projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring: clamp to_submit in io_submit_sqes()
[linux-block.git] / fs / io_uring.c
CommitLineData
2b188cc1
JA		 1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared application/kernel submission and completion ring pairs, for
4 * supporting fast/efficient IO.
5 *
6 * A note on the read/write ordering memory barriers that are matched between
1e84b97b
SB		 7 * the application and kernel side.
8 *
9 * After the application reads the CQ ring tail, it must use an
10 * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses
11 * before writing the tail (using smp_load_acquire to read the tail will
12 * do). It also needs a smp_mb() before updating CQ head (ordering the
13 * entry load(s) with the head store), pairing with an implicit barrier
14 * through a control-dependency in io_get_cqring (smp_store_release to
15 * store head will do). Failure to do so could lead to reading invalid
16 * CQ entries.
17 *
18 * Likewise, the application must use an appropriate smp_wmb() before
19 * writing the SQ tail (ordering SQ entry stores with the tail store),
20 * which pairs with smp_load_acquire in io_get_sqring (smp_store_release
21 * to store the tail will do). And it needs a barrier ordering the SQ
22 * head load before writing new SQ entries (smp_load_acquire to read
23 * head will do).
24 *
25 * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application
26 * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after*
27 * updating the SQ tail; a full memory barrier smp_mb() is needed
28 * between.
2b188cc1
JA		 29 *
30 * Also see the examples in the liburing library:
31 *
32 * git://git.kernel.dk/liburing
33 *
34 * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens
35 * from data shared between the kernel and application. This is done both
36 * for ordering purposes, but also to ensure that once a value is loaded from
37 * data that the application could potentially modify, it remains stable.
38 *
39 * Copyright (C) 2018-2019 Jens Axboe
c992fe29 40 * Copyright (c) 2018-2019 Christoph Hellwig
2b188cc1
JA		 41 */
42#include <linux/kernel.h>
43#include <linux/init.h>
44#include <linux/errno.h>
45#include <linux/syscalls.h>
46#include <linux/compat.h>
47#include <linux/refcount.h>
48#include <linux/uio.h>
49
50#include <linux/sched/signal.h>
51#include <linux/fs.h>
52#include <linux/file.h>
53#include <linux/fdtable.h>
54#include <linux/mm.h>
55#include <linux/mman.h>
56#include <linux/mmu_context.h>
57#include <linux/percpu.h>
58#include <linux/slab.h>
6c271ce2 59#include <linux/kthread.h>
2b188cc1 60#include <linux/blkdev.h>
edafccee 61#include <linux/bvec.h>
2b188cc1
JA		 62#include <linux/net.h>
63#include <net/sock.h>
64#include <net/af_unix.h>
6b06314c 65#include <net/scm.h>
2b188cc1
JA		 66#include <linux/anon_inodes.h>
67#include <linux/sched/mm.h>
68#include <linux/uaccess.h>
69#include <linux/nospec.h>
edafccee
JA		 70#include <linux/sizes.h>
71#include <linux/hugetlb.h>
aa4c3967 72#include <linux/highmem.h>
15b71abe
JA		 73#include <linux/namei.h>
74#include <linux/fsnotify.h>
4840e418 75#include <linux/fadvise.h>
2b188cc1 76
c826bd7a
DD		 77#define CREATE_TRACE_POINTS
78#include <trace/events/io_uring.h>
79
2b188cc1
JA		 80#include <uapi/linux/io_uring.h>
81
82#include "internal.h"
561fb04a 83#include "io-wq.h"
2b188cc1 84
5277deaa 85#define IORING_MAX_ENTRIES 32768
33a107f0 86#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
65e19f54
JA		 87
88/*
89 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
90 */
91#define IORING_FILE_TABLE_SHIFT 9
92#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
93#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
94#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
2b188cc1
JA		 95
96struct io_uring {
97 u32 head ____cacheline_aligned_in_smp;
98 u32 tail ____cacheline_aligned_in_smp;
99};
100
1e84b97b 101/*
75b28aff
HV		 102 * This data is shared with the application through the mmap at offsets
103 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
1e84b97b
SB		 104 *
105 * The offsets to the member fields are published through struct
106 * io_sqring_offsets when calling io_uring_setup.
