_ctx *ctx) |
5626 | { | |
5627 | struct io_buffer *buf; | |
5628 | struct page *page; | |
5629 | int bufs_in_page; | |
5630 | ||
5631 | /* | |
5632 | * Completions that don't happen inline (eg not under uring_lock) will | |
5633 | * add to ->io_buffers_comp. If we don't have any free buffers, check | |
5634 | * the completion list and splice those entries first. | |
5635 | */ | |
5636 | if (!list_empty_careful(&ctx->io_buffers_comp)) { | |
5637 | spin_lock(&ctx->completion_lock); | |
5638 | if (!list_empty(&ctx->io_buffers_comp)) { | |
5639 | list_splice_init(&ctx->io_buffers_comp, | |
5640 | &ctx->io_buffers_cache); | |
5641 | spin_unlock(&ctx->completion_lock); | |
5642 | return 0; | |
5643 | } | |
5644 | spin_unlock(&ctx->completion_lock); | |
5645 | } | |
5646 | ||
5647 | /* | |
5648 | * No free buffers and no completion entries either. Allocate a new | |
5649 | * page worth of buffer entries and add those to our freelist. | |
5650 | */ | |
5651 | page = alloc_page(GFP_KERNEL_ACCOUNT); | |
5652 | if (!page) | |
5653 | return -ENOMEM; | |
5654 | ||
5655 | list_add(&page->lru, &ctx->io_buffers_pages); | |
5656 | ||
5657 | buf = page_address(page); | |
5658 | bufs_in_page = PAGE_SIZE / sizeof(*buf); | |
5659 | while (bufs_in_page) { | |
5660 | list_add_tail(&buf->list, &ctx->io_buffers_cache); | |
5661 | buf++; | |
5662 | bufs_in_page--; | |
5663 | } | |
5664 | ||
5665 | return 0; | |
5666 | } | |
5667 | ||
5668 | static int io_add_buffers(struct io_ring_ctx *ctx, struct io_provide_buf *pbuf, | |
dbc7d452 | 5669 | struct io_buffer_list *bl) |
ddf0322d JA |
5670 | { |
5671 | struct io_buffer *buf; | |
5672 | u64 addr = pbuf->addr; | |
5673 | int i, bid = pbuf->bid; | |
5674 | ||
5675 | for (i = 0; i < pbuf->nbufs; i++) { | |
cc3cec83 JA |
5676 | if (list_empty(&ctx->io_buffers_cache) && |
5677 | io_refill_buffer_cache(ctx)) | |
ddf0322d | 5678 | break; |
cc3cec83 JA |
5679 | buf = list_first_entry(&ctx->io_buffers_cache, struct io_buffer, |
5680 | list); | |
dbc7d452 | 5681 | list_move_tail(&buf->list, &bl->buf_list); |
ddf0322d | 5682 | buf->addr = addr; |
d1f82808 | 5683 | buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); |
ddf0322d | 5684 | buf->bid = bid; |
b1c62645 | 5685 | buf->bgid = pbuf->bgid; |
ddf0322d JA |
5686 | addr += pbuf->len; |
5687 | bid++; | |
f240762f | 5688 | cond_resched(); |
ddf0322d JA |
5689 | } |
5690 | ||
dbc7d452 | 5691 | return i ? 0 : -ENOMEM; |
ddf0322d JA |
5692 | } |
5693 | ||
9cfc7e94 JA |
5694 | static __cold int io_init_bl_list(struct io_ring_ctx *ctx) |
5695 | { | |
5696 | int i; | |
5697 | ||
5698 | ctx->io_bl = kcalloc(BGID_ARRAY, sizeof(struct io_buffer_list), | |
5699 | GFP_KERNEL); | |
5700 | if (!ctx->io_bl) | |
5701 | return -ENOMEM; | |
5702 | ||
5703 | for (i = 0; i < BGID_ARRAY; i++) { | |
5704 | INIT_LIST_HEAD(&ctx->io_bl[i].buf_list); | |
5705 | ctx->io_bl[i].bgid = i; | |
5706 | } | |
5707 | ||
5708 | return 0; | |
5709 | } | |
5710 | ||
889fca73 | 5711 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
5712 | { |
5713 | struct io_provide_buf *p = &req->pbuf; | |
5714 | struct io_ring_ctx *ctx = req->ctx; | |
dbc7d452 | 5715 | struct io_buffer_list *bl; |
ddf0322d JA |
5716 | int ret = 0; |
5717 | ||
f8929630 | 5718 | io_ring_submit_lock(ctx, issue_flags); |
ddf0322d | 5719 | |
9cfc7e94 JA |
5720 | if (unlikely(p->bgid < BGID_ARRAY && !ctx->io_bl)) { |
5721 | ret = io_init_bl_list(ctx); | |
5722 | if (ret) | |
5723 | goto err; | |
5724 | } | |
ddf0322d | 5725 | |
dbc7d452 JA |
5726 | bl = io_buffer_get_list(ctx, p->bgid); |
5727 | if (unlikely(!bl)) { | |
c7fb1942 | 5728 | bl = kzalloc(sizeof(*bl), GFP_KERNEL); |
dbc7d452 JA |
5729 | if (!bl) { |
5730 | ret = -ENOMEM; | |
5731 | goto err; | |
5732 | } | |
1d0dbbfa | 5733 | INIT_LIST_HEAD(&bl->buf_list); |
9cfc7e94 JA |
5734 | ret = io_buffer_add_list(ctx, bl, p->bgid); |
5735 | if (ret) { | |
5736 | kfree(bl); | |
5737 | goto err; | |
5738 | } | |
ddf0322d | 5739 | } |
c7fb1942 JA |
5740 | /* can't add buffers via this command for a mapped buffer ring */ |
5741 | if (bl->buf_nr_pages) { | |
5742 | ret = -EINVAL; | |
5743 | goto err; | |
ddf0322d | 5744 | } |
dbc7d452 JA |
5745 | |
5746 | ret = io_add_buffers(ctx, p, bl); | |
5747 | err: | |
ddf0322d | 5748 | if (ret < 0) |
93d2bcd2 | 5749 | req_set_fail(req); |
9fb8cb49 PB |
5750 | /* complete before unlock, IOPOLL may need the lock */ |
5751 | __io_req_complete(req, issue_flags, ret, 0); | |
f8929630 | 5752 | io_ring_submit_unlock(ctx, issue_flags); |
ddf0322d | 5753 | return 0; |
cebdb986 JA |
5754 | } |
5755 | ||
3e4827b0 JA |
5756 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
5757 | const struct io_uring_sqe *sqe) | |
5758 | { | |
5759 | #if defined(CONFIG_EPOLL) | |
73911426 | 5760 | if (sqe->buf_index || sqe->splice_fd_in) |
3232dd02 | 5761 | return -EINVAL; |
3e4827b0 JA |
5762 | |
5763 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
5764 | req->epoll.op = READ_ONCE(sqe->len); | |
5765 | req->epoll.fd = READ_ONCE(sqe->off); | |
5766 | ||
5767 | if (ep_op_has_event(req->epoll.op)) { | |
5768 | struct epoll_event __user *ev; | |
5769 | ||
5770 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
5771 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
5772 | return -EFAULT; | |
5773 | } | |
5774 | ||
5775 | return 0; | |
5776 | #else | |
5777 | return -EOPNOTSUPP; | |
5778 | #endif | |
5779 | } | |
5780 | ||
889fca73 | 5781 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
5782 | { |
5783 | #if defined(CONFIG_EPOLL) | |
5784 | struct io_epoll *ie = &req->epoll; | |
5785 | int ret; | |
45d189c6 | 5786 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
5787 | |
5788 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
5789 | if (force_nonblock && ret == -EAGAIN) | |
5790 | return -EAGAIN; | |
5791 | ||
5792 | if (ret < 0) | |
93d2bcd2 | 5793 | req_set_fail(req); |
889fca73 | 5794 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
5795 | return 0; |
5796 | #else | |
5797 | return -EOPNOTSUPP; | |
5798 | #endif | |
5799 | } | |
5800 | ||
c1ca757b JA |
5801 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5802 | { | |
5803 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
73911426 | 5804 | if (sqe->buf_index || sqe->off || sqe->splice_fd_in) |
3232dd02 | 5805 | return -EINVAL; |
c1ca757b JA |
5806 | |
5807 | req->madvise.addr = READ_ONCE(sqe->addr); | |
5808 | req->madvise.len = READ_ONCE(sqe->len); | |
5809 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
5810 | return 0; | |
5811 | #else | |
5812 | return -EOPNOTSUPP; | |
5813 | #endif | |
5814 | } | |
5815 | ||
45d189c6 | 5816 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
5817 | { |
5818 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
5819 | struct io_madvise *ma = &req->madvise; | |
5820 | int ret; | |
5821 | ||
45d189c6 | 5822 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
5823 | return -EAGAIN; |
5824 | ||
0726b01e | 5825 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
e1e16097 | 5826 | io_req_complete(req, ret); |
c1ca757b JA |
5827 | return 0; |
5828 | #else | |
5829 | return -EOPNOTSUPP; | |
5830 | #endif | |
5831 | } | |
5832 | ||
4840e418 JA |
5833 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5834 | { | |
73911426 | 5835 | if (sqe->buf_index || sqe->addr || sqe->splice_fd_in) |
3232dd02 | 5836 | return -EINVAL; |
4840e418 JA |
5837 | |
5838 | req->fadvise.offset = READ_ONCE(sqe->off); | |
5839 | req->fadvise.len = READ_ONCE(sqe->len); | |
5840 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
5841 | return 0; | |
5842 | } | |
5843 | ||
45d189c6 | 5844 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
5845 | { |
5846 | struct io_fadvise *fa = &req->fadvise; | |
5847 | int ret; | |
5848 | ||
45d189c6 | 5849 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
5850 | switch (fa->advice) { |
5851 | case POSIX_FADV_NORMAL: | |
5852 | case POSIX_FADV_RANDOM: | |
5853 | case POSIX_FADV_SEQUENTIAL: | |
5854 | break; | |
5855 | default: | |
5856 | return -EAGAIN; | |
5857 | } | |
5858 | } | |
4840e418 JA |
5859 | |
5860 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
5861 | if (ret < 0) | |
93d2bcd2 | 5862 | req_set_fail(req); |
0bdf3398 | 5863 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
5864 | return 0; |
5865 | } | |
5866 | ||
eddc7ef5 JA |
5867 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5868 | { | |
1b6fe6e0 SR |
5869 | const char __user *path; |
5870 | ||
73911426 | 5871 | if (sqe->buf_index || sqe->splice_fd_in) |
eddc7ef5 | 5872 | return -EINVAL; |
9c280f90 | 5873 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 5874 | return -EBADF; |
eddc7ef5 | 5875 | |
1d9e1288 BM |
5876 | req->statx.dfd = READ_ONCE(sqe->fd); |
5877 | req->statx.mask = READ_ONCE(sqe->len); | |
1b6fe6e0 | 5878 | path = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
5879 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
5880 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 | 5881 | |
1b6fe6e0 SR |
5882 | req->statx.filename = getname_flags(path, |
5883 | getname_statx_lookup_flags(req->statx.flags), | |
5884 | NULL); | |
5885 | ||
5886 | if (IS_ERR(req->statx.filename)) { | |
5887 | int ret = PTR_ERR(req->statx.filename); | |
5888 | ||
5889 | req->statx.filename = NULL; | |
5890 | return ret; | |
5891 | } | |
5892 | ||
5893 | req->flags |= REQ_F_NEED_CLEANUP; | |
eddc7ef5 JA |
5894 | return 0; |
5895 | } | |
5896 | ||
45d189c6 | 5897 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 5898 | { |
1d9e1288 | 5899 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
5900 | int ret; |
5901 | ||
59d70013 | 5902 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
5903 | return -EAGAIN; |
5904 | ||
e62753e4 BM |
5905 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
5906 | ctx->buffer); | |
e1e16097 | 5907 | io_req_complete(req, ret); |
eddc7ef5 JA |
5908 | return 0; |
5909 | } | |
5910 | ||
b5dba59e JA |
5911 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5912 | { | |
d884b649 | 5913 | if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) |
b5dba59e | 5914 | return -EINVAL; |
9c280f90 | 5915 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 5916 | return -EBADF; |
b5dba59e JA |
5917 | |
5918 | req->close.fd = READ_ONCE(sqe->fd); | |
7df778be | 5919 | req->close.file_slot = READ_ONCE(sqe->file_index); |
d884b649 | 5920 | if (req->close.file_slot && req->close.fd) |
7df778be PB |
5921 | return -EINVAL; |
5922 | ||
b5dba59e | 5923 | return 0; |
b5dba59e JA |
5924 | } |
5925 | ||
889fca73 | 5926 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 5927 | { |
9eac1904 | 5928 | struct files_struct *files = current->files; |
3af73b28 | 5929 | struct io_close *close = &req->close; |
9eac1904 | 5930 | struct fdtable *fdt; |
40a19260 | 5931 | struct file *file; |
a1fde923 | 5932 | int ret = -EBADF; |
b5dba59e | 5933 | |
7df778be PB |
5934 | if (req->close.file_slot) { |
5935 | ret = io_close_fixed(req, issue_flags); | |
d884b649 | 5936 | goto err; |
7df778be PB |
5937 | } |
5938 | ||
9eac1904 JA |
5939 | spin_lock(&files->file_lock); |
5940 | fdt = files_fdtable(files); | |
5941 | if (close->fd >= fdt->max_fds) { | |
5942 | spin_unlock(&files->file_lock); | |
5943 | goto err; | |
5944 | } | |
0bf1dbee CH |
5945 | file = rcu_dereference_protected(fdt->fd[close->fd], |
5946 | lockdep_is_held(&files->file_lock)); | |
a1fde923 | 5947 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 | 5948 | spin_unlock(&files->file_lock); |
9eac1904 | 5949 | goto err; |
3af73b28 | 5950 | } |
b5dba59e JA |
5951 | |
5952 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 5953 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 5954 | spin_unlock(&files->file_lock); |
0bf0eefd | 5955 | return -EAGAIN; |
a2100672 | 5956 | } |
b5dba59e | 5957 | |
6319194e | 5958 | file = __close_fd_get_file(close->fd); |
9eac1904 | 5959 | spin_unlock(&files->file_lock); |
6319194e | 5960 | if (!file) |
9eac1904 | 5961 | goto err; |
9eac1904 | 5962 | |
3af73b28 | 5963 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
5964 | ret = filp_close(file, current->files); |
5965 | err: | |
3af73b28 | 5966 | if (ret < 0) |
93d2bcd2 | 5967 | req_set_fail(req); |
889fca73 | 5968 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 5969 | return 0; |
b5dba59e JA |
5970 | } |
5971 | ||
1155c76a | 5972 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 | 5973 | { |
73911426 | 5974 | if (unlikely(sqe->addr || sqe->buf_index || sqe->splice_fd_in)) |
5d17b4a4 JA |
5975 | return -EINVAL; |
5976 | ||
8ed8d3c3 JA |
5977 | req->sync.off = READ_ONCE(sqe->off); |
5978 | req->sync.len = READ_ONCE(sqe->len); | |
5979 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
5980 | return 0; |
5981 | } | |
5982 | ||
45d189c6 | 5983 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 5984 | { |
8ed8d3c3 JA |
5985 | int ret; |
5986 | ||
ac45abc0 | 5987 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 5988 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
5989 | return -EAGAIN; |
5990 | ||
9adbd45d | 5991 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 | 5992 | req->sync.flags); |
e1e16097 | 5993 | io_req_complete(req, ret); |
5d17b4a4 JA |
5994 | return 0; |
5995 | } | |
5996 | ||
469956e8 | 5997 | #if defined(CONFIG_NET) |
1151a7cc JA |
5998 | static int io_shutdown_prep(struct io_kiocb *req, |
5999 | const struct io_uring_sqe *sqe) | |
6000 | { | |
6001 | if (unlikely(sqe->off || sqe->addr || sqe->rw_flags || | |
6002 | sqe->buf_index || sqe->splice_fd_in)) | |
6003 | return -EINVAL; | |
6004 | ||
6005 | req->shutdown.how = READ_ONCE(sqe->len); | |
6006 | return 0; | |
6007 | } | |
6008 | ||
6009 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) | |
6010 | { | |
6011 | struct socket *sock; | |
6012 | int ret; | |
6013 | ||
6014 | if (issue_flags & IO_URING_F_NONBLOCK) | |
6015 | return -EAGAIN; | |
6016 | ||
6017 | sock = sock_from_file(req->file); | |
6018 | if (unlikely(!sock)) | |
6019 | return -ENOTSOCK; | |
6020 | ||
6021 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
6022 | io_req_complete(req, ret); | |
6023 | return 0; | |
6024 | } | |
6025 | ||
4c3c0943 JA |
6026 | static bool io_net_retry(struct socket *sock, int flags) |
6027 | { | |
6028 | if (!(flags & MSG_WAITALL)) | |
6029 | return false; | |
6030 | return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET; | |
6031 | } | |
6032 | ||
02d27d89 PB |
6033 | static int io_setup_async_msg(struct io_kiocb *req, |
6034 | struct io_async_msghdr *kmsg) | |
6035 | { | |
e8c2bc1f JA |
6036 | struct io_async_msghdr *async_msg = req->async_data; |
6037 | ||
6038 | if (async_msg) | |
02d27d89 | 6039 | return -EAGAIN; |
e8c2bc1f | 6040 | if (io_alloc_async_data(req)) { |
257e84a5 | 6041 | kfree(kmsg->free_iov); |
02d27d89 PB |
6042 | return -ENOMEM; |
6043 | } | |
e8c2bc1f | 6044 | async_msg = req->async_data; |
02d27d89 | 6045 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 6046 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 6047 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
6048 | /* if were using fast_iov, set it to the new one */ |
6049 | if (!async_msg->free_iov) | |
6050 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
6051 | ||
02d27d89 PB |
6052 | return -EAGAIN; |
6053 | } | |
6054 | ||
2ae523ed PB |
6055 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
6056 | struct io_async_msghdr *iomsg) | |
6057 | { | |
2ae523ed | 6058 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 6059 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 6060 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 6061 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
6062 | } |
6063 | ||
93642ef8 PB |
6064 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
6065 | { | |
6066 | int ret; | |
6067 | ||
93642ef8 PB |
6068 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
6069 | if (!ret) | |
6070 | req->flags |= REQ_F_NEED_CLEANUP; | |
6071 | return ret; | |
6072 | } | |
6073 | ||
3529d8c2 | 6074 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 6075 | { |
e47293fd | 6076 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 6077 | |
0455d4cc | 6078 | if (unlikely(sqe->file_index)) |
d2b6f48b | 6079 | return -EINVAL; |
588faa1e JA |
6080 | if (unlikely(sqe->addr2 || sqe->file_index)) |
6081 | return -EINVAL; | |
d2b6f48b | 6082 | |
270a5940 | 6083 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 6084 | sr->len = READ_ONCE(sqe->len); |
0455d4cc JA |
6085 | sr->flags = READ_ONCE(sqe->addr2); |
6086 | if (sr->flags & ~IORING_RECVSEND_POLL_FIRST) | |
6087 | return -EINVAL; | |
04411806 PB |
6088 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
6089 | if (sr->msg_flags & MSG_DONTWAIT) | |
6090 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 6091 | |
d8768362 JA |
6092 | #ifdef CONFIG_COMPAT |
6093 | if (req->ctx->compat) | |
6094 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
6095 | #endif | |
4c3c0943 | 6096 | sr->done_io = 0; |
93642ef8 | 6097 | return 0; |
03b1230c JA |
6098 | } |
6099 | ||
889fca73 | 6100 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 6101 | { |
6b754c8b | 6102 | struct io_async_msghdr iomsg, *kmsg; |
4c3c0943 | 6103 | struct io_sr_msg *sr = &req->sr_msg; |
0fa03c62 | 6104 | struct socket *sock; |
7a7cacba | 6105 | unsigned flags; |
0031275d | 6106 | int min_ret = 0; |
0fa03c62 JA |
6107 | int ret; |
6108 | ||
dba4a925 | 6109 | sock = sock_from_file(req->file); |
7a7cacba | 6110 | if (unlikely(!sock)) |
dba4a925 | 6111 | return -ENOTSOCK; |
3529d8c2 | 6112 | |
d886e185 PB |
6113 | if (req_has_async_data(req)) { |
6114 | kmsg = req->async_data; | |
6115 | } else { | |
7a7cacba PB |
6116 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
6117 | if (ret) | |
6118 | return ret; | |
6119 | kmsg = &iomsg; | |
0fa03c62 | 6120 | } |
0fa03c62 | 6121 | |
0455d4cc JA |
6122 | if (!(req->flags & REQ_F_POLLED) && |
6123 | (sr->flags & IORING_RECVSEND_POLL_FIRST)) | |
6124 | return io_setup_async_msg(req, kmsg); | |
6125 | ||
0a352aaa | 6126 | flags = sr->msg_flags; |
04411806 | 6127 | if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 6128 | flags |= MSG_DONTWAIT; |
0031275d SM |
6129 | if (flags & MSG_WAITALL) |
6130 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
6131 | ||
7a7cacba | 6132 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
0fa03c62 | 6133 | |
7297ce3d PB |
6134 | if (ret < min_ret) { |
6135 | if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) | |
6136 | return io_setup_async_msg(req, kmsg); | |
6137 | if (ret == -ERESTARTSYS) | |
6138 | ret = -EINTR; | |
4c3c0943 JA |
6139 | if (ret > 0 && io_net_retry(sock, flags)) { |
6140 | sr->done_io += ret; | |
6141 | req->flags |= REQ_F_PARTIAL_IO; | |
6142 | return io_setup_async_msg(req, kmsg); | |
6143 | } | |
7297ce3d PB |
6144 | req_set_fail(req); |
6145 | } | |
257e84a5 PB |
6146 | /* fast path, check for non-NULL to avoid function call */ |
6147 | if (kmsg->free_iov) | |
6148 | kfree(kmsg->free_iov); | |
99bc4c38 | 6149 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4c3c0943 JA |
6150 | if (ret >= 0) |
6151 | ret += sr->done_io; | |
6152 | else if (sr->done_io) | |
6153 | ret = sr->done_io; | |
889fca73 | 6154 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 6155 | return 0; |
03b1230c | 6156 | } |
aa1fa28f | 6157 | |
889fca73 | 6158 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 6159 | { |
7a7cacba PB |
6160 | struct io_sr_msg *sr = &req->sr_msg; |
6161 | struct msghdr msg; | |
6162 | struct iovec iov; | |
fddaface | 6163 | struct socket *sock; |
7a7cacba | 6164 | unsigned flags; |
0031275d | 6165 | int min_ret = 0; |
fddaface JA |
6166 | int ret; |
6167 | ||
0455d4cc JA |
6168 | if (!(req->flags & REQ_F_POLLED) && |
6169 | (sr->flags & IORING_RECVSEND_POLL_FIRST)) | |
6170 | return -EAGAIN; | |
6171 | ||
dba4a925 | 6172 | sock = sock_from_file(req->file); |
7a7cacba | 6173 | if (unlikely(!sock)) |
dba4a925 | 6174 | return -ENOTSOCK; |
fddaface | 6175 | |
7a7cacba PB |
6176 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
6177 | if (unlikely(ret)) | |
14db8411 | 6178 | return ret; |
fddaface | 6179 | |
7a7cacba PB |
6180 | msg.msg_name = NULL; |
6181 | msg.msg_control = NULL; | |
6182 | msg.msg_controllen = 0; | |
6183 | msg.msg_namelen = 0; | |
fddaface | 6184 | |
0a352aaa | 6185 | flags = sr->msg_flags; |
04411806 | 6186 | if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 6187 | flags |= MSG_DONTWAIT; |
0031275d SM |
6188 | if (flags & MSG_WAITALL) |
6189 | min_ret = iov_iter_count(&msg.msg_iter); | |
6190 | ||
7a7cacba PB |
6191 | msg.msg_flags = flags; |
6192 | ret = sock_sendmsg(sock, &msg); | |
7297ce3d PB |
6193 | if (ret < min_ret) { |
6194 | if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) | |
6195 | return -EAGAIN; | |
6196 | if (ret == -ERESTARTSYS) | |
6197 | ret = -EINTR; | |
4c3c0943 JA |
6198 | if (ret > 0 && io_net_retry(sock, flags)) { |
6199 | sr->len -= ret; | |
6200 | sr->buf += ret; | |
6201 | sr->done_io += ret; | |
6202 | req->flags |= REQ_F_PARTIAL_IO; | |
6203 | return -EAGAIN; | |
6204 | } | |
93d2bcd2 | 6205 | req_set_fail(req); |
7297ce3d | 6206 | } |
4c3c0943 JA |
6207 | if (ret >= 0) |
6208 | ret += sr->done_io; | |
6209 | else if (sr->done_io) | |
6210 | ret = sr->done_io; | |
889fca73 | 6211 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 6212 | return 0; |
fddaface JA |
6213 | } |
6214 | ||
1400e697 PB |
6215 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
6216 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
6217 | { |
6218 | struct io_sr_msg *sr = &req->sr_msg; | |
6219 | struct iovec __user *uiov; | |
6220 | size_t iov_len; | |
6221 | int ret; | |
6222 | ||
1400e697 PB |
6223 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
6224 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
6225 | if (ret) |
6226 | return ret; | |
6227 | ||
6228 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
6229 | if (iov_len > 1) | |
6230 | return -EINVAL; | |
5476dfed | 6231 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 6232 | return -EFAULT; |
5476dfed | 6233 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 6234 | iomsg->free_iov = NULL; |
52de1fe1 | 6235 | } else { |
257e84a5 | 6236 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 6237 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 6238 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 6239 | false); |
52de1fe1 JA |
6240 | if (ret > 0) |
6241 | ret = 0; | |
6242 | } | |
6243 | ||
6244 | return ret; | |
6245 | } | |
6246 | ||
6247 | #ifdef CONFIG_COMPAT | |
6248 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 6249 | struct io_async_msghdr *iomsg) |
52de1fe1 | 6250 | { |
52de1fe1 JA |
6251 | struct io_sr_msg *sr = &req->sr_msg; |
6252 | struct compat_iovec __user *uiov; | |
6253 | compat_uptr_t ptr; | |
6254 | compat_size_t len; | |
6255 | int ret; | |
6256 | ||
4af3417a PB |
6257 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
6258 | &ptr, &len); | |
52de1fe1 JA |
6259 | if (ret) |
6260 | return ret; | |
6261 | ||
6262 | uiov = compat_ptr(ptr); | |
6263 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
6264 | compat_ssize_t clen; | |
6265 | ||
6266 | if (len > 1) | |
6267 | return -EINVAL; | |
6268 | if (!access_ok(uiov, sizeof(*uiov))) | |
6269 | return -EFAULT; | |
6270 | if (__get_user(clen, &uiov->iov_len)) | |
6271 | return -EFAULT; | |
6272 | if (clen < 0) | |
6273 | return -EINVAL; | |
2d280bc8 | 6274 | sr->len = clen; |
257e84a5 | 6275 | iomsg->free_iov = NULL; |
52de1fe1 | 6276 | } else { |
257e84a5 | 6277 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 6278 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 6279 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 6280 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
6281 | if (ret < 0) |
6282 | return ret; | |
6283 | } | |
6284 | ||
6285 | return 0; | |
6286 | } | |
6287 | #endif | |
6288 | ||
1400e697 PB |
6289 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
6290 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 6291 | { |
1400e697 | 6292 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
6293 | |
6294 | #ifdef CONFIG_COMPAT | |
6295 | if (req->ctx->compat) | |
1400e697 | 6296 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 6297 | #endif |
52de1fe1 | 6298 | |
1400e697 | 6299 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
6300 | } |
6301 | ||
93642ef8 | 6302 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 6303 | { |
99bc4c38 | 6304 | int ret; |
3529d8c2 | 6305 | |
93642ef8 PB |
6306 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
6307 | if (!ret) | |
6308 | req->flags |= REQ_F_NEED_CLEANUP; | |
6309 | return ret; | |
6310 | } | |
6311 | ||
6312 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
6313 | { | |
6314 | struct io_sr_msg *sr = &req->sr_msg; | |
6315 | ||
0455d4cc | 6316 | if (unlikely(sqe->file_index)) |
d2b6f48b | 6317 | return -EINVAL; |
5a1e99b6 JA |
6318 | if (unlikely(sqe->addr2 || sqe->file_index)) |
6319 | return -EINVAL; | |
d2b6f48b | 6320 | |
270a5940 | 6321 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 6322 | sr->len = READ_ONCE(sqe->len); |
0455d4cc JA |
6323 | sr->flags = READ_ONCE(sqe->addr2); |
6324 | if (sr->flags & ~IORING_RECVSEND_POLL_FIRST) | |
6325 | return -EINVAL; | |
04411806 PB |
6326 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
6327 | if (sr->msg_flags & MSG_DONTWAIT) | |
6328 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 6329 | |
d8768362 JA |
6330 | #ifdef CONFIG_COMPAT |
6331 | if (req->ctx->compat) | |
6332 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
6333 | #endif | |
7ba89d2a | 6334 | sr->done_io = 0; |
93642ef8 | 6335 | return 0; |
aa1fa28f JA |
6336 | } |
6337 | ||
889fca73 | 6338 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 6339 | { |
6b754c8b | 6340 | struct io_async_msghdr iomsg, *kmsg; |
7ba89d2a | 6341 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 6342 | struct socket *sock; |
f548a12e | 6343 | unsigned int cflags; |
7a7cacba | 6344 | unsigned flags; |
d1fd1c20 | 6345 | int ret, min_ret = 0; |
45d189c6 | 6346 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 6347 | |
dba4a925 | 6348 | sock = sock_from_file(req->file); |
7a7cacba | 6349 | if (unlikely(!sock)) |
dba4a925 | 6350 | return -ENOTSOCK; |
3529d8c2 | 6351 | |
d886e185 PB |
6352 | if (req_has_async_data(req)) { |
6353 | kmsg = req->async_data; | |
6354 | } else { | |
7a7cacba PB |
6355 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
6356 | if (ret) | |
681fda8d | 6357 | return ret; |
7a7cacba PB |
6358 | kmsg = &iomsg; |
6359 | } | |
03b1230c | 6360 | |
0455d4cc JA |
6361 | if (!(req->flags & REQ_F_POLLED) && |
6362 | (sr->flags & IORING_RECVSEND_POLL_FIRST)) | |
6363 | return io_setup_async_msg(req, kmsg); | |
6364 | ||
b66e65f4 | 6365 | if (io_do_buffer_select(req)) { |
c54d52c2 JA |
6366 | void __user *buf; |
6367 | ||
4e906702 | 6368 | buf = io_buffer_select(req, &sr->len, issue_flags); |
984824db CH |
6369 | if (!buf) |
6370 | return -ENOBUFS; | |
c54d52c2 | 6371 | kmsg->fast_iov[0].iov_base = buf; |
0a352aaa JA |
6372 | kmsg->fast_iov[0].iov_len = sr->len; |
6373 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1, | |
6374 | sr->len); | |
7a7cacba | 6375 | } |
52de1fe1 | 6376 | |
0a352aaa | 6377 | flags = sr->msg_flags; |
04411806 | 6378 | if (force_nonblock) |
7a7cacba | 6379 | flags |= MSG_DONTWAIT; |
0031275d SM |
6380 | if (flags & MSG_WAITALL) |
6381 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
6382 | ||
f548a12e | 6383 | kmsg->msg.msg_get_inq = 1; |
0a352aaa | 6384 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg, kmsg->uaddr, flags); |
7297ce3d PB |
6385 | if (ret < min_ret) { |
6386 | if (ret == -EAGAIN && force_nonblock) | |
6387 | return io_setup_async_msg(req, kmsg); | |
6388 | if (ret == -ERESTARTSYS) | |
6389 | ret = -EINTR; | |
7ba89d2a JA |
6390 | if (ret > 0 && io_net_retry(sock, flags)) { |
6391 | sr->done_io += ret; | |
8a3e8ee5 | 6392 | req->flags |= REQ_F_PARTIAL_IO; |
7ba89d2a JA |
6393 | return io_setup_async_msg(req, kmsg); |
6394 | } | |
7297ce3d PB |
6395 | req_set_fail(req); |
6396 | } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) { | |
6397 | req_set_fail(req); | |
6398 | } | |
03b1230c | 6399 | |
257e84a5 PB |
6400 | /* fast path, check for non-NULL to avoid function call */ |
6401 | if (kmsg->free_iov) | |
6402 | kfree(kmsg->free_iov); | |
99bc4c38 | 6403 | req->flags &= ~REQ_F_NEED_CLEANUP; |
7ba89d2a JA |
6404 | if (ret >= 0) |
6405 | ret += sr->done_io; | |
6406 | else if (sr->done_io) | |
6407 | ret = sr->done_io; | |
f548a12e JA |
6408 | cflags = io_put_kbuf(req, issue_flags); |
6409 | if (kmsg->msg.msg_inq) | |
6410 | cflags |= IORING_CQE_F_SOCK_NONEMPTY; | |
6411 | __io_req_complete(req, issue_flags, ret, cflags); | |
03b1230c | 6412 | return 0; |
0fa03c62 | 6413 | } |
5d17b4a4 | 6414 | |
889fca73 | 6415 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 6416 | { |
7a7cacba PB |
6417 | struct io_sr_msg *sr = &req->sr_msg; |
6418 | struct msghdr msg; | |
fddaface | 6419 | struct socket *sock; |
7a7cacba | 6420 | struct iovec iov; |
f548a12e | 6421 | unsigned int cflags; |
7a7cacba | 6422 | unsigned flags; |
d1fd1c20 | 6423 | int ret, min_ret = 0; |
45d189c6 | 6424 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 6425 | |
0455d4cc JA |
6426 | if (!(req->flags & REQ_F_POLLED) && |
6427 | (sr->flags & IORING_RECVSEND_POLL_FIRST)) | |
6428 | return -EAGAIN; | |
6429 | ||
dba4a925 | 6430 | sock = sock_from_file(req->file); |
7a7cacba | 6431 | if (unlikely(!sock)) |
dba4a925 | 6432 | return -ENOTSOCK; |
fddaface | 6433 | |
b66e65f4 | 6434 | if (io_do_buffer_select(req)) { |
c54d52c2 JA |
6435 | void __user *buf; |
6436 | ||
4e906702 | 6437 | buf = io_buffer_select(req, &sr->len, issue_flags); |
984824db CH |
6438 | if (!buf) |
6439 | return -ENOBUFS; | |
c54d52c2 | 6440 | sr->buf = buf; |
bc02ef33 | 6441 | } |
bcda7baa | 6442 | |
c54d52c2 | 6443 | ret = import_single_range(READ, sr->buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
6444 | if (unlikely(ret)) |
6445 | goto out_free; | |
fddaface | 6446 | |
7a7cacba | 6447 | msg.msg_name = NULL; |
f548a12e | 6448 | msg.msg_namelen = 0; |
7a7cacba | 6449 | msg.msg_control = NULL; |
f548a12e JA |
6450 | msg.msg_get_inq = 1; |
6451 | msg.msg_flags = 0; | |
7a7cacba | 6452 | msg.msg_controllen = 0; |
7a7cacba | 6453 | msg.msg_iocb = NULL; |
fddaface | 6454 | |
0a352aaa | 6455 | flags = sr->msg_flags; |
04411806 | 6456 | if (force_nonblock) |
7a7cacba | 6457 | flags |= MSG_DONTWAIT; |
0031275d SM |
6458 | if (flags & MSG_WAITALL) |
6459 | min_ret = iov_iter_count(&msg.msg_iter); | |
6460 | ||
7a7cacba | 6461 | ret = sock_recvmsg(sock, &msg, flags); |
7297ce3d PB |
6462 | if (ret < min_ret) { |
6463 | if (ret == -EAGAIN && force_nonblock) | |
6464 | return -EAGAIN; | |
6465 | if (ret == -ERESTARTSYS) | |
6466 | ret = -EINTR; | |
7ba89d2a JA |
6467 | if (ret > 0 && io_net_retry(sock, flags)) { |
6468 | sr->len -= ret; | |
6469 | sr->buf += ret; | |
6470 | sr->done_io += ret; | |
8a3e8ee5 | 6471 | req->flags |= REQ_F_PARTIAL_IO; |
7ba89d2a JA |
6472 | return -EAGAIN; |
6473 | } | |
7297ce3d PB |
6474 | req_set_fail(req); |
6475 | } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) { | |
0d7c1153 | 6476 | out_free: |
93d2bcd2 | 6477 | req_set_fail(req); |
7297ce3d | 6478 | } |
cc3cec83 | 6479 | |
7ba89d2a JA |
6480 | if (ret >= 0) |
6481 | ret += sr->done_io; | |
6482 | else if (sr->done_io) | |
6483 | ret = sr->done_io; | |
f548a12e JA |
6484 | cflags = io_put_kbuf(req, issue_flags); |
6485 | if (msg.msg_inq) | |
6486 | cflags |= IORING_CQE_F_SOCK_NONEMPTY; | |
6487 | __io_req_complete(req, issue_flags, ret, cflags); | |
fddaface | 6488 | return 0; |
fddaface JA |
6489 | } |
6490 | ||
3529d8c2 | 6491 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 6492 | { |
8ed8d3c3 | 6493 | struct io_accept *accept = &req->accept; |
4e86a2c9 | 6494 | unsigned flags; |
8ed8d3c3 | 6495 | |
73911426 | 6496 | if (sqe->len || sqe->buf_index) |
17f2fe35 JA |
6497 | return -EINVAL; |
6498 | ||
d55e5f5b JA |
6499 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
6500 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 6501 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 6502 | accept->nofile = rlimit(RLIMIT_NOFILE); |
4e86a2c9 HX |
6503 | flags = READ_ONCE(sqe->ioprio); |
6504 | if (flags & ~IORING_ACCEPT_MULTISHOT) | |
6505 | return -EINVAL; | |
a7083ad5 | 6506 | |
aaa4db12 | 6507 | accept->file_slot = READ_ONCE(sqe->file_index); |
4e86a2c9 HX |
6508 | if (accept->file_slot) { |
6509 | if (accept->flags & SOCK_CLOEXEC) | |
6510 | return -EINVAL; | |
6511 | if (flags & IORING_ACCEPT_MULTISHOT && | |
6512 | accept->file_slot != IORING_FILE_INDEX_ALLOC) | |
6513 | return -EINVAL; | |
6514 | } | |
a7083ad5 PB |
6515 | if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) |
6516 | return -EINVAL; | |
6517 | if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) | |
6518 | accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; | |
4e86a2c9 HX |
6519 | if (flags & IORING_ACCEPT_MULTISHOT) |
6520 | req->flags |= REQ_F_APOLL_MULTISHOT; | |
8ed8d3c3 | 6521 | return 0; |
8ed8d3c3 | 6522 | } |
17f2fe35 | 6523 | |
889fca73 | 6524 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 6525 | { |
4e86a2c9 | 6526 | struct io_ring_ctx *ctx = req->ctx; |
8ed8d3c3 | 6527 | struct io_accept *accept = &req->accept; |
45d189c6 | 6528 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 6529 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
aaa4db12 | 6530 | bool fixed = !!accept->file_slot; |
a7083ad5 PB |
6531 | struct file *file; |
6532 | int ret, fd; | |
8ed8d3c3 | 6533 | |
4e86a2c9 | 6534 | retry: |
aaa4db12 PB |
6535 | if (!fixed) { |
6536 | fd = __get_unused_fd_flags(accept->flags, accept->nofile); | |
6537 | if (unlikely(fd < 0)) | |
6538 | return fd; | |
6539 | } | |
a7083ad5 PB |
6540 | file = do_accept(req->file, file_flags, accept->addr, accept->addr_len, |
6541 | accept->flags); | |
6542 | if (IS_ERR(file)) { | |
aaa4db12 PB |
6543 | if (!fixed) |
6544 | put_unused_fd(fd); | |
a7083ad5 | 6545 | ret = PTR_ERR(file); |
4e86a2c9 HX |
6546 | if (ret == -EAGAIN && force_nonblock) { |
6547 | /* | |
6548 | * if it's multishot and polled, we don't need to | |
6549 | * return EAGAIN to arm the poll infra since it | |
6550 | * has already been done | |
6551 | */ | |
6552 | if ((req->flags & IO_APOLL_MULTI_POLLED) == | |
6553 | IO_APOLL_MULTI_POLLED) | |
6554 | ret = 0; | |
6555 | return ret; | |
6556 | } | |
ac45abc0 PB |
6557 | if (ret == -ERESTARTSYS) |
6558 | ret = -EINTR; | |
93d2bcd2 | 6559 | req_set_fail(req); |
aaa4db12 | 6560 | } else if (!fixed) { |
a7083ad5 PB |
6561 | fd_install(fd, file); |
6562 | ret = fd; | |
aaa4db12 | 6563 | } else { |
c30c3e00 JA |
6564 | ret = io_fixed_fd_install(req, issue_flags, file, |
6565 | accept->file_slot); | |
ac45abc0 | 6566 | } |
4e86a2c9 HX |
6567 | |
6568 | if (!(req->flags & REQ_F_APOLL_MULTISHOT)) { | |
6569 | __io_req_complete(req, issue_flags, ret, 0); | |
6570 | return 0; | |
6571 | } | |
6572 | if (ret >= 0) { | |
6573 | bool filled; | |
6574 | ||
6575 | spin_lock(&ctx->completion_lock); | |
6576 | filled = io_fill_cqe_aux(ctx, req->cqe.user_data, ret, | |
6577 | IORING_CQE_F_MORE); | |
6578 | io_commit_cqring(ctx); | |
6579 | spin_unlock(&ctx->completion_lock); | |
6580 | if (filled) { | |
6581 | io_cqring_ev_posted(ctx); | |
6582 | goto retry; | |
6583 | } | |
6584 | ret = -ECANCELED; | |
6585 | } | |
6586 | ||
6587 | return ret; | |
8ed8d3c3 JA |
6588 | } |
6589 | ||
1374e08e JA |
6590 | static int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
6591 | { | |
6592 | struct io_socket *sock = &req->sock; | |
6593 | ||
ee692a21 | 6594 | if (sqe->addr || sqe->rw_flags || sqe->buf_index) |
1374e08e JA |
6595 | return -EINVAL; |
6596 | ||
6597 | sock->domain = READ_ONCE(sqe->fd); | |
6598 | sock->type = READ_ONCE(sqe->off); | |
6599 | sock->protocol = READ_ONCE(sqe->len); | |
6600 | sock->file_slot = READ_ONCE(sqe->file_index); | |
6601 | sock->nofile = rlimit(RLIMIT_NOFILE); | |
6602 | ||
6603 | sock->flags = sock->type & ~SOCK_TYPE_MASK; | |
6604 | if (sock->file_slot && (sock->flags & SOCK_CLOEXEC)) | |
6605 | return -EINVAL; | |
6606 | if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) | |
6607 | return -EINVAL; | |
6608 | return 0; | |
6609 | } | |
6610 | ||
6611 | static int io_socket(struct io_kiocb *req, unsigned int issue_flags) | |
6612 | { | |
6613 | struct io_socket *sock = &req->sock; | |
6614 | bool fixed = !!sock->file_slot; | |
6615 | struct file *file; | |
6616 | int ret, fd; | |
6617 | ||
6618 | if (!fixed) { | |
6619 | fd = __get_unused_fd_flags(sock->flags, sock->nofile); | |
6620 | if (unlikely(fd < 0)) | |
6621 | return fd; | |
6622 | } | |