107 */
75b28aff 108struct io_rings {
1e84b97b
SB		 109 /*
110 * Head and tail offsets into the ring; the offsets need to be
111 * masked to get valid indices.
112 *
75b28aff
HV		 113 * The kernel controls head of the sq ring and the tail of the cq ring,
114 * and the application controls tail of the sq ring and the head of the
115 * cq ring.
1e84b97b 116 */
75b28aff 117 struct io_uring sq, cq;
1e84b97b 118 /*
75b28aff 119 * Bitmasks to apply to head and tail offsets (constant, equals
1e84b97b
SB		 120 * ring_entries - 1)
121 */
75b28aff
HV		 122 u32 sq_ring_mask, cq_ring_mask;
123 /* Ring sizes (constant, power of 2) */
124 u32 sq_ring_entries, cq_ring_entries;
1e84b97b
SB		 125 /*
126 * Number of invalid entries dropped by the kernel due to
127 * invalid index stored in array
128 *
129 * Written by the kernel, shouldn't be modified by the
130 * application (i.e. get number of "new events" by comparing to
131 * cached value).
132 *
133 * After a new SQ head value was read by the application this
134 * counter includes all submissions that were dropped reaching
135 * the new SQ head (and possibly more).
136 */
75b28aff 137 u32 sq_dropped;
1e84b97b
SB		 138 /*
139 * Runtime flags
140 *
141 * Written by the kernel, shouldn't be modified by the
142 * application.
143 *
144 * The application needs a full memory barrier before checking
145 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
146 */
75b28aff 147 u32 sq_flags;
1e84b97b
SB		 148 /*
149 * Number of completion events lost because the queue was full;
150 * this should be avoided by the application by making sure
0b4295b5 151 * there are not more requests pending than there is space in
1e84b97b
SB		 152 * the completion queue.
153 *
154 * Written by the kernel, shouldn't be modified by the
155 * application (i.e. get number of "new events" by comparing to
156 * cached value).
157 *
158 * As completion events come in out of order this counter is not
159 * ordered with any other data.
160 */
75b28aff 161 u32 cq_overflow;
1e84b97b
SB		 162 /*
163 * Ring buffer of completion events.
164 *
165 * The kernel writes completion events fresh every time they are
166 * produced, so the application is allowed to modify pending
167 * entries.
168 */
75b28aff 169 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
2b188cc1
JA		 170};
171
edafccee
JA		 172struct io_mapped_ubuf {
173 u64 ubuf;
174 size_t len;
175 struct bio_vec *bvec;
176 unsigned int nr_bvecs;
177};
178
65e19f54
JA		 179struct fixed_file_table {
180 struct file **files;
31b51510
JA		 181};
182
05f3fb3c
JA		 183enum {
184 FFD_F_ATOMIC,
185};
186
187struct fixed_file_data {
188 struct fixed_file_table *table;
189 struct io_ring_ctx *ctx;
190
191 struct percpu_ref refs;
192 struct llist_head put_llist;
193 unsigned long state;
194 struct work_struct ref_work;
195 struct completion done;
196};
197
2b188cc1
JA		 198struct io_ring_ctx {
199 struct {
200 struct percpu_ref refs;
201 } ____cacheline_aligned_in_smp;
202
203 struct {
204 unsigned int flags;
205 bool compat;
206 bool account_mem;
1d7bb1d5 207 bool cq_overflow_flushed;
1b4a51b6 208 bool drain_next;
2b188cc1 209
75b28aff
HV		 210 /*
211 * Ring buffer of indices into array of io_uring_sqe, which is
212 * mmapped by the application using the IORING_OFF_SQES offset.
213 *
214 * This indirection could e.g. be used to assign fixed
215 * io_uring_sqe entries to operations and only submit them to
216 * the queue when needed.
217 *
218 * The kernel modifies neither the indices array nor the entries
219 * array.