6623 | file = __sys_socket_file(sock->domain, sock->type, sock->protocol); | |
6624 | if (IS_ERR(file)) { | |
6625 | if (!fixed) | |
6626 | put_unused_fd(fd); | |
6627 | ret = PTR_ERR(file); | |
6628 | if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK)) | |
6629 | return -EAGAIN; | |
6630 | if (ret == -ERESTARTSYS) | |
6631 | ret = -EINTR; | |
6632 | req_set_fail(req); | |
6633 | } else if (!fixed) { | |
6634 | fd_install(fd, file); | |
6635 | ret = fd; | |
6636 | } else { | |
fa82dd10 JA |
6637 | ret = io_fixed_fd_install(req, issue_flags, file, |
6638 | sock->file_slot); | |
1374e08e JA |
6639 | } |
6640 | __io_req_complete(req, issue_flags, ret, 0); | |
6641 | return 0; | |
6642 | } | |
6643 | ||
93642ef8 PB |
6644 | static int io_connect_prep_async(struct io_kiocb *req) |
6645 | { | |
6646 | struct io_async_connect *io = req->async_data; | |
6647 | struct io_connect *conn = &req->connect; | |
6648 | ||
6649 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
6650 | } | |
6651 | ||
3529d8c2 | 6652 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 6653 | { |
3529d8c2 | 6654 | struct io_connect *conn = &req->connect; |
f499a021 | 6655 | |
73911426 | 6656 | if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in) |
3fbb51c1 JA |
6657 | return -EINVAL; |
6658 | ||
3529d8c2 JA |
6659 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
6660 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 6661 | return 0; |
f499a021 JA |
6662 | } |
6663 | ||
889fca73 | 6664 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 6665 | { |
e8c2bc1f | 6666 | struct io_async_connect __io, *io; |
f8e85cf2 | 6667 | unsigned file_flags; |
3fbb51c1 | 6668 | int ret; |
45d189c6 | 6669 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 6670 | |
d886e185 | 6671 | if (req_has_async_data(req)) { |
e8c2bc1f | 6672 | io = req->async_data; |
f499a021 | 6673 | } else { |
3529d8c2 JA |
6674 | ret = move_addr_to_kernel(req->connect.addr, |
6675 | req->connect.addr_len, | |
e8c2bc1f | 6676 | &__io.address); |
f499a021 JA |
6677 | if (ret) |
6678 | goto out; | |
6679 | io = &__io; | |
6680 | } | |
6681 | ||
3fbb51c1 JA |
6682 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
6683 | ||
e8c2bc1f | 6684 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 6685 | req->connect.addr_len, file_flags); |
87f80d62 | 6686 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
d886e185 | 6687 | if (req_has_async_data(req)) |
b7bb4f7d | 6688 | return -EAGAIN; |
e8c2bc1f | 6689 | if (io_alloc_async_data(req)) { |
f499a021 JA |
6690 | ret = -ENOMEM; |
6691 | goto out; | |
6692 | } | |
e8c2bc1f | 6693 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 6694 | return -EAGAIN; |
f499a021 | 6695 | } |
f8e85cf2 JA |
6696 | if (ret == -ERESTARTSYS) |
6697 | ret = -EINTR; | |
f499a021 | 6698 | out: |
4e88d6e7 | 6699 | if (ret < 0) |
93d2bcd2 | 6700 | req_set_fail(req); |
889fca73 | 6701 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 6702 | return 0; |
469956e8 Y |
6703 | } |
6704 | #else /* !CONFIG_NET */ | |
99a10081 JA |
6705 | #define IO_NETOP_FN(op) \ |
6706 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
6707 | { \ | |
6708 | return -EOPNOTSUPP; \ | |
6709 | } | |
6710 | ||
6711 | #define IO_NETOP_PREP(op) \ | |
6712 | IO_NETOP_FN(op) \ | |
6713 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
6714 | { \ | |
6715 | return -EOPNOTSUPP; \ | |
6716 | } \ | |
6717 | ||
6718 | #define IO_NETOP_PREP_ASYNC(op) \ | |
6719 | IO_NETOP_PREP(op) \ | |
6720 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
6721 | { \ | |
6722 | return -EOPNOTSUPP; \ | |
6723 | } | |
6724 | ||
6725 | IO_NETOP_PREP_ASYNC(sendmsg); | |
6726 | IO_NETOP_PREP_ASYNC(recvmsg); | |
6727 | IO_NETOP_PREP_ASYNC(connect); | |
6728 | IO_NETOP_PREP(accept); | |
1374e08e | 6729 | IO_NETOP_PREP(socket); |
1151a7cc | 6730 | IO_NETOP_PREP(shutdown); |
99a10081 JA |
6731 | IO_NETOP_FN(send); |
6732 | IO_NETOP_FN(recv); | |
469956e8 | 6733 | #endif /* CONFIG_NET */ |
f8e85cf2 | 6734 | |
d7718a9d JA |
6735 | struct io_poll_table { |
6736 | struct poll_table_struct pt; | |
6737 | struct io_kiocb *req; | |
68b11e8b | 6738 | int nr_entries; |
d7718a9d JA |
6739 | int error; |
6740 | }; | |
ce593a6c | 6741 | |
aa43477b | 6742 | #define IO_POLL_CANCEL_FLAG BIT(31) |
e2c0cb7c | 6743 | #define IO_POLL_REF_MASK GENMASK(30, 0) |
6d816e08 | 6744 | |
aa43477b PB |
6745 | /* |
6746 | * If refs part of ->poll_refs (see IO_POLL_REF_MASK) is 0, it's free. We can | |
6747 | * bump it and acquire ownership. It's disallowed to modify requests while not | |
6748 | * owning it, that prevents from races for enqueueing task_work's and b/w | |
6749 | * arming poll and wakeups. | |
6750 | */ | |
6751 | static inline bool io_poll_get_ownership(struct io_kiocb *req) | |
6752 | { | |
6753 | return !(atomic_fetch_inc(&req->poll_refs) & IO_POLL_REF_MASK); | |
d7718a9d JA |
6754 | } |
6755 | ||
aa43477b | 6756 | static void io_poll_mark_cancelled(struct io_kiocb *req) |
74ce6ce4 | 6757 | { |
aa43477b | 6758 | atomic_or(IO_POLL_CANCEL_FLAG, &req->poll_refs); |
74ce6ce4 JA |
6759 | } |
6760 | ||
d4e7cd36 | 6761 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 6762 | { |
e8c2bc1f | 6763 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 6764 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 6765 | return req->async_data; |
d4e7cd36 JA |
6766 | return req->apoll->double_poll; |
6767 | } | |
6768 | ||
6769 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
6770 | { | |
6771 | if (req->opcode == IORING_OP_POLL_ADD) | |
6772 | return &req->poll; | |
6773 | return &req->apoll->poll; | |
6774 | } | |
6775 | ||
5641897a | 6776 | static void io_poll_req_insert(struct io_kiocb *req) |
d4e7cd36 | 6777 | { |
5641897a PB |
6778 | struct io_ring_ctx *ctx = req->ctx; |
6779 | struct hlist_head *list; | |
18bceab1 | 6780 | |
cef216fc | 6781 | list = &ctx->cancel_hash[hash_long(req->cqe.user_data, ctx->cancel_hash_bits)]; |
5641897a | 6782 | hlist_add_head(&req->hash_node, list); |
18bceab1 JA |
6783 | } |
6784 | ||
5641897a PB |
6785 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, |
6786 | wait_queue_func_t wake_func) | |
18bceab1 | 6787 | { |
5641897a | 6788 | poll->head = NULL; |
5641897a PB |
6789 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
6790 | /* mask in events that we always want/need */ | |
6791 | poll->events = events | IO_POLL_UNMASK; | |
6792 | INIT_LIST_HEAD(&poll->wait.entry); | |
6793 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
18bceab1 JA |
6794 | } |
6795 | ||
aa43477b | 6796 | static inline void io_poll_remove_entry(struct io_poll_iocb *poll) |
18bceab1 | 6797 | { |
791f3465 | 6798 | struct wait_queue_head *head = smp_load_acquire(&poll->head); |
18bceab1 | 6799 | |
791f3465 PB |
6800 | if (head) { |
6801 | spin_lock_irq(&head->lock); | |
6802 | list_del_init(&poll->wait.entry); | |
6803 | poll->head = NULL; | |
6804 | spin_unlock_irq(&head->lock); | |
6805 | } | |
aa43477b | 6806 | } |
18bceab1 | 6807 | |
aa43477b PB |
6808 | static void io_poll_remove_entries(struct io_kiocb *req) |
6809 | { | |
91eac1c6 JA |
6810 | /* |
6811 | * Nothing to do if neither of those flags are set. Avoid dipping | |
6812 | * into the poll/apoll/double cachelines if we can. | |
6813 | */ | |
6814 | if (!(req->flags & (REQ_F_SINGLE_POLL | REQ_F_DOUBLE_POLL))) | |
6815 | return; | |
18bceab1 | 6816 | |
791f3465 PB |
6817 | /* |
6818 | * While we hold the waitqueue lock and the waitqueue is nonempty, | |
6819 | * wake_up_pollfree() will wait for us. However, taking the waitqueue | |
6820 | * lock in the first place can race with the waitqueue being freed. | |
6821 | * | |
6822 | * We solve this as eventpoll does: by taking advantage of the fact that | |
6823 | * all users of wake_up_pollfree() will RCU-delay the actual free. If | |
6824 | * we enter rcu_read_lock() and see that the pointer to the queue is | |
6825 | * non-NULL, we can then lock it without the memory being freed out from | |
6826 | * under us. | |
6827 | * | |
6828 | * Keep holding rcu_read_lock() as long as we hold the queue lock, in | |
6829 | * case the caller deletes the entry from the queue, leaving it empty. | |
6830 | * In that case, only RCU prevents the queue memory from being freed. | |
6831 | */ | |
6832 | rcu_read_lock(); | |
91eac1c6 JA |
6833 | if (req->flags & REQ_F_SINGLE_POLL) |
6834 | io_poll_remove_entry(io_poll_get_single(req)); | |
6835 | if (req->flags & REQ_F_DOUBLE_POLL) | |
6836 | io_poll_remove_entry(io_poll_get_double(req)); | |
791f3465 | 6837 | rcu_read_unlock(); |
18bceab1 JA |
6838 | } |
6839 | ||
dbc2564c | 6840 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags); |
aa43477b PB |
6841 | /* |
6842 | * All poll tw should go through this. Checks for poll events, manages | |
6843 | * references, does rewait, etc. | |
6844 | * | |
6845 | * Returns a negative error on failure. >0 when no action require, which is | |
6846 | * either spurious wakeup or multishot CQE is served. 0 when it's done with | |
cef216fc | 6847 | * the request, then the mask is stored in req->cqe.res. |
aa43477b | 6848 | */ |
dbc2564c | 6849 | static int io_poll_check_events(struct io_kiocb *req, bool *locked) |
18bceab1 | 6850 | { |
74ce6ce4 | 6851 | struct io_ring_ctx *ctx = req->ctx; |
dbc2564c | 6852 | int v, ret; |
18bceab1 | 6853 | |
316319e8 | 6854 | /* req->task == current here, checking PF_EXITING is safe */ |
e09ee510 | 6855 | if (unlikely(req->task->flags & PF_EXITING)) |
f2219057 | 6856 | return -ECANCELED; |
18bceab1 | 6857 | |
aa43477b PB |
6858 | do { |
6859 | v = atomic_read(&req->poll_refs); | |
74ce6ce4 | 6860 | |
aa43477b PB |
6861 | /* tw handler should be the owner, and so have some references */ |
6862 | if (WARN_ON_ONCE(!(v & IO_POLL_REF_MASK))) | |
6863 | return 0; | |
6864 | if (v & IO_POLL_CANCEL_FLAG) | |
6865 | return -ECANCELED; | |
8706e04e | 6866 | |
cef216fc | 6867 | if (!req->cqe.res) { |
2804ecd8 | 6868 | struct poll_table_struct pt = { ._key = req->apoll_events }; |
cef216fc | 6869 | req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events; |
c8b5e260 | 6870 | } |
74ce6ce4 | 6871 | |
dbc2564c HX |
6872 | if ((unlikely(!req->cqe.res))) |
6873 | continue; | |
6874 | if (req->apoll_events & EPOLLONESHOT) | |
6875 | return 0; | |
6876 | ||
6877 | /* multishot, just fill a CQE and proceed */ | |
6878 | if (!(req->flags & REQ_F_APOLL_MULTISHOT)) { | |
6879 | __poll_t mask = mangle_poll(req->cqe.res & | |
6880 | req->apoll_events); | |
aa43477b | 6881 | bool filled; |
18bceab1 | 6882 | |
aa43477b | 6883 | spin_lock(&ctx->completion_lock); |
dbc2564c HX |
6884 | filled = io_fill_cqe_aux(ctx, req->cqe.user_data, |
6885 | mask, IORING_CQE_F_MORE); | |
aa43477b PB |
6886 | io_commit_cqring(ctx); |
6887 | spin_unlock(&ctx->completion_lock); | |
dbc2564c HX |
6888 | if (filled) { |
6889 | io_cqring_ev_posted(ctx); | |
6890 | continue; | |
6891 | } | |
6892 | return -ECANCELED; | |
aa43477b | 6893 | } |
18bceab1 | 6894 | |
dbc2564c HX |
6895 | io_tw_lock(req->ctx, locked); |
6896 | if (unlikely(req->task->flags & PF_EXITING)) | |
6897 | return -EFAULT; | |
6898 | ret = io_issue_sqe(req, | |
6899 | IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); | |
6900 | if (ret) | |
6901 | return ret; | |
6902 | ||
aa43477b PB |
6903 | /* |
6904 | * Release all references, retry if someone tried to restart | |
6905 | * task_work while we were executing it. | |
6906 | */ | |
6907 | } while (atomic_sub_return(v & IO_POLL_REF_MASK, &req->poll_refs)); | |
18bceab1 | 6908 | |
18bceab1 JA |
6909 | return 1; |
6910 | } | |
6911 | ||
aa43477b | 6912 | static void io_poll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 | 6913 | { |
18bceab1 | 6914 | struct io_ring_ctx *ctx = req->ctx; |
aa43477b | 6915 | int ret; |
18bceab1 | 6916 | |
dbc2564c | 6917 | ret = io_poll_check_events(req, locked); |
aa43477b PB |
6918 | if (ret > 0) |
6919 | return; | |
6920 | ||
6921 | if (!ret) { | |
cef216fc | 6922 | req->cqe.res = mangle_poll(req->cqe.res & req->poll.events); |
e27414be | 6923 | } else { |
cef216fc | 6924 | req->cqe.res = ret; |
aa43477b | 6925 | req_set_fail(req); |
a62682f9 | 6926 | } |
aa43477b PB |
6927 | |
6928 | io_poll_remove_entries(req); | |
6929 | spin_lock(&ctx->completion_lock); | |
6930 | hash_del(&req->hash_node); | |
cef216fc | 6931 | __io_req_complete_post(req, req->cqe.res, 0); |
aa43477b PB |
6932 | io_commit_cqring(ctx); |
6933 | spin_unlock(&ctx->completion_lock); | |
6934 | io_cqring_ev_posted(ctx); | |
18bceab1 JA |
6935 | } |
6936 | ||
aa43477b | 6937 | static void io_apoll_task_func(struct io_kiocb *req, bool *locked) |
18bceab1 JA |
6938 | { |
6939 | struct io_ring_ctx *ctx = req->ctx; | |
aa43477b | 6940 | int ret; |
18bceab1 | 6941 | |
dbc2564c | 6942 | ret = io_poll_check_events(req, locked); |
aa43477b PB |
6943 | if (ret > 0) |
6944 | return; | |
18bceab1 | 6945 | |
aa43477b PB |
6946 | io_poll_remove_entries(req); |
6947 | spin_lock(&ctx->completion_lock); | |
6948 | hash_del(&req->hash_node); | |
6949 | spin_unlock(&ctx->completion_lock); | |
18bceab1 | 6950 | |
aa43477b PB |
6951 | if (!ret) |
6952 | io_req_task_submit(req, locked); | |
6953 | else | |
6954 | io_req_complete_failed(req, ret); | |
18bceab1 JA |
6955 | } |
6956 | ||
58f5c8d3 | 6957 | static void __io_poll_execute(struct io_kiocb *req, int mask, __poll_t events) |
aa43477b | 6958 | { |
cef216fc | 6959 | req->cqe.res = mask; |
81459350 JA |
6960 | /* |
6961 | * This is useful for poll that is armed on behalf of another | |
6962 | * request, and where the wakeup path could be on a different | |
6963 | * CPU. We want to avoid pulling in req->apoll->events for that | |
6964 | * case. | |
6965 | */ | |
2804ecd8 | 6966 | req->apoll_events = events; |
aa43477b PB |
6967 | if (req->opcode == IORING_OP_POLL_ADD) |
6968 | req->io_task_work.func = io_poll_task_func; | |
6969 | else | |
6970 | req->io_task_work.func = io_apoll_task_func; | |
6971 | ||
cef216fc | 6972 | trace_io_uring_task_add(req->ctx, req, req->cqe.user_data, req->opcode, mask); |
3fe07bcd | 6973 | io_req_task_work_add(req); |
aa43477b PB |
6974 | } |
6975 | ||
58f5c8d3 CH |
6976 | static inline void io_poll_execute(struct io_kiocb *req, int res, |
6977 | __poll_t events) | |
aa43477b PB |
6978 | { |
6979 | if (io_poll_get_ownership(req)) | |
81459350 | 6980 | __io_poll_execute(req, res, events); |
aa43477b PB |
6981 | } |
6982 | ||
6983 | static void io_poll_cancel_req(struct io_kiocb *req) | |
6984 | { | |
6985 | io_poll_mark_cancelled(req); | |
6986 | /* kick tw, which should complete the request */ | |
81459350 | 6987 | io_poll_execute(req, 0, 0); |
aa43477b PB |
6988 | } |
6989 | ||
d89a4fac JA |
6990 | #define wqe_to_req(wait) ((void *)((unsigned long) (wait)->private & ~1)) |
6991 | #define wqe_is_double(wait) ((unsigned long) (wait)->private & 1) | |
a294bef5 | 6992 | #define IO_ASYNC_POLL_COMMON (EPOLLONESHOT | EPOLLPRI) |
d89a4fac | 6993 | |
aa43477b PB |
6994 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
6995 | void *key) | |
18bceab1 | 6996 | { |
d89a4fac | 6997 | struct io_kiocb *req = wqe_to_req(wait); |
aa43477b PB |
6998 | struct io_poll_iocb *poll = container_of(wait, struct io_poll_iocb, |
6999 | wait); | |
18bceab1 JA |
7000 | __poll_t mask = key_to_poll(key); |
7001 | ||
791f3465 PB |
7002 | if (unlikely(mask & POLLFREE)) { |
7003 | io_poll_mark_cancelled(req); | |
7004 | /* we have to kick tw in case it's not already */ | |
81459350 | 7005 | io_poll_execute(req, 0, poll->events); |
791f3465 PB |
7006 | |
7007 | /* | |
7008 | * If the waitqueue is being freed early but someone is already | |
7009 | * holds ownership over it, we have to tear down the request as | |
7010 | * best we can. That means immediately removing the request from | |
7011 | * its waitqueue and preventing all further accesses to the | |
7012 | * waitqueue via the request. | |
7013 | */ | |
7014 | list_del_init(&poll->wait.entry); | |
7015 | ||
7016 | /* | |
7017 | * Careful: this *must* be the last step, since as soon | |
7018 | * as req->head is NULL'ed out, the request can be | |
7019 | * completed and freed, since aio_poll_complete_work() | |
7020 | * will no longer need to take the waitqueue lock. | |
7021 | */ | |
7022 | smp_store_release(&poll->head, NULL); | |
7023 | return 1; | |
7024 | } | |
7025 | ||
aa43477b | 7026 | /* for instances that support it check for an event match first */ |
1b1d7b4b | 7027 | if (mask && !(mask & (poll->events & ~IO_ASYNC_POLL_COMMON))) |
18bceab1 JA |
7028 | return 0; |
7029 | ||
eb0089d6 PB |
7030 | if (io_poll_get_ownership(req)) { |
7031 | /* optional, saves extra locking for removal in tw handler */ | |
7032 | if (mask && poll->events & EPOLLONESHOT) { | |
7033 | list_del_init(&poll->wait.entry); | |
7034 | poll->head = NULL; | |
d89a4fac JA |
7035 | if (wqe_is_double(wait)) |
7036 | req->flags &= ~REQ_F_DOUBLE_POLL; | |
7037 | else | |
7038 | req->flags &= ~REQ_F_SINGLE_POLL; | |
eb0089d6 | 7039 | } |
81459350 | 7040 | __io_poll_execute(req, mask, poll->events); |
eb0089d6 | 7041 | } |
18bceab1 | 7042 | return 1; |
18bceab1 JA |
7043 | } |
7044 | ||
7045 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
7046 | struct wait_queue_head *head, |
7047 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
7048 | { |
7049 | struct io_kiocb *req = pt->req; | |
d89a4fac | 7050 | unsigned long wqe_private = (unsigned long) req; |
18bceab1 JA |
7051 | |
7052 | /* | |
68b11e8b PB |
7053 | * The file being polled uses multiple waitqueues for poll handling |
7054 | * (e.g. one for read, one for write). Setup a separate io_poll_iocb | |
7055 | * if this happens. | |
18bceab1 | 7056 | */ |
68b11e8b | 7057 | if (unlikely(pt->nr_entries)) { |
aa43477b | 7058 | struct io_poll_iocb *first = poll; |
58852d4d | 7059 | |
23a65db8 | 7060 | /* double add on the same waitqueue head, ignore */ |
aa43477b | 7061 | if (first->head == head) |
23a65db8 | 7062 | return; |
18bceab1 | 7063 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 7064 | if (*poll_ptr) { |
23a65db8 PB |
7065 | if ((*poll_ptr)->head == head) |
7066 | return; | |
18bceab1 JA |
7067 | pt->error = -EINVAL; |
7068 | return; | |
7069 | } | |
aa43477b | 7070 | |
18bceab1 JA |
7071 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
7072 | if (!poll) { | |
7073 | pt->error = -ENOMEM; | |
7074 | return; | |
7075 | } | |
d89a4fac JA |
7076 | /* mark as double wq entry */ |
7077 | wqe_private |= 1; | |
91eac1c6 | 7078 | req->flags |= REQ_F_DOUBLE_POLL; |
aa43477b | 7079 | io_init_poll_iocb(poll, first->events, first->wait.func); |
807abcb0 | 7080 | *poll_ptr = poll; |
d886e185 PB |
7081 | if (req->opcode == IORING_OP_POLL_ADD) |
7082 | req->flags |= REQ_F_ASYNC_DATA; | |
18bceab1 JA |
7083 | } |
7084 | ||
91eac1c6 | 7085 | req->flags |= REQ_F_SINGLE_POLL; |
68b11e8b | 7086 | pt->nr_entries++; |
18bceab1 | 7087 | poll->head = head; |
d89a4fac | 7088 | poll->wait.private = (void *) wqe_private; |
a31eb4a2 JX |
7089 | |
7090 | if (poll->events & EPOLLEXCLUSIVE) | |
7091 | add_wait_queue_exclusive(head, &poll->wait); | |
7092 | else | |
7093 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
7094 | } |
7095 | ||
aa43477b | 7096 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
18bceab1 JA |
7097 | struct poll_table_struct *p) |
7098 | { | |
7099 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
d7718a9d | 7100 | |
aa43477b PB |
7101 | __io_queue_proc(&pt->req->poll, pt, head, |
7102 | (struct io_poll_iocb **) &pt->req->async_data); | |
d7718a9d JA |
7103 | } |
7104 | ||
aa43477b PB |
7105 | static int __io_arm_poll_handler(struct io_kiocb *req, |
7106 | struct io_poll_iocb *poll, | |
7107 | struct io_poll_table *ipt, __poll_t mask) | |
d7718a9d JA |
7108 | { |
7109 | struct io_ring_ctx *ctx = req->ctx; | |
aa43477b | 7110 | int v; |
d7718a9d | 7111 | |
4d52f338 | 7112 | INIT_HLIST_NODE(&req->hash_node); |
8e29da69 | 7113 | req->work.cancel_seq = atomic_read(&ctx->cancel_seq); |
aa43477b | 7114 | io_init_poll_iocb(poll, mask, io_poll_wake); |
b90cd197 | 7115 | poll->file = req->file; |
d7718a9d JA |
7116 | |
7117 | ipt->pt._key = mask; | |
7118 | ipt->req = req; | |
68b11e8b PB |
7119 | ipt->error = 0; |
7120 | ipt->nr_entries = 0; | |
d7718a9d | 7121 | |
aa43477b PB |
7122 | /* |
7123 | * Take the ownership to delay any tw execution up until we're done | |
7124 | * with poll arming. see io_poll_get_ownership(). | |
7125 | */ | |
7126 | atomic_set(&req->poll_refs, 1); | |
d7718a9d | 7127 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
aa43477b PB |
7128 | |
7129 | if (mask && (poll->events & EPOLLONESHOT)) { | |
7130 | io_poll_remove_entries(req); | |
7131 | /* no one else has access to the req, forget about the ref */ | |
7132 | return mask; | |
7133 | } | |
7134 | if (!mask && unlikely(ipt->error || !ipt->nr_entries)) { | |
7135 | io_poll_remove_entries(req); | |
7136 | if (!ipt->error) | |
7137 | ipt->error = -EINVAL; | |
7138 | return 0; | |
7139 | } | |
d7718a9d | 7140 | |
79ebeaee | 7141 | spin_lock(&ctx->completion_lock); |
aa43477b PB |
7142 | io_poll_req_insert(req); |
7143 | spin_unlock(&ctx->completion_lock); | |
7144 | ||
7145 | if (mask) { | |
7146 | /* can't multishot if failed, just queue the event we've got */ | |
7147 | if (unlikely(ipt->error || !ipt->nr_entries)) | |
7148 | poll->events |= EPOLLONESHOT; | |
81459350 | 7149 | __io_poll_execute(req, mask, poll->events); |
aa43477b | 7150 | return 0; |
d7718a9d JA |
7151 | } |
7152 | ||
aa43477b PB |
7153 | /* |
7154 | * Release ownership. If someone tried to queue a tw while it was | |
7155 | * locked, kick it off for them. | |
7156 | */ | |
7157 | v = atomic_dec_return(&req->poll_refs); | |
7158 | if (unlikely(v & IO_POLL_REF_MASK)) | |
81459350 | 7159 | __io_poll_execute(req, 0, poll->events); |
aa43477b PB |
7160 | return 0; |
7161 | } | |
7162 | ||
7163 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
7164 | struct poll_table_struct *p) | |
7165 | { | |
7166 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
7167 | struct async_poll *apoll = pt->req->apoll; | |
7168 | ||
7169 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); | |
d7718a9d JA |
7170 | } |
7171 | ||
59b735ae OL |
7172 | enum { |
7173 | IO_APOLL_OK, | |
7174 | IO_APOLL_ABORTED, | |
7175 | IO_APOLL_READY | |
7176 | }; | |
7177 | ||
4d9237e3 | 7178 | static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags) |
d7718a9d JA |
7179 | { |
7180 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
7181 | struct io_ring_ctx *ctx = req->ctx; | |
7182 | struct async_poll *apoll; | |
7183 | struct io_poll_table ipt; | |
dbc2564c | 7184 | __poll_t mask = POLLPRI | POLLERR; |
aa43477b | 7185 | int ret; |
d7718a9d | 7186 | |
b2d9c3da PB |
7187 | if (!def->pollin && !def->pollout) |
7188 | return IO_APOLL_ABORTED; | |
10c87333 | 7189 | if (!file_can_poll(req->file)) |
658d0a40 | 7190 | return IO_APOLL_ABORTED; |
10c87333 | 7191 | if ((req->flags & (REQ_F_POLLED|REQ_F_PARTIAL_IO)) == REQ_F_POLLED) |
658d0a40 | 7192 | return IO_APOLL_ABORTED; |
dbc2564c HX |
7193 | if (!(req->flags & REQ_F_APOLL_MULTISHOT)) |
7194 | mask |= EPOLLONESHOT; | |
b2d9c3da PB |
7195 | |
7196 | if (def->pollin) { | |
a294bef5 | 7197 | mask |= EPOLLIN | EPOLLRDNORM; |
b2d9c3da PB |
7198 | |
7199 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
7200 | if ((req->opcode == IORING_OP_RECVMSG) && | |
7201 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
a294bef5 | 7202 | mask &= ~EPOLLIN; |
b2d9c3da | 7203 | } else { |
a294bef5 | 7204 | mask |= EPOLLOUT | EPOLLWRNORM; |
b2d9c3da | 7205 | } |
52dd8640 DY |
7206 | if (def->poll_exclusive) |
7207 | mask |= EPOLLEXCLUSIVE; | |
10c87333 JA |
7208 | if (req->flags & REQ_F_POLLED) { |
7209 | apoll = req->apoll; | |
7210 | } else if (!(issue_flags & IO_URING_F_UNLOCKED) && | |
7211 | !list_empty(&ctx->apoll_cache)) { | |
4d9237e3 JA |
7212 | apoll = list_first_entry(&ctx->apoll_cache, struct async_poll, |
7213 | poll.wait.entry); | |
7214 | list_del_init(&apoll->poll.wait.entry); | |
7215 | } else { | |
7216 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
7217 | if (unlikely(!apoll)) | |
7218 | return IO_APOLL_ABORTED; | |
7219 | } | |
807abcb0 | 7220 | apoll->double_poll = NULL; |
d7718a9d | 7221 | req->apoll = apoll; |
b2d9c3da | 7222 | req->flags |= REQ_F_POLLED; |
d7718a9d JA |
7223 | ipt.pt._qproc = io_async_queue_proc; |
7224 | ||
4d55f238 | 7225 | io_kbuf_recycle(req, issue_flags); |
abdad709 | 7226 | |
aa43477b | 7227 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask); |
41a5169c HX |
7228 | if (ret || ipt.error) |
7229 | return ret ? IO_APOLL_READY : IO_APOLL_ABORTED; | |
7230 | ||
cef216fc | 7231 | trace_io_uring_poll_arm(ctx, req, req->cqe.user_data, req->opcode, |
236daeae | 7232 | mask, apoll->poll.events); |
59b735ae | 7233 | return IO_APOLL_OK; |
d7718a9d JA |
7234 | } |
7235 | ||
76e1b642 JA |
7236 | /* |
7237 | * Returns true if we found and killed one or more poll requests | |
7238 | */ | |
c072481d PB |
7239 | static __cold bool io_poll_remove_all(struct io_ring_ctx *ctx, |
7240 | struct task_struct *tsk, bool cancel_all) | |
221c5eb2 | 7241 | { |
78076bb6 | 7242 | struct hlist_node *tmp; |
221c5eb2 | 7243 | struct io_kiocb *req; |
aa43477b PB |
7244 | bool found = false; |
7245 | int i; | |
221c5eb2 | 7246 | |
79ebeaee | 7247 | spin_lock(&ctx->completion_lock); |
78076bb6 JA |
7248 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
7249 | struct hlist_head *list; | |
7250 | ||
7251 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 7252 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
42a7b4ed | 7253 | if (io_match_task_safe(req, tsk, cancel_all)) { |
61bc84c4 | 7254 | hlist_del_init(&req->hash_node); |
aa43477b PB |
7255 | io_poll_cancel_req(req); |
7256 | found = true; | |
7257 | } | |
f3606e3a | 7258 | } |
221c5eb2 | 7259 | } |
79ebeaee | 7260 | spin_unlock(&ctx->completion_lock); |
aa43477b | 7261 | return found; |
221c5eb2 JA |
7262 | } |
7263 | ||
b21432b4 JA |
7264 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, bool poll_only, |
7265 | struct io_cancel_data *cd) | |
e07785b0 | 7266 | __must_hold(&ctx->completion_lock) |
47f46768 | 7267 | { |
78076bb6 | 7268 | struct hlist_head *list; |
47f46768 JA |
7269 | struct io_kiocb *req; |
7270 | ||
b21432b4 | 7271 | list = &ctx->cancel_hash[hash_long(cd->data, ctx->cancel_hash_bits)]; |
78076bb6 | 7272 | hlist_for_each_entry(req, list, hash_node) { |
b21432b4 | 7273 | if (cd->data != req->cqe.user_data) |
b41e9852 | 7274 | continue; |
9ba5fac8 PB |
7275 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
7276 | continue; | |
8e29da69 JA |
7277 | if (cd->flags & IORING_ASYNC_CANCEL_ALL) { |
7278 | if (cd->seq == req->work.cancel_seq) | |
7279 | continue; | |
7280 | req->work.cancel_seq = cd->seq; | |
7281 | } | |
b2cb805f | 7282 | return req; |
47f46768 | 7283 | } |
b2cb805f JA |
7284 | return NULL; |
7285 | } | |
7286 | ||
4bf94615 JA |
7287 | static struct io_kiocb *io_poll_file_find(struct io_ring_ctx *ctx, |
7288 | struct io_cancel_data *cd) | |
7289 | __must_hold(&ctx->completion_lock) | |
7290 | { | |
7291 | struct io_kiocb *req; | |
7292 | int i; | |
7293 | ||
7294 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7295 | struct hlist_head *list; | |
7296 | ||
7297 | list = &ctx->cancel_hash[i]; | |
7298 | hlist_for_each_entry(req, list, hash_node) { | |
970f256e JA |
7299 | if (!(cd->flags & IORING_ASYNC_CANCEL_ANY) && |
7300 | req->file != cd->file) | |
4bf94615 JA |
7301 | continue; |
7302 | if (cd->seq == req->work.cancel_seq) | |
7303 | continue; | |
7304 | req->work.cancel_seq = cd->seq; | |
7305 | return req; | |
7306 | } | |
7307 | } | |
7308 | return NULL; | |
7309 | } | |
7310 | ||
aa43477b PB |
7311 | static bool io_poll_disarm(struct io_kiocb *req) |
7312 | __must_hold(&ctx->completion_lock) | |
7313 | { | |
7314 | if (!io_poll_get_ownership(req)) | |
7315 | return false; | |
7316 | io_poll_remove_entries(req); | |
7317 | hash_del(&req->hash_node); | |
7318 | return true; | |
7319 | } | |
7320 | ||
b21432b4 | 7321 | static int io_poll_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd) |
e07785b0 | 7322 | __must_hold(&ctx->completion_lock) |
b2cb805f | 7323 | { |
4bf94615 | 7324 | struct io_kiocb *req; |
b2cb805f | 7325 | |
970f256e | 7326 | if (cd->flags & (IORING_ASYNC_CANCEL_FD|IORING_ASYNC_CANCEL_ANY)) |
4bf94615 JA |
7327 | req = io_poll_file_find(ctx, cd); |
7328 | else | |
7329 | req = io_poll_find(ctx, false, cd); | |
b2cb805f JA |
7330 | if (!req) |
7331 | return -ENOENT; | |
aa43477b PB |
7332 | io_poll_cancel_req(req); |
7333 | return 0; | |
47f46768 JA |
7334 | } |
7335 | ||
9096af3e PB |
7336 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
7337 | unsigned int flags) | |
7338 | { | |
7339 | u32 events; | |
47f46768 | 7340 | |
9096af3e PB |
7341 | events = READ_ONCE(sqe->poll32_events); |
7342 | #ifdef __BIG_ENDIAN | |
7343 | events = swahw32(events); | |
7344 | #endif | |
7345 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
7346 | events |= EPOLLONESHOT; | |
7347 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
47f46768 JA |
7348 | } |
7349 | ||
54739cc6 | 7350 | static int io_poll_remove_prep(struct io_kiocb *req, |
3529d8c2 | 7351 | const struct io_uring_sqe *sqe) |
0969e783 | 7352 | { |
c5de0036 PB |
7353 | struct io_poll_update *upd = &req->poll_update; |
7354 | u32 flags; | |
7355 | ||
73911426 | 7356 | if (sqe->buf_index || sqe->splice_fd_in) |
c5de0036 PB |
7357 | return -EINVAL; |
7358 | flags = READ_ONCE(sqe->len); | |
7359 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
7360 | IORING_POLL_ADD_MULTI)) | |
7361 | return -EINVAL; | |
7362 | /* meaningless without update */ | |
7363 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
7364 | return -EINVAL; |
7365 | ||
c5de0036 PB |
7366 | upd->old_user_data = READ_ONCE(sqe->addr); |
7367 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
7368 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 7369 | |
c5de0036 PB |
7370 | upd->new_user_data = READ_ONCE(sqe->off); |
7371 | if (!upd->update_user_data && upd->new_user_data) | |
7372 | return -EINVAL; | |
7373 | if (upd->update_events) | |
7374 | upd->events = io_poll_parse_events(sqe, flags); | |
7375 | else if (sqe->poll32_events) | |
7376 | return -EINVAL; | |
221c5eb2 | 7377 | |
221c5eb2 JA |
7378 | return 0; |
7379 | } | |
7380 | ||
3529d8c2 | 7381 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
7382 | { |
7383 | struct io_poll_iocb *poll = &req->poll; | |
c5de0036 | 7384 | u32 flags; |
221c5eb2 | 7385 | |
73911426 | 7386 | if (sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
7387 | return -EINVAL; |
7388 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 7389 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 | 7390 | return -EINVAL; |
04c76b41 PB |
7391 | if ((flags & IORING_POLL_ADD_MULTI) && (req->flags & REQ_F_CQE_SKIP)) |
7392 | return -EINVAL; | |
221c5eb2 | 7393 | |
48dcd38d | 7394 | io_req_set_refcount(req); |
2804ecd8 | 7395 | req->apoll_events = poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
7396 | return 0; |
7397 | } | |
7398 | ||
61e98203 | 7399 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
7400 | { |
7401 | struct io_poll_iocb *poll = &req->poll; | |
0969e783 | 7402 | struct io_poll_table ipt; |
aa43477b | 7403 | int ret; |
0969e783 | 7404 | |
d7718a9d | 7405 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 7406 | |
aa43477b PB |
7407 | ret = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events); |
7408 | ret = ret ?: ipt.error; | |
7409 | if (ret) | |
7410 | __io_req_complete(req, issue_flags, ret, 0); | |
7411 | return 0; | |
221c5eb2 JA |
7412 | } |
7413 | ||
54739cc6 | 7414 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 | 7415 | { |
b21432b4 | 7416 | struct io_cancel_data cd = { .data = req->poll_update.old_user_data, }; |
b69de288 JA |
7417 | struct io_ring_ctx *ctx = req->ctx; |
7418 | struct io_kiocb *preq; | |
2bbb146d | 7419 | int ret2, ret = 0; |
cc8e9ba7 | 7420 | bool locked; |
b69de288 | 7421 | |
79ebeaee | 7422 | spin_lock(&ctx->completion_lock); |
b21432b4 | 7423 | preq = io_poll_find(ctx, true, &cd); |
aa43477b | 7424 | if (!preq || !io_poll_disarm(preq)) { |
79ebeaee | 7425 | spin_unlock(&ctx->completion_lock); |
aa43477b | 7426 | ret = preq ? -EALREADY : -ENOENT; |
2bbb146d | 7427 | goto out; |
b69de288 | 7428 | } |
79ebeaee | 7429 | spin_unlock(&ctx->completion_lock); |
cb3b200e | 7430 | |
2bbb146d PB |
7431 | if (req->poll_update.update_events || req->poll_update.update_user_data) { |
7432 | /* only mask one event flags, keep behavior flags */ | |
7433 | if (req->poll_update.update_events) { | |
7434 | preq->poll.events &= ~0xffff; | |
7435 | preq->poll.events |= req->poll_update.events & 0xffff; | |
7436 | preq->poll.events |= IO_POLL_UNMASK; | |
cb3b200e | 7437 | } |
2bbb146d | 7438 | if (req->poll_update.update_user_data) |
cef216fc | 7439 | preq->cqe.user_data = req->poll_update.new_user_data; |
b69de288 | 7440 | |
2bbb146d PB |
7441 | ret2 = io_poll_add(preq, issue_flags); |
7442 | /* successfully updated, don't complete poll request */ | |
7443 | if (!ret2) | |
7444 | goto out; | |
b69de288 | 7445 | } |
6224590d | 7446 | |
2bbb146d | 7447 | req_set_fail(preq); |
cef216fc | 7448 | preq->cqe.res = -ECANCELED; |
cc8e9ba7 PB |
7449 | locked = !(issue_flags & IO_URING_F_UNLOCKED); |
7450 | io_req_task_complete(preq, &locked); | |
2bbb146d PB |
7451 | out: |
7452 | if (ret < 0) | |
6224590d | 7453 | req_set_fail(req); |
2bbb146d | 7454 | /* complete update request, we're done with it */ |
cc8e9ba7 | 7455 | __io_req_complete(req, issue_flags, ret, 0); |
b69de288 | 7456 | return 0; |
89850fce JA |
7457 | } |
7458 | ||
5262f567 JA |
7459 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
7460 | { | |
ad8a48ac JA |
7461 | struct io_timeout_data *data = container_of(timer, |
7462 | struct io_timeout_data, timer); | |
7463 | struct io_kiocb *req = data->req; | |
7464 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
7465 | unsigned long flags; |
7466 | ||
89850fce | 7467 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
a71976f3 | 7468 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
7469 | atomic_set(&req->ctx->cq_timeouts, |
7470 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
89850fce | 7471 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); |
01cec8c1 | 7472 | |
a90c8bf6 PB |
7473 | if (!(data->flags & IORING_TIMEOUT_ETIME_SUCCESS)) |
7474 | req_set_fail(req); | |
7475 | ||
cef216fc | 7476 | req->cqe.res = -ETIME; |
a90c8bf6 | 7477 | req->io_task_work.func = io_req_task_complete; |
3fe07bcd | 7478 | io_req_task_work_add(req); |
5262f567 JA |
7479 | return HRTIMER_NORESTART; |
7480 | } | |
7481 | ||
fbd15848 | 7482 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
b21432b4 | 7483 | struct io_cancel_data *cd) |
89850fce | 7484 | __must_hold(&ctx->timeout_lock) |
f254ac04 | 7485 | { |
fbd15848 | 7486 | struct io_timeout_data *io; |
47f46768 | 7487 | struct io_kiocb *req; |
fd9c7bc5 | 7488 | bool found = false; |
f254ac04 | 7489 | |
135fcde8 | 7490 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
970f256e JA |
7491 | if (!(cd->flags & IORING_ASYNC_CANCEL_ANY) && |
7492 | cd->data != req->cqe.user_data) | |
8e29da69 | 7493 | continue; |
970f256e | 7494 | if (cd->flags & (IORING_ASYNC_CANCEL_ALL|IORING_ASYNC_CANCEL_ANY)) { |
8e29da69 JA |
7495 | if (cd->seq == req->work.cancel_seq) |
7496 | continue; | |
7497 | req->work.cancel_seq = cd->seq; | |
7498 | } | |
7499 | found = true; | |
7500 | break; | |
47f46768 | 7501 | } |
fd9c7bc5 PB |
7502 | if (!found) |
7503 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
7504 | |
7505 | io = req->async_data; | |
fd9c7bc5 | 7506 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 7507 | return ERR_PTR(-EALREADY); |
a71976f3 | 7508 | list_del_init(&req->timeout.list); |
fbd15848 PB |
7509 | return req; |
7510 | } | |
47f46768 | 7511 | |
b21432b4 | 7512 | static int io_timeout_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd) |
ec3c3d0f | 7513 | __must_hold(&ctx->completion_lock) |
fbd15848 | 7514 | { |
3645c200 PB |
7515 | struct io_kiocb *req; |
7516 | ||
7517 | spin_lock_irq(&ctx->timeout_lock); | |
b21432b4 | 7518 | req = io_timeout_extract(ctx, cd); |
3645c200 | 7519 | spin_unlock_irq(&ctx->timeout_lock); |
fbd15848 PB |
7520 | |
7521 | if (IS_ERR(req)) | |
7522 | return PTR_ERR(req); | |
6695490d | 7523 | io_req_task_queue_fail(req, -ECANCELED); |
f254ac04 JA |
7524 | return 0; |
7525 | } | |
7526 | ||
50c1df2b JA |
7527 | static clockid_t io_timeout_get_clock(struct io_timeout_data *data) |
7528 | { | |
7529 | switch (data->flags & IORING_TIMEOUT_CLOCK_MASK) { | |
7530 | case IORING_TIMEOUT_BOOTTIME: | |
7531 | return CLOCK_BOOTTIME; | |
7532 | case IORING_TIMEOUT_REALTIME: | |
7533 | return CLOCK_REALTIME; | |
7534 | default: | |
7535 | /* can't happen, vetted at prep time */ | |
7536 | WARN_ON_ONCE(1); | |
7537 | fallthrough; | |
7538 | case 0: | |
7539 | return CLOCK_MONOTONIC; | |
7540 | } | |
7541 | } | |
7542 | ||
f1042b6c PB |
7543 | static int io_linked_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
7544 | struct timespec64 *ts, enum hrtimer_mode mode) | |
7545 | __must_hold(&ctx->timeout_lock) | |
7546 | { | |
7547 | struct io_timeout_data *io; | |
7548 | struct io_kiocb *req; | |
7549 | bool found = false; | |
7550 | ||
7551 | list_for_each_entry(req, &ctx->ltimeout_list, timeout.list) { | |
cef216fc | 7552 | found = user_data == req->cqe.user_data; |
f1042b6c PB |
7553 | if (found) |
7554 | break; | |
7555 | } | |
7556 | if (!found) | |
7557 | return -ENOENT; | |
7558 | ||
7559 | io = req->async_data; | |
7560 | if (hrtimer_try_to_cancel(&io->timer) == -1) | |
7561 | return -EALREADY; | |
7562 | hrtimer_init(&io->timer, io_timeout_get_clock(io), mode); | |
7563 | io->timer.function = io_link_timeout_fn; | |
7564 | hrtimer_start(&io->timer, timespec64_to_ktime(*ts), mode); | |
7565 | return 0; | |
7566 | } | |
7567 | ||
9c8e11b3 PB |
7568 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
7569 | struct timespec64 *ts, enum hrtimer_mode mode) | |
89850fce | 7570 | __must_hold(&ctx->timeout_lock) |
47f46768 | 7571 | { |
b21432b4 JA |
7572 | struct io_cancel_data cd = { .data = user_data, }; |
7573 | struct io_kiocb *req = io_timeout_extract(ctx, &cd); | |
9c8e11b3 | 7574 | struct io_timeout_data *data; |
47f46768 | 7575 | |
9c8e11b3 PB |
7576 | if (IS_ERR(req)) |
7577 | return PTR_ERR(req); | |
47f46768 | 7578 | |
9c8e11b3 PB |
7579 | req->timeout.off = 0; /* noseq */ |
7580 | data = req->async_data; | |
7581 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
50c1df2b | 7582 | hrtimer_init(&data->timer, io_timeout_get_clock(data), mode); |
9c8e11b3 PB |
7583 | data->timer.function = io_timeout_fn; |
7584 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
7585 | return 0; | |
47f46768 JA |
7586 | } |
7587 | ||
3529d8c2 JA |
7588 | static int io_timeout_remove_prep(struct io_kiocb *req, |
7589 | const struct io_uring_sqe *sqe) | |
b29472ee | 7590 | { |
9c8e11b3 PB |
7591 | struct io_timeout_rem *tr = &req->timeout_rem; |
7592 | ||
61710e43 DA |
7593 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
7594 | return -EINVAL; | |
73911426 | 7595 | if (sqe->buf_index || sqe->len || sqe->splice_fd_in) |
b29472ee JA |
7596 | return -EINVAL; |
7597 | ||
f1042b6c | 7598 | tr->ltimeout = false; |
9c8e11b3 PB |
7599 | tr->addr = READ_ONCE(sqe->addr); |
7600 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
f1042b6c PB |
7601 | if (tr->flags & IORING_TIMEOUT_UPDATE_MASK) { |
7602 | if (hweight32(tr->flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
7603 | return -EINVAL; | |
7604 | if (tr->flags & IORING_LINK_TIMEOUT_UPDATE) | |
7605 | tr->ltimeout = true; | |
7606 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE_MASK|IORING_TIMEOUT_ABS)) | |
9c8e11b3 PB |
7607 | return -EINVAL; |
7608 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
7609 | return -EFAULT; | |
2087009c YB |
7610 | if (tr->ts.tv_sec < 0 || tr->ts.tv_nsec < 0) |
7611 | return -EINVAL; | |
9c8e11b3 PB |
7612 | } else if (tr->flags) { |
7613 | /* timeout removal doesn't support flags */ | |
b29472ee | 7614 | return -EINVAL; |
9c8e11b3 | 7615 | } |
b29472ee | 7616 | |
b29472ee JA |
7617 | return 0; |
7618 | } | |
7619 | ||
8662daec PB |
7620 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
7621 | { | |
7622 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
7623 | : HRTIMER_MODE_REL; | |
7624 | } | |
7625 | ||
11365043 JA |
7626 | /* |
7627 | * Remove or update an existing timeout command | |
7628 | */ | |
61e98203 | 7629 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 7630 | { |
9c8e11b3 | 7631 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 7632 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 7633 | int ret; |
11365043 | 7634 | |
ec3c3d0f | 7635 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { |
b21432b4 JA |
7636 | struct io_cancel_data cd = { .data = tr->addr, }; |
7637 | ||
ec3c3d0f | 7638 | spin_lock(&ctx->completion_lock); |
b21432b4 | 7639 | ret = io_timeout_cancel(ctx, &cd); |
ec3c3d0f PB |
7640 | spin_unlock(&ctx->completion_lock); |
7641 | } else { | |
f1042b6c PB |
7642 | enum hrtimer_mode mode = io_translate_timeout_mode(tr->flags); |
7643 | ||
ec3c3d0f | 7644 | spin_lock_irq(&ctx->timeout_lock); |
f1042b6c PB |
7645 | if (tr->ltimeout) |
7646 | ret = io_linked_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
7647 | else | |
7648 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); | |
ec3c3d0f PB |
7649 | spin_unlock_irq(&ctx->timeout_lock); |
7650 | } | |
11365043 | 7651 | |
4e88d6e7 | 7652 | if (ret < 0) |
93d2bcd2 | 7653 | req_set_fail(req); |
505657bc | 7654 | io_req_complete_post(req, ret, 0); |
11365043 | 7655 | return 0; |
5262f567 JA |
7656 | } |
7657 | ||
ecddc25d JA |
7658 | static int __io_timeout_prep(struct io_kiocb *req, |
7659 | const struct io_uring_sqe *sqe, | |
7660 | bool is_timeout_link) | |
5262f567 | 7661 | { |
ad8a48ac | 7662 | struct io_timeout_data *data; |
a41525ab | 7663 | unsigned flags; |
56080b02 | 7664 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 7665 | |
73911426 | 7666 | if (sqe->buf_index || sqe->len != 1 || sqe->splice_fd_in) |
a41525ab | 7667 | return -EINVAL; |
56080b02 | 7668 | if (off && is_timeout_link) |
2d28390a | 7669 | return -EINVAL; |
a41525ab | 7670 | flags = READ_ONCE(sqe->timeout_flags); |
6224590d PB |
7671 | if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_CLOCK_MASK | |
7672 | IORING_TIMEOUT_ETIME_SUCCESS)) | |
50c1df2b JA |
7673 | return -EINVAL; |
7674 | /* more than one clock specified is invalid, obviously */ | |
7675 | if (hweight32(flags & IORING_TIMEOUT_CLOCK_MASK) > 1) | |
5262f567 | 7676 | return -EINVAL; |
bdf20073 | 7677 | |
ef9dd637 | 7678 | INIT_LIST_HEAD(&req->timeout.list); |
bfe68a22 | 7679 | req->timeout.off = off; |
f18ee4cf PB |
7680 | if (unlikely(off && !req->ctx->off_timeout_used)) |
7681 | req->ctx->off_timeout_used = true; | |
26a61679 | 7682 | |
d6a644a7 PB |
7683 | if (WARN_ON_ONCE(req_has_async_data(req))) |
7684 | return -EFAULT; | |
7685 | if (io_alloc_async_data(req)) | |
26a61679 JA |
7686 | return -ENOMEM; |
7687 | ||
e8c2bc1f | 7688 | data = req->async_data; |
ad8a48ac | 7689 | data->req = req; |
50c1df2b | 7690 | data->flags = flags; |
ad8a48ac JA |
7691 | |
7692 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
7693 | return -EFAULT; |
7694 | ||
f6223ff7 YB |
7695 | if (data->ts.tv_sec < 0 || data->ts.tv_nsec < 0) |
7696 | return -EINVAL; | |
7697 | ||
e677edbc | 7698 | INIT_LIST_HEAD(&req->timeout.list); |
8662daec | 7699 | data->mode = io_translate_timeout_mode(flags); |
50c1df2b | 7700 | hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode); |
b97e736a PB |
7701 | |
7702 | if (is_timeout_link) { | |
7703 | struct io_submit_link *link = &req->ctx->submit_state.link; | |
7704 | ||
7705 | if (!link->head) | |
7706 | return -EINVAL; | |
7707 | if (link->last->opcode == IORING_OP_LINK_TIMEOUT) | |
7708 | return -EINVAL; | |
4d13d1a4 PB |
7709 | req->timeout.head = link->last; |
7710 | link->last->flags |= REQ_F_ARM_LTIMEOUT; | |
b97e736a | 7711 | } |
ad8a48ac JA |
7712 | return 0; |
7713 | } | |
7714 | ||
ecddc25d JA |
7715 | static int io_timeout_prep(struct io_kiocb *req, |
7716 | const struct io_uring_sqe *sqe) | |
7717 | { | |
7718 | return __io_timeout_prep(req, sqe, false); | |
7719 | } | |
7720 | ||
7721 | static int io_link_timeout_prep(struct io_kiocb *req, | |
7722 | const struct io_uring_sqe *sqe) | |
7723 | { | |
7724 | return __io_timeout_prep(req, sqe, true); | |
7725 | } | |
7726 | ||
61e98203 | 7727 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 7728 | { |
ad8a48ac | 7729 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 7730 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 7731 | struct list_head *entry; |
bfe68a22 | 7732 | u32 tail, off = req->timeout.off; |
ad8a48ac | 7733 | |
89850fce | 7734 | spin_lock_irq(&ctx->timeout_lock); |
93bd25bb | 7735 | |
5262f567 JA |
7736 | /* |
7737 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
7738 | * timeout event to be satisfied. If it isn't set, then this is |
7739 | * a pure timeout request, sequence isn't used. | |
5262f567 | 7740 | */ |
8eb7e2d0 | 7741 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
7742 | entry = ctx->timeout_list.prev; |
7743 | goto add; | |
7744 | } | |
5262f567 | 7745 | |
bfe68a22 PB |
7746 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
7747 | req->timeout.target_seq = tail + off; | |
5262f567 | 7748 | |
f010505b MDG |
7749 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
7750 | * This is safe because ->completion_lock is held, and submissions | |
7751 | * and completions are never mixed in the same ->completion_lock section. | |
7752 | */ | |
7753 | ctx->cq_last_tm_flush = tail; | |
7754 | ||
5262f567 JA |
7755 | /* |
7756 | * Insertion sort, ensuring the first entry in the list is always | |
7757 | * the one we need first. | |
7758 | */ | |
5262f567 | 7759 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
7760 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
7761 | timeout.list); | |
5262f567 | 7762 | |
8eb7e2d0 | 7763 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 7764 | continue; |
bfe68a22 PB |
7765 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
7766 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
7767 | break; |
7768 | } | |
93bd25bb | 7769 | add: |
135fcde8 | 7770 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
7771 | data->timer.function = io_timeout_fn; |
7772 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
89850fce | 7773 | spin_unlock_irq(&ctx->timeout_lock); |
5262f567 JA |
7774 | return 0; |
7775 | } | |
5262f567 | 7776 | |
62755e35 JA |
7777 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
7778 | { | |
7779 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 7780 | struct io_cancel_data *cd = data; |
62755e35 | 7781 | |
8e29da69 JA |
7782 | if (req->ctx != cd->ctx) |
7783 | return false; | |
970f256e JA |
7784 | if (cd->flags & IORING_ASYNC_CANCEL_ANY) { |
7785 | ; | |
7786 | } else if (cd->flags & IORING_ASYNC_CANCEL_FD) { | |
4bf94615 JA |
7787 | if (req->file != cd->file) |
7788 | return false; | |
7789 | } else { | |
7790 | if (req->cqe.user_data != cd->data) | |
7791 | return false; | |
7792 | } | |
970f256e | 7793 | if (cd->flags & (IORING_ASYNC_CANCEL_ALL|IORING_ASYNC_CANCEL_ANY)) { |
8e29da69 JA |
7794 | if (cd->seq == req->work.cancel_seq) |
7795 | return false; | |
7796 | req->work.cancel_seq = cd->seq; | |
7797 | } | |
7798 | return true; | |
62755e35 JA |
7799 | } |
7800 | ||
b21432b4 JA |
7801 | static int io_async_cancel_one(struct io_uring_task *tctx, |
7802 | struct io_cancel_data *cd) | |
62755e35 | 7803 | { |
62755e35 | 7804 | enum io_wq_cancel cancel_ret; |
62755e35 | 7805 | int ret = 0; |
970f256e | 7806 | bool all; |
62755e35 | 7807 | |
f458dd84 | 7808 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
7809 | return -ENOENT; |
7810 | ||
970f256e JA |
7811 | all = cd->flags & (IORING_ASYNC_CANCEL_ALL|IORING_ASYNC_CANCEL_ANY); |
7812 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, cd, all); | |
62755e35 JA |
7813 | switch (cancel_ret) { |
7814 | case IO_WQ_CANCEL_OK: | |
7815 | ret = 0; | |
7816 | break; | |
7817 | case IO_WQ_CANCEL_RUNNING: | |
7818 | ret = -EALREADY; | |
7819 | break; | |
7820 | case IO_WQ_CANCEL_NOTFOUND: | |
7821 | ret = -ENOENT; | |
7822 | break; | |
7823 | } | |
7824 | ||
e977d6d3 JA |
7825 | return ret; |
7826 | } | |
7827 | ||
b21432b4 | 7828 | static int io_try_cancel(struct io_kiocb *req, struct io_cancel_data *cd) |
47f46768 | 7829 | { |
8cb01fac | 7830 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 JA |
7831 | int ret; |
7832 | ||
dadebc35 | 7833 | WARN_ON_ONCE(!io_wq_current_is_worker() && req->task != current); |
8cb01fac | 7834 | |
b21432b4 | 7835 | ret = io_async_cancel_one(req->task->io_uring, cd); |
ccbf7261 JA |
7836 | /* |
7837 | * Fall-through even for -EALREADY, as we may have poll armed | |
7838 | * that need unarming. | |
7839 | */ | |
7840 | if (!ret) | |
7841 | return 0; | |
505657bc PB |
7842 | |
7843 | spin_lock(&ctx->completion_lock); | |
b21432b4 | 7844 | ret = io_poll_cancel(ctx, cd); |
ccbf7261 JA |
7845 | if (ret != -ENOENT) |
7846 | goto out; | |
4bf94615 JA |
7847 | if (!(cd->flags & IORING_ASYNC_CANCEL_FD)) |
7848 | ret = io_timeout_cancel(ctx, cd); | |
505657bc PB |
7849 | out: |
7850 | spin_unlock(&ctx->completion_lock); | |
7851 | return ret; | |
47f46768 JA |
7852 | } |
7853 | ||
970f256e JA |
7854 | #define CANCEL_FLAGS (IORING_ASYNC_CANCEL_ALL | IORING_ASYNC_CANCEL_FD | \ |
7855 | IORING_ASYNC_CANCEL_ANY) | |
7856 | ||
3529d8c2 JA |
7857 | static int io_async_cancel_prep(struct io_kiocb *req, |
7858 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 7859 | { |
4bf94615 | 7860 | if (unlikely(req->flags & REQ_F_BUFFER_SELECT)) |
61710e43 | 7861 | return -EINVAL; |
73911426 | 7862 | if (sqe->off || sqe->len || sqe->splice_fd_in) |
e977d6d3 JA |
7863 | return -EINVAL; |
7864 | ||
fbf23849 | 7865 | req->cancel.addr = READ_ONCE(sqe->addr); |
8e29da69 | 7866 | req->cancel.flags = READ_ONCE(sqe->cancel_flags); |
970f256e | 7867 | if (req->cancel.flags & ~CANCEL_FLAGS) |
8e29da69 | 7868 | return -EINVAL; |
970f256e JA |
7869 | if (req->cancel.flags & IORING_ASYNC_CANCEL_FD) { |
7870 | if (req->cancel.flags & IORING_ASYNC_CANCEL_ANY) | |
7871 | return -EINVAL; | |
4bf94615 | 7872 | req->cancel.fd = READ_ONCE(sqe->fd); |
970f256e | 7873 | } |
8e29da69 | 7874 | |
fbf23849 JA |
7875 | return 0; |
7876 | } | |
7877 | ||
8e29da69 JA |
7878 | static int __io_async_cancel(struct io_cancel_data *cd, struct io_kiocb *req, |
7879 | unsigned int issue_flags) | |
fbf23849 | 7880 | { |
970f256e | 7881 | bool all = cd->flags & (IORING_ASYNC_CANCEL_ALL|IORING_ASYNC_CANCEL_ANY); |
8e29da69 | 7882 | struct io_ring_ctx *ctx = cd->ctx; |
58f99373 | 7883 | struct io_tctx_node *node; |
8e29da69 | 7884 | int ret, nr = 0; |
58f99373 | 7885 | |
8e29da69 JA |
7886 | do { |
7887 | ret = io_try_cancel(req, cd); | |
7888 | if (ret == -ENOENT) | |
7889 | break; | |
970f256e | 7890 | if (!all) |
8e29da69 JA |
7891 | return ret; |
7892 | nr++; | |
7893 | } while (1); | |
58f99373 PB |
7894 | |
7895 | /* slow path, try all io-wq's */ | |
f8929630 | 7896 | io_ring_submit_lock(ctx, issue_flags); |
58f99373 PB |
7897 | ret = -ENOENT; |
7898 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
7899 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 7900 | |
8e29da69 JA |
7901 | ret = io_async_cancel_one(tctx, cd); |
7902 | if (ret != -ENOENT) { | |
970f256e | 7903 | if (!all) |
8e29da69 JA |
7904 | break; |
7905 | nr++; | |
7906 | } | |
58f99373 | 7907 | } |
f8929630 | 7908 | io_ring_submit_unlock(ctx, issue_flags); |
970f256e | 7909 | return all ? nr : ret; |
8e29da69 JA |
7910 | } |
7911 | ||
7912 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) | |
7913 | { | |
7914 | struct io_cancel_data cd = { | |
7915 | .ctx = req->ctx, | |
7916 | .data = req->cancel.addr, | |
7917 | .flags = req->cancel.flags, | |
7918 | .seq = atomic_inc_return(&req->ctx->cancel_seq), | |
7919 | }; | |
7920 | int ret; | |
7921 | ||
4bf94615 JA |
7922 | if (cd.flags & IORING_ASYNC_CANCEL_FD) { |
7923 | if (req->flags & REQ_F_FIXED_FILE) | |
7924 | req->file = io_file_get_fixed(req, req->cancel.fd, | |
7925 | issue_flags); | |
7926 | else | |
7927 | req->file = io_file_get_normal(req, req->cancel.fd); | |
7928 | if (!req->file) { | |
7929 | ret = -EBADF; | |
7930 | goto done; | |
7931 | } | |
7932 | cd.file = req->file; | |
58f99373 | 7933 | } |
4bf94615 | 7934 | |
8e29da69 | 7935 | ret = __io_async_cancel(&cd, req, issue_flags); |
58f99373 | 7936 | done: |
58f99373 | 7937 | if (ret < 0) |
93d2bcd2 | 7938 | req_set_fail(req); |
505657bc | 7939 | io_req_complete_post(req, ret, 0); |
5262f567 JA |
7940 | return 0; |
7941 | } | |
7942 | ||
54739cc6 | 7943 | static int io_files_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
7944 | const struct io_uring_sqe *sqe) |
7945 | { | |
61710e43 DA |
7946 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
7947 | return -EINVAL; | |
73911426 | 7948 | if (sqe->rw_flags || sqe->splice_fd_in) |
05f3fb3c JA |
7949 | return -EINVAL; |
7950 | ||
269bbe5f BM |
7951 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
7952 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
7953 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 7954 | return -EINVAL; |
269bbe5f | 7955 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
7956 | return 0; |
7957 | } | |
7958 | ||
a7c41b46 XW |
7959 | static int io_files_update_with_index_alloc(struct io_kiocb *req, |
7960 | unsigned int issue_flags) | |
7961 | { | |
7962 | __s32 __user *fds = u64_to_user_ptr(req->rsrc_update.arg); | |
7963 | unsigned int done; | |
7964 | struct file *file; | |
7965 | int ret, fd; | |
7966 | ||
7967 | for (done = 0; done < req->rsrc_update.nr_args; done++) { | |
7968 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7969 | ret = -EFAULT; | |
7970 | break; | |
7971 | } | |
7972 | ||
7973 | file = fget(fd); | |
7974 | if (!file) { | |
7975 | ret = -EBADF; | |
7976 | break; | |
7977 | } | |
7978 | ret = io_fixed_fd_install(req, issue_flags, file, | |
7979 | IORING_FILE_INDEX_ALLOC); | |
7980 | if (ret < 0) | |
7981 | break; | |
7982 | if (copy_to_user(&fds[done], &ret, sizeof(ret))) { | |
a7c41b46 | 7983 | __io_close_fixed(req, issue_flags, ret); |
e71d7c56 | 7984 | ret = -EFAULT; |
a7c41b46 XW |
7985 | break; |
7986 | } | |
7987 | } | |
7988 | ||
7989 | if (done) | |
7990 | return done; | |
7991 | return ret; | |
7992 | } | |
7993 | ||
889fca73 | 7994 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
7995 | { |
7996 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 7997 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 7998 | int ret; |
fbf23849 | 7999 | |
269bbe5f BM |
8000 | up.offset = req->rsrc_update.offset; |
8001 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
8002 | up.nr = 0; |
8003 | up.tags = 0; | |
615cee49 | 8004 | up.resv = 0; |
d8a3ba9c | 8005 | up.resv2 = 0; |
05f3fb3c | 8006 | |
a7c41b46 XW |
8007 | if (req->rsrc_update.offset == IORING_FILE_INDEX_ALLOC) { |
8008 | ret = io_files_update_with_index_alloc(req, issue_flags); | |
8009 | } else { | |
8010 | io_ring_submit_lock(ctx, issue_flags); | |
8011 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, | |
8012 | &up, req->rsrc_update.nr_args); | |
8013 | io_ring_submit_unlock(ctx, issue_flags); | |
8014 | } | |
05f3fb3c JA |
8015 | |
8016 | if (ret < 0) | |
93d2bcd2 | 8017 | req_set_fail(req); |
889fca73 | 8018 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
8019 | return 0; |
8020 | } | |
8021 | ||
bfe76559 | 8022 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 8023 | { |
d625c6ee | 8024 | switch (req->opcode) { |
e781573e | 8025 | case IORING_OP_NOP: |
2bb04df7 | 8026 | return io_nop_prep(req, sqe); |
f67676d1 JA |
8027 | case IORING_OP_READV: |
8028 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 8029 | case IORING_OP_READ: |
f67676d1 JA |
8030 | case IORING_OP_WRITEV: |
8031 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 8032 | case IORING_OP_WRITE: |
584b0180 | 8033 | return io_prep_rw(req, sqe); |
0969e783 | 8034 | case IORING_OP_POLL_ADD: |
bfe76559 | 8035 | return io_poll_add_prep(req, sqe); |
0969e783 | 8036 | case IORING_OP_POLL_REMOVE: |
54739cc6 | 8037 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 8038 | case IORING_OP_FSYNC: |
1155c76a | 8039 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 8040 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 8041 | return io_sfr_prep(req, sqe); |
03b1230c | 8042 | case IORING_OP_SENDMSG: |
fddaface | 8043 | case IORING_OP_SEND: |
bfe76559 | 8044 | return io_sendmsg_prep(req, sqe); |
03b1230c | 8045 | case IORING_OP_RECVMSG: |
fddaface | 8046 | case IORING_OP_RECV: |
bfe76559 | 8047 | return io_recvmsg_prep(req, sqe); |
f499a021 | 8048 | case IORING_OP_CONNECT: |
bfe76559 | 8049 | return io_connect_prep(req, sqe); |
2d28390a | 8050 | case IORING_OP_TIMEOUT: |
ecddc25d | 8051 | return io_timeout_prep(req, sqe); |
b29472ee | 8052 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 8053 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 8054 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 8055 | return io_async_cancel_prep(req, sqe); |
2d28390a | 8056 | case IORING_OP_LINK_TIMEOUT: |
ecddc25d | 8057 | return io_link_timeout_prep(req, sqe); |
8ed8d3c3 | 8058 | case IORING_OP_ACCEPT: |
bfe76559 | 8059 | return io_accept_prep(req, sqe); |
d63d1b5e | 8060 | case IORING_OP_FALLOCATE: |
bfe76559 | 8061 | return io_fallocate_prep(req, sqe); |
15b71abe | 8062 | case IORING_OP_OPENAT: |
bfe76559 | 8063 | return io_openat_prep(req, sqe); |
b5dba59e | 8064 | case IORING_OP_CLOSE: |
bfe76559 | 8065 | return io_close_prep(req, sqe); |
05f3fb3c | 8066 | case IORING_OP_FILES_UPDATE: |
54739cc6 | 8067 | return io_files_update_prep(req, sqe); |
eddc7ef5 | 8068 | case IORING_OP_STATX: |
bfe76559 | 8069 | return io_statx_prep(req, sqe); |
4840e418 | 8070 | case IORING_OP_FADVISE: |
bfe76559 | 8071 | return io_fadvise_prep(req, sqe); |
c1ca757b | 8072 | case IORING_OP_MADVISE: |
bfe76559 | 8073 | return io_madvise_prep(req, sqe); |
cebdb986 | 8074 | case IORING_OP_OPENAT2: |
bfe76559 | 8075 | return io_openat2_prep(req, sqe); |
3e4827b0 | 8076 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 8077 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 8078 | case IORING_OP_SPLICE: |
bfe76559 | 8079 | return io_splice_prep(req, sqe); |
ddf0322d | 8080 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 8081 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 8082 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 8083 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 8084 | case IORING_OP_TEE: |
bfe76559 | 8085 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
8086 | case IORING_OP_SHUTDOWN: |
8087 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
8088 | case IORING_OP_RENAMEAT: |
8089 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
8090 | case IORING_OP_UNLINKAT: |
8091 | return io_unlinkat_prep(req, sqe); | |
e34a02dc DK |
8092 | case IORING_OP_MKDIRAT: |
8093 | return io_mkdirat_prep(req, sqe); | |
7a8721f8 DK |
8094 | case IORING_OP_SYMLINKAT: |
8095 | return io_symlinkat_prep(req, sqe); | |
cf30da90 DK |
8096 | case IORING_OP_LINKAT: |
8097 | return io_linkat_prep(req, sqe); | |
4f57f06c JA |
8098 | case IORING_OP_MSG_RING: |
8099 | return io_msg_ring_prep(req, sqe); | |
e9621e2b SR |
8100 | case IORING_OP_FSETXATTR: |
8101 | return io_fsetxattr_prep(req, sqe); | |
8102 | case IORING_OP_SETXATTR: | |
8103 | return io_setxattr_prep(req, sqe); | |
a56834e0 SR |
8104 | case IORING_OP_FGETXATTR: |
8105 | return io_fgetxattr_prep(req, sqe); | |
8106 | case IORING_OP_GETXATTR: | |
8107 | return io_getxattr_prep(req, sqe); | |
1374e08e JA |
8108 | case IORING_OP_SOCKET: |
8109 | return io_socket_prep(req, sqe); | |
ee692a21 JA |
8110 | case IORING_OP_URING_CMD: |
8111 | return io_uring_cmd_prep(req, sqe); | |
f67676d1 JA |
8112 | } |
8113 | ||
bfe76559 PB |
8114 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
8115 | req->opcode); | |
bd54b6fe | 8116 | return -EINVAL; |
bfe76559 PB |
8117 | } |
8118 | ||
93642ef8 | 8119 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 8120 | { |
a196c78b JA |
8121 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
8122 | ||
8123 | /* assign early for deferred execution for non-fixed file */ | |
8124 | if (def->needs_file && !(req->flags & REQ_F_FIXED_FILE)) | |
3a166bdb | 8125 | req->file = io_file_get_normal(req, req->cqe.fd); |
a196c78b | 8126 | if (!def->needs_async_setup) |
b7e298d2 | 8127 | return 0; |
d886e185 | 8128 | if (WARN_ON_ONCE(req_has_async_data(req))) |
b7e298d2 PB |
8129 | return -EFAULT; |
8130 | if (io_alloc_async_data(req)) | |
8131 | return -EAGAIN; | |
8132 | ||
93642ef8 PB |
8133 | switch (req->opcode) { |
8134 | case IORING_OP_READV: | |
157dc813 | 8135 | return io_readv_prep_async(req); |
93642ef8 | 8136 | case IORING_OP_WRITEV: |
157dc813 | 8137 | return io_writev_prep_async(req); |
93642ef8 | 8138 | case IORING_OP_SENDMSG: |
93642ef8 PB |
8139 | return io_sendmsg_prep_async(req); |
8140 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
8141 | return io_recvmsg_prep_async(req); |
8142 | case IORING_OP_CONNECT: | |
8143 | return io_connect_prep_async(req); | |
ee692a21 JA |
8144 | case IORING_OP_URING_CMD: |
8145 | return io_uring_cmd_prep_async(req); | |
93642ef8 | 8146 | } |
b7e298d2 PB |
8147 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
8148 | req->opcode); | |
8149 | return -EFAULT; | |
f67676d1 JA |
8150 | } |
8151 | ||
9cf7c104 PB |
8152 | static u32 io_get_sequence(struct io_kiocb *req) |
8153 | { | |
a3dbdf54 | 8154 | u32 seq = req->ctx->cached_sq_head; |
963c6abb | 8155 | struct io_kiocb *cur; |
9cf7c104 | 8156 | |
a3dbdf54 | 8157 | /* need original cached_sq_head, but it was increased for each req */ |
963c6abb | 8158 | io_for_each_link(cur, req) |
a3dbdf54 PB |
8159 | seq--; |
8160 | return seq; | |
9cf7c104 PB |
8161 | } |
8162 | ||
c072481d | 8163 | static __cold void io_drain_req(struct io_kiocb *req) |
de0617e4 | 8164 | { |
a197f664 | 8165 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 8166 | struct io_defer_entry *de; |
f67676d1 | 8167 | int ret; |
e0eb71dc | 8168 | u32 seq = io_get_sequence(req); |
3c19966d | 8169 | |
9d858b21 | 8170 | /* Still need defer if there is pending req in defer list. */ |
e302f104 | 8171 | spin_lock(&ctx->completion_lock); |
5e371265 | 8172 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) { |
e302f104 | 8173 | spin_unlock(&ctx->completion_lock); |
e0eb71dc | 8174 | queue: |
10c66904 | 8175 | ctx->drain_active = false; |
e0eb71dc PB |
8176 | io_req_task_queue(req); |
8177 | return; | |
10c66904 | 8178 | } |
e302f104 | 8179 | spin_unlock(&ctx->completion_lock); |
9cf7c104 | 8180 | |
b7e298d2 | 8181 | ret = io_req_prep_async(req); |
e0eb71dc PB |
8182 | if (ret) { |
8183 | fail: | |
8184 | io_req_complete_failed(req, ret); | |
8185 | return; | |
8186 | } | |
cbdcb435 | 8187 | io_prep_async_link(req); |
27dc8338 | 8188 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
76cc33d7 | 8189 | if (!de) { |
1b48773f | 8190 | ret = -ENOMEM; |
e0eb71dc | 8191 | goto fail; |
76cc33d7 | 8192 | } |
2d28390a | 8193 | |
79ebeaee | 8194 | spin_lock(&ctx->completion_lock); |
9cf7c104 | 8195 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
79ebeaee | 8196 | spin_unlock(&ctx->completion_lock); |
27dc8338 | 8197 | kfree(de); |
e0eb71dc | 8198 | goto queue; |
de0617e4 JA |
8199 | } |
8200 | ||
cef216fc | 8201 | trace_io_uring_defer(ctx, req, req->cqe.user_data, req->opcode); |
27dc8338 | 8202 | de->req = req; |
9cf7c104 | 8203 | de->seq = seq; |
27dc8338 | 8204 | list_add_tail(&de->list, &ctx->defer_list); |
79ebeaee | 8205 | spin_unlock(&ctx->completion_lock); |
de0617e4 JA |
8206 | } |
8207 | ||
68fb8979 | 8208 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 8209 | { |
8197b053 PB |
8210 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
8211 | spin_lock(&req->ctx->completion_lock); | |
cc3cec83 | 8212 | io_put_kbuf_comp(req); |
8197b053 PB |
8213 | spin_unlock(&req->ctx->completion_lock); |
8214 | } | |
99bc4c38 | 8215 | |
0e1b6fe3 PB |
8216 | if (req->flags & REQ_F_NEED_CLEANUP) { |
8217 | switch (req->opcode) { | |
8218 | case IORING_OP_READV: | |
8219 | case IORING_OP_READ_FIXED: | |
8220 | case IORING_OP_READ: | |
8221 | case IORING_OP_WRITEV: | |
8222 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
8223 | case IORING_OP_WRITE: { |
8224 | struct io_async_rw *io = req->async_data; | |
1dacb4df PB |
8225 | |
8226 | kfree(io->free_iovec); | |
0e1b6fe3 | 8227 | break; |
e8c2bc1f | 8228 | } |
0e1b6fe3 | 8229 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
8230 | case IORING_OP_SENDMSG: { |
8231 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
8232 | |
8233 | kfree(io->free_iov); | |
0e1b6fe3 | 8234 | break; |
e8c2bc1f | 8235 | } |
f3cd4850 JA |
8236 | case IORING_OP_OPENAT: |
8237 | case IORING_OP_OPENAT2: | |
8238 | if (req->open.filename) | |
8239 | putname(req->open.filename); | |
8240 | break; | |
80a261fd JA |
8241 | case IORING_OP_RENAMEAT: |
8242 | putname(req->rename.oldpath); | |
8243 | putname(req->rename.newpath); | |
8244 | break; | |
14a1143b JA |
8245 | case IORING_OP_UNLINKAT: |
8246 | putname(req->unlink.filename); | |
8247 | break; | |
e34a02dc DK |
8248 | case IORING_OP_MKDIRAT: |
8249 | putname(req->mkdir.filename); | |
8250 | break; | |
7a8721f8 DK |
8251 | case IORING_OP_SYMLINKAT: |
8252 | putname(req->symlink.oldpath); | |
8253 | putname(req->symlink.newpath); | |
8254 | break; | |
cf30da90 DK |
8255 | case IORING_OP_LINKAT: |
8256 | putname(req->hardlink.oldpath); | |
8257 | putname(req->hardlink.newpath); | |
8258 | break; | |
1b6fe6e0 SR |
8259 | case IORING_OP_STATX: |
8260 | if (req->statx.filename) | |
8261 | putname(req->statx.filename); | |
8262 | break; | |
e9621e2b SR |
8263 | case IORING_OP_SETXATTR: |
8264 | case IORING_OP_FSETXATTR: | |
a56834e0 SR |
8265 | case IORING_OP_GETXATTR: |
8266 | case IORING_OP_FGETXATTR: | |
e9621e2b SR |
8267 | __io_xattr_finish(req); |
8268 | break; | |
0e1b6fe3 | 8269 | } |
99bc4c38 | 8270 | } |
75652a30 JA |
8271 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
8272 | kfree(req->apoll->double_poll); | |
8273 | kfree(req->apoll); | |
8274 | req->apoll = NULL; | |
8275 | } | |
9cae36a0 JA |
8276 | if (req->flags & REQ_F_INFLIGHT) { |
8277 | struct io_uring_task *tctx = req->task->io_uring; | |
8278 | ||
8279 | atomic_dec(&tctx->inflight_tracked); | |
8280 | } | |
c854357b | 8281 | if (req->flags & REQ_F_CREDS) |
b8e64b53 | 8282 | put_cred(req->creds); |
d886e185 PB |
8283 | if (req->flags & REQ_F_ASYNC_DATA) { |
8284 | kfree(req->async_data); | |
8285 | req->async_data = NULL; | |
8286 | } | |
c854357b | 8287 | req->flags &= ~IO_REQ_CLEAN_FLAGS; |
99bc4c38 PB |
8288 | } |
8289 | ||
6bf9c47a JA |
8290 | static bool io_assign_file(struct io_kiocb *req, unsigned int issue_flags) |
8291 | { | |
8292 | if (req->file || !io_op_defs[req->opcode].needs_file) | |
8293 | return true; | |
8294 | ||
8295 | if (req->flags & REQ_F_FIXED_FILE) | |
cef216fc | 8296 | req->file = io_file_get_fixed(req, req->cqe.fd, issue_flags); |
6bf9c47a | 8297 | else |
cef216fc | 8298 | req->file = io_file_get_normal(req, req->cqe.fd); |
6bf9c47a | 8299 | |
772f5e00 | 8300 | return !!req->file; |
6bf9c47a JA |
8301 | } |
8302 | ||
889fca73 | 8303 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 8304 | { |
fcde59fe | 8305 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5730b27e | 8306 | const struct cred *creds = NULL; |
d625c6ee | 8307 | int ret; |
2b188cc1 | 8308 | |
70152140 JA |
8309 | if (unlikely(!io_assign_file(req, issue_flags))) |
8310 | return -EBADF; | |
8311 | ||
6878b40e | 8312 | if (unlikely((req->flags & REQ_F_CREDS) && req->creds != current_cred())) |
c10d1f98 | 8313 | creds = override_creds(req->creds); |
5730b27e | 8314 | |
fcde59fe | 8315 | if (!def->audit_skip) |
5bd2182d PM |
8316 | audit_uring_entry(req->opcode); |
8317 | ||
d625c6ee | 8318 | switch (req->opcode) { |
2b188cc1 | 8319 | case IORING_OP_NOP: |
889fca73 | 8320 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
8321 | break; |
8322 | case IORING_OP_READV: | |
edafccee | 8323 | case IORING_OP_READ_FIXED: |
3a6820f2 | 8324 | case IORING_OP_READ: |
889fca73 | 8325 | ret = io_read(req, issue_flags); |
edafccee | 8326 | break; |
3529d8c2 | 8327 | case IORING_OP_WRITEV: |
edafccee | 8328 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 8329 | case IORING_OP_WRITE: |
889fca73 | 8330 | ret = io_write(req, issue_flags); |
2b188cc1 | 8331 | break; |
c992fe29 | 8332 | case IORING_OP_FSYNC: |
45d189c6 | 8333 | ret = io_fsync(req, issue_flags); |
c992fe29 | 8334 | break; |
221c5eb2 | 8335 | case IORING_OP_POLL_ADD: |
61e98203 | 8336 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
8337 | break; |
8338 | case IORING_OP_POLL_REMOVE: | |
54739cc6 | 8339 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 8340 | break; |
5d17b4a4 | 8341 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 8342 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 8343 | break; |
0fa03c62 | 8344 | case IORING_OP_SENDMSG: |
889fca73 | 8345 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 8346 | break; |
fddaface | 8347 | case IORING_OP_SEND: |
889fca73 | 8348 | ret = io_send(req, issue_flags); |
0fa03c62 | 8349 | break; |
aa1fa28f | 8350 | case IORING_OP_RECVMSG: |
889fca73 | 8351 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 8352 | break; |
fddaface | 8353 | case IORING_OP_RECV: |
889fca73 | 8354 | ret = io_recv(req, issue_flags); |
aa1fa28f | 8355 | break; |
5262f567 | 8356 | case IORING_OP_TIMEOUT: |
61e98203 | 8357 | ret = io_timeout(req, issue_flags); |
5262f567 | 8358 | break; |
11365043 | 8359 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 8360 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 8361 | break; |
17f2fe35 | 8362 | case IORING_OP_ACCEPT: |
889fca73 | 8363 | ret = io_accept(req, issue_flags); |
17f2fe35 | 8364 | break; |
f8e85cf2 | 8365 | case IORING_OP_CONNECT: |
889fca73 | 8366 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 8367 | break; |
62755e35 | 8368 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 8369 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 8370 | break; |
d63d1b5e | 8371 | case IORING_OP_FALLOCATE: |
45d189c6 | 8372 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 8373 | break; |
15b71abe | 8374 | case IORING_OP_OPENAT: |
45d189c6 | 8375 | ret = io_openat(req, issue_flags); |
15b71abe | 8376 | break; |
b5dba59e | 8377 | case IORING_OP_CLOSE: |
889fca73 | 8378 | ret = io_close(req, issue_flags); |
b5dba59e | 8379 | break; |
05f3fb3c | 8380 | case IORING_OP_FILES_UPDATE: |
889fca73 | 8381 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 8382 | break; |
eddc7ef5 | 8383 | case IORING_OP_STATX: |
45d189c6 | 8384 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 8385 | break; |
4840e418 | 8386 | case IORING_OP_FADVISE: |
45d189c6 | 8387 | ret = io_fadvise(req, issue_flags); |
4840e418 | 8388 | break; |
c1ca757b | 8389 | case IORING_OP_MADVISE: |
45d189c6 | 8390 | ret = io_madvise(req, issue_flags); |
c1ca757b | 8391 | break; |
cebdb986 | 8392 | case IORING_OP_OPENAT2: |
45d189c6 | 8393 | ret = io_openat2(req, issue_flags); |
cebdb986 | 8394 | break; |
3e4827b0 | 8395 | case IORING_OP_EPOLL_CTL: |
889fca73 | 8396 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 8397 | break; |
7d67af2c | 8398 | case IORING_OP_SPLICE: |
45d189c6 | 8399 | ret = io_splice(req, issue_flags); |
7d67af2c | 8400 | break; |
ddf0322d | 8401 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 8402 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 8403 | break; |
067524e9 | 8404 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 8405 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 8406 | break; |
f2a8d5c7 | 8407 | case IORING_OP_TEE: |
45d189c6 | 8408 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 8409 | break; |
36f4fa68 | 8410 | case IORING_OP_SHUTDOWN: |
45d189c6 | 8411 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 8412 | break; |
80a261fd | 8413 | case IORING_OP_RENAMEAT: |
45d189c6 | 8414 | ret = io_renameat(req, issue_flags); |
80a261fd | 8415 | break; |
14a1143b | 8416 | case IORING_OP_UNLINKAT: |
45d189c6 | 8417 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 8418 | break; |
e34a02dc DK |
8419 | case IORING_OP_MKDIRAT: |
8420 | ret = io_mkdirat(req, issue_flags); | |
8421 | break; | |
7a8721f8 DK |
8422 | case IORING_OP_SYMLINKAT: |
8423 | ret = io_symlinkat(req, issue_flags); | |
8424 | break; | |
cf30da90 DK |
8425 | case IORING_OP_LINKAT: |
8426 | ret = io_linkat(req, issue_flags); | |
8427 | break; | |
4f57f06c JA |
8428 | case IORING_OP_MSG_RING: |
8429 | ret = io_msg_ring(req, issue_flags); | |
8430 | break; | |
e9621e2b SR |
8431 | case IORING_OP_FSETXATTR: |
8432 | ret = io_fsetxattr(req, issue_flags); | |
8433 | break; | |
8434 | case IORING_OP_SETXATTR: | |
8435 | ret = io_setxattr(req, issue_flags); | |
8436 | break; | |
a56834e0 SR |
8437 | case IORING_OP_FGETXATTR: |
8438 | ret = io_fgetxattr(req, issue_flags); | |
8439 | break; | |
8440 | case IORING_OP_GETXATTR: | |
8441 | ret = io_getxattr(req, issue_flags); | |
8442 | break; | |
1374e08e JA |
8443 | case IORING_OP_SOCKET: |
8444 | ret = io_socket(req, issue_flags); | |
8445 | break; | |
ee692a21 JA |
8446 | case IORING_OP_URING_CMD: |