220 */
221 u32 *sq_array;
2b188cc1
JA		 222 unsigned cached_sq_head;
223 unsigned sq_entries;
224 unsigned sq_mask;
6c271ce2 225 unsigned sq_thread_idle;
498ccd9e 226 unsigned cached_sq_dropped;
206aefde 227 atomic_t cached_cq_overflow;
ad3eb2c8 228 unsigned long sq_check_overflow;
de0617e4
JA		 229
230 struct list_head defer_list;
5262f567 231 struct list_head timeout_list;
1d7bb1d5 232 struct list_head cq_overflow_list;
fcb323cc
JA		 233
234 wait_queue_head_t inflight_wait;
ad3eb2c8 235 struct io_uring_sqe *sq_sqes;
2b188cc1
JA		 236 } ____cacheline_aligned_in_smp;
237
206aefde
JA		 238 struct io_rings *rings;
239
2b188cc1 240 /* IO offload */
561fb04a 241 struct io_wq *io_wq;
6c271ce2 242 struct task_struct *sqo_thread; /* if using sq thread polling */
2b188cc1 243 struct mm_struct *sqo_mm;
6c271ce2 244 wait_queue_head_t sqo_wait;
75b28aff 245
6b06314c
JA		 246 /*
247 * If used, fixed file set. Writers must ensure that ->refs is dead,
248 * readers must ensure that ->refs is alive as long as the file* is
249 * used. Only updated through io_uring_register(2).
250 */
05f3fb3c 251 struct fixed_file_data *file_data;
6b06314c
JA		 252 unsigned nr_user_files;
253
edafccee
JA		 254 /* if used, fixed mapped user buffers */
255 unsigned nr_user_bufs;
256 struct io_mapped_ubuf *user_bufs;
257
2b188cc1
JA		 258 struct user_struct *user;
259
0b8c0ec7 260 const struct cred *creds;
181e448d 261
206aefde
JA		 262 /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */
263 struct completion *completions;
264
0ddf92e8
JA		 265 /* if all else fails... */
266 struct io_kiocb *fallback_req;
267
206aefde
JA		 268#if defined(CONFIG_UNIX)
269 struct socket *ring_sock;
270#endif
271
272 struct {
273 unsigned cached_cq_tail;
274 unsigned cq_entries;
275 unsigned cq_mask;
276 atomic_t cq_timeouts;
ad3eb2c8 277 unsigned long cq_check_overflow;
206aefde
JA		 278 struct wait_queue_head cq_wait;
279 struct fasync_struct *cq_fasync;
280 struct eventfd_ctx *cq_ev_fd;
281 } ____cacheline_aligned_in_smp;
2b188cc1
JA		 282
283 struct {
284 struct mutex uring_lock;
285 wait_queue_head_t wait;
286 } ____cacheline_aligned_in_smp;
287
288 struct {
289 spinlock_t completion_lock;
e94f141b
JA		 290 struct llist_head poll_llist;
291
def596e9
JA		 292 /*
293 * ->poll_list is protected by the ctx->uring_lock for
294 * io_uring instances that don't use IORING_SETUP_SQPOLL.
295 * For SQPOLL, only the single threaded io_sq_thread() will
296 * manipulate the list, hence no extra locking is needed there.
297 */
298 struct list_head poll_list;
78076bb6
JA		 299 struct hlist_head *cancel_hash;
300 unsigned cancel_hash_bits;
e94f141b 301 bool poll_multi_file;
31b51510 302
fcb323cc
JA		 303 spinlock_t inflight_lock;
304 struct list_head inflight_list;
2b188cc1 305 } ____cacheline_aligned_in_smp;
2b188cc1
JA		 306};
307
09bb8394
JA		 308/*
309 * First field must be the file pointer in all the
310 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
311 */
221c5eb2
JA		 312struct io_poll_iocb {
313 struct file *file;
0969e783
JA		 314 union {
315 struct wait_queue_head *head;
316 u64 addr;
317 };
221c5eb2 318 __poll_t events;
8c838788 319 bool done;
221c5eb2 320 bool canceled;
392edb45 321 struct wait_queue_entry wait;
221c5eb2
JA		 322};
323
b5dba59e
JA		 324struct io_close {
325 struct file *file;
326 struct file *put_file;
327 int fd;
328};
329
ad8a48ac
JA		 330struct io_timeout_data {
331 struct io_kiocb *req;