8447 | ret = io_uring_cmd(req, issue_flags); | |
8448 | break; | |
2b188cc1 JA |
8449 | default: |
8450 | ret = -EINVAL; | |
8451 | break; | |
8452 | } | |
8453 | ||
fcde59fe | 8454 | if (!def->audit_skip) |
5bd2182d PM |
8455 | audit_uring_exit(!ret, ret); |
8456 | ||
5730b27e JA |
8457 | if (creds) |
8458 | revert_creds(creds); | |
def596e9 JA |
8459 | if (ret) |
8460 | return ret; | |
b532576e | 8461 | /* If the op doesn't have a file, we're not polling for it */ |
9983028e | 8462 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && req->file) |
9882131c | 8463 | io_iopoll_req_issued(req, issue_flags); |
def596e9 JA |
8464 | |
8465 | return 0; | |
2b188cc1 JA |
8466 | } |
8467 | ||
ebc11b6c PB |
8468 | static struct io_wq_work *io_wq_free_work(struct io_wq_work *work) |
8469 | { | |
8470 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8471 | ||
8472 | req = io_put_req_find_next(req); | |
8473 | return req ? &req->work : NULL; | |
8474 | } | |
8475 | ||
5280f7e5 | 8476 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
8477 | { |
8478 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6bf9c47a | 8479 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
d01905db PB |
8480 | unsigned int issue_flags = IO_URING_F_UNLOCKED; |
8481 | bool needs_poll = false; | |
6bf9c47a | 8482 | int ret = 0, err = -ECANCELED; |
2b188cc1 | 8483 | |
48dcd38d PB |
8484 | /* one will be dropped by ->io_free_work() after returning to io-wq */ |
8485 | if (!(req->flags & REQ_F_REFCOUNT)) | |
8486 | __io_req_set_refcount(req, 2); | |
8487 | else | |
8488 | req_ref_get(req); | |
5d5901a3 | 8489 | |
cb2d344c | 8490 | io_arm_ltimeout(req); |
6bf9c47a | 8491 | |
dadebc35 | 8492 | /* either cancelled or io-wq is dying, so don't touch tctx->iowq */ |
d01905db | 8493 | if (work->flags & IO_WQ_WORK_CANCEL) { |
0f8da75b | 8494 | fail: |
6bf9c47a | 8495 | io_req_task_queue_fail(req, err); |
d01905db PB |
8496 | return; |
8497 | } | |
0f8da75b PB |
8498 | if (!io_assign_file(req, issue_flags)) { |
8499 | err = -EBADF; | |
8500 | work->flags |= IO_WQ_WORK_CANCEL; | |
8501 | goto fail; | |
8502 | } | |
31b51510 | 8503 | |
d01905db | 8504 | if (req->flags & REQ_F_FORCE_ASYNC) { |
afb7f56f PB |
8505 | bool opcode_poll = def->pollin || def->pollout; |
8506 | ||
8507 | if (opcode_poll && file_can_poll(req->file)) { | |
8508 | needs_poll = true; | |
d01905db | 8509 | issue_flags |= IO_URING_F_NONBLOCK; |
afb7f56f | 8510 | } |
561fb04a | 8511 | } |
31b51510 | 8512 | |
d01905db PB |
8513 | do { |
8514 | ret = io_issue_sqe(req, issue_flags); | |
8515 | if (ret != -EAGAIN) | |
8516 | break; | |
8517 | /* | |
8518 | * We can get EAGAIN for iopolled IO even though we're | |
8519 | * forcing a sync submission from here, since we can't | |
8520 | * wait for request slots on the block side. | |
8521 | */ | |
8522 | if (!needs_poll) { | |
e0deb6a0 PB |
8523 | if (!(req->ctx->flags & IORING_SETUP_IOPOLL)) |
8524 | break; | |
d01905db PB |
8525 | cond_resched(); |
8526 | continue; | |
90fa0288 HX |
8527 | } |
8528 | ||
4d9237e3 | 8529 | if (io_arm_poll_handler(req, issue_flags) == IO_APOLL_OK) |
d01905db PB |
8530 | return; |
8531 | /* aborted or ready, in either case retry blocking */ | |
8532 | needs_poll = false; | |
8533 | issue_flags &= ~IO_URING_F_NONBLOCK; | |
8534 | } while (1); | |
31b51510 | 8535 | |
a3df7698 | 8536 | /* avoid locking problems by failing it from a clean context */ |
5d5901a3 | 8537 | if (ret) |
a3df7698 | 8538 | io_req_task_queue_fail(req, ret); |
2b188cc1 JA |
8539 | } |
8540 | ||
aeca241b | 8541 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
042b0d85 | 8542 | unsigned i) |
65e19f54 | 8543 | { |
042b0d85 | 8544 | return &table->files[i]; |
dafecf19 PB |
8545 | } |
8546 | ||
65e19f54 JA |
8547 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
8548 | int index) | |
8549 | { | |
aeca241b | 8550 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
65e19f54 | 8551 | |
a04b0ac0 | 8552 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
8553 | } |
8554 | ||
a04b0ac0 | 8555 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
8556 | { |
8557 | unsigned long file_ptr = (unsigned long) file; | |
8558 | ||
88459b50 | 8559 | file_ptr |= io_file_get_flags(file); |
a04b0ac0 | 8560 | file_slot->file_ptr = file_ptr; |
65e19f54 JA |
8561 | } |
8562 | ||
5106dd6e JA |
8563 | static inline struct file *io_file_get_fixed(struct io_kiocb *req, int fd, |
8564 | unsigned int issue_flags) | |
09bb8394 | 8565 | { |
5106dd6e JA |
8566 | struct io_ring_ctx *ctx = req->ctx; |
8567 | struct file *file = NULL; | |
ac177053 | 8568 | unsigned long file_ptr; |
09bb8394 | 8569 | |
93f052cb | 8570 | io_ring_submit_lock(ctx, issue_flags); |
5106dd6e | 8571 | |
ac177053 | 8572 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
5106dd6e | 8573 | goto out; |
ac177053 PB |
8574 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8575 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; | |
8576 | file = (struct file *) (file_ptr & FFS_MASK); | |
8577 | file_ptr &= ~FFS_MASK; | |
8578 | /* mask in overlapping REQ_F and FFS bits */ | |
35645ac3 | 8579 | req->flags |= (file_ptr << REQ_F_SUPPORT_NOWAIT_BIT); |
5106dd6e | 8580 | io_req_set_rsrc_node(req, ctx, 0); |
d78bd8ad | 8581 | WARN_ON_ONCE(file && !test_bit(fd, ctx->file_table.bitmap)); |
5106dd6e | 8582 | out: |
93f052cb | 8583 | io_ring_submit_unlock(ctx, issue_flags); |
ac177053 PB |
8584 | return file; |
8585 | } | |
d44f554e | 8586 | |
5106dd6e | 8587 | static struct file *io_file_get_normal(struct io_kiocb *req, int fd) |
ac177053 | 8588 | { |
62906e89 | 8589 | struct file *file = fget(fd); |
ac177053 | 8590 | |
cef216fc | 8591 | trace_io_uring_file_get(req->ctx, req, req->cqe.user_data, fd); |
09bb8394 | 8592 | |
ac177053 | 8593 | /* we don't allow fixed io_uring files */ |
d5361233 | 8594 | if (file && file->f_op == &io_uring_fops) |
9cae36a0 | 8595 | io_req_track_inflight(req); |
8371adf5 | 8596 | return file; |
09bb8394 JA |
8597 | } |
8598 | ||
f237c30a | 8599 | static void io_req_task_link_timeout(struct io_kiocb *req, bool *locked) |
89b263f6 JA |
8600 | { |
8601 | struct io_kiocb *prev = req->timeout.prev; | |
617a8948 | 8602 | int ret = -ENOENT; |
89b263f6 JA |
8603 | |
8604 | if (prev) { | |
b21432b4 JA |
8605 | if (!(req->task->flags & PF_EXITING)) { |
8606 | struct io_cancel_data cd = { | |
8607 | .ctx = req->ctx, | |
8608 | .data = prev->cqe.user_data, | |
8609 | }; | |
8610 | ||
8611 | ret = io_try_cancel(req, &cd); | |
8612 | } | |
505657bc | 8613 | io_req_complete_post(req, ret ?: -ETIME, 0); |
89b263f6 | 8614 | io_put_req(prev); |
89b263f6 JA |
8615 | } else { |
8616 | io_req_complete_post(req, -ETIME, 0); | |
8617 | } | |
8618 | } | |
8619 | ||
2665abfd | 8620 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 8621 | { |
ad8a48ac JA |
8622 | struct io_timeout_data *data = container_of(timer, |
8623 | struct io_timeout_data, timer); | |
90cd7e42 | 8624 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 8625 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 8626 | unsigned long flags; |
2665abfd | 8627 | |
89b263f6 | 8628 | spin_lock_irqsave(&ctx->timeout_lock, flags); |
90cd7e42 PB |
8629 | prev = req->timeout.head; |
8630 | req->timeout.head = NULL; | |
2665abfd JA |
8631 | |
8632 | /* | |
8633 | * We don't expect the list to be empty, that will only happen if we | |
8634 | * race with the completion of the linked work. | |
8635 | */ | |
447c19f3 | 8636 | if (prev) { |
f2f87370 | 8637 | io_remove_next_linked(prev); |
447c19f3 PB |
8638 | if (!req_ref_inc_not_zero(prev)) |
8639 | prev = NULL; | |
8640 | } | |
ef9dd637 | 8641 | list_del(&req->timeout.list); |
89b263f6 JA |
8642 | req->timeout.prev = prev; |
8643 | spin_unlock_irqrestore(&ctx->timeout_lock, flags); | |
2665abfd | 8644 | |
89b263f6 | 8645 | req->io_task_work.func = io_req_task_link_timeout; |
3fe07bcd | 8646 | io_req_task_work_add(req); |
2665abfd JA |
8647 | return HRTIMER_NORESTART; |
8648 | } | |
8649 | ||
de968c18 | 8650 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 8651 | { |
de968c18 PB |
8652 | struct io_ring_ctx *ctx = req->ctx; |
8653 | ||
89b263f6 | 8654 | spin_lock_irq(&ctx->timeout_lock); |
76a46e06 | 8655 | /* |
f2f87370 PB |
8656 | * If the back reference is NULL, then our linked request finished |
8657 | * before we got a chance to setup the timer | |
76a46e06 | 8658 | */ |
90cd7e42 | 8659 | if (req->timeout.head) { |
e8c2bc1f | 8660 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 8661 | |
ad8a48ac JA |
8662 | data->timer.function = io_link_timeout_fn; |
8663 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
8664 | data->mode); | |
ef9dd637 | 8665 | list_add_tail(&req->timeout.list, &ctx->ltimeout_list); |
2665abfd | 8666 | } |
89b263f6 | 8667 | spin_unlock_irq(&ctx->timeout_lock); |
2665abfd | 8668 | /* drop submission reference */ |
76a46e06 JA |
8669 | io_put_req(req); |
8670 | } | |
2665abfd | 8671 | |
7bfa9bad | 8672 | static void io_queue_async(struct io_kiocb *req, int ret) |
d475a9a6 PB |
8673 | __must_hold(&req->ctx->uring_lock) |
8674 | { | |
7bfa9bad PB |
8675 | struct io_kiocb *linked_timeout; |
8676 | ||
8677 | if (ret != -EAGAIN || (req->flags & REQ_F_NOWAIT)) { | |
8678 | io_req_complete_failed(req, ret); | |
8679 | return; | |
8680 | } | |
8681 | ||
8682 | linked_timeout = io_prep_linked_timeout(req); | |
d475a9a6 | 8683 | |
4d9237e3 | 8684 | switch (io_arm_poll_handler(req, 0)) { |
d475a9a6 | 8685 | case IO_APOLL_READY: |
d475a9a6 PB |
8686 | io_req_task_queue(req); |
8687 | break; | |
8688 | case IO_APOLL_ABORTED: | |
8689 | /* | |
8690 | * Queued up for async execution, worker will release | |
8691 | * submit reference when the iocb is actually submitted. | |
8692 | */ | |
6436c770 | 8693 | io_kbuf_recycle(req, 0); |
77955efb | 8694 | io_queue_iowq(req, NULL); |
d475a9a6 | 8695 | break; |
b1c62645 | 8696 | case IO_APOLL_OK: |
b1c62645 | 8697 | break; |
d475a9a6 PB |
8698 | } |
8699 | ||
8700 | if (linked_timeout) | |
8701 | io_queue_linked_timeout(linked_timeout); | |
8702 | } | |
8703 | ||
cbc2e203 | 8704 | static inline void io_queue_sqe(struct io_kiocb *req) |
282cdc86 | 8705 | __must_hold(&req->ctx->uring_lock) |
2b188cc1 | 8706 | { |
e0c5c576 | 8707 | int ret; |
2b188cc1 | 8708 | |
c5eef2b9 | 8709 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 8710 | |
fff4e40e PB |
8711 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
8712 | io_req_add_compl_list(req); | |
d9f9d284 | 8713 | return; |
fff4e40e | 8714 | } |
491381ce JA |
8715 | /* |
8716 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
8717 | * doesn't support non-blocking read/write attempts | |
8718 | */ | |
7bfa9bad | 8719 | if (likely(!ret)) |
cb2d344c | 8720 | io_arm_ltimeout(req); |
7bfa9bad PB |
8721 | else |
8722 | io_queue_async(req, ret); | |
2b188cc1 JA |
8723 | } |
8724 | ||
4652fe3f | 8725 | static void io_queue_sqe_fallback(struct io_kiocb *req) |
282cdc86 | 8726 | __must_hold(&req->ctx->uring_lock) |
4fe2c963 | 8727 | { |
17b147f6 PB |
8728 | if (unlikely(req->flags & REQ_F_FAIL)) { |
8729 | /* | |
8730 | * We don't submit, fail them all, for that replace hardlinks | |
8731 | * with normal links. Extra REQ_F_LINK is tolerated. | |
8732 | */ | |
8733 | req->flags &= ~REQ_F_HARDLINK; | |
8734 | req->flags |= REQ_F_LINK; | |
8735 | io_req_complete_failed(req, req->cqe.res); | |
e0eb71dc PB |
8736 | } else if (unlikely(req->ctx->drain_active)) { |
8737 | io_drain_req(req); | |
76cc33d7 PB |
8738 | } else { |
8739 | int ret = io_req_prep_async(req); | |
8740 | ||
8741 | if (unlikely(ret)) | |
8742 | io_req_complete_failed(req, ret); | |
8743 | else | |
77955efb | 8744 | io_queue_iowq(req, NULL); |
ce35a47a | 8745 | } |
4fe2c963 JL |
8746 | } |
8747 | ||
b16fed66 PB |
8748 | /* |
8749 | * Check SQE restrictions (opcode and flags). | |
8750 | * | |
8751 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
8752 | */ | |
8753 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
8754 | struct io_kiocb *req, | |
8755 | unsigned int sqe_flags) | |
4fe2c963 | 8756 | { |
b16fed66 PB |
8757 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) |
8758 | return false; | |
8759 | ||
8760 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
8761 | ctx->restrictions.sqe_flags_required) | |
8762 | return false; | |
8763 | ||
8764 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
8765 | ctx->restrictions.sqe_flags_required)) | |
8766 | return false; | |
8767 | ||
8768 | return true; | |
4fe2c963 JL |
8769 | } |
8770 | ||
22b2ca31 PB |
8771 | static void io_init_req_drain(struct io_kiocb *req) |
8772 | { | |
8773 | struct io_ring_ctx *ctx = req->ctx; | |
8774 | struct io_kiocb *head = ctx->submit_state.link.head; | |
8775 | ||
8776 | ctx->drain_active = true; | |
8777 | if (head) { | |
8778 | /* | |
8779 | * If we need to drain a request in the middle of a link, drain | |
8780 | * the head request and the next request/link after the current | |
8781 | * link. Considering sequential execution of links, | |
b6c7db32 | 8782 | * REQ_F_IO_DRAIN will be maintained for every request of our |
22b2ca31 PB |
8783 | * link. |
8784 | */ | |
b6c7db32 | 8785 | head->flags |= REQ_F_IO_DRAIN | REQ_F_FORCE_ASYNC; |
22b2ca31 PB |
8786 | ctx->drain_next = true; |
8787 | } | |
8788 | } | |
8789 | ||
b16fed66 PB |
8790 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
8791 | const struct io_uring_sqe *sqe) | |
282cdc86 | 8792 | __must_hold(&ctx->uring_lock) |
b16fed66 | 8793 | { |
fcde59fe | 8794 | const struct io_op_def *def; |
b16fed66 | 8795 | unsigned int sqe_flags; |
fc0ae024 | 8796 | int personality; |
4a04d1d1 | 8797 | u8 opcode; |
b16fed66 | 8798 | |
864ea921 | 8799 | /* req is partially pre-initialised, see io_preinit_req() */ |
4a04d1d1 | 8800 | req->opcode = opcode = READ_ONCE(sqe->opcode); |
b16fed66 PB |
8801 | /* same numerical values with corresponding REQ_F_*, safe to copy */ |
8802 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
cef216fc | 8803 | req->cqe.user_data = READ_ONCE(sqe->user_data); |
b16fed66 | 8804 | req->file = NULL; |
c1bdf8ed | 8805 | req->rsrc_node = NULL; |
b16fed66 | 8806 | req->task = current; |
b16fed66 | 8807 | |
4a04d1d1 PB |
8808 | if (unlikely(opcode >= IORING_OP_LAST)) { |
8809 | req->opcode = 0; | |
b16fed66 | 8810 | return -EINVAL; |
4a04d1d1 | 8811 | } |
fcde59fe | 8812 | def = &io_op_defs[opcode]; |
68fe256a PB |
8813 | if (unlikely(sqe_flags & ~SQE_COMMON_FLAGS)) { |
8814 | /* enforce forwards compatibility on users */ | |
8815 | if (sqe_flags & ~SQE_VALID_FLAGS) | |
8816 | return -EINVAL; | |
4e906702 | 8817 | if (sqe_flags & IOSQE_BUFFER_SELECT) { |
fcde59fe | 8818 | if (!def->buffer_select) |
4e906702 JA |
8819 | return -EOPNOTSUPP; |
8820 | req->buf_index = READ_ONCE(sqe->buf_group); | |
8821 | } | |
5562a8d7 PB |
8822 | if (sqe_flags & IOSQE_CQE_SKIP_SUCCESS) |
8823 | ctx->drain_disabled = true; | |
8824 | if (sqe_flags & IOSQE_IO_DRAIN) { | |
8825 | if (ctx->drain_disabled) | |
8826 | return -EOPNOTSUPP; | |
22b2ca31 | 8827 | io_init_req_drain(req); |
5562a8d7 | 8828 | } |
2a56a9bd PB |
8829 | } |
8830 | if (unlikely(ctx->restricted || ctx->drain_active || ctx->drain_next)) { | |
8831 | if (ctx->restricted && !io_check_restriction(ctx, req, sqe_flags)) | |
8832 | return -EACCES; | |
8833 | /* knock it to the slow queue path, will be drained there */ | |
8834 | if (ctx->drain_active) | |
8835 | req->flags |= REQ_F_FORCE_ASYNC; | |
8836 | /* if there is no link, we're at "next" request and need to drain */ | |
8837 | if (unlikely(ctx->drain_next) && !ctx->submit_state.link.head) { | |
8838 | ctx->drain_next = false; | |
8839 | ctx->drain_active = true; | |
b6c7db32 | 8840 | req->flags |= REQ_F_IO_DRAIN | REQ_F_FORCE_ASYNC; |
2a56a9bd | 8841 | } |
68fe256a | 8842 | } |
b16fed66 | 8843 | |
fcde59fe | 8844 | if (!def->ioprio && sqe->ioprio) |
73911426 | 8845 | return -EINVAL; |
fcde59fe | 8846 | if (!def->iopoll && (ctx->flags & IORING_SETUP_IOPOLL)) |
73911426 JA |
8847 | return -EINVAL; |
8848 | ||
fcde59fe | 8849 | if (def->needs_file) { |
6d63416d PB |
8850 | struct io_submit_state *state = &ctx->submit_state; |
8851 | ||
cef216fc | 8852 | req->cqe.fd = READ_ONCE(sqe->fd); |
6bf9c47a | 8853 | |
6d63416d PB |
8854 | /* |
8855 | * Plug now if we have more than 2 IO left after this, and the | |
8856 | * target is potentially a read/write to block based storage. | |
8857 | */ | |
fcde59fe | 8858 | if (state->need_plug && def->plug) { |
6d63416d PB |
8859 | state->plug_started = true; |
8860 | state->need_plug = false; | |
5ca7a8b3 | 8861 | blk_start_plug_nr_ios(&state->plug, state->submit_nr); |
6d63416d | 8862 | } |
b16fed66 | 8863 | } |
863e0560 | 8864 | |
003e8dcc JA |
8865 | personality = READ_ONCE(sqe->personality); |
8866 | if (personality) { | |
cdab10bf LT |
8867 | int ret; |
8868 | ||
c10d1f98 PB |
8869 | req->creds = xa_load(&ctx->personalities, personality); |
8870 | if (!req->creds) | |
003e8dcc | 8871 | return -EINVAL; |
c10d1f98 | 8872 | get_cred(req->creds); |
cdc1404a PM |
8873 | ret = security_uring_override_creds(req->creds); |
8874 | if (ret) { | |
8875 | put_cred(req->creds); | |
8876 | return ret; | |
8877 | } | |
b8e64b53 | 8878 | req->flags |= REQ_F_CREDS; |
003e8dcc | 8879 | } |
b16fed66 | 8880 | |
fc0ae024 | 8881 | return io_req_prep(req, sqe); |
b16fed66 PB |
8882 | } |
8883 | ||
df3becde PB |
8884 | static __cold int io_submit_fail_init(const struct io_uring_sqe *sqe, |
8885 | struct io_kiocb *req, int ret) | |
8886 | { | |
8887 | struct io_ring_ctx *ctx = req->ctx; | |
8888 | struct io_submit_link *link = &ctx->submit_state.link; | |
8889 | struct io_kiocb *head = link->head; | |
8890 | ||
8891 | trace_io_uring_req_failed(sqe, ctx, req, ret); | |
8892 | ||
8893 | /* | |
8894 | * Avoid breaking links in the middle as it renders links with SQPOLL | |
8895 | * unusable. Instead of failing eagerly, continue assembling the link if | |
8896 | * applicable and mark the head with REQ_F_FAIL. The link flushing code | |
8897 | * should find the flag and handle the rest. | |
8898 | */ | |
8899 | req_fail_link_node(req, ret); | |
8900 | if (head && !(head->flags & REQ_F_FAIL)) | |
8901 | req_fail_link_node(head, -ECANCELED); | |
8902 | ||
8903 | if (!(req->flags & IO_REQ_LINK_FLAGS)) { | |
8904 | if (head) { | |
8905 | link->last->link = req; | |
8906 | link->head = NULL; | |
8907 | req = head; | |
8908 | } | |
8909 | io_queue_sqe_fallback(req); | |
8910 | return ret; | |
8911 | } | |
8912 | ||
8913 | if (head) | |
8914 | link->last->link = req; | |
8915 | else | |
8916 | link->head = req; | |
8917 | link->last = req; | |
8918 | return 0; | |
8919 | } | |
8920 | ||
8921 | static inline int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
a1ab7b35 | 8922 | const struct io_uring_sqe *sqe) |
282cdc86 | 8923 | __must_hold(&ctx->uring_lock) |
9e645e11 | 8924 | { |
a1ab7b35 | 8925 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 8926 | int ret; |
9e645e11 | 8927 | |
a6b8cadc | 8928 | ret = io_init_req(ctx, req, sqe); |
df3becde PB |
8929 | if (unlikely(ret)) |
8930 | return io_submit_fail_init(sqe, req, ret); | |
441b8a78 | 8931 | |
be7053b7 | 8932 | /* don't need @sqe from now on */ |
cef216fc | 8933 | trace_io_uring_submit_sqe(ctx, req, req->cqe.user_data, req->opcode, |
236daeae OL |
8934 | req->flags, true, |
8935 | ctx->flags & IORING_SETUP_SQPOLL); | |
a6b8cadc | 8936 | |
9e645e11 JA |
8937 | /* |
8938 | * If we already have a head request, queue this one for async | |
8939 | * submittal once the head completes. If we don't have a head but | |
8940 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
8941 | * submitted sync once the chain is complete. If none of those | |
8942 | * conditions are true (normal request), then just queue it. | |
8943 | */ | |
924a07e4 | 8944 | if (unlikely(link->head)) { |
df3becde PB |
8945 | ret = io_req_prep_async(req); |
8946 | if (unlikely(ret)) | |
8947 | return io_submit_fail_init(sqe, req, ret); | |
8948 | ||
8949 | trace_io_uring_link(ctx, req, link->head); | |
f2f87370 | 8950 | link->last->link = req; |
863e0560 | 8951 | link->last = req; |
32fe525b | 8952 | |
da1a08c5 | 8953 | if (req->flags & IO_REQ_LINK_FLAGS) |
f15a3431 | 8954 | return 0; |
df3becde PB |
8955 | /* last request of the link, flush it */ |
8956 | req = link->head; | |
f15a3431 | 8957 | link->head = NULL; |
924a07e4 PB |
8958 | if (req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)) |
8959 | goto fallback; | |
8960 | ||
8961 | } else if (unlikely(req->flags & (IO_REQ_LINK_FLAGS | | |
8962 | REQ_F_FORCE_ASYNC | REQ_F_FAIL))) { | |
8963 | if (req->flags & IO_REQ_LINK_FLAGS) { | |
8964 | link->head = req; | |
8965 | link->last = req; | |
8966 | } else { | |
8967 | fallback: | |
8968 | io_queue_sqe_fallback(req); | |
8969 | } | |
f15a3431 | 8970 | return 0; |
9e645e11 | 8971 | } |
2e6e1fde | 8972 | |
f15a3431 | 8973 | io_queue_sqe(req); |
1d4240cc | 8974 | return 0; |
9e645e11 JA |
8975 | } |
8976 | ||
9a56a232 JA |
8977 | /* |
8978 | * Batched submission is done, ensure local IO is flushed out. | |
8979 | */ | |
553deffd | 8980 | static void io_submit_state_end(struct io_ring_ctx *ctx) |
9a56a232 | 8981 | { |
553deffd PB |
8982 | struct io_submit_state *state = &ctx->submit_state; |
8983 | ||
e126391c PB |
8984 | if (unlikely(state->link.head)) |
8985 | io_queue_sqe_fallback(state->link.head); | |
553deffd | 8986 | /* flush only after queuing links as they can generate completions */ |
c450178d | 8987 | io_submit_flush_completions(ctx); |
27926b68 JA |
8988 | if (state->plug_started) |
8989 | blk_finish_plug(&state->plug); | |
9a56a232 JA |
8990 | } |
8991 | ||
8992 | /* | |
8993 | * Start submission side cache. | |
8994 | */ | |
8995 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 8996 | unsigned int max_ios) |
9a56a232 | 8997 | { |
27926b68 | 8998 | state->plug_started = false; |
4b628aeb | 8999 | state->need_plug = max_ios > 2; |
5ca7a8b3 | 9000 | state->submit_nr = max_ios; |
a1ab7b35 PB |
9001 | /* set only head, no need to init link_last in advance */ |
9002 | state->link.head = NULL; | |
9a56a232 JA |
9003 | } |
9004 | ||
2b188cc1 JA |
9005 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
9006 | { | |
75b28aff | 9007 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 9008 | |
caf582c6 PB |
9009 | /* |
9010 | * Ensure any loads from the SQEs are done at this point, | |
9011 | * since once we write the new head, the application could | |
9012 | * write new data to them. | |
9013 | */ | |
9014 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
9015 | } |
9016 | ||
2b188cc1 | 9017 | /* |
dd9ae8a0 | 9018 | * Fetch an sqe, if one is available. Note this returns a pointer to memory |
2b188cc1 JA |
9019 | * that is mapped by userspace. This means that care needs to be taken to |
9020 | * ensure that reads are stable, as we cannot rely on userspace always | |
9021 | * being a good citizen. If members of the sqe are validated and then later | |
9022 | * used, it's important that those reads are done through READ_ONCE() to | |
9023 | * prevent a re-load down the line. | |
9024 | */ | |
709b302f | 9025 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 9026 | { |
ea5ab3b5 | 9027 | unsigned head, mask = ctx->sq_entries - 1; |
17d3aeb3 | 9028 | unsigned sq_idx = ctx->cached_sq_head++ & mask; |
2b188cc1 JA |
9029 | |
9030 | /* | |
9031 | * The cached sq head (or cq tail) serves two purposes: | |
9032 | * | |
9033 | * 1) allows us to batch the cost of updating the user visible | |
9034 | * head updates. | |
9035 | * 2) allows the kernel side to track the head on its own, even | |
9036 | * though the application is the one updating it. | |
9037 | */ | |
17d3aeb3 | 9038 | head = READ_ONCE(ctx->sq_array[sq_idx]); |
ebdeb7c0 JA |
9039 | if (likely(head < ctx->sq_entries)) { |
9040 | /* double index for 128-byte SQEs, twice as long */ | |
9041 | if (ctx->flags & IORING_SETUP_SQE128) | |
9042 | head <<= 1; | |
709b302f | 9043 | return &ctx->sq_sqes[head]; |
ebdeb7c0 | 9044 | } |
2b188cc1 JA |
9045 | |
9046 | /* drop invalid entries */ | |
15641e42 PB |
9047 | ctx->cq_extra--; |
9048 | WRITE_ONCE(ctx->rings->sq_dropped, | |
9049 | READ_ONCE(ctx->rings->sq_dropped) + 1); | |
709b302f PB |
9050 | return NULL; |
9051 | } | |
9052 | ||
0f212204 | 9053 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
282cdc86 | 9054 | __must_hold(&ctx->uring_lock) |
6c271ce2 | 9055 | { |
69629809 | 9056 | unsigned int entries = io_sqring_entries(ctx); |
8e6971a8 PB |
9057 | unsigned int left; |
9058 | int ret; | |
6c271ce2 | 9059 | |
51d48dab | 9060 | if (unlikely(!entries)) |
69629809 | 9061 | return 0; |
ee7d46d9 | 9062 | /* make sure SQ entry isn't read before tail */ |
8e6971a8 PB |
9063 | ret = left = min3(nr, ctx->sq_entries, entries); |
9064 | io_get_task_refs(left); | |
9065 | io_submit_state_start(&ctx->submit_state, left); | |
6c271ce2 | 9066 | |
69629809 | 9067 | do { |
3529d8c2 | 9068 | const struct io_uring_sqe *sqe; |
196be95c | 9069 | struct io_kiocb *req; |
fb5ccc98 | 9070 | |
8e6971a8 | 9071 | if (unlikely(!io_alloc_req_refill(ctx))) |
fb5ccc98 | 9072 | break; |
a33ae9ce | 9073 | req = io_alloc_req(ctx); |
4fccfcbb PB |
9074 | sqe = io_get_sqe(ctx); |
9075 | if (unlikely(!sqe)) { | |
fa05457a | 9076 | io_req_add_to_cache(req, ctx); |
4fccfcbb PB |
9077 | break; |
9078 | } | |
6c271ce2 | 9079 | |
1cd15904 PB |
9080 | /* |
9081 | * Continue submitting even for sqe failure if the | |
9082 | * ring was setup with IORING_SETUP_SUBMIT_ALL | |
9083 | */ | |
9084 | if (unlikely(io_submit_sqe(ctx, req, sqe)) && | |
9085 | !(ctx->flags & IORING_SETUP_SUBMIT_ALL)) { | |
9086 | left--; | |
9087 | break; | |
bcbb7bf6 | 9088 | } |
1cd15904 | 9089 | } while (--left); |
9466f437 | 9090 | |
8e6971a8 PB |
9091 | if (unlikely(left)) { |
9092 | ret -= left; | |
9093 | /* try again if it submitted nothing and can't allocate a req */ | |
9094 | if (!ret && io_req_cache_empty(ctx)) | |
9095 | ret = -EAGAIN; | |
9096 | current->io_uring->cached_refs += left; | |
9466f437 | 9097 | } |
6c271ce2 | 9098 | |
553deffd | 9099 | io_submit_state_end(ctx); |
ae9428ca PB |
9100 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
9101 | io_commit_sqring(ctx); | |
8e6971a8 | 9102 | return ret; |
6c271ce2 JA |
9103 | } |
9104 | ||
e4b6d902 PB |
9105 | static inline bool io_sqd_events_pending(struct io_sq_data *sqd) |
9106 | { | |
9107 | return READ_ONCE(sqd->state); | |
9108 | } | |
9109 | ||
08369246 | 9110 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 9111 | { |
c8d1ba58 | 9112 | unsigned int to_submit; |
bdcd3eab | 9113 | int ret = 0; |
6c271ce2 | 9114 | |
c8d1ba58 | 9115 | to_submit = io_sqring_entries(ctx); |
e95eee2d | 9116 | /* if we're handling multiple rings, cap submit size for fairness */ |
4ce8ad95 OL |
9117 | if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE) |
9118 | to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE; | |
e95eee2d | 9119 | |
5eef4e87 | 9120 | if (!wq_list_empty(&ctx->iopoll_list) || to_submit) { |
948e1947 PB |
9121 | const struct cred *creds = NULL; |
9122 | ||
9123 | if (ctx->sq_creds != current_cred()) | |
9124 | creds = override_creds(ctx->sq_creds); | |
a4c0b3de | 9125 | |
c8d1ba58 | 9126 | mutex_lock(&ctx->uring_lock); |
5eef4e87 | 9127 | if (!wq_list_empty(&ctx->iopoll_list)) |
5ba3c874 | 9128 | io_do_iopoll(ctx, true); |
906a3c6f | 9129 | |
3b763ba1 PB |
9130 | /* |
9131 | * Don't submit if refs are dying, good for io_uring_register(), | |
9132 | * but also it is relied upon by io_ring_exit_work() | |
9133 | */ | |
0298ef96 PB |
9134 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
9135 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 9136 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 | 9137 | mutex_unlock(&ctx->uring_lock); |
cb318216 | 9138 | |
acfb381d PB |
9139 | if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) |
9140 | wake_up(&ctx->sqo_sq_wait); | |
948e1947 PB |
9141 | if (creds) |
9142 | revert_creds(creds); | |
acfb381d | 9143 | } |
6c271ce2 | 9144 | |
08369246 XW |
9145 | return ret; |
9146 | } | |
6c271ce2 | 9147 | |
c072481d | 9148 | static __cold void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
08369246 XW |
9149 | { |
9150 | struct io_ring_ctx *ctx; | |
9151 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 9152 | |
c9dca27d PB |
9153 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
9154 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 9155 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 9156 | } |
6c271ce2 | 9157 | |
e4b6d902 PB |
9158 | static bool io_sqd_handle_event(struct io_sq_data *sqd) |
9159 | { | |
9160 | bool did_sig = false; | |
9161 | struct ksignal ksig; | |
9162 | ||
9163 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || | |
9164 | signal_pending(current)) { | |
9165 | mutex_unlock(&sqd->lock); | |
9166 | if (signal_pending(current)) | |
9167 | did_sig = get_signal(&ksig); | |
9168 | cond_resched(); | |
9169 | mutex_lock(&sqd->lock); | |
9170 | } | |
e4b6d902 PB |
9171 | return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
9172 | } | |
9173 | ||
c8d1ba58 JA |
9174 | static int io_sq_thread(void *data) |
9175 | { | |
69fb2131 JA |
9176 | struct io_sq_data *sqd = data; |
9177 | struct io_ring_ctx *ctx; | |
a0d9205f | 9178 | unsigned long timeout = 0; |
37d1e2e3 | 9179 | char buf[TASK_COMM_LEN]; |
08369246 | 9180 | DEFINE_WAIT(wait); |
6c271ce2 | 9181 | |
696ee88a | 9182 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 9183 | set_task_comm(current, buf); |
37d1e2e3 JA |
9184 | |
9185 | if (sqd->sq_cpu != -1) | |
9186 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
9187 | else | |
9188 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
9189 | current->flags |= PF_NO_SETAFFINITY; | |
9190 | ||
5bd2182d PM |
9191 | audit_alloc_kernel(current); |
9192 | ||
09a6f4ef | 9193 | mutex_lock(&sqd->lock); |
e4b6d902 | 9194 | while (1) { |
1a924a80 | 9195 | bool cap_entries, sqt_spin = false; |
c1edbf5f | 9196 | |
e4b6d902 PB |
9197 | if (io_sqd_events_pending(sqd) || signal_pending(current)) { |
9198 | if (io_sqd_handle_event(sqd)) | |
c7d95613 | 9199 | break; |
08369246 XW |
9200 | timeout = jiffies + sqd->sq_thread_idle; |
9201 | } | |
e4b6d902 | 9202 | |
e95eee2d | 9203 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 9204 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
948e1947 | 9205 | int ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a | 9206 | |
5eef4e87 | 9207 | if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list))) |
08369246 | 9208 | sqt_spin = true; |
69fb2131 | 9209 | } |
dd432ea5 PB |
9210 | if (io_run_task_work()) |
9211 | sqt_spin = true; | |
6c271ce2 | 9212 | |
08369246 | 9213 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 9214 | cond_resched(); |
08369246 XW |
9215 | if (sqt_spin) |
9216 | timeout = jiffies + sqd->sq_thread_idle; | |
9217 | continue; | |
9218 | } | |
9219 | ||
08369246 | 9220 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
7f62d40d | 9221 | if (!io_sqd_events_pending(sqd) && !task_work_pending(current)) { |
1a924a80 PB |
9222 | bool needs_sched = true; |
9223 | ||
724cb4f9 | 9224 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
3a4b89a2 JA |
9225 | atomic_or(IORING_SQ_NEED_WAKEUP, |
9226 | &ctx->rings->sq_flags); | |
724cb4f9 | 9227 | if ((ctx->flags & IORING_SETUP_IOPOLL) && |
5eef4e87 | 9228 | !wq_list_empty(&ctx->iopoll_list)) { |
724cb4f9 HX |
9229 | needs_sched = false; |
9230 | break; | |
9231 | } | |
649bb75d AK |
9232 | |
9233 | /* | |
9234 | * Ensure the store of the wakeup flag is not | |
9235 | * reordered with the load of the SQ tail | |
9236 | */ | |
f2e030dd | 9237 | smp_mb__after_atomic(); |
649bb75d | 9238 | |
724cb4f9 HX |
9239 | if (io_sqring_entries(ctx)) { |
9240 | needs_sched = false; | |
9241 | break; | |
9242 | } | |
9243 | } | |
9244 | ||
9245 | if (needs_sched) { | |
9246 | mutex_unlock(&sqd->lock); | |
9247 | schedule(); | |
9248 | mutex_lock(&sqd->lock); | |
9249 | } | |
69fb2131 | 9250 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
3a4b89a2 JA |
9251 | atomic_andnot(IORING_SQ_NEED_WAKEUP, |
9252 | &ctx->rings->sq_flags); | |
6c271ce2 | 9253 | } |
08369246 XW |
9254 | |
9255 | finish_wait(&sqd->wait, &wait); | |
9256 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 9257 | } |
28cea78a | 9258 | |
78cc687b | 9259 | io_uring_cancel_generic(true, sqd); |
37d1e2e3 | 9260 | sqd->thread = NULL; |
05962f95 | 9261 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
3a4b89a2 | 9262 | atomic_or(IORING_SQ_NEED_WAKEUP, &ctx->rings->sq_flags); |
521d6a73 | 9263 | io_run_task_work(); |
734551df PB |
9264 | mutex_unlock(&sqd->lock); |
9265 | ||
5bd2182d PM |
9266 | audit_free(current); |
9267 | ||
37d1e2e3 JA |
9268 | complete(&sqd->exited); |
9269 | do_exit(0); | |
6c271ce2 JA |
9270 | } |
9271 | ||
bda52162 JA |
9272 | struct io_wait_queue { |
9273 | struct wait_queue_entry wq; | |
9274 | struct io_ring_ctx *ctx; | |
5fd46178 | 9275 | unsigned cq_tail; |
bda52162 JA |
9276 | unsigned nr_timeouts; |
9277 | }; | |
9278 | ||
6c503150 | 9279 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
9280 | { |
9281 | struct io_ring_ctx *ctx = iowq->ctx; | |
5fd46178 | 9282 | int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; |
bda52162 JA |
9283 | |
9284 | /* | |
d195a66e | 9285 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
9286 | * started waiting. For timeouts, we always want to return to userspace, |
9287 | * regardless of event count. | |
9288 | */ | |
5fd46178 | 9289 | return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
bda52162 JA |
9290 | } |
9291 | ||
9292 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
9293 | int wake_flags, void *key) | |
9294 | { | |
9295 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
9296 | wq); | |
9297 | ||
6c503150 PB |
9298 | /* |
9299 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
9300 | * the task, and the next invocation will do it. | |
9301 | */ | |
10988a0a DY |
9302 | if (io_should_wake(iowq) || |
9303 | test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &iowq->ctx->check_cq)) | |
6c503150 PB |
9304 | return autoremove_wake_function(curr, mode, wake_flags, key); |
9305 | return -1; | |
bda52162 JA |
9306 | } |
9307 | ||
af9c1a44 JA |
9308 | static int io_run_task_work_sig(void) |
9309 | { | |
9310 | if (io_run_task_work()) | |
9311 | return 1; | |
0b8cfa97 | 9312 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 9313 | return -ERESTARTSYS; |
c5020bc8 OL |
9314 | if (task_sigpending(current)) |
9315 | return -EINTR; | |
9316 | return 0; | |
af9c1a44 JA |
9317 | } |
9318 | ||
eeb60b9a PB |
9319 | /* when returns >0, the caller should retry */ |
9320 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
9321 | struct io_wait_queue *iowq, | |
22833966 | 9322 | ktime_t timeout) |
eeb60b9a PB |
9323 | { |
9324 | int ret; | |
155bc950 | 9325 | unsigned long check_cq; |
eeb60b9a PB |
9326 | |
9327 | /* make sure we run task_work before checking for signals */ | |
9328 | ret = io_run_task_work_sig(); | |
9329 | if (ret || io_should_wake(iowq)) | |
9330 | return ret; | |
155bc950 | 9331 | check_cq = READ_ONCE(ctx->check_cq); |
eeb60b9a | 9332 | /* let the caller flush overflows, retry */ |
155bc950 | 9333 | if (check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT)) |
eeb60b9a | 9334 | return 1; |
155bc950 DY |
9335 | if (unlikely(check_cq & BIT(IO_CHECK_CQ_DROPPED_BIT))) |
9336 | return -EBADR; | |
22833966 JA |
9337 | if (!schedule_hrtimeout(&timeout, HRTIMER_MODE_ABS)) |
9338 | return -ETIME; | |
9339 | return 1; | |
eeb60b9a PB |
9340 | } |
9341 | ||
2b188cc1 JA |
9342 | /* |
9343 | * Wait until events become available, if we don't already have some. The | |
9344 | * application must reap them itself, as they reside on the shared cq ring. | |
9345 | */ | |
9346 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
9347 | const sigset_t __user *sig, size_t sigsz, |
9348 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 9349 | { |
90291099 | 9350 | struct io_wait_queue iowq; |
75b28aff | 9351 | struct io_rings *rings = ctx->rings; |
22833966 | 9352 | ktime_t timeout = KTIME_MAX; |
c1d5a224 | 9353 | int ret; |
2b188cc1 | 9354 | |
b41e9852 | 9355 | do { |
90f67366 | 9356 | io_cqring_overflow_flush(ctx); |
6c503150 | 9357 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 9358 | return 0; |
4c6e277c | 9359 | if (!io_run_task_work()) |
b41e9852 | 9360 | break; |
b41e9852 | 9361 | } while (1); |
2b188cc1 JA |
9362 | |
9363 | if (sig) { | |
9e75ad5d AB |
9364 | #ifdef CONFIG_COMPAT |
9365 | if (in_compat_syscall()) | |
9366 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 9367 | sigsz); |
9e75ad5d AB |
9368 | else |
9369 | #endif | |
b772434b | 9370 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 9371 | |
2b188cc1 JA |
9372 | if (ret) |
9373 | return ret; | |
9374 | } | |
9375 | ||
950e79dd OL |
9376 | if (uts) { |
9377 | struct timespec64 ts; | |
9378 | ||
9379 | if (get_timespec64(&ts, uts)) | |
9380 | return -EFAULT; | |
9381 | timeout = ktime_add_ns(timespec64_to_ktime(ts), ktime_get_ns()); | |
9382 | } | |
9383 | ||
90291099 PB |
9384 | init_waitqueue_func_entry(&iowq.wq, io_wake_function); |
9385 | iowq.wq.private = current; | |
9386 | INIT_LIST_HEAD(&iowq.wq.entry); | |
9387 | iowq.ctx = ctx; | |
bda52162 | 9388 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
5fd46178 | 9389 | iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; |
90291099 | 9390 | |
c826bd7a | 9391 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 9392 | do { |
ca0a2651 | 9393 | /* if we can't even flush overflow, don't wait for more */ |
90f67366 | 9394 | if (!io_cqring_overflow_flush(ctx)) { |
ca0a2651 JA |
9395 | ret = -EBUSY; |
9396 | break; | |
9397 | } | |
311997b3 | 9398 | prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, |
bda52162 | 9399 | TASK_INTERRUPTIBLE); |
22833966 | 9400 | ret = io_cqring_wait_schedule(ctx, &iowq, timeout); |
ca0a2651 | 9401 | cond_resched(); |
eeb60b9a | 9402 | } while (ret > 0); |
bda52162 | 9403 | |
b4f20bb4 | 9404 | finish_wait(&ctx->cq_wait, &iowq.wq); |
b7db41c9 | 9405 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 9406 | |
75b28aff | 9407 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
9408 | } |
9409 | ||
9123c8ff | 9410 | static void io_free_page_table(void **table, size_t size) |
05f3fb3c | 9411 | { |
9123c8ff | 9412 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); |
05f3fb3c | 9413 | |
846a4ef2 | 9414 | for (i = 0; i < nr_tables; i++) |
9123c8ff PB |
9415 | kfree(table[i]); |
9416 | kfree(table); | |
9417 | } | |
9418 | ||
c072481d | 9419 | static __cold void **io_alloc_page_table(size_t size) |
9123c8ff PB |
9420 | { |
9421 | unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE); | |
9422 | size_t init_size = size; | |
9423 | void **table; | |
9424 | ||
0bea96f5 | 9425 | table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
9426 | if (!table) |
9427 | return NULL; | |
9428 | ||
9429 | for (i = 0; i < nr_tables; i++) { | |
27f6b318 | 9430 | unsigned int this_size = min_t(size_t, size, PAGE_SIZE); |
9123c8ff | 9431 | |
0bea96f5 | 9432 | table[i] = kzalloc(this_size, GFP_KERNEL_ACCOUNT); |
9123c8ff PB |
9433 | if (!table[i]) { |
9434 | io_free_page_table(table, init_size); | |
9435 | return NULL; | |
9436 | } | |
9437 | size -= this_size; | |
9438 | } | |
9439 | return table; | |
05f3fb3c JA |
9440 | } |
9441 | ||
28a9fe25 | 9442 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
1642b445 | 9443 | { |
28a9fe25 PB |
9444 | percpu_ref_exit(&ref_node->refs); |
9445 | kfree(ref_node); | |
1642b445 PB |
9446 | } |
9447 | ||
c072481d | 9448 | static __cold void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
b9bd2bea PB |
9449 | { |
9450 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); | |
9451 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; | |
9452 | unsigned long flags; | |
9453 | bool first_add = false; | |
b36a2050 | 9454 | unsigned long delay = HZ; |
b9bd2bea PB |
9455 | |
9456 | spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); | |
9457 | node->done = true; | |
9458 | ||
b36a2050 DY |
9459 | /* if we are mid-quiesce then do not delay */ |
9460 | if (node->rsrc_data->quiesce) | |
9461 | delay = 0; | |
9462 | ||
b9bd2bea PB |
9463 | while (!list_empty(&ctx->rsrc_ref_list)) { |
9464 | node = list_first_entry(&ctx->rsrc_ref_list, | |
9465 | struct io_rsrc_node, node); | |
9466 | /* recycle ref nodes in order */ | |
9467 | if (!node->done) | |
9468 | break; | |
9469 | list_del(&node->node); | |
9470 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
9471 | } | |
9472 | spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); | |
9473 | ||
9474 | if (first_add) | |
b36a2050 | 9475 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
b9bd2bea PB |
9476 | } |
9477 | ||
f6133fbd | 9478 | static struct io_rsrc_node *io_rsrc_node_alloc(void) |
b9bd2bea PB |
9479 | { |
9480 | struct io_rsrc_node *ref_node; | |
9481 | ||
9482 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); | |
9483 | if (!ref_node) | |
9484 | return NULL; | |
9485 | ||
9486 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, | |
9487 | 0, GFP_KERNEL)) { | |
9488 | kfree(ref_node); | |
9489 | return NULL; | |
9490 | } | |
9491 | INIT_LIST_HEAD(&ref_node->node); | |
9492 | INIT_LIST_HEAD(&ref_node->rsrc_list); | |
9493 | ref_node->done = false; | |
9494 | return ref_node; | |
9495 | } | |
9496 | ||
a7f0ed5a PB |
9497 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
9498 | struct io_rsrc_data *data_to_kill) | |
ab409402 | 9499 | __must_hold(&ctx->uring_lock) |
6b06314c | 9500 | { |
a7f0ed5a PB |
9501 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
9502 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
6b06314c | 9503 | |
ab409402 PB |
9504 | io_rsrc_refs_drop(ctx); |
9505 | ||
a7f0ed5a PB |
9506 | if (data_to_kill) { |
9507 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 9508 | |
a7f0ed5a | 9509 | rsrc_node->rsrc_data = data_to_kill; |
4956b9ea | 9510 | spin_lock_irq(&ctx->rsrc_ref_lock); |
a7f0ed5a | 9511 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); |
4956b9ea | 9512 | spin_unlock_irq(&ctx->rsrc_ref_lock); |
82fbcfa9 | 9513 | |
3e942498 | 9514 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
9515 | percpu_ref_kill(&rsrc_node->refs); |
9516 | ctx->rsrc_node = NULL; | |
9517 | } | |
6b06314c | 9518 | |
a7f0ed5a PB |
9519 | if (!ctx->rsrc_node) { |
9520 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
9521 | ctx->rsrc_backup_node = NULL; | |
9522 | } | |
8bad28d8 HX |
9523 | } |
9524 | ||
a7f0ed5a | 9525 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
9526 | { |
9527 | if (ctx->rsrc_backup_node) | |
9528 | return 0; | |
f6133fbd | 9529 | ctx->rsrc_backup_node = io_rsrc_node_alloc(); |
8dd03afe | 9530 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
8bad28d8 HX |
9531 | } |
9532 | ||
c072481d PB |
9533 | static __cold int io_rsrc_ref_quiesce(struct io_rsrc_data *data, |
9534 | struct io_ring_ctx *ctx) | |
8bad28d8 HX |
9535 | { |
9536 | int ret; | |
05589553 | 9537 | |
215c3902 | 9538 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
9539 | if (data->quiesce) |
9540 | return -ENXIO; | |
05589553 | 9541 | |
8bad28d8 | 9542 | data->quiesce = true; |
1ffc5422 | 9543 | do { |
a7f0ed5a | 9544 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 9545 | if (ret) |
f2303b1f | 9546 | break; |
a7f0ed5a | 9547 | io_rsrc_node_switch(ctx, data); |
f2303b1f | 9548 | |
3e942498 PB |
9549 | /* kill initial ref, already quiesced if zero */ |
9550 | if (atomic_dec_and_test(&data->refs)) | |
9551 | break; | |
c018db4a | 9552 | mutex_unlock(&ctx->uring_lock); |
8bad28d8 | 9553 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 | 9554 | ret = wait_for_completion_interruptible(&data->done); |
c018db4a JA |
9555 | if (!ret) { |
9556 | mutex_lock(&ctx->uring_lock); | |
80912cef DY |
9557 | if (atomic_read(&data->refs) > 0) { |
9558 | /* | |
9559 | * it has been revived by another thread while | |
9560 | * we were unlocked | |
9561 | */ | |
9562 | mutex_unlock(&ctx->uring_lock); | |
9563 | } else { | |
9564 | break; | |
9565 | } | |
c018db4a | 9566 | } |
8bad28d8 | 9567 | |
3e942498 PB |
9568 | atomic_inc(&data->refs); |
9569 | /* wait for all works potentially completing data->done */ | |
9570 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 9571 | reinit_completion(&data->done); |
8dd03afe | 9572 | |
1ffc5422 | 9573 | ret = io_run_task_work_sig(); |
8bad28d8 | 9574 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 9575 | } while (ret >= 0); |
8bad28d8 | 9576 | data->quiesce = false; |
05f3fb3c | 9577 | |
8bad28d8 | 9578 | return ret; |
d7954b2b BM |
9579 | } |
9580 | ||
2d091d62 PB |
9581 | static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx) |
9582 | { | |
9583 | unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK; | |
9584 | unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT; | |
9585 | ||
9586 | return &data->tags[table_idx][off]; | |
9587 | } | |
9588 | ||
44b31f2f | 9589 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 | 9590 | { |
2d091d62 PB |
9591 | size_t size = data->nr * sizeof(data->tags[0][0]); |
9592 | ||
9593 | if (data->tags) | |
9594 | io_free_page_table((void **)data->tags, size); | |
44b31f2f PB |
9595 | kfree(data); |
9596 | } | |
9597 | ||
c072481d PB |
9598 | static __cold int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put, |
9599 | u64 __user *utags, unsigned nr, | |
9600 | struct io_rsrc_data **pdata) | |
1ad555c6 | 9601 | { |
b895c9a6 | 9602 | struct io_rsrc_data *data; |
2d091d62 | 9603 | int ret = -ENOMEM; |
d878c816 | 9604 | unsigned i; |
1ad555c6 BM |
9605 | |
9606 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
9607 | if (!data) | |
d878c816 | 9608 | return -ENOMEM; |
2d091d62 | 9609 | data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); |
b60c8dce | 9610 | if (!data->tags) { |
1ad555c6 | 9611 | kfree(data); |
d878c816 PB |
9612 | return -ENOMEM; |
9613 | } | |
2d091d62 PB |
9614 | |
9615 | data->nr = nr; | |
9616 | data->ctx = ctx; | |
9617 | data->do_put = do_put; | |
d878c816 | 9618 | if (utags) { |
2d091d62 | 9619 | ret = -EFAULT; |
d878c816 | 9620 | for (i = 0; i < nr; i++) { |
fdd1dc31 CIK |
9621 | u64 *tag_slot = io_get_tag_slot(data, i); |
9622 | ||
9623 | if (copy_from_user(tag_slot, &utags[i], | |
9624 | sizeof(*tag_slot))) | |
2d091d62 | 9625 | goto fail; |
d878c816 | 9626 | } |
1ad555c6 | 9627 | } |
b60c8dce | 9628 | |
3e942498 | 9629 | atomic_set(&data->refs, 1); |
1ad555c6 | 9630 | init_completion(&data->done); |
d878c816 PB |
9631 | *pdata = data; |
9632 | return 0; | |
2d091d62 PB |
9633 | fail: |
9634 | io_rsrc_data_free(data); | |
9635 | return ret; | |
1ad555c6 BM |
9636 | } |
9637 | ||
9123c8ff PB |
9638 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
9639 | { | |
0bea96f5 PB |
9640 | table->files = kvcalloc(nr_files, sizeof(table->files[0]), |
9641 | GFP_KERNEL_ACCOUNT); | |
d78bd8ad JA |
9642 | if (unlikely(!table->files)) |
9643 | return false; | |
9644 | ||
9645 | table->bitmap = bitmap_zalloc(nr_files, GFP_KERNEL_ACCOUNT); | |
9646 | if (unlikely(!table->bitmap)) { | |
9647 | kvfree(table->files); | |
9648 | return false; | |
9649 | } | |
9650 | ||
9651 | return true; | |
9123c8ff PB |
9652 | } |
9653 | ||
042b0d85 | 9654 | static void io_free_file_tables(struct io_file_table *table) |
9123c8ff | 9655 | { |
042b0d85 | 9656 | kvfree(table->files); |
d78bd8ad | 9657 | bitmap_free(table->bitmap); |
9123c8ff | 9658 | table->files = NULL; |
d78bd8ad JA |
9659 | table->bitmap = NULL; |
9660 | } | |
9661 | ||
9662 | static inline void io_file_bitmap_set(struct io_file_table *table, int bit) | |
9663 | { | |
9664 | WARN_ON_ONCE(test_bit(bit, table->bitmap)); | |
9665 | __set_bit(bit, table->bitmap); | |
4278a0de | 9666 | table->alloc_hint = bit + 1; |
d78bd8ad JA |
9667 | } |
9668 | ||
9669 | static inline void io_file_bitmap_clear(struct io_file_table *table, int bit) | |
9670 | { | |
9671 | __clear_bit(bit, table->bitmap); | |
b70b8e33 | 9672 | table->alloc_hint = bit; |
9123c8ff PB |
9673 | } |
9674 | ||
fff4db76 | 9675 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
1ad555c6 | 9676 | { |
69cc1b6f | 9677 | #if !defined(IO_URING_SCM_ALL) |
1f59bc0f PB |
9678 | int i; |
9679 | ||
9680 | for (i = 0; i < ctx->nr_user_files; i++) { | |
9681 | struct file *file = io_file_from_index(ctx, i); | |
9682 | ||
5e45690a JA |
9683 | if (!file) |
9684 | continue; | |
9685 | if (io_fixed_file_slot(&ctx->file_table, i)->file_ptr & FFS_SCM) | |
1f59bc0f | 9686 | continue; |
d78bd8ad | 9687 | io_file_bitmap_clear(&ctx->file_table, i); |
1f59bc0f PB |
9688 | fput(file); |
9689 | } | |
5e45690a | 9690 | #endif |
1f59bc0f | 9691 | |
fff4db76 PB |
9692 | #if defined(CONFIG_UNIX) |
9693 | if (ctx->ring_sock) { | |
9694 | struct sock *sock = ctx->ring_sock->sk; | |
9695 | struct sk_buff *skb; | |
9696 | ||
9697 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
9698 | kfree_skb(skb); | |
9699 | } | |
fff4db76 | 9700 | #endif |
042b0d85 | 9701 | io_free_file_tables(&ctx->file_table); |
44b31f2f | 9702 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
9703 | ctx->file_data = NULL; |
9704 | ctx->nr_user_files = 0; | |
1ad555c6 BM |
9705 | } |
9706 | ||
d7954b2b BM |
9707 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
9708 | { | |
b0380bf6 | 9709 | unsigned nr = ctx->nr_user_files; |
d7954b2b BM |
9710 | int ret; |
9711 | ||
08480400 | 9712 | if (!ctx->file_data) |
d7954b2b | 9713 | return -ENXIO; |
b0380bf6 PB |
9714 | |
9715 | /* | |
9716 | * Quiesce may unlock ->uring_lock, and while it's not held | |
9717 | * prevent new requests using the table. | |
9718 | */ | |
9719 | ctx->nr_user_files = 0; | |
08480400 | 9720 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
b0380bf6 | 9721 | ctx->nr_user_files = nr; |
08480400 PB |
9722 | if (!ret) |
9723 | __io_sqe_files_unregister(ctx); | |
9724 | return ret; | |
6b06314c JA |
9725 | } |
9726 | ||
37d1e2e3 | 9727 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 9728 | __releases(&sqd->lock) |
37d1e2e3 | 9729 | { |
521d6a73 PB |
9730 | WARN_ON_ONCE(sqd->thread == current); |
9731 | ||
9e138a48 PB |
9732 | /* |
9733 | * Do the dance but not conditional clear_bit() because it'd race with | |
9734 | * other threads incrementing park_pending and setting the bit. | |
9735 | */ | |
37d1e2e3 | 9736 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
9737 | if (atomic_dec_return(&sqd->park_pending)) |
9738 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 9739 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
9740 | } |
9741 | ||
86e0d676 | 9742 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 9743 | __acquires(&sqd->lock) |
37d1e2e3 | 9744 | { |
521d6a73 PB |
9745 | WARN_ON_ONCE(sqd->thread == current); |
9746 | ||
9e138a48 | 9747 | atomic_inc(&sqd->park_pending); |
86e0d676 | 9748 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 9749 | mutex_lock(&sqd->lock); |
05962f95 | 9750 | if (sqd->thread) |
86e0d676 | 9751 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
9752 | } |
9753 | ||
9754 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
9755 | { | |
521d6a73 | 9756 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 9757 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 9758 | |
05962f95 | 9759 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 9760 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
9761 | if (sqd->thread) |
9762 | wake_up_process(sqd->thread); | |
09a6f4ef | 9763 | mutex_unlock(&sqd->lock); |
05962f95 | 9764 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
9765 | } |
9766 | ||
534ca6d6 | 9767 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 9768 | { |
534ca6d6 | 9769 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
9770 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
9771 | ||
37d1e2e3 JA |
9772 | io_sq_thread_stop(sqd); |
9773 | kfree(sqd); | |
9774 | } | |
9775 | } | |
9776 | ||
9777 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
9778 | { | |
9779 | struct io_sq_data *sqd = ctx->sq_data; | |
9780 | ||
9781 | if (sqd) { | |
05962f95 | 9782 | io_sq_thread_park(sqd); |
521d6a73 | 9783 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 9784 | io_sqd_update_thread_idle(sqd); |
05962f95 | 9785 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
9786 | |
9787 | io_put_sq_data(sqd); | |
9788 | ctx->sq_data = NULL; | |
534ca6d6 JA |
9789 | } |
9790 | } | |
9791 | ||
aa06165d JA |
9792 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
9793 | { | |
9794 | struct io_ring_ctx *ctx_attach; | |
9795 | struct io_sq_data *sqd; | |
9796 | struct fd f; | |
9797 | ||
9798 | f = fdget(p->wq_fd); | |
9799 | if (!f.file) | |
9800 | return ERR_PTR(-ENXIO); | |
9801 | if (f.file->f_op != &io_uring_fops) { | |
9802 | fdput(f); | |
9803 | return ERR_PTR(-EINVAL); | |
9804 | } | |
9805 | ||
9806 | ctx_attach = f.file->private_data; | |
9807 | sqd = ctx_attach->sq_data; | |
9808 | if (!sqd) { | |
9809 | fdput(f); | |
9810 | return ERR_PTR(-EINVAL); | |
9811 | } | |
5c2469e0 JA |
9812 | if (sqd->task_tgid != current->tgid) { |
9813 | fdput(f); | |
9814 | return ERR_PTR(-EPERM); | |
9815 | } | |
aa06165d JA |
9816 | |
9817 | refcount_inc(&sqd->refs); | |
9818 | fdput(f); | |
9819 | return sqd; | |
9820 | } | |
9821 | ||
26984fbf PB |
9822 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
9823 | bool *attached) | |
534ca6d6 JA |
9824 | { |
9825 | struct io_sq_data *sqd; | |
9826 | ||
26984fbf | 9827 | *attached = false; |
5c2469e0 JA |
9828 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
9829 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
9830 | if (!IS_ERR(sqd)) { |
9831 | *attached = true; | |
5c2469e0 | 9832 | return sqd; |
26984fbf | 9833 | } |
5c2469e0 JA |
9834 | /* fall through for EPERM case, setup new sqd/task */ |
9835 | if (PTR_ERR(sqd) != -EPERM) | |
9836 | return sqd; | |
9837 | } | |
aa06165d | 9838 | |
534ca6d6 JA |
9839 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
9840 | if (!sqd) | |
9841 | return ERR_PTR(-ENOMEM); | |
9842 | ||
9e138a48 | 9843 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 9844 | refcount_set(&sqd->refs, 1); |
69fb2131 | 9845 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 9846 | mutex_init(&sqd->lock); |
534ca6d6 | 9847 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 9848 | init_completion(&sqd->exited); |
534ca6d6 JA |
9849 | return sqd; |
9850 | } | |
9851 | ||
6b06314c JA |
9852 | /* |
9853 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
9854 | * the io_uring can be safely unregistered on process exit, even if we have | |
1f59bc0f PB |
9855 | * loops in the file referencing. We account only files that can hold other |
9856 | * files because otherwise they can't form a loop and so are not interesting | |
9857 | * for GC. | |
6b06314c | 9858 | */ |
8b3171bd | 9859 | static int io_scm_file_account(struct io_ring_ctx *ctx, struct file *file) |
6b06314c | 9860 | { |
73b25d3b | 9861 | #if defined(CONFIG_UNIX) |
6b06314c | 9862 | struct sock *sk = ctx->ring_sock->sk; |
73b25d3b | 9863 | struct sk_buff_head *head = &sk->sk_receive_queue; |
6b06314c JA |
9864 | struct scm_fp_list *fpl; |
9865 | struct sk_buff *skb; | |
6b06314c | 9866 | |
73b25d3b PB |
9867 | if (likely(!io_file_need_scm(file))) |
9868 | return 0; | |
6b06314c | 9869 | |
73b25d3b PB |
9870 | /* |
9871 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
9872 | * file set. If there's no room, fall back to allocating a new skb | |
9873 | * and filling it in. | |
9874 | */ | |
9875 | spin_lock_irq(&head->lock); | |
9876 | skb = skb_peek(head); | |
9877 | if (skb && UNIXCB(skb).fp->count < SCM_MAX_FD) | |
9878 | __skb_unlink(skb, head); | |
9879 | else | |
9880 | skb = NULL; | |
9881 | spin_unlock_irq(&head->lock); | |
6b06314c | 9882 | |
6b06314c | 9883 | if (!skb) { |
73b25d3b PB |
9884 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
9885 | if (!fpl) | |
9886 | return -ENOMEM; | |
6b06314c | 9887 | |
73b25d3b PB |
9888 | skb = alloc_skb(0, GFP_KERNEL); |
9889 | if (!skb) { | |
9890 | kfree(fpl); | |
9891 | return -ENOMEM; | |
9892 | } | |
6b06314c | 9893 | |
73b25d3b PB |
9894 | fpl->user = get_uid(current_user()); |
9895 | fpl->max = SCM_MAX_FD; | |
9896 | fpl->count = 0; | |
65e19f54 | 9897 | |
73b25d3b PB |
9898 | UNIXCB(skb).fp = fpl; |
9899 | skb->sk = sk; | |
9900 | skb->destructor = unix_destruct_scm; | |
9901 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); | |
6b06314c JA |
9902 | } |
9903 | ||
73b25d3b PB |
9904 | fpl = UNIXCB(skb).fp; |
9905 | fpl->fp[fpl->count++] = get_file(file); | |
9906 | unix_inflight(fpl->user, file); | |
9907 | skb_queue_head(head, skb); | |
dca58c6a | 9908 | fput(file); |
73b25d3b | 9909 | #endif |
6b06314c JA |
9910 | return 0; |
9911 | } | |
6b06314c | 9912 | |
47e90392 | 9913 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 9914 | { |
50238531 | 9915 | struct file *file = prsrc->file; |
05f3fb3c JA |
9916 | #if defined(CONFIG_UNIX) |
9917 | struct sock *sock = ctx->ring_sock->sk; | |
9918 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
9919 | struct sk_buff *skb; | |
9920 | int i; | |
9921 | ||
1f59bc0f PB |
9922 | if (!io_file_need_scm(file)) { |
9923 | fput(file); | |
9924 | return; | |
9925 | } | |
9926 | ||
05f3fb3c JA |
9927 | __skb_queue_head_init(&list); |
9928 | ||
9929 | /* | |
9930 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
9931 | * remove this entry and rearrange the file array. | |
9932 | */ | |
9933 | skb = skb_dequeue(head); | |
9934 | while (skb) { | |
9935 | struct scm_fp_list *fp; | |
9936 | ||
9937 | fp = UNIXCB(skb).fp; | |
9938 | for (i = 0; i < fp->count; i++) { | |
9939 | int left; | |
9940 | ||
9941 | if (fp->fp[i] != file) | |
9942 | continue; | |
9943 | ||
9944 | unix_notinflight(fp->user, fp->fp[i]); | |
9945 | left = fp->count - 1 - i; | |
9946 | if (left) { | |
9947 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
9948 | left * sizeof(struct file *)); | |
9949 | } | |
9950 | fp->count--; | |
9951 | if (!fp->count) { | |
9952 | kfree_skb(skb); | |
9953 | skb = NULL; | |
9954 | } else { | |
9955 | __skb_queue_tail(&list, skb); | |
9956 | } | |
9957 | fput(file); | |
9958 | file = NULL; | |
9959 | break; | |
9960 | } | |
9961 | ||
9962 | if (!file) | |
9963 | break; | |
9964 | ||
9965 | __skb_queue_tail(&list, skb); | |
9966 | ||
9967 | skb = skb_dequeue(head); | |
9968 | } | |
9969 | ||
9970 | if (skb_peek(&list)) { | |
9971 | spin_lock_irq(&head->lock); | |
9972 | while ((skb = __skb_dequeue(&list)) != NULL) | |
9973 | __skb_queue_tail(head, skb); | |
9974 | spin_unlock_irq(&head->lock); | |
9975 | } | |
9976 | #else | |
9977 | fput(file); | |
9978 | #endif | |
9979 | } | |
9980 | ||
b895c9a6 | 9981 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 9982 | { |
b895c9a6 | 9983 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
9984 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
9985 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 9986 | |
269bbe5f BM |
9987 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
9988 | list_del(&prsrc->list); | |
b60c8dce PB |
9989 | |
9990 | if (prsrc->tag) { | |
f8929630 PB |
9991 | if (ctx->flags & IORING_SETUP_IOPOLL) |
9992 | mutex_lock(&ctx->uring_lock); | |
b60c8dce | 9993 | |
79ebeaee | 9994 | spin_lock(&ctx->completion_lock); |
913a571a | 9995 | io_fill_cqe_aux(ctx, prsrc->tag, 0, 0); |
b60c8dce | 9996 | io_commit_cqring(ctx); |
79ebeaee | 9997 | spin_unlock(&ctx->completion_lock); |
b60c8dce | 9998 | io_cqring_ev_posted(ctx); |
f8929630 PB |
9999 | |
10000 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
10001 | mutex_unlock(&ctx->uring_lock); | |
b60c8dce PB |
10002 | } |
10003 | ||
40ae0ff7 | 10004 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 10005 | kfree(prsrc); |
65e19f54 | 10006 | } |
05589553 | 10007 | |
28a9fe25 | 10008 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
10009 | if (atomic_dec_and_test(&rsrc_data->refs)) |
10010 | complete(&rsrc_data->done); | |
2faf852d | 10011 | } |
65e19f54 | 10012 | |
269bbe5f | 10013 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
10014 | { |
10015 | struct io_ring_ctx *ctx; | |
10016 | struct llist_node *node; | |
10017 | ||
269bbe5f BM |
10018 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
10019 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
10020 | |
10021 | while (node) { | |
b895c9a6 | 10022 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
10023 | struct llist_node *next = node->next; |
10024 | ||
b895c9a6 | 10025 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 10026 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
10027 | node = next; |
10028 | } | |
10029 | } | |
10030 | ||
6b06314c | 10031 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 10032 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
10033 | { |
10034 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 10035 | struct file *file; |
f3baed39 | 10036 | int fd, ret; |
846a4ef2 | 10037 | unsigned i; |
6b06314c | 10038 | |
05f3fb3c | 10039 | if (ctx->file_data) |
6b06314c JA |
10040 | return -EBUSY; |
10041 | if (!nr_args) | |
10042 | return -EINVAL; | |
10043 | if (nr_args > IORING_MAX_FIXED_FILES) | |
10044 | return -EMFILE; | |
3a1b8a4e PB |
10045 | if (nr_args > rlimit(RLIMIT_NOFILE)) |
10046 | return -EMFILE; | |
a7f0ed5a | 10047 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
10048 | if (ret) |
10049 | return ret; | |
d878c816 PB |
10050 | ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args, |
10051 | &ctx->file_data); | |
10052 | if (ret) | |
10053 | return ret; | |
6b06314c | 10054 | |
a03a2a20 PB |
10055 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) { |
10056 | io_rsrc_data_free(ctx->file_data); | |
10057 | ctx->file_data = NULL; | |
10058 | return -ENOMEM; | |
10059 | } | |
65e19f54 | 10060 | |
08a45173 | 10061 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
a03a2a20 PB |
10062 | struct io_fixed_file *file_slot; |
10063 | ||
a8da73a3 | 10064 | if (fds && copy_from_user(&fd, &fds[i], sizeof(fd))) { |
600cf3f8 | 10065 | ret = -EFAULT; |
a03a2a20 | 10066 | goto fail; |
600cf3f8 | 10067 | } |
08a45173 | 10068 | /* allow sparse sets */ |
a8da73a3 | 10069 | if (!fds || fd == -1) { |
792e3582 | 10070 | ret = -EINVAL; |
2d091d62 | 10071 | if (unlikely(*io_get_tag_slot(ctx->file_data, i))) |
a03a2a20 | 10072 | goto fail; |
08a45173 | 10073 | continue; |
792e3582 | 10074 | } |
6b06314c | 10075 | |
05f3fb3c | 10076 | file = fget(fd); |
6b06314c | 10077 | ret = -EBADF; |
792e3582 | 10078 | if (unlikely(!file)) |
a03a2a20 | 10079 | goto fail; |
05f3fb3c | 10080 | |
6b06314c JA |
10081 | /* |
10082 | * Don't allow io_uring instances to be registered. If UNIX | |
10083 | * isn't enabled, then this causes a reference cycle and this | |
10084 | * instance can never get freed. If UNIX is enabled we'll | |
10085 | * handle it just fine, but there's still no point in allowing | |
10086 | * a ring fd as it doesn't support regular read/write anyway. | |
10087 | */ | |
05f3fb3c JA |
10088 | if (file->f_op == &io_uring_fops) { |
10089 | fput(file); | |
a03a2a20 | 10090 | goto fail; |
6b06314c | 10091 | } |
8b3171bd | 10092 | ret = io_scm_file_account(ctx, file); |
a03a2a20 | 10093 | if (ret) { |
600cf3f8 | 10094 | fput(file); |
a03a2a20 | 10095 | goto fail; |
c3a31e60 | 10096 | } |
e390510a PB |
10097 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
10098 | io_fixed_file_set(file_slot, file); | |
d78bd8ad | 10099 | io_file_bitmap_set(&ctx->file_table, i); |
c3a31e60 JA |
10100 | } |
10101 | ||
a7f0ed5a | 10102 | io_rsrc_node_switch(ctx, NULL); |
c3a31e60 | 10103 | return 0; |
a03a2a20 PB |
10104 | fail: |
10105 | __io_sqe_files_unregister(ctx); | |
6b06314c | 10106 | return ret; |
c3a31e60 JA |
10107 | } |
10108 | ||
9c7b0ba8 PB |
10109 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
10110 | struct io_rsrc_node *node, void *rsrc) | |
10111 | { | |
8f0a2480 | 10112 | u64 *tag_slot = io_get_tag_slot(data, idx); |
9c7b0ba8 PB |
10113 | struct io_rsrc_put *prsrc; |
10114 | ||
10115 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); | |
10116 | if (!prsrc) | |
10117 | return -ENOMEM; | |
10118 | ||
8f0a2480 PB |
10119 | prsrc->tag = *tag_slot; |
10120 | *tag_slot = 0; | |
9c7b0ba8 PB |
10121 | prsrc->rsrc = rsrc; |
10122 | list_add(&prsrc->list, &node->rsrc_list); | |
10123 | return 0; | |
10124 | } | |
10125 | ||
b9445598 PB |
10126 | static int io_install_fixed_file(struct io_kiocb *req, struct file *file, |
10127 | unsigned int issue_flags, u32 slot_index) | |
61c1b44a | 10128 | __must_hold(&req->ctx->uring_lock) |
b9445598 PB |
10129 | { |
10130 | struct io_ring_ctx *ctx = req->ctx; | |
9c7b0ba8 | 10131 | bool needs_switch = false; |
b9445598 | 10132 | struct io_fixed_file *file_slot; |
61c1b44a | 10133 | int ret; |
b9445598 | 10134 | |
b9445598 | 10135 | if (file->f_op == &io_uring_fops) |
61c1b44a | 10136 | return -EBADF; |
b9445598 | 10137 | if (!ctx->file_data) |
61c1b44a | 10138 | return -ENXIO; |
b9445598 | 10139 | if (slot_index >= ctx->nr_user_files) |
61c1b44a | 10140 | return -EINVAL; |
b9445598 PB |
10141 | |
10142 | slot_index = array_index_nospec(slot_index, ctx->nr_user_files); | |
10143 | file_slot = io_fixed_file_slot(&ctx->file_table, slot_index); | |
9c7b0ba8 PB |
10144 | |
10145 | if (file_slot->file_ptr) { | |
10146 | struct file *old_file; | |
10147 | ||
10148 | ret = io_rsrc_node_switch_start(ctx); | |
10149 | if (ret) | |
10150 | goto err; | |
10151 | ||
10152 | old_file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
10153 | ret = io_queue_rsrc_removal(ctx->file_data, slot_index, | |
10154 | ctx->rsrc_node, old_file); | |
10155 | if (ret) | |
10156 | goto err; | |
10157 | file_slot->file_ptr = 0; | |
d78bd8ad | 10158 | io_file_bitmap_clear(&ctx->file_table, slot_index); |
9c7b0ba8 PB |
10159 | needs_switch = true; |
10160 | } | |
b9445598 | 10161 | |
8b3171bd | 10162 | ret = io_scm_file_account(ctx, file); |
e390510a PB |
10163 | if (!ret) { |
10164 | *io_get_tag_slot(ctx->file_data, slot_index) = 0; | |
10165 | io_fixed_file_set(file_slot, file); | |
d78bd8ad | 10166 | io_file_bitmap_set(&ctx->file_table, slot_index); |
b9445598 | 10167 | } |
b9445598 | 10168 | err: |
9c7b0ba8 PB |
10169 | if (needs_switch) |
10170 | io_rsrc_node_switch(ctx, ctx->file_data); | |
b9445598 PB |
10171 | if (ret) |
10172 | fput(file); | |
10173 | return ret; | |
10174 | } | |
10175 | ||
a7c41b46 XW |
10176 | static int __io_close_fixed(struct io_kiocb *req, unsigned int issue_flags, |
10177 | unsigned int offset) | |
7df778be | 10178 | { |
7df778be PB |
10179 | struct io_ring_ctx *ctx = req->ctx; |
10180 | struct io_fixed_file *file_slot; | |
10181 | struct file *file; | |
4cdd158b | 10182 | int ret; |
7df778be | 10183 | |
f8929630 | 10184 | io_ring_submit_lock(ctx, issue_flags); |
7df778be PB |
10185 | ret = -ENXIO; |
10186 | if (unlikely(!ctx->file_data)) | |
10187 | goto out; | |
10188 | ret = -EINVAL; | |
10189 | if (offset >= ctx->nr_user_files) | |
10190 | goto out; | |
10191 | ret = io_rsrc_node_switch_start(ctx); | |
10192 | if (ret) | |
10193 | goto out; | |
10194 | ||
4cdd158b PB |
10195 | offset = array_index_nospec(offset, ctx->nr_user_files); |
10196 | file_slot = io_fixed_file_slot(&ctx->file_table, offset); | |
7df778be PB |
10197 | ret = -EBADF; |
10198 | if (!file_slot->file_ptr) | |
10199 | goto out; | |
10200 | ||
10201 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
10202 | ret = io_queue_rsrc_removal(ctx->file_data, offset, ctx->rsrc_node, file); | |
10203 | if (ret) | |
10204 | goto out; | |
10205 | ||
10206 | file_slot->file_ptr = 0; | |
d78bd8ad | 10207 | io_file_bitmap_clear(&ctx->file_table, offset); |
7df778be PB |
10208 | io_rsrc_node_switch(ctx, ctx->file_data); |
10209 | ret = 0; | |
10210 | out: | |
f8929630 | 10211 | io_ring_submit_unlock(ctx, issue_flags); |
7df778be PB |
10212 | return ret; |
10213 | } | |
10214 | ||
a7c41b46 XW |
10215 | static inline int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) |
10216 | { | |
10217 | return __io_close_fixed(req, issue_flags, req->close.file_slot - 1); | |
10218 | } | |
10219 | ||
05f3fb3c | 10220 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
c3bdad02 | 10221 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
10222 | unsigned nr_args) |
10223 | { | |
c3bdad02 | 10224 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 10225 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 10226 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
10227 | struct io_fixed_file *file_slot; |
10228 | struct file *file; | |
98f0b3b4 PB |
10229 | int fd, i, err = 0; |
10230 | unsigned int done; | |
05589553 | 10231 | bool needs_switch = false; |
c3a31e60 | 10232 | |
98f0b3b4 PB |
10233 | if (!ctx->file_data) |
10234 | return -ENXIO; | |
10235 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 JA |
10236 | return -EINVAL; |
10237 | ||
67973b93 | 10238 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
10239 | u64 tag = 0; |
10240 | ||
10241 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
10242 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
10243 | err = -EFAULT; |
10244 | break; | |
10245 | } | |
c3bdad02 PB |
10246 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
10247 | err = -EINVAL; | |
10248 | break; | |
10249 | } | |
4e0377a1 | 10250 | if (fd == IORING_REGISTER_FILES_SKIP) |
10251 | continue; | |
10252 | ||
67973b93 | 10253 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 10254 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 10255 | |
a04b0ac0 PB |
10256 | if (file_slot->file_ptr) { |
10257 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
4cdd158b | 10258 | err = io_queue_rsrc_removal(data, i, ctx->rsrc_node, file); |
a5318d3c HD |
10259 | if (err) |
10260 | break; | |
a04b0ac0 | 10261 | file_slot->file_ptr = 0; |
d78bd8ad | 10262 | io_file_bitmap_clear(&ctx->file_table, i); |
05589553 | 10263 | needs_switch = true; |
c3a31e60 JA |
10264 | } |
10265 | if (fd != -1) { | |
c3a31e60 JA |
10266 | file = fget(fd); |
10267 | if (!file) { | |
10268 | err = -EBADF; | |
10269 | break; | |
10270 | } | |
10271 | /* | |
10272 | * Don't allow io_uring instances to be registered. If | |
10273 | * UNIX isn't enabled, then this causes a reference | |
10274 | * cycle and this instance can never get freed. If UNIX | |
10275 | * is enabled we'll handle it just fine, but there's | |
10276 | * still no point in allowing a ring fd as it doesn't | |
10277 | * support regular read/write anyway. | |
10278 | */ | |
10279 | if (file->f_op == &io_uring_fops) { | |
10280 | fput(file); | |
10281 | err = -EBADF; | |
10282 | break; | |
10283 | } | |
8b3171bd | 10284 | err = io_scm_file_account(ctx, file); |
f3bd9dae YY |
10285 | if (err) { |
10286 | fput(file); | |
c3a31e60 | 10287 | break; |
f3bd9dae | 10288 | } |
e390510a PB |
10289 | *io_get_tag_slot(data, i) = tag; |
10290 | io_fixed_file_set(file_slot, file); | |
d78bd8ad | 10291 | io_file_bitmap_set(&ctx->file_table, i); |
c3a31e60 | 10292 | } |
05f3fb3c JA |
10293 | } |
10294 | ||
a7f0ed5a PB |
10295 | if (needs_switch) |
10296 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
10297 | return done ? done : err; |
10298 | } | |
05589553 | 10299 | |
685fe7fe JA |
10300 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
10301 | struct task_struct *task) | |
24369c2e | 10302 | { |
e941894e | 10303 | struct io_wq_hash *hash; |
24369c2e | 10304 | struct io_wq_data data; |
24369c2e | 10305 | unsigned int concurrency; |
24369c2e | 10306 | |
362a9e65 | 10307 | mutex_lock(&ctx->uring_lock); |
e941894e JA |
10308 | hash = ctx->hash_map; |
10309 | if (!hash) { | |
10310 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
362a9e65 YY |
10311 | if (!hash) { |
10312 | mutex_unlock(&ctx->uring_lock); | |
e941894e | 10313 | return ERR_PTR(-ENOMEM); |
362a9e65 | 10314 | } |
e941894e JA |
10315 | refcount_set(&hash->refs, 1); |
10316 | init_waitqueue_head(&hash->wait); | |
10317 | ctx->hash_map = hash; | |
24369c2e | 10318 | } |
362a9e65 | 10319 | mutex_unlock(&ctx->uring_lock); |
24369c2e | 10320 | |
e941894e | 10321 | data.hash = hash; |
685fe7fe | 10322 | data.task = task; |
ebc11b6c | 10323 | data.free_work = io_wq_free_work; |
f5fa38c5 | 10324 | data.do_work = io_wq_submit_work; |
24369c2e | 10325 | |
d25e3a3d JA |
10326 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
10327 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 10328 | |
5aa75ed5 | 10329 | return io_wq_create(concurrency, &data); |
24369c2e PB |
10330 | } |
10331 | ||
c072481d PB |
10332 | static __cold int io_uring_alloc_task_context(struct task_struct *task, |
10333 | struct io_ring_ctx *ctx) | |
0f212204 JA |
10334 | { |
10335 | struct io_uring_task *tctx; | |
d8a6df10 | 10336 | int ret; |
0f212204 | 10337 | |
09899b19 | 10338 | tctx = kzalloc(sizeof(*tctx), GFP_KERNEL); |
0f212204 JA |
10339 | if (unlikely(!tctx)) |
10340 | return -ENOMEM; | |
10341 | ||
e7a6c00d JA |
10342 | tctx->registered_rings = kcalloc(IO_RINGFD_REG_MAX, |
10343 | sizeof(struct file *), GFP_KERNEL); | |
10344 | if (unlikely(!tctx->registered_rings)) { | |
10345 | kfree(tctx); | |
10346 | return -ENOMEM; | |
10347 | } | |
10348 | ||
d8a6df10 JA |
10349 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
10350 | if (unlikely(ret)) { | |
e7a6c00d | 10351 | kfree(tctx->registered_rings); |
d8a6df10 JA |
10352 | kfree(tctx); |
10353 | return ret; | |
10354 | } | |
10355 | ||
685fe7fe | 10356 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
10357 | if (IS_ERR(tctx->io_wq)) { |
10358 | ret = PTR_ERR(tctx->io_wq); | |
10359 | percpu_counter_destroy(&tctx->inflight); | |
e7a6c00d | 10360 | kfree(tctx->registered_rings); |
5aa75ed5 JA |
10361 | kfree(tctx); |
10362 | return ret; | |
10363 | } | |
10364 | ||
0f212204 JA |
10365 | xa_init(&tctx->xa); |
10366 | init_waitqueue_head(&tctx->wait); | |
fdaf083c | 10367 | atomic_set(&tctx->in_idle, 0); |
9cae36a0 | 10368 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 10369 | task->io_uring = tctx; |
7cbf1722 JA |
10370 | spin_lock_init(&tctx->task_lock); |
10371 | INIT_WQ_LIST(&tctx->task_list); | |
3fe07bcd | 10372 | INIT_WQ_LIST(&tctx->prio_task_list); |
7cbf1722 | 10373 | init_task_work(&tctx->task_work, tctx_task_work); |
0f212204 JA |
10374 | return 0; |
10375 | } | |
10376 | ||
10377 | void __io_uring_free(struct task_struct *tsk) | |
10378 | { | |
10379 | struct io_uring_task *tctx = tsk->io_uring; | |
10380 | ||
10381 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e | 10382 | WARN_ON_ONCE(tctx->io_wq); |
09899b19 | 10383 | WARN_ON_ONCE(tctx->cached_refs); |
ef8eaa4e | 10384 | |
e7a6c00d | 10385 | kfree(tctx->registered_rings); |
d8a6df10 | 10386 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
10387 | kfree(tctx); |
10388 | tsk->io_uring = NULL; | |
10389 | } | |
10390 | ||
c072481d PB |
10391 | static __cold int io_sq_offload_create(struct io_ring_ctx *ctx, |
10392 | struct io_uring_params *p) | |
2b188cc1 JA |
10393 | { |
10394 | int ret; | |
10395 | ||
d25e3a3d JA |
10396 | /* Retain compatibility with failing for an invalid attach attempt */ |
10397 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
10398 | IORING_SETUP_ATTACH_WQ) { | |
10399 | struct fd f; | |
10400 | ||
10401 | f = fdget(p->wq_fd); | |
10402 | if (!f.file) | |
10403 | return -ENXIO; | |
0cc936f7 JA |
10404 | if (f.file->f_op != &io_uring_fops) { |
10405 | fdput(f); | |
f2a48dd0 | 10406 | return -EINVAL; |
0cc936f7 JA |
10407 | } |
10408 | fdput(f); | |
d25e3a3d | 10409 | } |
6c271ce2 | 10410 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 10411 | struct task_struct *tsk; |
534ca6d6 | 10412 | struct io_sq_data *sqd; |
26984fbf | 10413 | bool attached; |
534ca6d6 | 10414 | |
cdc1404a PM |
10415 | ret = security_uring_sqpoll(); |
10416 | if (ret) | |
10417 | return ret; | |
10418 | ||
26984fbf | 10419 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
10420 | if (IS_ERR(sqd)) { |
10421 | ret = PTR_ERR(sqd); | |
10422 | goto err; | |
10423 | } | |
69fb2131 | 10424 | |
7c30f36a | 10425 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 10426 | ctx->sq_data = sqd; |
917257da JA |
10427 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
10428 | if (!ctx->sq_thread_idle) | |
10429 | ctx->sq_thread_idle = HZ; | |
10430 | ||
78d7f6ba | 10431 | io_sq_thread_park(sqd); |
de75a3d3 PB |
10432 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
10433 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 10434 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 10435 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
10436 | io_sq_thread_unpark(sqd); |
10437 | ||
de75a3d3 PB |
10438 | if (ret < 0) |
10439 | goto err; | |
10440 | if (attached) | |
5aa75ed5 | 10441 | return 0; |
aa06165d | 10442 | |
6c271ce2 | 10443 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 10444 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 10445 | |
917257da | 10446 | ret = -EINVAL; |
f2a48dd0 | 10447 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 10448 | goto err_sqpoll; |
37d1e2e3 | 10449 | sqd->sq_cpu = cpu; |
6c271ce2 | 10450 | } else { |
37d1e2e3 | 10451 | sqd->sq_cpu = -1; |
6c271ce2 | 10452 | } |
37d1e2e3 JA |
10453 | |
10454 | sqd->task_pid = current->pid; | |
5c2469e0 | 10455 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
10456 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
10457 | if (IS_ERR(tsk)) { | |
10458 | ret = PTR_ERR(tsk); | |
e8f98f24 | 10459 | goto err_sqpoll; |
6c271ce2 | 10460 | } |
97a73a0f | 10461 | |
46fe18b1 | 10462 | sqd->thread = tsk; |
97a73a0f | 10463 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 10464 | wake_up_new_task(tsk); |
0f212204 JA |
10465 | if (ret) |
10466 | goto err; | |
6c271ce2 JA |
10467 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
10468 | /* Can't have SQ_AFF without SQPOLL */ | |
10469 | ret = -EINVAL; | |
10470 | goto err; | |
10471 | } | |
10472 | ||
2b188cc1 | 10473 | return 0; |
f2a48dd0 PB |
10474 | err_sqpoll: |
10475 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 10476 | err: |
37d1e2e3 | 10477 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
10478 | return ret; |
10479 | } | |
10480 | ||
a087e2b5 BM |
10481 | static inline void __io_unaccount_mem(struct user_struct *user, |
10482 | unsigned long nr_pages) | |
2b188cc1 JA |
10483 | { |
10484 | atomic_long_sub(nr_pages, &user->locked_vm); | |
10485 | } | |
10486 | ||
a087e2b5 BM |
10487 | static inline int __io_account_mem(struct user_struct *user, |
10488 | unsigned long nr_pages) | |
2b188cc1 JA |
10489 | { |
10490 | unsigned long page_limit, cur_pages, new_pages; | |
10491 | ||
10492 | /* Don't allow more pages than we can safely lock */ | |
10493 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
10494 | ||
10495 | do { | |
10496 | cur_pages = atomic_long_read(&user->locked_vm); | |
10497 | new_pages = cur_pages + nr_pages; | |
10498 | if (new_pages > page_limit) | |
10499 | return -ENOMEM; | |
10500 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
10501 | new_pages) != cur_pages); | |
10502 | ||
10503 | return 0; | |
10504 | } | |
10505 | ||
26bfa89e | 10506 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 10507 | { |
62e398be | 10508 | if (ctx->user) |
a087e2b5 | 10509 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 10510 | |
26bfa89e JA |
10511 | if (ctx->mm_account) |
10512 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
10513 | } |
10514 | ||
26bfa89e | 10515 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 10516 | { |
30975825 BM |
10517 | int ret; |
10518 | ||
62e398be | 10519 | if (ctx->user) { |
30975825 BM |
10520 | ret = __io_account_mem(ctx->user, nr_pages); |
10521 | if (ret) | |
10522 | return ret; | |
10523 | } | |
10524 | ||
26bfa89e JA |
10525 | if (ctx->mm_account) |
10526 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
10527 | |
10528 | return 0; | |
10529 | } | |
10530 | ||
2b188cc1 JA |
10531 | static void io_mem_free(void *ptr) |
10532 | { | |
52e04ef4 MR |
10533 | struct page *page; |
10534 | ||
10535 | if (!ptr) | |
10536 | return; | |
2b188cc1 | 10537 | |
52e04ef4 | 10538 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
10539 | if (put_page_testzero(page)) |
10540 | free_compound_page(page); | |
10541 | } | |
10542 | ||
10543 | static void *io_mem_alloc(size_t size) | |
10544 | { | |
0a3f1e0b | 10545 | gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP; |
2b188cc1 | 10546 | |
0a3f1e0b | 10547 | return (void *) __get_free_pages(gfp, get_order(size)); |
2b188cc1 JA |
10548 | } |
10549 | ||
baf9cb64 SR |
10550 | static unsigned long rings_size(struct io_ring_ctx *ctx, unsigned int sq_entries, |
10551 | unsigned int cq_entries, size_t *sq_offset) | |
75b28aff HV |
10552 | { |
10553 | struct io_rings *rings; | |
10554 | size_t off, sq_array_size; | |
10555 | ||
10556 | off = struct_size(rings, cqes, cq_entries); | |
10557 | if (off == SIZE_MAX) | |
10558 | return SIZE_MAX; | |
baf9cb64 SR |
10559 | if (ctx->flags & IORING_SETUP_CQE32) { |
10560 | if (check_shl_overflow(off, 1, &off)) | |
10561 | return SIZE_MAX; | |
10562 | } | |
75b28aff HV |
10563 | |
10564 | #ifdef CONFIG_SMP | |
10565 | off = ALIGN(off, SMP_CACHE_BYTES); | |
10566 | if (off == 0) | |
10567 | return SIZE_MAX; | |
10568 | #endif | |
10569 | ||
b36200f5 DV |
10570 | if (sq_offset) |
10571 | *sq_offset = off; | |
10572 | ||
75b28aff HV |
10573 | sq_array_size = array_size(sizeof(u32), sq_entries); |
10574 | if (sq_array_size == SIZE_MAX) | |
10575 | return SIZE_MAX; | |
10576 | ||
10577 | if (check_add_overflow(off, sq_array_size, &off)) | |
10578 | return SIZE_MAX; | |
10579 | ||
75b28aff HV |
10580 | return off; |
10581 | } | |
10582 | ||
41edf1a5 | 10583 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 10584 | { |
41edf1a5 | 10585 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
10586 | unsigned int i; |
10587 | ||
6224843d PB |
10588 | if (imu != ctx->dummy_ubuf) { |
10589 | for (i = 0; i < imu->nr_bvecs; i++) | |
10590 | unpin_user_page(imu->bvec[i].bv_page); | |
10591 | if (imu->acct_pages) | |
10592 | io_unaccount_mem(ctx, imu->acct_pages); | |
10593 | kvfree(imu); | |
10594 | } | |
41edf1a5 | 10595 | *slot = NULL; |
7f61a1e9 PB |
10596 | } |
10597 | ||
bd54b6fe | 10598 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 10599 | { |
634d00df PB |
10600 | io_buffer_unmap(ctx, &prsrc->buf); |
10601 | prsrc->buf = NULL; | |
bd54b6fe | 10602 | } |
edafccee | 10603 | |
bd54b6fe BM |
10604 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
10605 | { | |
10606 | unsigned int i; | |
edafccee | 10607 | |
7f61a1e9 PB |
10608 | for (i = 0; i < ctx->nr_user_bufs; i++) |
10609 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 10610 | kfree(ctx->user_bufs); |
bb6659cc | 10611 | io_rsrc_data_free(ctx->buf_data); |
edafccee | 10612 | ctx->user_bufs = NULL; |
bd54b6fe | 10613 | ctx->buf_data = NULL; |
edafccee | 10614 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
10615 | } |
10616 | ||
0a96bbe4 | 10617 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 10618 | { |
d11d31fc | 10619 | unsigned nr = ctx->nr_user_bufs; |
bd54b6fe | 10620 | int ret; |
edafccee | 10621 | |
bd54b6fe | 10622 | if (!ctx->buf_data) |
edafccee JA |
10623 | return -ENXIO; |
10624 | ||
d11d31fc PB |
10625 | /* |
10626 | * Quiesce may unlock ->uring_lock, and while it's not held | |
10627 | * prevent new requests using the table. | |
10628 | */ | |
10629 | ctx->nr_user_bufs = 0; | |
bd54b6fe | 10630 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); |
d11d31fc | 10631 | ctx->nr_user_bufs = nr; |
bd54b6fe BM |
10632 | if (!ret) |
10633 | __io_sqe_buffers_unregister(ctx); | |
10634 | return ret; | |
edafccee JA |
10635 | } |
10636 | ||
10637 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
10638 | void __user *arg, unsigned index) | |
10639 | { | |
10640 | struct iovec __user *src; | |
10641 | ||
10642 | #ifdef CONFIG_COMPAT | |
10643 | if (ctx->compat) { | |
10644 | struct compat_iovec __user *ciovs; | |
10645 | struct compat_iovec ciov; | |
10646 | ||
10647 | ciovs = (struct compat_iovec __user *) arg; | |
10648 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
10649 | return -EFAULT; | |
10650 | ||
d55e5f5b | 10651 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
10652 | dst->iov_len = ciov.iov_len; |
10653 | return 0; | |
10654 | } | |
10655 | #endif | |
10656 | src = (struct iovec __user *) arg; | |
10657 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
10658 | return -EFAULT; | |
10659 | return 0; | |
10660 | } | |
10661 | ||
de293938 JA |
10662 | /* |
10663 | * Not super efficient, but this is just a registration time. And we do cache | |
10664 | * the last compound head, so generally we'll only do a full search if we don't | |
10665 | * match that one. | |
10666 | * | |
10667 | * We check if the given compound head page has already been accounted, to | |
10668 | * avoid double accounting it. This allows us to account the full size of the | |
10669 | * page, not just the constituent pages of a huge page. | |
10670 | */ | |
10671 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
10672 | int nr_pages, struct page *hpage) | |
10673 | { | |
10674 | int i, j; | |
10675 | ||
10676 | /* check current page array */ | |
10677 | for (i = 0; i < nr_pages; i++) { | |
10678 | if (!PageCompound(pages[i])) | |
10679 | continue; | |
10680 | if (compound_head(pages[i]) == hpage) | |
10681 | return true; | |
10682 | } | |
10683 | ||
10684 | /* check previously registered pages */ | |
10685 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 10686 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
10687 | |
10688 | for (j = 0; j < imu->nr_bvecs; j++) { | |
10689 | if (!PageCompound(imu->bvec[j].bv_page)) | |
10690 | continue; | |
10691 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
10692 | return true; | |
10693 | } | |
10694 | } | |
10695 | ||
10696 | return false; | |
10697 | } | |
10698 | ||
10699 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
10700 | int nr_pages, struct io_mapped_ubuf *imu, | |
10701 | struct page **last_hpage) | |
10702 | { | |
10703 | int i, ret; | |
10704 | ||
216e5835 | 10705 | imu->acct_pages = 0; |
de293938 JA |
10706 | for (i = 0; i < nr_pages; i++) { |
10707 | if (!PageCompound(pages[i])) { | |
10708 | imu->acct_pages++; | |
10709 | } else { | |
10710 | struct page *hpage; | |
10711 | ||
10712 | hpage = compound_head(pages[i]); | |
10713 | if (hpage == *last_hpage) | |
10714 | continue; | |
10715 | *last_hpage = hpage; | |
10716 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
10717 | continue; | |
10718 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
10719 | } | |
10720 | } | |
10721 | ||
10722 | if (!imu->acct_pages) | |
10723 | return 0; | |
10724 | ||
26bfa89e | 10725 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
10726 | if (ret) |
10727 | imu->acct_pages = 0; | |
10728 | return ret; | |
10729 | } | |
10730 | ||
d8c2237d JA |
10731 | static struct page **io_pin_pages(unsigned long ubuf, unsigned long len, |
10732 | int *npages) | |
edafccee | 10733 | { |
d8c2237d | 10734 | unsigned long start, end, nr_pages; |
edafccee JA |
10735 | struct vm_area_struct **vmas = NULL; |
10736 | struct page **pages = NULL; | |
d8c2237d | 10737 | int i, pret, ret = -ENOMEM; |
6224843d | 10738 | |
d8c2237d | 10739 | end = (ubuf + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
0a96bbe4 BM |
10740 | start = ubuf >> PAGE_SHIFT; |
10741 | nr_pages = end - start; | |
10742 | ||
0a96bbe4 BM |
10743 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); |
10744 | if (!pages) | |
10745 | goto done; | |
10746 | ||
10747 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
10748 | GFP_KERNEL); | |
10749 | if (!vmas) | |
10750 | goto done; | |
edafccee | 10751 | |
0a96bbe4 BM |
10752 | ret = 0; |
10753 | mmap_read_lock(current->mm); | |
10754 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
10755 | pages, vmas); | |
10756 | if (pret == nr_pages) { | |
10757 | /* don't support file backed memory */ | |
10758 | for (i = 0; i < nr_pages; i++) { | |
10759 | struct vm_area_struct *vma = vmas[i]; | |
10760 | ||
40dad765 PB |
10761 | if (vma_is_shmem(vma)) |
10762 | continue; | |
0a96bbe4 BM |
10763 | if (vma->vm_file && |
10764 | !is_file_hugepages(vma->vm_file)) { | |
10765 | ret = -EOPNOTSUPP; | |
10766 | break; | |
10767 | } | |
10768 | } | |
d8c2237d | 10769 | *npages = nr_pages; |
0a96bbe4 BM |
10770 | } else { |
10771 | ret = pret < 0 ? pret : -EFAULT; | |
10772 | } | |
10773 | mmap_read_unlock(current->mm); | |
10774 | if (ret) { | |
10775 | /* | |
10776 | * if we did partial map, or found file backed vmas, | |
10777 | * release any pages we did get | |
10778 | */ | |
10779 | if (pret > 0) | |
10780 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
10781 | goto done; |
10782 | } | |
d8c2237d JA |
10783 | ret = 0; |
10784 | done: | |
10785 | kvfree(vmas); | |
10786 | if (ret < 0) { | |
10787 | kvfree(pages); | |
10788 | pages = ERR_PTR(ret); | |
10789 | } | |
10790 | return pages; | |
10791 | } | |
0a96bbe4 | 10792 | |
d8c2237d JA |
10793 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
10794 | struct io_mapped_ubuf **pimu, | |
10795 | struct page **last_hpage) | |
10796 | { | |
10797 | struct io_mapped_ubuf *imu = NULL; | |
10798 | struct page **pages = NULL; | |
10799 | unsigned long off; | |
10800 | size_t size; | |
10801 | int ret, nr_pages, i; | |
10802 | ||
10803 | if (!iov->iov_base) { | |
10804 | *pimu = ctx->dummy_ubuf; | |
10805 | return 0; | |
10806 | } | |
10807 | ||
10808 | *pimu = NULL; | |
10809 | ret = -ENOMEM; | |
10810 | ||
10811 | pages = io_pin_pages((unsigned long) iov->iov_base, iov->iov_len, | |
10812 | &nr_pages); | |
10813 | if (IS_ERR(pages)) { | |
10814 | ret = PTR_ERR(pages); | |
10815 | pages = NULL; | |
10816 | goto done; | |
10817 | } | |
10818 | ||
10819 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); | |
10820 | if (!imu) | |
10821 | goto done; | |
0a96bbe4 | 10822 | |
d8c2237d | 10823 | ret = io_buffer_account_pin(ctx, pages, nr_pages, imu, last_hpage); |
0a96bbe4 | 10824 | if (ret) { |
d8c2237d | 10825 | unpin_user_pages(pages, nr_pages); |
0a96bbe4 BM |
10826 | goto done; |
10827 | } | |
10828 | ||
d8c2237d | 10829 | off = (unsigned long) iov->iov_base & ~PAGE_MASK; |
0a96bbe4 BM |
10830 | size = iov->iov_len; |
10831 | for (i = 0; i < nr_pages; i++) { | |
10832 | size_t vec_len; | |
10833 | ||
10834 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
10835 | imu->bvec[i].bv_page = pages[i]; | |
10836 | imu->bvec[i].bv_len = vec_len; | |
10837 | imu->bvec[i].bv_offset = off; | |
10838 | off = 0; | |
10839 | size -= vec_len; | |
10840 | } | |
10841 | /* store original address for later verification */ | |
d8c2237d JA |
10842 | imu->ubuf = (unsigned long) iov->iov_base; |
10843 | imu->ubuf_end = imu->ubuf + iov->iov_len; | |
0a96bbe4 | 10844 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 10845 | *pimu = imu; |
0a96bbe4 BM |
10846 | ret = 0; |
10847 | done: | |
41edf1a5 PB |
10848 | if (ret) |
10849 | kvfree(imu); | |
0a96bbe4 | 10850 | kvfree(pages); |
0a96bbe4 BM |
10851 | return ret; |
10852 | } | |
10853 | ||
2b358604 | 10854 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 10855 | { |
87094465 PB |
10856 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
10857 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 10858 | } |
edafccee | 10859 | |
2b358604 BM |
10860 | static int io_buffer_validate(struct iovec *iov) |
10861 | { | |
50e96989 PB |
10862 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
10863 | ||
2b358604 BM |
10864 | /* |
10865 | * Don't impose further limits on the size and buffer | |
10866 | * constraints here, we'll -EINVAL later when IO is | |
10867 | * submitted if they are wrong. | |
10868 | */ | |
6224843d PB |
10869 | if (!iov->iov_base) |
10870 | return iov->iov_len ? -EFAULT : 0; | |
10871 | if (!iov->iov_len) | |
2b358604 | 10872 | return -EFAULT; |
edafccee | 10873 | |
2b358604 BM |
10874 | /* arbitrary limit, but we need something */ |
10875 | if (iov->iov_len > SZ_1G) | |
10876 | return -EFAULT; | |
edafccee | 10877 | |
50e96989 PB |
10878 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
10879 | return -EOVERFLOW; | |
10880 | ||
2b358604 BM |
10881 | return 0; |
10882 | } | |
edafccee | 10883 | |
2b358604 | 10884 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
634d00df | 10885 | unsigned int nr_args, u64 __user *tags) |
2b358604 | 10886 | { |
bd54b6fe BM |
10887 | struct page *last_hpage = NULL; |
10888 | struct io_rsrc_data *data; | |
2b358604 BM |
10889 | int i, ret; |
10890 | struct iovec iov; | |
edafccee | 10891 | |
87094465 PB |
10892 | if (ctx->user_bufs) |
10893 | return -EBUSY; | |
489809e2 | 10894 | if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS) |
87094465 | 10895 | return -EINVAL; |
bd54b6fe | 10896 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
10897 | if (ret) |
10898 | return ret; | |
d878c816 PB |
10899 | ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data); |
10900 | if (ret) | |
10901 | return ret; | |
bd54b6fe BM |
10902 | ret = io_buffers_map_alloc(ctx, nr_args); |
10903 | if (ret) { | |
bb6659cc | 10904 | io_rsrc_data_free(data); |
bd54b6fe BM |
10905 | return ret; |
10906 | } | |
edafccee | 10907 | |
87094465 | 10908 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
0184f08e PB |
10909 | if (arg) { |
10910 | ret = io_copy_iov(ctx, &iov, arg, i); | |
10911 | if (ret) | |
10912 | break; | |
10913 | ret = io_buffer_validate(&iov); | |
10914 | if (ret) | |
10915 | break; | |
10916 | } else { | |
10917 | memset(&iov, 0, sizeof(iov)); | |
10918 | } | |
10919 | ||
2d091d62 | 10920 | if (!iov.iov_base && *io_get_tag_slot(data, i)) { |
cf3770e7 CIK |
10921 | ret = -EINVAL; |
10922 | break; | |
10923 | } | |
edafccee | 10924 | |
41edf1a5 PB |
10925 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], |
10926 | &last_hpage); | |
0a96bbe4 BM |
10927 | if (ret) |
10928 | break; | |
edafccee | 10929 | } |
0a96bbe4 | 10930 | |
bd54b6fe | 10931 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 10932 | |
bd54b6fe BM |
10933 | ctx->buf_data = data; |
10934 | if (ret) | |
10935 | __io_sqe_buffers_unregister(ctx); | |
10936 | else | |
10937 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
10938 | return ret; |
10939 | } | |
10940 | ||
634d00df PB |
10941 | static int __io_sqe_buffers_update(struct io_ring_ctx *ctx, |
10942 | struct io_uring_rsrc_update2 *up, | |
10943 | unsigned int nr_args) | |
10944 | { | |
10945 | u64 __user *tags = u64_to_user_ptr(up->tags); | |
10946 | struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); | |
634d00df PB |
10947 | struct page *last_hpage = NULL; |
10948 | bool needs_switch = false; | |
10949 | __u32 done; | |
10950 | int i, err; | |
10951 | ||
10952 | if (!ctx->buf_data) | |
10953 | return -ENXIO; | |
10954 | if (up->offset + nr_args > ctx->nr_user_bufs) | |
10955 | return -EINVAL; | |
10956 | ||
10957 | for (done = 0; done < nr_args; done++) { | |
0b8c0e7c PB |
10958 | struct io_mapped_ubuf *imu; |
10959 | int offset = up->offset + done; | |
634d00df PB |
10960 | u64 tag = 0; |
10961 | ||
10962 | err = io_copy_iov(ctx, &iov, iovs, done); | |
10963 | if (err) | |
10964 | break; | |
10965 | if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) { | |
10966 | err = -EFAULT; | |
10967 | break; | |
10968 | } | |
0b8c0e7c PB |
10969 | err = io_buffer_validate(&iov); |
10970 | if (err) | |
10971 | break; | |
cf3770e7 CIK |
10972 | if (!iov.iov_base && tag) { |
10973 | err = -EINVAL; | |
10974 | break; | |
10975 | } | |
0b8c0e7c PB |
10976 | err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage); |
10977 | if (err) | |
10978 | break; | |
634d00df | 10979 | |
0b8c0e7c | 10980 | i = array_index_nospec(offset, ctx->nr_user_bufs); |
6224843d | 10981 | if (ctx->user_bufs[i] != ctx->dummy_ubuf) { |
4cdd158b | 10982 | err = io_queue_rsrc_removal(ctx->buf_data, i, |
0b8c0e7c PB |
10983 | ctx->rsrc_node, ctx->user_bufs[i]); |
10984 | if (unlikely(err)) { | |
10985 | io_buffer_unmap(ctx, &imu); | |
634d00df | 10986 | break; |
0b8c0e7c | 10987 | } |
634d00df PB |
10988 | ctx->user_bufs[i] = NULL; |
10989 | needs_switch = true; | |
10990 | } | |
10991 | ||
0b8c0e7c | 10992 | ctx->user_bufs[i] = imu; |
2d091d62 | 10993 | *io_get_tag_slot(ctx->buf_data, offset) = tag; |
634d00df PB |
10994 | } |
10995 | ||
10996 | if (needs_switch) | |
10997 | io_rsrc_node_switch(ctx, ctx->buf_data); | |
10998 | return done ? done : err; | |
10999 | } | |
11000 | ||
c75312dd UA |
11001 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg, |
11002 | unsigned int eventfd_async) | |
9b402849 | 11003 | { |
77bc59b4 | 11004 | struct io_ev_fd *ev_fd; |
9b402849 | 11005 | __s32 __user *fds = arg; |
f0a4e62b | 11006 | int fd; |
9b402849 | 11007 | |
77bc59b4 UA |
11008 | ev_fd = rcu_dereference_protected(ctx->io_ev_fd, |
11009 | lockdep_is_held(&ctx->uring_lock)); | |
11010 | if (ev_fd) | |
9b402849 JA |
11011 | return -EBUSY; |
11012 | ||
11013 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
11014 | return -EFAULT; | |
11015 | ||
77bc59b4 UA |
11016 | ev_fd = kmalloc(sizeof(*ev_fd), GFP_KERNEL); |
11017 | if (!ev_fd) | |
11018 | return -ENOMEM; | |
fe7e3257 | 11019 | |
77bc59b4 UA |
11020 | ev_fd->cq_ev_fd = eventfd_ctx_fdget(fd); |
11021 | if (IS_ERR(ev_fd->cq_ev_fd)) { | |
f0a4e62b | 11022 | int ret = PTR_ERR(ev_fd->cq_ev_fd); |
77bc59b4 | 11023 | kfree(ev_fd); |
9b402849 JA |
11024 | return ret; |
11025 | } | |
c75312dd | 11026 | ev_fd->eventfd_async = eventfd_async; |
9aa8dfde | 11027 | ctx->has_evfd = true; |
77bc59b4 | 11028 | rcu_assign_pointer(ctx->io_ev_fd, ev_fd); |
f0a4e62b | 11029 | return 0; |
77bc59b4 UA |
11030 | } |
11031 | ||
11032 | static void io_eventfd_put(struct rcu_head *rcu) | |
11033 | { | |
11034 | struct io_ev_fd *ev_fd = container_of(rcu, struct io_ev_fd, rcu); | |
11035 | ||
11036 | eventfd_ctx_put(ev_fd->cq_ev_fd); | |
11037 | kfree(ev_fd); | |
9b402849 JA |
11038 | } |
11039 | ||
11040 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
11041 | { | |
77bc59b4 UA |
11042 | struct io_ev_fd *ev_fd; |
11043 | ||
11044 | ev_fd = rcu_dereference_protected(ctx->io_ev_fd, | |
11045 | lockdep_is_held(&ctx->uring_lock)); | |
11046 | if (ev_fd) { | |
9aa8dfde | 11047 | ctx->has_evfd = false; |
77bc59b4 UA |
11048 | rcu_assign_pointer(ctx->io_ev_fd, NULL); |
11049 | call_rcu(&ev_fd->rcu, io_eventfd_put); | |
9b402849 JA |
11050 | return 0; |
11051 | } | |
11052 | ||
11053 | return -ENXIO; | |
11054 | } | |
11055 | ||
5a2e745d JA |
11056 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
11057 | { | |
9cfc7e94 JA |
11058 | struct io_buffer_list *bl; |
11059 | unsigned long index; | |
dbc7d452 JA |
11060 | int i; |
11061 | ||
9cfc7e94 JA |
11062 | for (i = 0; i < BGID_ARRAY; i++) { |
11063 | if (!ctx->io_bl) | |
11064 | break; | |
11065 | __io_remove_buffers(ctx, &ctx->io_bl[i], -1U); | |
11066 | } | |
dbc7d452 | 11067 | |
9cfc7e94 JA |
11068 | xa_for_each(&ctx->io_bl_xa, index, bl) { |
11069 | xa_erase(&ctx->io_bl_xa, bl->bgid); | |
11070 | __io_remove_buffers(ctx, bl, -1U); | |
21870e02 | 11071 | kfree(bl); |
dbc7d452 | 11072 | } |
cc3cec83 JA |
11073 | |
11074 | while (!list_empty(&ctx->io_buffers_pages)) { | |
11075 | struct page *page; | |
11076 | ||
11077 | page = list_first_entry(&ctx->io_buffers_pages, struct page, lru); | |
11078 | list_del_init(&page->lru); | |
11079 | __free_page(page); | |
11080 | } | |
5a2e745d JA |
11081 | } |
11082 | ||
4010fec4 | 11083 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 11084 | { |
cd0ca2e0 | 11085 | struct io_submit_state *state = &ctx->submit_state; |
37f0e767 | 11086 | int nr = 0; |
bf019da7 | 11087 | |
9a4fdbd8 | 11088 | mutex_lock(&ctx->uring_lock); |
cd0ca2e0 | 11089 | io_flush_cached_locked_reqs(ctx, state); |
9a4fdbd8 | 11090 | |
88ab95be | 11091 | while (!io_req_cache_empty(ctx)) { |
c2b6c6bc PB |
11092 | struct io_wq_work_node *node; |
11093 | struct io_kiocb *req; | |
9a4fdbd8 | 11094 | |
c2b6c6bc PB |
11095 | node = wq_stack_extract(&state->free_list); |
11096 | req = container_of(node, struct io_kiocb, comp_list); | |
11097 | kmem_cache_free(req_cachep, req); | |
37f0e767 | 11098 | nr++; |
c2b6c6bc | 11099 | } |
37f0e767 PB |
11100 | if (nr) |
11101 | percpu_ref_put_many(&ctx->refs, nr); | |
9a4fdbd8 JA |
11102 | mutex_unlock(&ctx->uring_lock); |
11103 | } | |
11104 | ||
43597aac | 11105 | static void io_wait_rsrc_data(struct io_rsrc_data *data) |
2b188cc1 | 11106 | { |
43597aac | 11107 | if (data && !atomic_dec_and_test(&data->refs)) |
bd54b6fe | 11108 | wait_for_completion(&data->done); |
bd54b6fe | 11109 | } |
04fc6c80 | 11110 | |
4d9237e3 JA |
11111 | static void io_flush_apoll_cache(struct io_ring_ctx *ctx) |
11112 | { | |
11113 | struct async_poll *apoll; | |
11114 | ||
11115 | while (!list_empty(&ctx->apoll_cache)) { | |
11116 | apoll = list_first_entry(&ctx->apoll_cache, struct async_poll, | |
11117 | poll.wait.entry); | |
11118 | list_del(&apoll->poll.wait.entry); | |
11119 | kfree(apoll); | |
11120 | } | |
11121 | } | |
11122 | ||
c072481d | 11123 | static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx) |
2b188cc1 | 11124 | { |
37d1e2e3 | 11125 | io_sq_thread_finish(ctx); |
2aede0e4 | 11126 | |
37d1e2e3 | 11127 | if (ctx->mm_account) { |
2aede0e4 JA |
11128 | mmdrop(ctx->mm_account); |
11129 | ctx->mm_account = NULL; | |
30975825 | 11130 | } |
def596e9 | 11131 | |
ab409402 | 11132 | io_rsrc_refs_drop(ctx); |
43597aac PB |
11133 | /* __io_rsrc_put_work() may need uring_lock to progress, wait w/o it */ |
11134 | io_wait_rsrc_data(ctx->buf_data); | |
11135 | io_wait_rsrc_data(ctx->file_data); | |
11136 | ||
8bad28d8 | 11137 | mutex_lock(&ctx->uring_lock); |
43597aac | 11138 | if (ctx->buf_data) |
bd54b6fe | 11139 | __io_sqe_buffers_unregister(ctx); |
43597aac | 11140 | if (ctx->file_data) |
08480400 | 11141 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
11142 | if (ctx->rings) |
11143 | __io_cqring_overflow_flush(ctx, true); | |
9b402849 | 11144 | io_eventfd_unregister(ctx); |
4d9237e3 | 11145 | io_flush_apoll_cache(ctx); |
77bc59b4 | 11146 | mutex_unlock(&ctx->uring_lock); |
5a2e745d | 11147 | io_destroy_buffers(ctx); |
07db298a PB |
11148 | if (ctx->sq_creds) |
11149 | put_cred(ctx->sq_creds); | |
def596e9 | 11150 | |
a7f0ed5a PB |
11151 | /* there are no registered resources left, nobody uses it */ |
11152 | if (ctx->rsrc_node) | |
11153 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 11154 | if (ctx->rsrc_backup_node) |
b895c9a6 | 11155 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a | 11156 | flush_delayed_work(&ctx->rsrc_put_work); |
756ab7c0 | 11157 | flush_delayed_work(&ctx->fallback_work); |
a7f0ed5a PB |
11158 | |
11159 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
11160 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
def596e9 | 11161 | |
2b188cc1 | 11162 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
11163 | if (ctx->ring_sock) { |
11164 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 11165 | sock_release(ctx->ring_sock); |
355e8d26 | 11166 | } |
2b188cc1 | 11167 | #endif |
ef9dd637 | 11168 | WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list)); |
2b188cc1 | 11169 | |
75b28aff | 11170 | io_mem_free(ctx->rings); |
2b188cc1 | 11171 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
11172 | |
11173 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 11174 | free_uid(ctx->user); |
4010fec4 | 11175 | io_req_caches_free(ctx); |
e941894e JA |
11176 | if (ctx->hash_map) |
11177 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 11178 | kfree(ctx->cancel_hash); |
6224843d | 11179 | kfree(ctx->dummy_ubuf); |
9cfc7e94 JA |
11180 | kfree(ctx->io_bl); |
11181 | xa_destroy(&ctx->io_bl_xa); | |
2b188cc1 JA |
11182 | kfree(ctx); |
11183 | } | |
11184 | ||
11185 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
11186 | { | |
11187 | struct io_ring_ctx *ctx = file->private_data; | |
11188 | __poll_t mask = 0; | |
11189 | ||
d60aa65b | 11190 | poll_wait(file, &ctx->cq_wait, wait); |
4f7067c3 SB |
11191 | /* |
11192 | * synchronizes with barrier from wq_has_sleeper call in | |
11193 | * io_commit_cqring | |
11194 | */ | |
2b188cc1 | 11195 | smp_rmb(); |
90554200 | 11196 | if (!io_sqring_full(ctx)) |
2b188cc1 | 11197 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
11198 | |
11199 | /* | |
11200 | * Don't flush cqring overflow list here, just do a simple check. | |
11201 | * Otherwise there could possible be ABBA deadlock: | |
11202 | * CPU0 CPU1 | |
11203 | * ---- ---- | |
11204 | * lock(&ctx->uring_lock); | |
11205 | * lock(&ep->mtx); | |
11206 | * lock(&ctx->uring_lock); | |
11207 | * lock(&ep->mtx); | |
11208 | * | |
11209 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
11210 | * pushs them to do the flush. | |
11211 | */ | |
10988a0a DY |
11212 | if (io_cqring_events(ctx) || |
11213 | test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq)) | |
2b188cc1 JA |
11214 | mask |= EPOLLIN | EPOLLRDNORM; |
11215 | ||
11216 | return mask; | |
11217 | } | |
11218 | ||
0bead8cd | 11219 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 11220 | { |
4379bf8b | 11221 | const struct cred *creds; |
071698e1 | 11222 | |
61cf9370 | 11223 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
11224 | if (creds) { |
11225 | put_cred(creds); | |
0bead8cd | 11226 | return 0; |
1e6fa521 | 11227 | } |
0bead8cd YD |
11228 | |
11229 | return -EINVAL; | |
11230 | } | |
11231 | ||
d56d938b PB |
11232 | struct io_tctx_exit { |
11233 | struct callback_head task_work; | |
11234 | struct completion completion; | |
baf186c4 | 11235 | struct io_ring_ctx *ctx; |
d56d938b PB |
11236 | }; |
11237 | ||
c072481d | 11238 | static __cold void io_tctx_exit_cb(struct callback_head *cb) |
d56d938b PB |
11239 | { |
11240 | struct io_uring_task *tctx = current->io_uring; | |
11241 | struct io_tctx_exit *work; | |
11242 | ||
11243 | work = container_of(cb, struct io_tctx_exit, task_work); | |
11244 | /* | |
11245 | * When @in_idle, we're in cancellation and it's racy to remove the | |
11246 | * node. It'll be removed by the end of cancellation, just ignore it. | |
11247 | */ | |
11248 | if (!atomic_read(&tctx->in_idle)) | |
eef51daa | 11249 | io_uring_del_tctx_node((unsigned long)work->ctx); |
d56d938b PB |
11250 | complete(&work->completion); |
11251 | } | |
11252 | ||
c072481d | 11253 | static __cold bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
28090c13 PB |
11254 | { |
11255 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
11256 | ||
11257 | return req->ctx == data; | |
11258 | } | |
11259 | ||
c072481d | 11260 | static __cold void io_ring_exit_work(struct work_struct *work) |
85faa7b8 | 11261 | { |
d56d938b | 11262 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 11263 | unsigned long timeout = jiffies + HZ * 60 * 5; |
58d3be2c | 11264 | unsigned long interval = HZ / 20; |
d56d938b PB |
11265 | struct io_tctx_exit exit; |
11266 | struct io_tctx_node *node; | |
11267 | int ret; | |
85faa7b8 | 11268 | |
56952e91 JA |
11269 | /* |
11270 | * If we're doing polled IO and end up having requests being | |
11271 | * submitted async (out-of-line), then completions can come in while | |
11272 | * we're waiting for refs to drop. We need to reap these manually, | |
11273 | * as nobody else will be looking for them. | |
11274 | */ | |
b2edc0a7 | 11275 | do { |
3dd0c97a | 11276 | io_uring_try_cancel_requests(ctx, NULL, true); |
28090c13 PB |
11277 | if (ctx->sq_data) { |
11278 | struct io_sq_data *sqd = ctx->sq_data; | |
11279 | struct task_struct *tsk; | |
11280 | ||
11281 | io_sq_thread_park(sqd); | |
11282 | tsk = sqd->thread; | |
11283 | if (tsk && tsk->io_uring && tsk->io_uring->io_wq) | |
11284 | io_wq_cancel_cb(tsk->io_uring->io_wq, | |
11285 | io_cancel_ctx_cb, ctx, true); | |
11286 | io_sq_thread_unpark(sqd); | |
11287 | } | |
b5bb3a24 | 11288 | |
37f0e767 PB |
11289 | io_req_caches_free(ctx); |
11290 | ||
58d3be2c PB |
11291 | if (WARN_ON_ONCE(time_after(jiffies, timeout))) { |
11292 | /* there is little hope left, don't run it too often */ | |
11293 | interval = HZ * 60; | |
11294 | } | |
11295 | } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); | |
d56d938b | 11296 | |
7f00651a PB |
11297 | init_completion(&exit.completion); |
11298 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
11299 | exit.ctx = ctx; | |
89b5066e PB |
11300 | /* |
11301 | * Some may use context even when all refs and requests have been put, | |
11302 | * and they are free to do so while still holding uring_lock or | |
5b0a6acc | 11303 | * completion_lock, see io_req_task_submit(). Apart from other work, |
89b5066e PB |
11304 | * this lock/unlock section also waits them to finish. |
11305 | */ | |
d56d938b PB |
11306 | mutex_lock(&ctx->uring_lock); |
11307 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
11308 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
11309 | ||
d56d938b PB |
11310 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
11311 | ctx_node); | |
7f00651a PB |
11312 | /* don't spin on a single task if cancellation failed */ |
11313 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
11314 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
11315 | if (WARN_ON_ONCE(ret)) | |
11316 | continue; | |
d56d938b PB |
11317 | |
11318 | mutex_unlock(&ctx->uring_lock); | |
11319 | wait_for_completion(&exit.completion); | |
d56d938b PB |
11320 | mutex_lock(&ctx->uring_lock); |
11321 | } | |
11322 | mutex_unlock(&ctx->uring_lock); | |
79ebeaee JA |
11323 | spin_lock(&ctx->completion_lock); |
11324 | spin_unlock(&ctx->completion_lock); | |
d56d938b | 11325 | |
85faa7b8 JA |
11326 | io_ring_ctx_free(ctx); |
11327 | } | |
11328 | ||
80c4cbdb | 11329 | /* Returns true if we found and killed one or more timeouts */ |
c072481d PB |
11330 | static __cold bool io_kill_timeouts(struct io_ring_ctx *ctx, |
11331 | struct task_struct *tsk, bool cancel_all) | |
80c4cbdb PB |
11332 | { |
11333 | struct io_kiocb *req, *tmp; | |
11334 | int canceled = 0; | |
11335 | ||
79ebeaee JA |
11336 | spin_lock(&ctx->completion_lock); |
11337 | spin_lock_irq(&ctx->timeout_lock); | |
80c4cbdb | 11338 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
3dd0c97a | 11339 | if (io_match_task(req, tsk, cancel_all)) { |
80c4cbdb PB |
11340 | io_kill_timeout(req, -ECANCELED); |
11341 | canceled++; | |
11342 | } | |
11343 | } | |
79ebeaee | 11344 | spin_unlock_irq(&ctx->timeout_lock); |
60053be8 | 11345 | io_commit_cqring(ctx); |
79ebeaee | 11346 | spin_unlock(&ctx->completion_lock); |
80c4cbdb PB |
11347 | if (canceled != 0) |
11348 | io_cqring_ev_posted(ctx); | |
11349 | return canceled != 0; | |
11350 | } | |
11351 | ||
c072481d | 11352 | static __cold void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
2b188cc1 | 11353 | { |
61cf9370 MWO |
11354 | unsigned long index; |
11355 | struct creds *creds; | |
11356 | ||
2b188cc1 JA |
11357 | mutex_lock(&ctx->uring_lock); |
11358 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 11359 | if (ctx->rings) |
6c2450ae | 11360 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
11361 | xa_for_each(&ctx->personalities, index, creds) |
11362 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
11363 | mutex_unlock(&ctx->uring_lock); |
11364 | ||
60053be8 PB |
11365 | /* failed during ring init, it couldn't have issued any requests */ |
11366 | if (ctx->rings) { | |
11367 | io_kill_timeouts(ctx, NULL, true); | |
11368 | io_poll_remove_all(ctx, NULL, true); | |
11369 | /* if we failed setting up the ctx, we might not have any rings */ | |
11370 | io_iopoll_try_reap_events(ctx); | |
11371 | } | |
309fc03a | 11372 | |
85faa7b8 | 11373 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
11374 | /* |
11375 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
11376 | * if we're exiting a ton of rings at the same time. It just adds | |
11377 | * noise and overhead, there's no discernable change in runtime | |
11378 | * over using system_wq. | |
11379 | */ | |
11380 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
11381 | } |
11382 | ||
11383 | static int io_uring_release(struct inode *inode, struct file *file) | |
11384 | { | |
11385 | struct io_ring_ctx *ctx = file->private_data; | |
11386 | ||
11387 | file->private_data = NULL; | |
11388 | io_ring_ctx_wait_and_kill(ctx); | |
11389 | return 0; | |
11390 | } | |
11391 | ||
f6edbabb PB |
11392 | struct io_task_cancel { |
11393 | struct task_struct *task; | |
3dd0c97a | 11394 | bool all; |
f6edbabb | 11395 | }; |
f254ac04 | 11396 | |
f6edbabb | 11397 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 11398 | { |
9a472ef7 | 11399 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 11400 | struct io_task_cancel *cancel = data; |
9a472ef7 | 11401 | |
6af3f48b | 11402 | return io_match_task_safe(req, cancel->task, cancel->all); |
b711d4ea JA |
11403 | } |
11404 | ||
c072481d PB |
11405 | static __cold bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
11406 | struct task_struct *task, | |
11407 | bool cancel_all) | |
b7ddce3c | 11408 | { |
e1915f76 | 11409 | struct io_defer_entry *de; |
b7ddce3c PB |
11410 | LIST_HEAD(list); |
11411 | ||
79ebeaee | 11412 | spin_lock(&ctx->completion_lock); |
b7ddce3c | 11413 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { |
6af3f48b | 11414 | if (io_match_task_safe(de->req, task, cancel_all)) { |
b7ddce3c PB |
11415 | list_cut_position(&list, &ctx->defer_list, &de->list); |
11416 | break; | |
11417 | } | |
11418 | } | |
79ebeaee | 11419 | spin_unlock(&ctx->completion_lock); |
e1915f76 PB |
11420 | if (list_empty(&list)) |
11421 | return false; | |
b7ddce3c PB |
11422 | |
11423 | while (!list_empty(&list)) { | |
11424 | de = list_first_entry(&list, struct io_defer_entry, list); | |
11425 | list_del_init(&de->list); | |
f41db273 | 11426 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
11427 | kfree(de); |
11428 | } | |
e1915f76 | 11429 | return true; |
b7ddce3c PB |
11430 | } |
11431 | ||
c072481d | 11432 | static __cold bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) |
1b00764f PB |
11433 | { |
11434 | struct io_tctx_node *node; | |
11435 | enum io_wq_cancel cret; | |
11436 | bool ret = false; | |
11437 | ||
11438 | mutex_lock(&ctx->uring_lock); | |
11439 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
11440 | struct io_uring_task *tctx = node->task->io_uring; | |
11441 | ||
11442 | /* | |
11443 | * io_wq will stay alive while we hold uring_lock, because it's | |
11444 | * killed after ctx nodes, which requires to take the lock. | |
11445 | */ | |
11446 | if (!tctx || !tctx->io_wq) | |
11447 | continue; | |
11448 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
11449 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
11450 | } | |
11451 | mutex_unlock(&ctx->uring_lock); | |
11452 | ||
11453 | return ret; | |
11454 | } | |
11455 | ||
c072481d PB |
11456 | static __cold void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
11457 | struct task_struct *task, | |
11458 | bool cancel_all) | |
9936c7c2 | 11459 | { |
3dd0c97a | 11460 | struct io_task_cancel cancel = { .task = task, .all = cancel_all, }; |
1b00764f | 11461 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 | 11462 | |
60053be8 PB |
11463 | /* failed during ring init, it couldn't have issued any requests */ |
11464 | if (!ctx->rings) | |
11465 | return; | |
11466 | ||
9936c7c2 PB |
11467 | while (1) { |
11468 | enum io_wq_cancel cret; | |
11469 | bool ret = false; | |
11470 | ||
1b00764f PB |
11471 | if (!task) { |
11472 | ret |= io_uring_try_cancel_iowq(ctx); | |
11473 | } else if (tctx && tctx->io_wq) { | |
11474 | /* | |
11475 | * Cancels requests of all rings, not only @ctx, but | |
11476 | * it's fine as the task is in exit/exec. | |
11477 | */ | |
5aa75ed5 | 11478 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
11479 | &cancel, true); |
11480 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
11481 | } | |
11482 | ||
11483 | /* SQPOLL thread does its own polling */ | |
3dd0c97a | 11484 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || |
d052d1d6 | 11485 | (ctx->sq_data && ctx->sq_data->thread == current)) { |
5eef4e87 | 11486 | while (!wq_list_empty(&ctx->iopoll_list)) { |
9936c7c2 PB |
11487 | io_iopoll_try_reap_events(ctx); |
11488 | ret = true; | |
11489 | } | |
11490 | } | |
11491 | ||
3dd0c97a PB |
11492 | ret |= io_cancel_defer_files(ctx, task, cancel_all); |
11493 | ret |= io_poll_remove_all(ctx, task, cancel_all); | |
11494 | ret |= io_kill_timeouts(ctx, task, cancel_all); | |
e5dc480d PB |
11495 | if (task) |
11496 | ret |= io_run_task_work(); | |
9936c7c2 PB |
11497 | if (!ret) |
11498 | break; | |
11499 | cond_resched(); | |
11500 | } | |
11501 | } | |
11502 | ||
eef51daa | 11503 | static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
0f212204 | 11504 | { |
236434c3 | 11505 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 11506 | struct io_tctx_node *node; |
a528b04e | 11507 | int ret; |
236434c3 MWO |
11508 | |
11509 | if (unlikely(!tctx)) { | |
5aa75ed5 | 11510 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
11511 | if (unlikely(ret)) |
11512 | return ret; | |
e139a1ec | 11513 | |
236434c3 | 11514 | tctx = current->io_uring; |
e139a1ec PB |
11515 | if (ctx->iowq_limits_set) { |
11516 | unsigned int limits[2] = { ctx->iowq_limits[0], | |
11517 | ctx->iowq_limits[1], }; | |
11518 | ||
11519 | ret = io_wq_max_workers(tctx->io_wq, limits); | |
11520 | if (ret) | |
11521 | return ret; | |
11522 | } | |
0f212204 | 11523 | } |
cf27f3b1 PB |
11524 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
11525 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
11526 | if (!node) | |
11527 | return -ENOMEM; | |
11528 | node->ctx = ctx; | |
11529 | node->task = current; | |
13bf43f5 | 11530 | |
cf27f3b1 PB |
11531 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
11532 | node, GFP_KERNEL)); | |
11533 | if (ret) { | |
11534 | kfree(node); | |
11535 | return ret; | |
0f212204 | 11536 | } |
cf27f3b1 PB |
11537 | |
11538 | mutex_lock(&ctx->uring_lock); | |
11539 | list_add(&node->ctx_node, &ctx->tctx_list); | |
11540 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 11541 | } |
cf27f3b1 | 11542 | tctx->last = ctx; |
0f212204 JA |
11543 | return 0; |
11544 | } | |
11545 | ||
cf27f3b1 PB |
11546 | /* |
11547 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
11548 | */ | |
eef51daa | 11549 | static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx) |
cf27f3b1 PB |
11550 | { |
11551 | struct io_uring_task *tctx = current->io_uring; | |
11552 | ||
11553 | if (likely(tctx && tctx->last == ctx)) | |
11554 | return 0; | |
eef51daa | 11555 | return __io_uring_add_tctx_node(ctx); |
cf27f3b1 PB |
11556 | } |
11557 | ||
0f212204 JA |
11558 | /* |
11559 | * Remove this io_uring_file -> task mapping. | |
11560 | */ | |
c072481d | 11561 | static __cold void io_uring_del_tctx_node(unsigned long index) |
0f212204 JA |
11562 | { |
11563 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 11564 | struct io_tctx_node *node; |
2941267b | 11565 | |
eebd2e37 PB |
11566 | if (!tctx) |
11567 | return; | |
13bf43f5 PB |
11568 | node = xa_erase(&tctx->xa, index); |
11569 | if (!node) | |
2941267b | 11570 | return; |
0f212204 | 11571 | |
13bf43f5 PB |
11572 | WARN_ON_ONCE(current != node->task); |
11573 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
11574 | ||
11575 | mutex_lock(&node->ctx->uring_lock); | |
11576 | list_del(&node->ctx_node); | |
11577 | mutex_unlock(&node->ctx->uring_lock); | |
11578 | ||
baf186c4 | 11579 | if (tctx->last == node->ctx) |
0f212204 | 11580 | tctx->last = NULL; |
13bf43f5 | 11581 | kfree(node); |
0f212204 JA |
11582 | } |
11583 | ||
c072481d | 11584 | static __cold void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 11585 | { |
ba5ef6dc | 11586 | struct io_wq *wq = tctx->io_wq; |
13bf43f5 | 11587 | struct io_tctx_node *node; |
de7f1d9e PB |
11588 | unsigned long index; |
11589 | ||
8bab4c09 | 11590 | xa_for_each(&tctx->xa, index, node) { |
eef51daa | 11591 | io_uring_del_tctx_node(index); |
8bab4c09 JA |
11592 | cond_resched(); |
11593 | } | |
b16ef427 ME |
11594 | if (wq) { |
11595 | /* | |
f6f9b278 | 11596 | * Must be after io_uring_del_tctx_node() (removes nodes under |
b16ef427 ME |
11597 | * uring_lock) to avoid race with io_uring_try_cancel_iowq(). |
11598 | */ | |
ba5ef6dc | 11599 | io_wq_put_and_exit(wq); |
dadebc35 | 11600 | tctx->io_wq = NULL; |
b16ef427 | 11601 | } |
de7f1d9e PB |
11602 | } |
11603 | ||
3f48cf18 | 11604 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 11605 | { |
3f48cf18 | 11606 | if (tracked) |
9cae36a0 | 11607 | return atomic_read(&tctx->inflight_tracked); |
521d6a73 PB |
11608 | return percpu_counter_sum(&tctx->inflight); |
11609 | } | |
11610 | ||
78cc687b PB |
11611 | /* |
11612 | * Find any io_uring ctx that this task has registered or done IO on, and cancel | |
78a78060 | 11613 | * requests. @sqd should be not-null IFF it's an SQPOLL thread cancellation. |
78cc687b | 11614 | */ |
c072481d PB |
11615 | static __cold void io_uring_cancel_generic(bool cancel_all, |
11616 | struct io_sq_data *sqd) | |
0e9ddb39 | 11617 | { |
521d6a73 | 11618 | struct io_uring_task *tctx = current->io_uring; |
734551df | 11619 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
11620 | s64 inflight; |
11621 | DEFINE_WAIT(wait); | |
fdaf083c | 11622 | |
78cc687b PB |
11623 | WARN_ON_ONCE(sqd && sqd->thread != current); |
11624 | ||
6d042ffb PO |
11625 | if (!current->io_uring) |
11626 | return; | |
17a91051 PB |
11627 | if (tctx->io_wq) |
11628 | io_wq_exit_start(tctx->io_wq); | |
11629 | ||
0e9ddb39 PB |
11630 | atomic_inc(&tctx->in_idle); |
11631 | do { | |
e9dbe221 | 11632 | io_uring_drop_tctx_refs(current); |
0e9ddb39 | 11633 | /* read completions before cancelations */ |
78cc687b | 11634 | inflight = tctx_inflight(tctx, !cancel_all); |
0e9ddb39 PB |
11635 | if (!inflight) |
11636 | break; | |
fdaf083c | 11637 | |
78cc687b PB |
11638 | if (!sqd) { |
11639 | struct io_tctx_node *node; | |
11640 | unsigned long index; | |
0f212204 | 11641 | |
78cc687b PB |
11642 | xa_for_each(&tctx->xa, index, node) { |
11643 | /* sqpoll task will cancel all its requests */ | |
11644 | if (node->ctx->sq_data) | |
11645 | continue; | |
11646 | io_uring_try_cancel_requests(node->ctx, current, | |
11647 | cancel_all); | |
11648 | } | |
11649 | } else { | |
11650 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) | |
11651 | io_uring_try_cancel_requests(ctx, current, | |
11652 | cancel_all); | |
11653 | } | |
17a91051 | 11654 | |
78a78060 JA |
11655 | prepare_to_wait(&tctx->wait, &wait, TASK_INTERRUPTIBLE); |
11656 | io_run_task_work(); | |
e9dbe221 | 11657 | io_uring_drop_tctx_refs(current); |
78a78060 | 11658 | |
0f212204 | 11659 | /* |
a1bb3cd5 PB |
11660 | * If we've seen completions, retry without waiting. This |
11661 | * avoids a race where a completion comes in before we did | |
11662 | * prepare_to_wait(). | |
0f212204 | 11663 | */ |
3dd0c97a | 11664 | if (inflight == tctx_inflight(tctx, !cancel_all)) |
a1bb3cd5 | 11665 | schedule(); |
f57555ed | 11666 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 11667 | } while (1); |
de7f1d9e | 11668 | |
8452d4a6 | 11669 | io_uring_clean_tctx(tctx); |
3dd0c97a | 11670 | if (cancel_all) { |
3cc7fdb9 PB |
11671 | /* |
11672 | * We shouldn't run task_works after cancel, so just leave | |
11673 | * ->in_idle set for normal exit. | |
11674 | */ | |
11675 | atomic_dec(&tctx->in_idle); | |
3f48cf18 PB |
11676 | /* for exec all current's requests should be gone, kill tctx */ |
11677 | __io_uring_free(current); | |
11678 | } | |
44e728b8 PB |
11679 | } |
11680 | ||
f552a27a | 11681 | void __io_uring_cancel(bool cancel_all) |
78cc687b | 11682 | { |
f552a27a | 11683 | io_uring_cancel_generic(cancel_all, NULL); |
78cc687b PB |
11684 | } |
11685 | ||
e7a6c00d JA |
11686 | void io_uring_unreg_ringfd(void) |
11687 | { | |
11688 | struct io_uring_task *tctx = current->io_uring; | |
11689 | int i; | |
11690 | ||
11691 | for (i = 0; i < IO_RINGFD_REG_MAX; i++) { | |
11692 | if (tctx->registered_rings[i]) { | |
11693 | fput(tctx->registered_rings[i]); | |
11694 | tctx->registered_rings[i] = NULL; | |
11695 | } | |
11696 | } | |
11697 | } | |
11698 | ||
11699 | static int io_ring_add_registered_fd(struct io_uring_task *tctx, int fd, | |
11700 | int start, int end) | |
11701 | { | |
11702 | struct file *file; | |
11703 | int offset; | |
11704 | ||
11705 | for (offset = start; offset < end; offset++) { | |
11706 | offset = array_index_nospec(offset, IO_RINGFD_REG_MAX); | |
11707 | if (tctx->registered_rings[offset]) | |
11708 | continue; | |
11709 | ||
11710 | file = fget(fd); | |
11711 | if (!file) { | |
11712 | return -EBADF; | |
11713 | } else if (file->f_op != &io_uring_fops) { | |
11714 | fput(file); | |
11715 | return -EOPNOTSUPP; | |
11716 | } | |
11717 | tctx->registered_rings[offset] = file; | |
11718 | return offset; | |
11719 | } | |
11720 | ||
11721 | return -EBUSY; | |
11722 | } | |
11723 | ||
11724 | /* | |
11725 | * Register a ring fd to avoid fdget/fdput for each io_uring_enter() | |
11726 | * invocation. User passes in an array of struct io_uring_rsrc_update | |
11727 | * with ->data set to the ring_fd, and ->offset given for the desired | |
11728 | * index. If no index is desired, application may set ->offset == -1U | |
11729 | * and we'll find an available index. Returns number of entries | |
11730 | * successfully processed, or < 0 on error if none were processed. | |
11731 | */ | |
11732 | static int io_ringfd_register(struct io_ring_ctx *ctx, void __user *__arg, | |
11733 | unsigned nr_args) | |
11734 | { | |
11735 | struct io_uring_rsrc_update __user *arg = __arg; | |
11736 | struct io_uring_rsrc_update reg; | |
11737 | struct io_uring_task *tctx; | |
11738 | int ret, i; | |
11739 | ||
11740 | if (!nr_args || nr_args > IO_RINGFD_REG_MAX) | |
11741 | return -EINVAL; | |
11742 | ||
11743 | mutex_unlock(&ctx->uring_lock); | |
11744 | ret = io_uring_add_tctx_node(ctx); | |
11745 | mutex_lock(&ctx->uring_lock); | |
11746 | if (ret) | |
11747 | return ret; | |
11748 | ||
11749 | tctx = current->io_uring; | |
11750 | for (i = 0; i < nr_args; i++) { | |
11751 | int start, end; | |
11752 | ||
11753 | if (copy_from_user(®, &arg[i], sizeof(reg))) { | |
11754 | ret = -EFAULT; | |
11755 | break; | |
11756 | } | |
11757 | ||
6fb53cf8 DY |
11758 | if (reg.resv) { |
11759 | ret = -EINVAL; | |
11760 | break; | |
11761 | } | |
11762 | ||
e7a6c00d JA |
11763 | if (reg.offset == -1U) { |
11764 | start = 0; | |
11765 | end = IO_RINGFD_REG_MAX; | |
11766 | } else { | |
11767 | if (reg.offset >= IO_RINGFD_REG_MAX) { | |
11768 | ret = -EINVAL; | |
11769 | break; | |
11770 | } | |
11771 | start = reg.offset; | |
11772 | end = start + 1; | |
11773 | } | |
11774 | ||
11775 | ret = io_ring_add_registered_fd(tctx, reg.data, start, end); | |
11776 | if (ret < 0) | |
11777 | break; | |
11778 | ||
11779 | reg.offset = ret; | |
11780 | if (copy_to_user(&arg[i], ®, sizeof(reg))) { | |
11781 | fput(tctx->registered_rings[reg.offset]); | |
11782 | tctx->registered_rings[reg.offset] = NULL; | |
11783 | ret = -EFAULT; | |
11784 | break; | |
11785 | } | |
11786 | } | |
11787 | ||
11788 | return i ? i : ret; | |
11789 | } | |
11790 | ||
11791 | static int io_ringfd_unregister(struct io_ring_ctx *ctx, void __user *__arg, | |
11792 | unsigned nr_args) | |
11793 | { | |
11794 | struct io_uring_rsrc_update __user *arg = __arg; | |
11795 | struct io_uring_task *tctx = current->io_uring; | |
11796 | struct io_uring_rsrc_update reg; | |
11797 | int ret = 0, i; | |
11798 | ||
11799 | if (!nr_args || nr_args > IO_RINGFD_REG_MAX) | |
11800 | return -EINVAL; | |
11801 | if (!tctx) | |
11802 | return 0; | |
11803 | ||
11804 | for (i = 0; i < nr_args; i++) { | |
11805 | if (copy_from_user(®, &arg[i], sizeof(reg))) { | |
11806 | ret = -EFAULT; | |
11807 | break; | |
11808 | } | |
303cc749 | 11809 | if (reg.resv || reg.data || reg.offset >= IO_RINGFD_REG_MAX) { |
e7a6c00d JA |
11810 | ret = -EINVAL; |
11811 | break; | |
11812 | } | |
11813 | ||
11814 | reg.offset = array_index_nospec(reg.offset, IO_RINGFD_REG_MAX); | |
11815 | if (tctx->registered_rings[reg.offset]) { | |
11816 | fput(tctx->registered_rings[reg.offset]); | |
11817 | tctx->registered_rings[reg.offset] = NULL; | |
11818 | } | |
11819 | } | |
11820 | ||
11821 | return i ? i : ret; | |
11822 | } | |
11823 | ||
6c5c240e RP |
11824 | static void *io_uring_validate_mmap_request(struct file *file, |
11825 | loff_t pgoff, size_t sz) | |
2b188cc1 | 11826 | { |
2b188cc1 | 11827 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 11828 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
11829 | struct page *page; |
11830 | void *ptr; | |
11831 | ||
11832 | switch (offset) { | |
11833 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
11834 | case IORING_OFF_CQ_RING: |
11835 | ptr = ctx->rings; | |
2b188cc1 JA |
11836 | break; |
11837 | case IORING_OFF_SQES: | |
11838 | ptr = ctx->sq_sqes; | |
11839 | break; | |
2b188cc1 | 11840 | default: |
6c5c240e | 11841 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
11842 | } |
11843 | ||
11844 | page = virt_to_head_page(ptr); | |
a50b854e | 11845 | if (sz > page_size(page)) |
6c5c240e RP |
11846 | return ERR_PTR(-EINVAL); |
11847 | ||
11848 | return ptr; | |
11849 | } | |
11850 | ||
11851 | #ifdef CONFIG_MMU | |
11852 | ||
c072481d | 11853 | static __cold int io_uring_mmap(struct file *file, struct vm_area_struct *vma) |
6c5c240e RP |
11854 | { |
11855 | size_t sz = vma->vm_end - vma->vm_start; | |
11856 | unsigned long pfn; | |
11857 | void *ptr; | |
11858 | ||
11859 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
11860 | if (IS_ERR(ptr)) | |
11861 | return PTR_ERR(ptr); | |
2b188cc1 JA |
11862 | |
11863 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
11864 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
11865 | } | |
11866 | ||
6c5c240e RP |
11867 | #else /* !CONFIG_MMU */ |
11868 | ||
11869 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
11870 | { | |
11871 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
11872 | } | |
11873 | ||
11874 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
11875 | { | |
11876 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
11877 | } | |
11878 | ||
11879 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
11880 | unsigned long addr, unsigned long len, | |
11881 | unsigned long pgoff, unsigned long flags) | |
11882 | { | |
11883 | void *ptr; | |
11884 | ||
11885 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
11886 | if (IS_ERR(ptr)) | |
11887 | return PTR_ERR(ptr); | |
11888 | ||
11889 | return (unsigned long) ptr; | |
11890 | } | |
11891 | ||
11892 | #endif /* !CONFIG_MMU */ | |
11893 | ||
d9d05217 | 11894 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
11895 | { |
11896 | DEFINE_WAIT(wait); | |
11897 | ||
11898 | do { | |
11899 | if (!io_sqring_full(ctx)) | |
11900 | break; | |
90554200 JA |
11901 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
11902 | ||
11903 | if (!io_sqring_full(ctx)) | |
11904 | break; | |
90554200 JA |
11905 | schedule(); |
11906 | } while (!signal_pending(current)); | |
11907 | ||
11908 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 11909 | return 0; |
90554200 JA |
11910 | } |
11911 | ||
f81440d3 PB |
11912 | static int io_validate_ext_arg(unsigned flags, const void __user *argp, size_t argsz) |
11913 | { | |
11914 | if (flags & IORING_ENTER_EXT_ARG) { | |
11915 | struct io_uring_getevents_arg arg; | |
11916 | ||
11917 | if (argsz != sizeof(arg)) | |
11918 | return -EINVAL; | |
11919 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
11920 | return -EFAULT; | |
11921 | } | |
11922 | return 0; | |
11923 | } | |
11924 | ||
c73ebb68 HX |
11925 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
11926 | struct __kernel_timespec __user **ts, | |
11927 | const sigset_t __user **sig) | |
11928 | { | |
11929 | struct io_uring_getevents_arg arg; | |
11930 | ||
11931 | /* | |
11932 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
11933 | * is just a pointer to the sigset_t. | |
11934 | */ | |
11935 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
11936 | *sig = (const sigset_t __user *) argp; | |
11937 | *ts = NULL; | |
11938 | return 0; | |
11939 | } | |
11940 | ||
11941 | /* | |
11942 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
11943 | * timespec and sigset_t pointers if good. | |
11944 | */ | |
11945 | if (*argsz != sizeof(arg)) | |
11946 | return -EINVAL; | |
11947 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
11948 | return -EFAULT; | |
d2347b96 DY |
11949 | if (arg.pad) |
11950 | return -EINVAL; | |
c73ebb68 HX |
11951 | *sig = u64_to_user_ptr(arg.sigmask); |
11952 | *argsz = arg.sigmask_sz; | |
11953 | *ts = u64_to_user_ptr(arg.ts); | |
11954 | return 0; | |
11955 | } | |
11956 | ||
2b188cc1 | 11957 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
11958 | u32, min_complete, u32, flags, const void __user *, argp, |
11959 | size_t, argsz) | |
2b188cc1 JA |
11960 | { |
11961 | struct io_ring_ctx *ctx; | |
2b188cc1 | 11962 | struct fd f; |
33f993da | 11963 | long ret; |
2b188cc1 | 11964 | |
4c6e277c | 11965 | io_run_task_work(); |
b41e9852 | 11966 | |
33f993da | 11967 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
e7a6c00d JA |
11968 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG | |
11969 | IORING_ENTER_REGISTERED_RING))) | |
2b188cc1 JA |
11970 | return -EINVAL; |
11971 | ||
e7a6c00d JA |
11972 | /* |
11973 | * Ring fd has been registered via IORING_REGISTER_RING_FDS, we | |
11974 | * need only dereference our task private array to find it. | |
11975 | */ | |
11976 | if (flags & IORING_ENTER_REGISTERED_RING) { | |
11977 | struct io_uring_task *tctx = current->io_uring; | |
11978 | ||
11979 | if (!tctx || fd >= IO_RINGFD_REG_MAX) | |
11980 | return -EINVAL; | |
11981 | fd = array_index_nospec(fd, IO_RINGFD_REG_MAX); | |
11982 | f.file = tctx->registered_rings[fd]; | |
4329490a | 11983 | f.flags = 0; |
e7a6c00d JA |
11984 | } else { |
11985 | f = fdget(fd); | |
e7a6c00d | 11986 | } |
2b188cc1 | 11987 | |
4329490a AV |
11988 | if (unlikely(!f.file)) |
11989 | return -EBADF; | |
11990 | ||
2b188cc1 | 11991 | ret = -EOPNOTSUPP; |
33f993da | 11992 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
11993 | goto out_fput; |
11994 | ||
11995 | ret = -ENXIO; | |
11996 | ctx = f.file->private_data; | |
33f993da | 11997 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
11998 | goto out_fput; |
11999 | ||
7e84e1c7 | 12000 | ret = -EBADFD; |
33f993da | 12001 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
12002 | goto out; |
12003 | ||
6c271ce2 JA |
12004 | /* |
12005 | * For SQ polling, the thread will do all submissions and completions. | |
12006 | * Just return the requested submit count, and wake the thread if | |
12007 | * we were asked to. | |
12008 | */ | |
b2a9eada | 12009 | ret = 0; |
6c271ce2 | 12010 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
90f67366 | 12011 | io_cqring_overflow_flush(ctx); |
89448c47 | 12012 | |
21f96522 JA |
12013 | if (unlikely(ctx->sq_data->thread == NULL)) { |
12014 | ret = -EOWNERDEAD; | |
04147488 | 12015 | goto out; |
21f96522 | 12016 | } |
6c271ce2 | 12017 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 12018 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
12019 | if (flags & IORING_ENTER_SQ_WAIT) { |
12020 | ret = io_sqpoll_wait_sq(ctx); | |
12021 | if (ret) | |
12022 | goto out; | |
12023 | } | |
3e813c90 | 12024 | ret = to_submit; |
b2a9eada | 12025 | } else if (to_submit) { |
eef51daa | 12026 | ret = io_uring_add_tctx_node(ctx); |
0f212204 JA |
12027 | if (unlikely(ret)) |
12028 | goto out; | |
7c504e65 | 12029 | |
2b188cc1 | 12030 | mutex_lock(&ctx->uring_lock); |
3e813c90 DY |
12031 | ret = io_submit_sqes(ctx, to_submit); |
12032 | if (ret != to_submit) { | |
d487b43c | 12033 | mutex_unlock(&ctx->uring_lock); |
7c504e65 | 12034 | goto out; |
d487b43c PB |
12035 | } |
12036 | if ((flags & IORING_ENTER_GETEVENTS) && ctx->syscall_iopoll) | |
12037 | goto iopoll_locked; | |
12038 | mutex_unlock(&ctx->uring_lock); | |
2b188cc1 JA |
12039 | } |
12040 | if (flags & IORING_ENTER_GETEVENTS) { | |
3e813c90 | 12041 | int ret2; |
773697b6 | 12042 | if (ctx->syscall_iopoll) { |
d487b43c PB |
12043 | /* |
12044 | * We disallow the app entering submit/complete with | |
12045 | * polling, but we still need to lock the ring to | |
12046 | * prevent racing with polled issue that got punted to | |
12047 | * a workqueue. | |
12048 | */ | |
12049 | mutex_lock(&ctx->uring_lock); | |
12050 | iopoll_locked: | |
3e813c90 DY |
12051 | ret2 = io_validate_ext_arg(flags, argp, argsz); |
12052 | if (likely(!ret2)) { | |
12053 | min_complete = min(min_complete, | |
12054 | ctx->cq_entries); | |
12055 | ret2 = io_iopoll_check(ctx, min_complete); | |
d487b43c PB |
12056 | } |
12057 | mutex_unlock(&ctx->uring_lock); | |
def596e9 | 12058 | } else { |
f81440d3 PB |
12059 | const sigset_t __user *sig; |
12060 | struct __kernel_timespec __user *ts; | |
12061 | ||
3e813c90 DY |
12062 | ret2 = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); |
12063 | if (likely(!ret2)) { | |
12064 | min_complete = min(min_complete, | |
12065 | ctx->cq_entries); | |
12066 | ret2 = io_cqring_wait(ctx, min_complete, sig, | |
12067 | argsz, ts); | |
12068 | } | |
def596e9 | 12069 | } |
c73ebb68 | 12070 | |
155bc950 | 12071 | if (!ret) { |
3e813c90 | 12072 | ret = ret2; |
2b188cc1 | 12073 | |
155bc950 DY |
12074 | /* |
12075 | * EBADR indicates that one or more CQE were dropped. | |
12076 | * Once the user has been informed we can clear the bit | |
12077 | * as they are obviously ok with those drops. | |
12078 | */ | |
12079 | if (unlikely(ret2 == -EBADR)) | |
12080 | clear_bit(IO_CHECK_CQ_DROPPED_BIT, | |
12081 | &ctx->check_cq); | |
def596e9 | 12082 | } |
2b188cc1 JA |
12083 | } |
12084 | ||
7c504e65 | 12085 | out: |
6805b32e | 12086 | percpu_ref_put(&ctx->refs); |
2b188cc1 | 12087 | out_fput: |
4329490a | 12088 | fdput(f); |
3e813c90 | 12089 | return ret; |
2b188cc1 JA |
12090 | } |
12091 | ||
bebdb65e | 12092 | #ifdef CONFIG_PROC_FS |
c072481d | 12093 | static __cold int io_uring_show_cred(struct seq_file *m, unsigned int id, |
61cf9370 | 12094 | const struct cred *cred) |
87ce955b | 12095 | { |
87ce955b JA |
12096 | struct user_namespace *uns = seq_user_ns(m); |
12097 | struct group_info *gi; | |
12098 | kernel_cap_t cap; | |
12099 | unsigned __capi; | |
12100 | int g; | |
12101 | ||
12102 | seq_printf(m, "%5d\n", id); | |
12103 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
12104 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
12105 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
12106 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
12107 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
12108 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
12109 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
12110 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
12111 | seq_puts(m, "\n\tGroups:\t"); | |
12112 | gi = cred->group_info; | |
12113 | for (g = 0; g < gi->ngroups; g++) { | |
12114 | seq_put_decimal_ull(m, g ? " " : "", | |
12115 | from_kgid_munged(uns, gi->gid[g])); | |
12116 | } | |
12117 | seq_puts(m, "\n\tCapEff:\t"); | |
12118 | cap = cred->cap_effective; | |
12119 | CAP_FOR_EACH_U32(__capi) | |
12120 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
12121 | seq_putc(m, '\n'); | |
12122 | return 0; | |
12123 | } | |
12124 | ||
c072481d PB |
12125 | static __cold void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, |
12126 | struct seq_file *m) | |
87ce955b | 12127 | { |
dbbe9c64 | 12128 | struct io_sq_data *sq = NULL; |
83f84356 HX |
12129 | struct io_overflow_cqe *ocqe; |
12130 | struct io_rings *r = ctx->rings; | |
12131 | unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1; | |
83f84356 HX |
12132 | unsigned int sq_head = READ_ONCE(r->sq.head); |
12133 | unsigned int sq_tail = READ_ONCE(r->sq.tail); | |
12134 | unsigned int cq_head = READ_ONCE(r->cq.head); | |
12135 | unsigned int cq_tail = READ_ONCE(r->cq.tail); | |
f9b3dfcc | 12136 | unsigned int cq_shift = 0; |
f75d1183 | 12137 | unsigned int sq_entries, cq_entries; |
fad8e0de | 12138 | bool has_lock; |
f9b3dfcc | 12139 | bool is_cqe32 = (ctx->flags & IORING_SETUP_CQE32); |
83f84356 HX |
12140 | unsigned int i; |
12141 | ||
f9b3dfcc SR |
12142 | if (is_cqe32) |
12143 | cq_shift = 1; | |
12144 | ||
83f84356 HX |
12145 | /* |
12146 | * we may get imprecise sqe and cqe info if uring is actively running | |
12147 | * since we get cached_sq_head and cached_cq_tail without uring_lock | |
12148 | * and sq_tail and cq_head are changed by userspace. But it's ok since | |
12149 | * we usually use these info when it is stuck. | |
12150 | */ | |
c0235652 | 12151 | seq_printf(m, "SqMask:\t0x%x\n", sq_mask); |
f75d1183 JA |
12152 | seq_printf(m, "SqHead:\t%u\n", sq_head); |
12153 | seq_printf(m, "SqTail:\t%u\n", sq_tail); | |
12154 | seq_printf(m, "CachedSqHead:\t%u\n", ctx->cached_sq_head); | |
12155 | seq_printf(m, "CqMask:\t0x%x\n", cq_mask); | |
12156 | seq_printf(m, "CqHead:\t%u\n", cq_head); | |
12157 | seq_printf(m, "CqTail:\t%u\n", cq_tail); | |
12158 | seq_printf(m, "CachedCqTail:\t%u\n", ctx->cached_cq_tail); | |
12159 | seq_printf(m, "SQEs:\t%u\n", sq_tail - ctx->cached_sq_head); | |
12160 | sq_entries = min(sq_tail - sq_head, ctx->sq_entries); | |
12161 | for (i = 0; i < sq_entries; i++) { | |
12162 | unsigned int entry = i + sq_head; | |
12163 | unsigned int sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]); | |
a1957780 | 12164 | struct io_uring_sqe *sqe; |
f75d1183 JA |
12165 | |
12166 | if (sq_idx > sq_mask) | |
12167 | continue; | |
12168 | sqe = &ctx->sq_sqes[sq_idx]; | |
12169 | seq_printf(m, "%5u: opcode:%d, fd:%d, flags:%x, user_data:%llu\n", | |
12170 | sq_idx, sqe->opcode, sqe->fd, sqe->flags, | |
12171 | sqe->user_data); | |
83f84356 | 12172 | } |
f75d1183 JA |
12173 | seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head); |
12174 | cq_entries = min(cq_tail - cq_head, ctx->cq_entries); | |
12175 | for (i = 0; i < cq_entries; i++) { | |
12176 | unsigned int entry = i + cq_head; | |
f9b3dfcc | 12177 | struct io_uring_cqe *cqe = &r->cqes[(entry & cq_mask) << cq_shift]; |
83f84356 | 12178 | |
f9b3dfcc SR |
12179 | if (!is_cqe32) { |
12180 | seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x\n", | |
f75d1183 JA |
12181 | entry & cq_mask, cqe->user_data, cqe->res, |
12182 | cqe->flags); | |
f9b3dfcc SR |
12183 | } else { |
12184 | seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x, " | |
12185 | "extra1:%llu, extra2:%llu\n", | |
12186 | entry & cq_mask, cqe->user_data, cqe->res, | |
12187 | cqe->flags, cqe->big_cqe[0], cqe->big_cqe[1]); | |
12188 | } | |
83f84356 | 12189 | } |
87ce955b | 12190 | |
fad8e0de JA |
12191 | /* |
12192 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
12193 | * since fdinfo case grabs it in the opposite direction of normal use | |
12194 | * cases. If we fail to get the lock, we just don't iterate any | |
12195 | * structures that could be going away outside the io_uring mutex. | |
12196 | */ | |
12197 | has_lock = mutex_trylock(&ctx->uring_lock); | |
12198 | ||
5f3f26f9 | 12199 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 12200 | sq = ctx->sq_data; |
5f3f26f9 JA |
12201 | if (!sq->thread) |
12202 | sq = NULL; | |
12203 | } | |
dbbe9c64 JQ |
12204 | |
12205 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
12206 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 12207 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 12208 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 12209 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 12210 | |
87ce955b JA |
12211 | if (f) |
12212 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
12213 | else | |
12214 | seq_printf(m, "%5u: <none>\n", i); | |
12215 | } | |
12216 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 12217 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 12218 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 12219 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 12220 | |
4751f53d | 12221 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 12222 | } |
61cf9370 MWO |
12223 | if (has_lock && !xa_empty(&ctx->personalities)) { |
12224 | unsigned long index; | |
12225 | const struct cred *cred; | |
12226 | ||
87ce955b | 12227 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
12228 | xa_for_each(&ctx->personalities, index, cred) |
12229 | io_uring_show_cred(m, index, cred); | |
87ce955b | 12230 | } |
83f84356 HX |
12231 | if (has_lock) |
12232 | mutex_unlock(&ctx->uring_lock); | |
12233 | ||
12234 | seq_puts(m, "PollList:\n"); | |
79ebeaee | 12235 | spin_lock(&ctx->completion_lock); |
d7718a9d JA |
12236 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
12237 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
12238 | struct io_kiocb *req; | |
12239 | ||
12240 | hlist_for_each_entry(req, list, hash_node) | |
12241 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7f62d40d | 12242 | task_work_pending(req->task)); |
d7718a9d | 12243 | } |
83f84356 HX |
12244 | |
12245 | seq_puts(m, "CqOverflowList:\n"); | |
12246 | list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) { | |
12247 | struct io_uring_cqe *cqe = &ocqe->cqe; | |
12248 | ||
12249 | seq_printf(m, " user_data=%llu, res=%d, flags=%x\n", | |
12250 | cqe->user_data, cqe->res, cqe->flags); | |
12251 | ||
12252 | } | |
12253 | ||
79ebeaee | 12254 | spin_unlock(&ctx->completion_lock); |
87ce955b JA |
12255 | } |
12256 | ||
c072481d | 12257 | static __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *f) |
87ce955b JA |
12258 | { |
12259 | struct io_ring_ctx *ctx = f->private_data; | |
12260 | ||
12261 | if (percpu_ref_tryget(&ctx->refs)) { | |
12262 | __io_uring_show_fdinfo(ctx, m); | |
12263 | percpu_ref_put(&ctx->refs); | |
12264 | } | |
12265 | } | |
bebdb65e | 12266 | #endif |
87ce955b | 12267 | |
2b188cc1 JA |
12268 | static const struct file_operations io_uring_fops = { |
12269 | .release = io_uring_release, | |
12270 | .mmap = io_uring_mmap, | |
6c5c240e RP |
12271 | #ifndef CONFIG_MMU |
12272 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
12273 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
12274 | #endif | |
2b188cc1 | 12275 | .poll = io_uring_poll, |
bebdb65e | 12276 | #ifdef CONFIG_PROC_FS |
87ce955b | 12277 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 12278 | #endif |
2b188cc1 JA |
12279 | }; |
12280 | ||
c072481d PB |
12281 | static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx, |
12282 | struct io_uring_params *p) | |
2b188cc1 | 12283 | { |
75b28aff HV |
12284 | struct io_rings *rings; |
12285 | size_t size, sq_array_offset; | |
2b188cc1 | 12286 | |
bd740481 JA |
12287 | /* make sure these are sane, as we already accounted them */ |
12288 | ctx->sq_entries = p->sq_entries; | |
12289 | ctx->cq_entries = p->cq_entries; | |
12290 | ||
baf9cb64 | 12291 | size = rings_size(ctx, p->sq_entries, p->cq_entries, &sq_array_offset); |
75b28aff HV |
12292 | if (size == SIZE_MAX) |
12293 | return -EOVERFLOW; | |
12294 | ||
12295 | rings = io_mem_alloc(size); | |
12296 | if (!rings) | |
2b188cc1 JA |
12297 | return -ENOMEM; |
12298 | ||
75b28aff HV |
12299 | ctx->rings = rings; |
12300 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
12301 | rings->sq_ring_mask = p->sq_entries - 1; | |
12302 | rings->cq_ring_mask = p->cq_entries - 1; | |
12303 | rings->sq_ring_entries = p->sq_entries; | |
12304 | rings->cq_ring_entries = p->cq_entries; | |
2b188cc1 | 12305 | |
ebdeb7c0 JA |
12306 | if (p->flags & IORING_SETUP_SQE128) |
12307 | size = array_size(2 * sizeof(struct io_uring_sqe), p->sq_entries); | |
12308 | else | |
12309 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
12310 | if (size == SIZE_MAX) { |
12311 | io_mem_free(ctx->rings); | |
12312 | ctx->rings = NULL; | |
2b188cc1 | 12313 | return -EOVERFLOW; |
eb065d30 | 12314 | } |
2b188cc1 JA |
12315 | |
12316 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
12317 | if (!ctx->sq_sqes) { |
12318 | io_mem_free(ctx->rings); | |
12319 | ctx->rings = NULL; | |
2b188cc1 | 12320 | return -ENOMEM; |
eb065d30 | 12321 | } |
2b188cc1 | 12322 | |
2b188cc1 JA |
12323 | return 0; |
12324 | } | |
12325 | ||
9faadcc8 PB |
12326 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
12327 | { | |
12328 | int ret, fd; | |
12329 | ||
12330 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
12331 | if (fd < 0) | |
12332 | return fd; | |
12333 | ||
eef51daa | 12334 | ret = io_uring_add_tctx_node(ctx); |
9faadcc8 PB |
12335 | if (ret) { |
12336 | put_unused_fd(fd); | |
12337 | return ret; | |
12338 | } | |
12339 | fd_install(fd, file); | |
12340 | return fd; | |
12341 | } | |
12342 | ||
2b188cc1 JA |
12343 | /* |
12344 | * Allocate an anonymous fd, this is what constitutes the application | |
12345 | * visible backing of an io_uring instance. The application mmaps this | |
12346 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
12347 | * we have to tie this fd to a socket for file garbage collection purposes. | |
12348 | */ | |
9faadcc8 | 12349 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
12350 | { |
12351 | struct file *file; | |
9faadcc8 | 12352 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
12353 | int ret; |
12354 | ||
2b188cc1 JA |
12355 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
12356 | &ctx->ring_sock); | |
12357 | if (ret) | |
9faadcc8 | 12358 | return ERR_PTR(ret); |
2b188cc1 JA |
12359 | #endif |
12360 | ||
91a9ab7c PM |
12361 | file = anon_inode_getfile_secure("[io_uring]", &io_uring_fops, ctx, |
12362 | O_RDWR | O_CLOEXEC, NULL); | |
2b188cc1 | 12363 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
12364 | if (IS_ERR(file)) { |
12365 | sock_release(ctx->ring_sock); | |
12366 | ctx->ring_sock = NULL; | |
12367 | } else { | |
12368 | ctx->ring_sock->file = file; | |
0f212204 | 12369 | } |
2b188cc1 | 12370 | #endif |
9faadcc8 | 12371 | return file; |
2b188cc1 JA |
12372 | } |
12373 | ||
c072481d PB |
12374 | static __cold int io_uring_create(unsigned entries, struct io_uring_params *p, |
12375 | struct io_uring_params __user *params) | |
2b188cc1 | 12376 | { |
2b188cc1 | 12377 | struct io_ring_ctx *ctx; |
9faadcc8 | 12378 | struct file *file; |
2b188cc1 JA |
12379 | int ret; |
12380 | ||
8110c1a6 | 12381 | if (!entries) |
2b188cc1 | 12382 | return -EINVAL; |
8110c1a6 JA |
12383 | if (entries > IORING_MAX_ENTRIES) { |
12384 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
12385 | return -EINVAL; | |
12386 | entries = IORING_MAX_ENTRIES; | |
12387 | } | |
2b188cc1 JA |
12388 | |
12389 | /* | |
12390 | * Use twice as many entries for the CQ ring. It's possible for the | |
12391 | * application to drive a higher depth than the size of the SQ ring, | |
12392 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
12393 | * some flexibility in overcommitting a bit. If the application has |
12394 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
12395 | * of CQ ring entries manually. | |
2b188cc1 JA |
12396 | */ |
12397 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
12398 | if (p->flags & IORING_SETUP_CQSIZE) { |
12399 | /* | |
12400 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
12401 | * to a power-of-two, if it isn't already. We do NOT impose | |
12402 | * any cq vs sq ring sizing. | |
12403 | */ | |
eb2667b3 | 12404 | if (!p->cq_entries) |
33a107f0 | 12405 | return -EINVAL; |
8110c1a6 JA |
12406 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
12407 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
12408 | return -EINVAL; | |
12409 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
12410 | } | |
eb2667b3 JQ |
12411 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
12412 | if (p->cq_entries < p->sq_entries) | |
12413 | return -EINVAL; | |
33a107f0 JA |
12414 | } else { |
12415 | p->cq_entries = 2 * p->sq_entries; | |
12416 | } | |
2b188cc1 | 12417 | |
2b188cc1 | 12418 | ctx = io_ring_ctx_alloc(p); |
62e398be | 12419 | if (!ctx) |
2b188cc1 | 12420 | return -ENOMEM; |
773697b6 PB |
12421 | |
12422 | /* | |
12423 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
12424 | * space applications don't need to do io completion events | |
12425 | * polling again, they can rely on io_sq_thread to do polling | |
12426 | * work, which can reduce cpu usage and uring_lock contention. | |
12427 | */ | |
12428 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
12429 | !(ctx->flags & IORING_SETUP_SQPOLL)) | |
12430 | ctx->syscall_iopoll = 1; | |
12431 | ||
2b188cc1 | 12432 | ctx->compat = in_compat_syscall(); |
62e398be JA |
12433 | if (!capable(CAP_IPC_LOCK)) |
12434 | ctx->user = get_uid(current_user()); | |
2aede0e4 | 12435 | |
9f010507 | 12436 | /* |
e1169f06 JA |
12437 | * For SQPOLL, we just need a wakeup, always. For !SQPOLL, if |
12438 | * COOP_TASKRUN is set, then IPIs are never needed by the app. | |
9f010507 | 12439 | */ |
e1169f06 JA |
12440 | ret = -EINVAL; |
12441 | if (ctx->flags & IORING_SETUP_SQPOLL) { | |
12442 | /* IPI related flags don't make sense with SQPOLL */ | |
ef060ea9 JA |
12443 | if (ctx->flags & (IORING_SETUP_COOP_TASKRUN | |
12444 | IORING_SETUP_TASKRUN_FLAG)) | |
e1169f06 | 12445 | goto err; |
9f010507 | 12446 | ctx->notify_method = TWA_SIGNAL_NO_IPI; |
e1169f06 JA |
12447 | } else if (ctx->flags & IORING_SETUP_COOP_TASKRUN) { |
12448 | ctx->notify_method = TWA_SIGNAL_NO_IPI; | |
12449 | } else { | |
ef060ea9 JA |
12450 | if (ctx->flags & IORING_SETUP_TASKRUN_FLAG) |
12451 | goto err; | |
9f010507 | 12452 | ctx->notify_method = TWA_SIGNAL; |
e1169f06 | 12453 | } |
9f010507 | 12454 | |
2aede0e4 JA |
12455 | /* |
12456 | * This is just grabbed for accounting purposes. When a process exits, | |
12457 | * the mm is exited and dropped before the files, hence we need to hang | |
12458 | * on to this mm purely for the purposes of being able to unaccount | |
12459 | * memory (locked/pinned vm). It's not used for anything else. | |
12460 | */ | |
6b7898eb | 12461 | mmgrab(current->mm); |
2aede0e4 | 12462 | ctx->mm_account = current->mm; |
6b7898eb | 12463 | |
2b188cc1 JA |
12464 | ret = io_allocate_scq_urings(ctx, p); |
12465 | if (ret) | |
12466 | goto err; | |
12467 | ||
7e84e1c7 | 12468 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
12469 | if (ret) |
12470 | goto err; | |
eae071c9 | 12471 | /* always set a rsrc node */ |
47b228ce PB |
12472 | ret = io_rsrc_node_switch_start(ctx); |
12473 | if (ret) | |
12474 | goto err; | |
eae071c9 | 12475 | io_rsrc_node_switch(ctx, NULL); |
2b188cc1 | 12476 | |
2b188cc1 | 12477 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
12478 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
12479 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
12480 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
12481 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
12482 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
12483 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
12484 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
12485 | |
12486 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
12487 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
12488 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
12489 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
12490 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
12491 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
12492 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 12493 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 12494 | |
7f13657d XW |
12495 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
12496 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 12497 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 12498 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9690557e | 12499 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS | |
c4212f3e JA |
12500 | IORING_FEAT_RSRC_TAGS | IORING_FEAT_CQE_SKIP | |
12501 | IORING_FEAT_LINKED_FILE; | |
7f13657d XW |
12502 | |
12503 | if (copy_to_user(params, p, sizeof(*p))) { | |
12504 | ret = -EFAULT; | |
12505 | goto err; | |
12506 | } | |
d1719f70 | 12507 | |
9faadcc8 PB |
12508 | file = io_uring_get_file(ctx); |
12509 | if (IS_ERR(file)) { | |
12510 | ret = PTR_ERR(file); | |
12511 | goto err; | |
12512 | } | |
12513 | ||
044c1ab3 JA |
12514 | /* |
12515 | * Install ring fd as the very last thing, so we don't risk someone | |
12516 | * having closed it before we finish setup | |
12517 | */ | |
9faadcc8 PB |
12518 | ret = io_uring_install_fd(ctx, file); |
12519 | if (ret < 0) { | |
12520 | /* fput will clean it up */ | |
12521 | fput(file); | |
12522 | return ret; | |
12523 | } | |
044c1ab3 | 12524 | |
c826bd7a | 12525 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
12526 | return ret; |
12527 | err: | |
12528 | io_ring_ctx_wait_and_kill(ctx); | |
12529 | return ret; | |
12530 | } | |
12531 | ||
12532 | /* | |
12533 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
12534 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
12535 | * params structure passed in. | |
12536 | */ | |
12537 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
12538 | { | |
12539 | struct io_uring_params p; | |
2b188cc1 JA |
12540 | int i; |
12541 | ||
12542 | if (copy_from_user(&p, params, sizeof(p))) | |
12543 | return -EFAULT; | |
12544 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
12545 | if (p.resv[i]) | |
12546 | return -EINVAL; | |
12547 | } | |
12548 | ||
6c271ce2 | 12549 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 12550 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 | 12551 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
e1169f06 | 12552 | IORING_SETUP_R_DISABLED | IORING_SETUP_SUBMIT_ALL | |
ebdeb7c0 | 12553 | IORING_SETUP_COOP_TASKRUN | IORING_SETUP_TASKRUN_FLAG | |
76c68fbf | 12554 | IORING_SETUP_SQE128 | IORING_SETUP_CQE32)) |
2b188cc1 JA |
12555 | return -EINVAL; |
12556 | ||
ef060ea9 | 12557 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
12558 | } |
12559 | ||
12560 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
12561 | struct io_uring_params __user *, params) | |
12562 | { | |
12563 | return io_uring_setup(entries, params); | |
12564 | } | |
12565 | ||
c072481d PB |
12566 | static __cold int io_probe(struct io_ring_ctx *ctx, void __user *arg, |
12567 | unsigned nr_args) | |
66f4af93 JA |
12568 | { |
12569 | struct io_uring_probe *p; | |
12570 | size_t size; | |
12571 | int i, ret; | |
12572 | ||
12573 | size = struct_size(p, ops, nr_args); | |
12574 | if (size == SIZE_MAX) | |
12575 | return -EOVERFLOW; | |
12576 | p = kzalloc(size, GFP_KERNEL); | |
12577 | if (!p) | |
12578 | return -ENOMEM; | |
12579 | ||
12580 | ret = -EFAULT; | |
12581 | if (copy_from_user(p, arg, size)) | |
12582 | goto out; | |
12583 | ret = -EINVAL; | |
12584 | if (memchr_inv(p, 0, size)) | |
12585 | goto out; | |
12586 | ||
12587 | p->last_op = IORING_OP_LAST - 1; | |
12588 | if (nr_args > IORING_OP_LAST) | |
12589 | nr_args = IORING_OP_LAST; | |
12590 | ||
12591 | for (i = 0; i < nr_args; i++) { | |
12592 | p->ops[i].op = i; | |
12593 | if (!io_op_defs[i].not_supported) | |
12594 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
12595 | } | |
12596 | p->ops_len = i; | |
12597 | ||
12598 | ret = 0; | |
12599 | if (copy_to_user(arg, p, size)) | |
12600 | ret = -EFAULT; | |
12601 | out: | |
12602 | kfree(p); | |
12603 | return ret; | |
12604 | } | |
12605 | ||
071698e1 JA |
12606 | static int io_register_personality(struct io_ring_ctx *ctx) |
12607 | { | |
4379bf8b | 12608 | const struct cred *creds; |
61cf9370 | 12609 | u32 id; |
1e6fa521 | 12610 | int ret; |
071698e1 | 12611 | |
4379bf8b | 12612 | creds = get_current_cred(); |
1e6fa521 | 12613 | |
61cf9370 MWO |
12614 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
12615 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
a30f895a JA |
12616 | if (ret < 0) { |
12617 | put_cred(creds); | |
12618 | return ret; | |
12619 | } | |
12620 | return id; | |
071698e1 JA |
12621 | } |
12622 | ||
c072481d PB |
12623 | static __cold int io_register_restrictions(struct io_ring_ctx *ctx, |
12624 | void __user *arg, unsigned int nr_args) | |
21b55dbc SG |
12625 | { |
12626 | struct io_uring_restriction *res; | |
12627 | size_t size; | |
12628 | int i, ret; | |
12629 | ||
7e84e1c7 SG |
12630 | /* Restrictions allowed only if rings started disabled */ |
12631 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
12632 | return -EBADFD; | |
12633 | ||
21b55dbc | 12634 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 12635 | if (ctx->restrictions.registered) |
21b55dbc SG |
12636 | return -EBUSY; |
12637 | ||
12638 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
12639 | return -EINVAL; | |
12640 | ||
12641 | size = array_size(nr_args, sizeof(*res)); | |
12642 | if (size == SIZE_MAX) | |
12643 | return -EOVERFLOW; | |
12644 | ||
12645 | res = memdup_user(arg, size); | |
12646 | if (IS_ERR(res)) | |
12647 | return PTR_ERR(res); | |
12648 | ||
12649 | ret = 0; | |
12650 | ||
12651 | for (i = 0; i < nr_args; i++) { | |
12652 | switch (res[i].opcode) { | |
12653 | case IORING_RESTRICTION_REGISTER_OP: | |
12654 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
12655 | ret = -EINVAL; | |
12656 | goto out; | |
12657 | } | |
12658 | ||
12659 | __set_bit(res[i].register_op, | |
12660 | ctx->restrictions.register_op); | |
12661 | break; | |
12662 | case IORING_RESTRICTION_SQE_OP: | |
12663 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
12664 | ret = -EINVAL; | |
12665 | goto out; | |
12666 | } | |
12667 | ||
12668 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
12669 | break; | |
12670 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
12671 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
12672 | break; | |
12673 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
12674 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
12675 | break; | |
12676 | default: | |
12677 | ret = -EINVAL; | |
12678 | goto out; | |
12679 | } | |
12680 | } | |
12681 | ||
12682 | out: | |
12683 | /* Reset all restrictions if an error happened */ | |
12684 | if (ret != 0) | |
12685 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
12686 | else | |
7e84e1c7 | 12687 | ctx->restrictions.registered = true; |
21b55dbc SG |
12688 | |
12689 | kfree(res); | |
12690 | return ret; | |
12691 | } | |
12692 | ||
7e84e1c7 SG |
12693 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
12694 | { | |
12695 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
12696 | return -EBADFD; | |
12697 | ||
12698 | if (ctx->restrictions.registered) | |
12699 | ctx->restricted = 1; | |
12700 | ||
0298ef96 PB |
12701 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
12702 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
12703 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
12704 | return 0; |
12705 | } | |
12706 | ||
fdecb662 | 12707 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 12708 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
12709 | unsigned nr_args) |
12710 | { | |
12711 | __u32 tmp; | |
12712 | int err; | |
12713 | ||
12714 | if (check_add_overflow(up->offset, nr_args, &tmp)) | |
12715 | return -EOVERFLOW; | |
12716 | err = io_rsrc_node_switch_start(ctx); | |
12717 | if (err) | |
12718 | return err; | |
12719 | ||
fdecb662 PB |
12720 | switch (type) { |
12721 | case IORING_RSRC_FILE: | |
98f0b3b4 | 12722 | return __io_sqe_files_update(ctx, up, nr_args); |
634d00df PB |
12723 | case IORING_RSRC_BUFFER: |
12724 | return __io_sqe_buffers_update(ctx, up, nr_args); | |
98f0b3b4 PB |
12725 | } |
12726 | return -EINVAL; | |
12727 | } | |
12728 | ||
c3bdad02 PB |
12729 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
12730 | unsigned nr_args) | |
98f0b3b4 | 12731 | { |
c3bdad02 | 12732 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
12733 | |
12734 | if (!nr_args) | |
12735 | return -EINVAL; | |
c3bdad02 PB |
12736 | memset(&up, 0, sizeof(up)); |
12737 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
12738 | return -EFAULT; | |
d8a3ba9c | 12739 | if (up.resv || up.resv2) |
565c5e61 | 12740 | return -EINVAL; |
c3bdad02 PB |
12741 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); |
12742 | } | |
12743 | ||
12744 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
992da01a | 12745 | unsigned size, unsigned type) |
c3bdad02 PB |
12746 | { |
12747 | struct io_uring_rsrc_update2 up; | |
12748 | ||
12749 | if (size != sizeof(up)) | |
12750 | return -EINVAL; | |
98f0b3b4 PB |
12751 | if (copy_from_user(&up, arg, sizeof(up))) |
12752 | return -EFAULT; | |
d8a3ba9c | 12753 | if (!up.nr || up.resv || up.resv2) |
98f0b3b4 | 12754 | return -EINVAL; |
992da01a | 12755 | return __io_register_rsrc_update(ctx, type, &up, up.nr); |
98f0b3b4 PB |
12756 | } |
12757 | ||
c072481d | 12758 | static __cold int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
992da01a | 12759 | unsigned int size, unsigned int type) |
792e3582 PB |
12760 | { |
12761 | struct io_uring_rsrc_register rr; | |
12762 | ||
12763 | /* keep it extendible */ | |
12764 | if (size != sizeof(rr)) | |
12765 | return -EINVAL; | |
12766 | ||
12767 | memset(&rr, 0, sizeof(rr)); | |
12768 | if (copy_from_user(&rr, arg, size)) | |
12769 | return -EFAULT; | |
a8da73a3 JA |
12770 | if (!rr.nr || rr.resv2) |
12771 | return -EINVAL; | |
12772 | if (rr.flags & ~IORING_RSRC_REGISTER_SPARSE) | |
792e3582 PB |
12773 | return -EINVAL; |
12774 | ||
992da01a | 12775 | switch (type) { |
792e3582 | 12776 | case IORING_RSRC_FILE: |
a8da73a3 JA |
12777 | if (rr.flags & IORING_RSRC_REGISTER_SPARSE && rr.data) |
12778 | break; | |
792e3582 PB |
12779 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), |
12780 | rr.nr, u64_to_user_ptr(rr.tags)); | |
634d00df | 12781 | case IORING_RSRC_BUFFER: |
0184f08e | 12782 | if (rr.flags & IORING_RSRC_REGISTER_SPARSE && rr.data) |
a8da73a3 | 12783 | break; |
634d00df PB |
12784 | return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data), |
12785 | rr.nr, u64_to_user_ptr(rr.tags)); | |
792e3582 PB |
12786 | } |
12787 | return -EINVAL; | |
12788 | } | |
12789 | ||
c072481d PB |
12790 | static __cold int io_register_iowq_aff(struct io_ring_ctx *ctx, |
12791 | void __user *arg, unsigned len) | |
fe76421d JA |
12792 | { |
12793 | struct io_uring_task *tctx = current->io_uring; | |
12794 | cpumask_var_t new_mask; | |
12795 | int ret; | |
12796 | ||
12797 | if (!tctx || !tctx->io_wq) | |
12798 | return -EINVAL; | |
12799 | ||
12800 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | |
12801 | return -ENOMEM; | |
12802 | ||
12803 | cpumask_clear(new_mask); | |
12804 | if (len > cpumask_size()) | |
12805 | len = cpumask_size(); | |
12806 | ||
0f5e4b83 ES |
12807 | if (in_compat_syscall()) { |
12808 | ret = compat_get_bitmap(cpumask_bits(new_mask), | |
12809 | (const compat_ulong_t __user *)arg, | |
12810 | len * 8 /* CHAR_BIT */); | |
12811 | } else { | |
12812 | ret = copy_from_user(new_mask, arg, len); | |
12813 | } | |
12814 | ||
12815 | if (ret) { | |
fe76421d JA |
12816 | free_cpumask_var(new_mask); |
12817 | return -EFAULT; | |
12818 | } | |
12819 | ||
12820 | ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); | |
12821 | free_cpumask_var(new_mask); | |
12822 | return ret; | |
12823 | } | |
12824 | ||
c072481d | 12825 | static __cold int io_unregister_iowq_aff(struct io_ring_ctx *ctx) |
fe76421d JA |
12826 | { |
12827 | struct io_uring_task *tctx = current->io_uring; | |
12828 | ||
12829 | if (!tctx || !tctx->io_wq) | |
12830 | return -EINVAL; | |
12831 | ||
12832 | return io_wq_cpu_affinity(tctx->io_wq, NULL); | |
12833 | } | |
12834 | ||
c072481d PB |
12835 | static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, |
12836 | void __user *arg) | |
b22fa62a | 12837 | __must_hold(&ctx->uring_lock) |
2e480058 | 12838 | { |
b22fa62a | 12839 | struct io_tctx_node *node; |
fa84693b JA |
12840 | struct io_uring_task *tctx = NULL; |
12841 | struct io_sq_data *sqd = NULL; | |
2e480058 JA |
12842 | __u32 new_count[2]; |
12843 | int i, ret; | |
12844 | ||
2e480058 JA |
12845 | if (copy_from_user(new_count, arg, sizeof(new_count))) |
12846 | return -EFAULT; | |
12847 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
12848 | if (new_count[i] > INT_MAX) | |
12849 | return -EINVAL; | |
12850 | ||
fa84693b JA |
12851 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
12852 | sqd = ctx->sq_data; | |
12853 | if (sqd) { | |
009ad9f0 JA |
12854 | /* |
12855 | * Observe the correct sqd->lock -> ctx->uring_lock | |
12856 | * ordering. Fine to drop uring_lock here, we hold | |
12857 | * a ref to the ctx. | |
12858 | */ | |
41d3a6bd | 12859 | refcount_inc(&sqd->refs); |
009ad9f0 | 12860 | mutex_unlock(&ctx->uring_lock); |
fa84693b | 12861 | mutex_lock(&sqd->lock); |
009ad9f0 | 12862 | mutex_lock(&ctx->uring_lock); |
41d3a6bd JA |
12863 | if (sqd->thread) |
12864 | tctx = sqd->thread->io_uring; | |
fa84693b JA |
12865 | } |
12866 | } else { | |
12867 | tctx = current->io_uring; | |
12868 | } | |
12869 | ||
e139a1ec | 12870 | BUILD_BUG_ON(sizeof(new_count) != sizeof(ctx->iowq_limits)); |
fa84693b | 12871 | |
bad119b9 PB |
12872 | for (i = 0; i < ARRAY_SIZE(new_count); i++) |
12873 | if (new_count[i]) | |
12874 | ctx->iowq_limits[i] = new_count[i]; | |
e139a1ec PB |
12875 | ctx->iowq_limits_set = true; |
12876 | ||
e139a1ec PB |
12877 | if (tctx && tctx->io_wq) { |
12878 | ret = io_wq_max_workers(tctx->io_wq, new_count); | |
12879 | if (ret) | |
12880 | goto err; | |
12881 | } else { | |
12882 | memset(new_count, 0, sizeof(new_count)); | |
12883 | } | |
fa84693b | 12884 | |
41d3a6bd | 12885 | if (sqd) { |
fa84693b | 12886 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
12887 | io_put_sq_data(sqd); |
12888 | } | |
2e480058 JA |
12889 | |
12890 | if (copy_to_user(arg, new_count, sizeof(new_count))) | |
12891 | return -EFAULT; | |
12892 | ||
b22fa62a PB |
12893 | /* that's it for SQPOLL, only the SQPOLL task creates requests */ |
12894 | if (sqd) | |
12895 | return 0; | |
12896 | ||
12897 | /* now propagate the restriction to all registered users */ | |
12898 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
12899 | struct io_uring_task *tctx = node->task->io_uring; | |
12900 | ||
12901 | if (WARN_ON_ONCE(!tctx->io_wq)) | |
12902 | continue; | |
12903 | ||
12904 | for (i = 0; i < ARRAY_SIZE(new_count); i++) | |
12905 | new_count[i] = ctx->iowq_limits[i]; | |
12906 | /* ignore errors, it always returns zero anyway */ | |
12907 | (void)io_wq_max_workers(tctx->io_wq, new_count); | |
12908 | } | |
2e480058 | 12909 | return 0; |
fa84693b | 12910 | err: |
41d3a6bd | 12911 | if (sqd) { |
fa84693b | 12912 | mutex_unlock(&sqd->lock); |
41d3a6bd JA |
12913 | io_put_sq_data(sqd); |
12914 | } | |
fa84693b | 12915 | return ret; |
2e480058 JA |
12916 | } |
12917 | ||
c7fb1942 JA |
12918 | static int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg) |
12919 | { | |
12920 | struct io_uring_buf_ring *br; | |
12921 | struct io_uring_buf_reg reg; | |
12922 | struct io_buffer_list *bl; | |
12923 | struct page **pages; | |
12924 | int nr_pages; | |
12925 | ||
12926 | if (copy_from_user(®, arg, sizeof(reg))) | |
12927 | return -EFAULT; | |
12928 | ||
12929 | if (reg.pad || reg.resv[0] || reg.resv[1] || reg.resv[2]) | |
12930 | return -EINVAL; | |
12931 | if (!reg.ring_addr) | |
12932 | return -EFAULT; | |
12933 | if (reg.ring_addr & ~PAGE_MASK) | |
12934 | return -EINVAL; | |
12935 | if (!is_power_of_2(reg.ring_entries)) | |
12936 | return -EINVAL; | |
12937 | ||
f9437ac0 DY |
12938 | /* cannot disambiguate full vs empty due to head/tail size */ |
12939 | if (reg.ring_entries >= 65536) | |
12940 | return -EINVAL; | |
12941 | ||
c7fb1942 JA |
12942 | if (unlikely(reg.bgid < BGID_ARRAY && !ctx->io_bl)) { |
12943 | int ret = io_init_bl_list(ctx); | |
12944 | if (ret) | |
12945 | return ret; | |
12946 | } | |
12947 | ||
12948 | bl = io_buffer_get_list(ctx, reg.bgid); | |
2fcabce2 JA |
12949 | if (bl) { |
12950 | /* if mapped buffer ring OR classic exists, don't allow */ | |
12951 | if (bl->buf_nr_pages || !list_empty(&bl->buf_list)) | |
12952 | return -EEXIST; | |
12953 | } else { | |
c7fb1942 JA |
12954 | bl = kzalloc(sizeof(*bl), GFP_KERNEL); |
12955 | if (!bl) | |
12956 | return -ENOMEM; | |
12957 | } | |
12958 | ||
12959 | pages = io_pin_pages(reg.ring_addr, | |
12960 | struct_size(br, bufs, reg.ring_entries), | |
12961 | &nr_pages); | |
12962 | if (IS_ERR(pages)) { | |
12963 | kfree(bl); | |
12964 | return PTR_ERR(pages); | |
12965 | } | |
12966 | ||
12967 | br = page_address(pages[0]); | |
12968 | bl->buf_pages = pages; | |
12969 | bl->buf_nr_pages = nr_pages; | |
12970 | bl->nr_entries = reg.ring_entries; | |
12971 | bl->buf_ring = br; | |
12972 | bl->mask = reg.ring_entries - 1; | |
12973 | io_buffer_add_list(ctx, bl, reg.bgid); | |
12974 | return 0; | |
12975 | } | |
12976 | ||
12977 | static int io_unregister_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg) | |
12978 | { | |
12979 | struct io_uring_buf_reg reg; | |
12980 | struct io_buffer_list *bl; | |
12981 | ||
12982 | if (copy_from_user(®, arg, sizeof(reg))) | |
12983 | return -EFAULT; | |
12984 | if (reg.pad || reg.resv[0] || reg.resv[1] || reg.resv[2]) | |
12985 | return -EINVAL; | |
12986 | ||
12987 | bl = io_buffer_get_list(ctx, reg.bgid); | |
12988 | if (!bl) | |
12989 | return -ENOENT; | |
12990 | if (!bl->buf_nr_pages) | |
12991 | return -EINVAL; | |
12992 | ||
12993 | __io_remove_buffers(ctx, bl, -1U); | |
12994 | if (bl->bgid >= BGID_ARRAY) { | |
12995 | xa_erase(&ctx->io_bl_xa, bl->bgid); | |
12996 | kfree(bl); | |
12997 | } | |
12998 | return 0; | |
12999 | } | |
13000 | ||
edafccee JA |
13001 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
13002 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
13003 | __releases(ctx->uring_lock) |
13004 | __acquires(ctx->uring_lock) | |
edafccee JA |
13005 | { |
13006 | int ret; | |
13007 | ||
35fa71a0 JA |
13008 | /* |
13009 | * We're inside the ring mutex, if the ref is already dying, then | |
13010 | * someone else killed the ctx or is already going through | |
13011 | * io_uring_register(). | |
13012 | */ | |
13013 | if (percpu_ref_is_dying(&ctx->refs)) | |
13014 | return -ENXIO; | |
13015 | ||
75c4021a PB |
13016 | if (ctx->restricted) { |
13017 | if (opcode >= IORING_REGISTER_LAST) | |
13018 | return -EINVAL; | |
13019 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
13020 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
13021 | return -EACCES; | |
13022 | } | |
13023 | ||
edafccee JA |
13024 | switch (opcode) { |
13025 | case IORING_REGISTER_BUFFERS: | |
0184f08e PB |
13026 | ret = -EFAULT; |
13027 | if (!arg) | |
13028 | break; | |
634d00df | 13029 | ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL); |
edafccee JA |
13030 | break; |
13031 | case IORING_UNREGISTER_BUFFERS: | |
13032 | ret = -EINVAL; | |
13033 | if (arg || nr_args) | |
13034 | break; | |
0a96bbe4 | 13035 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 13036 | break; |
6b06314c | 13037 | case IORING_REGISTER_FILES: |
a8da73a3 JA |
13038 | ret = -EFAULT; |
13039 | if (!arg) | |
13040 | break; | |
792e3582 | 13041 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
13042 | break; |
13043 | case IORING_UNREGISTER_FILES: | |
13044 | ret = -EINVAL; | |
13045 | if (arg || nr_args) | |
13046 | break; | |
13047 | ret = io_sqe_files_unregister(ctx); | |
13048 | break; | |
c3a31e60 | 13049 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 13050 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 13051 | break; |
9b402849 JA |
13052 | case IORING_REGISTER_EVENTFD: |
13053 | ret = -EINVAL; | |
13054 | if (nr_args != 1) | |
13055 | break; | |
c75312dd UA |
13056 | ret = io_eventfd_register(ctx, arg, 0); |
13057 | break; | |
13058 | case IORING_REGISTER_EVENTFD_ASYNC: | |
13059 | ret = -EINVAL; | |
13060 | if (nr_args != 1) | |
f2842ab5 | 13061 | break; |
c75312dd | 13062 | ret = io_eventfd_register(ctx, arg, 1); |
9b402849 JA |
13063 | break; |
13064 | case IORING_UNREGISTER_EVENTFD: | |
13065 | ret = -EINVAL; | |
13066 | if (arg || nr_args) | |
13067 | break; | |
13068 | ret = io_eventfd_unregister(ctx); | |
13069 | break; | |
66f4af93 JA |
13070 | case IORING_REGISTER_PROBE: |
13071 | ret = -EINVAL; | |
13072 | if (!arg || nr_args > 256) | |
13073 | break; | |
13074 | ret = io_probe(ctx, arg, nr_args); | |
13075 | break; | |
071698e1 JA |
13076 | case IORING_REGISTER_PERSONALITY: |
13077 | ret = -EINVAL; | |
13078 | if (arg || nr_args) | |
13079 | break; | |
13080 | ret = io_register_personality(ctx); | |
13081 | break; | |
13082 | case IORING_UNREGISTER_PERSONALITY: | |
13083 | ret = -EINVAL; | |
13084 | if (arg) | |
13085 | break; | |
13086 | ret = io_unregister_personality(ctx, nr_args); | |
13087 | break; | |
7e84e1c7 SG |
13088 | case IORING_REGISTER_ENABLE_RINGS: |
13089 | ret = -EINVAL; | |
13090 | if (arg || nr_args) | |
13091 | break; | |
13092 | ret = io_register_enable_rings(ctx); | |
13093 | break; | |
21b55dbc SG |
13094 | case IORING_REGISTER_RESTRICTIONS: |
13095 | ret = io_register_restrictions(ctx, arg, nr_args); | |
13096 | break; | |
992da01a PB |
13097 | case IORING_REGISTER_FILES2: |
13098 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE); | |
13099 | break; | |
13100 | case IORING_REGISTER_FILES_UPDATE2: | |
13101 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
13102 | IORING_RSRC_FILE); | |
13103 | break; | |
13104 | case IORING_REGISTER_BUFFERS2: | |
13105 | ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER); | |
792e3582 | 13106 | break; |
992da01a PB |
13107 | case IORING_REGISTER_BUFFERS_UPDATE: |
13108 | ret = io_register_rsrc_update(ctx, arg, nr_args, | |
13109 | IORING_RSRC_BUFFER); | |
c3bdad02 | 13110 | break; |
fe76421d JA |
13111 | case IORING_REGISTER_IOWQ_AFF: |
13112 | ret = -EINVAL; | |
13113 | if (!arg || !nr_args) | |
13114 | break; | |
13115 | ret = io_register_iowq_aff(ctx, arg, nr_args); | |
13116 | break; | |
13117 | case IORING_UNREGISTER_IOWQ_AFF: | |
13118 | ret = -EINVAL; | |
13119 | if (arg || nr_args) | |
13120 | break; | |
13121 | ret = io_unregister_iowq_aff(ctx); | |
13122 | break; | |
2e480058 JA |
13123 | case IORING_REGISTER_IOWQ_MAX_WORKERS: |
13124 | ret = -EINVAL; | |
13125 | if (!arg || nr_args != 2) | |
13126 | break; | |
13127 | ret = io_register_iowq_max_workers(ctx, arg); | |
13128 | break; | |
e7a6c00d JA |
13129 | case IORING_REGISTER_RING_FDS: |
13130 | ret = io_ringfd_register(ctx, arg, nr_args); | |
13131 | break; | |
13132 | case IORING_UNREGISTER_RING_FDS: | |
13133 | ret = io_ringfd_unregister(ctx, arg, nr_args); | |
13134 | break; | |
c7fb1942 JA |
13135 | case IORING_REGISTER_PBUF_RING: |
13136 | ret = -EINVAL; | |
13137 | if (!arg || nr_args != 1) | |
13138 | break; | |
13139 | ret = io_register_pbuf_ring(ctx, arg); | |
13140 | break; | |
13141 | case IORING_UNREGISTER_PBUF_RING: | |
13142 | ret = -EINVAL; | |
13143 | if (!arg || nr_args != 1) | |
13144 | break; | |
13145 | ret = io_unregister_pbuf_ring(ctx, arg); | |
13146 | break; | |
edafccee JA |
13147 | default: |
13148 | ret = -EINVAL; | |
13149 | break; | |
13150 | } | |
13151 | ||
edafccee JA |
13152 | return ret; |
13153 | } | |
13154 | ||
13155 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
13156 | void __user *, arg, unsigned int, nr_args) | |
13157 | { | |
13158 | struct io_ring_ctx *ctx; | |
13159 | long ret = -EBADF; | |
13160 | struct fd f; | |
13161 | ||
13162 | f = fdget(fd); | |
13163 | if (!f.file) | |
13164 | return -EBADF; | |
13165 | ||
13166 | ret = -EOPNOTSUPP; | |
13167 | if (f.file->f_op != &io_uring_fops) | |
13168 | goto out_fput; | |
13169 | ||
13170 | ctx = f.file->private_data; | |
13171 | ||
b6c23dd5 PB |
13172 | io_run_task_work(); |
13173 | ||
edafccee JA |
13174 | mutex_lock(&ctx->uring_lock); |
13175 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
13176 | mutex_unlock(&ctx->uring_lock); | |
2757be22 | 13177 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, ret); |
edafccee JA |
13178 | out_fput: |
13179 | fdput(f); | |
13180 | return ret; | |
13181 | } | |
13182 | ||
2b188cc1 JA |
13183 | static int __init io_uring_init(void) |
13184 | { | |
d7f62e82 SM |
13185 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
13186 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
13187 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
13188 | } while (0) | |
13189 | ||
13190 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
13191 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
13192 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
13193 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
13194 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
13195 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
13196 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
13197 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
13198 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
13199 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 13200 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
13201 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
13202 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
13203 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
13204 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
13205 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
13206 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
13207 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
13208 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
13209 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
13210 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
13211 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
13212 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
13213 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
13214 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
13215 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 13216 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
13217 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
13218 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
16340eab | 13219 | BUILD_BUG_SQE_ELEM(40, __u16, buf_group); |
d7f62e82 | 13220 | BUILD_BUG_SQE_ELEM(42, __u16, personality); |
7d67af2c | 13221 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
b9445598 | 13222 | BUILD_BUG_SQE_ELEM(44, __u32, file_index); |
e9621e2b | 13223 | BUILD_BUG_SQE_ELEM(48, __u64, addr3); |
d7f62e82 | 13224 | |
b0d658ec PB |
13225 | BUILD_BUG_ON(sizeof(struct io_uring_files_update) != |
13226 | sizeof(struct io_uring_rsrc_update)); | |
13227 | BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) > | |
13228 | sizeof(struct io_uring_rsrc_update2)); | |
90499ad0 PB |
13229 | |
13230 | /* ->buf_index is u16 */ | |
13231 | BUILD_BUG_ON(IORING_MAX_REG_BUFFERS >= (1u << 16)); | |
9cfc7e94 | 13232 | BUILD_BUG_ON(BGID_ARRAY * sizeof(struct io_buffer_list) > PAGE_SIZE); |
c7fb1942 JA |
13233 | BUILD_BUG_ON(offsetof(struct io_uring_buf_ring, bufs) != 0); |
13234 | BUILD_BUG_ON(offsetof(struct io_uring_buf, resv) != | |
13235 | offsetof(struct io_uring_buf_ring, tail)); | |
90499ad0 | 13236 | |
b0d658ec PB |
13237 | /* should fit into one byte */ |
13238 | BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8)); | |
68fe256a PB |
13239 | BUILD_BUG_ON(SQE_COMMON_FLAGS >= (1 << 8)); |
13240 | BUILD_BUG_ON((SQE_VALID_FLAGS | SQE_COMMON_FLAGS) != SQE_VALID_FLAGS); | |
b0d658ec | 13241 | |
d3656344 | 13242 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
32c2d33e | 13243 | BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int)); |
16340eab | 13244 | |
3a4b89a2 JA |
13245 | BUILD_BUG_ON(sizeof(atomic_t) != sizeof(u32)); |
13246 | ||
ee692a21 JA |
13247 | BUILD_BUG_ON(sizeof(struct io_uring_cmd) > 64); |
13248 | ||
91f245d5 JA |
13249 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
13250 | SLAB_ACCOUNT); | |
2b188cc1 JA |
13251 | return 0; |
13252 | }; | |
13253 | __initcall(io_uring_init); |