332 struct hrtimer timer;
333 struct timespec64 ts;
334 enum hrtimer_mode mode;
cc42e0ac 335 u32 seq_offset;
ad8a48ac
JA		 336};
337
8ed8d3c3
JA		 338struct io_accept {
339 struct file *file;
340 struct sockaddr __user *addr;
341 int __user *addr_len;
342 int flags;
343};
344
345struct io_sync {
346 struct file *file;
347 loff_t len;
348 loff_t off;
349 int flags;
d63d1b5e 350 int mode;
8ed8d3c3
JA		 351};
352
fbf23849
JA		 353struct io_cancel {
354 struct file *file;
355 u64 addr;
356};
357
b29472ee
JA		 358struct io_timeout {
359 struct file *file;
360 u64 addr;
361 int flags;
26a61679 362 unsigned count;
b29472ee
JA		 363};
364
9adbd45d
JA		 365struct io_rw {
366 /* NOTE: kiocb has the file as the first member, so don't do it here */
367 struct kiocb kiocb;
368 u64 addr;
369 u64 len;
370};
371
3fbb51c1
JA		 372struct io_connect {
373 struct file *file;
374 struct sockaddr __user *addr;
375 int addr_len;
376};
377
e47293fd
JA		 378struct io_sr_msg {
379 struct file *file;
380 struct user_msghdr __user *msg;
381 int msg_flags;
382};
383
15b71abe
JA		 384struct io_open {
385 struct file *file;
386 int dfd;
eddc7ef5
JA		 387 union {
388 umode_t mode;
389 unsigned mask;
390 };
15b71abe
JA		 391 const char __user *fname;
392 struct filename *filename;
eddc7ef5 393 struct statx __user *buffer;
15b71abe
JA		 394 int flags;
395};
396
05f3fb3c
JA		 397struct io_files_update {
398 struct file *file;
399 u64 arg;
400 u32 nr_args;
401 u32 offset;
402};
403
4840e418
JA		 404struct io_fadvise {
405 struct file *file;
406 u64 offset;
407 u32 len;
408 u32 advice;
409};
410
c1ca757b
JA		 411struct io_madvise {
412 struct file *file;
413 u64 addr;
414 u32 len;
415 u32 advice;
416};
417
f499a021
JA		 418struct io_async_connect {
419 struct sockaddr_storage address;
420};
421
03b1230c
JA		 422struct io_async_msghdr {
423 struct iovec fast_iov[UIO_FASTIOV];
424 struct iovec *iov;
425 struct sockaddr __user *uaddr;
426 struct msghdr msg;
427};
428
f67676d1
JA		 429struct io_async_rw {
430 struct iovec fast_iov[UIO_FASTIOV];
431 struct iovec *iov;
432 ssize_t nr_segs;
433 ssize_t size;
434};
435
15b71abe
JA		 436struct io_async_open {
437 struct filename *filename;
438};
439
1a6b74fc 440struct io_async_ctx {
f67676d1
JA		 441 union {
442 struct io_async_rw rw;
03b1230c 443 struct io_async_msghdr msg;
f499a021 444 struct io_async_connect connect;
2d28390a 445 struct io_timeout_data timeout;
15b71abe 446 struct io_async_open open;
f67676d1 447 };
1a6b74fc
JA		 448};
449
09bb8394
JA		 450/*
451 * NOTE! Each of the iocb union members has the file pointer
452 * as the first entry in their struct definition. So you can
453 * access the file pointer through any of the sub-structs,
454 * or directly as just 'ki_filp' in this struct.
455 */
2b188cc1 456struct io_kiocb {
221c5eb2 457 union {
09bb8394 458 struct file *file;
9adbd45d 459 struct io_rw rw;
221c5eb2 460 struct io_poll_iocb poll;
8ed8d3c3
JA		 461 struct io_accept accept;
462 struct io_sync sync;
fbf23849 463 struct io_cancel cancel;
b29472ee 464 struct io_timeout timeout;
3fbb51c1 465 struct io_connect connect;
e47293fd 466 struct io_sr_msg sr_msg;
15b71abe 467 struct io_open open;
b5dba59e 468 struct io_close close;
05f3fb3c 469 struct io_files_update files_update;
4840e418 470 struct io_fadvise fadvise;
c1ca757b 471 struct io_madvise madvise;
221c5eb2 472 };
2b188cc1 473
1a6b74fc 474 struct io_async_ctx *io;
e94f141b
JA		 475 union {
476 /*
477 * ring_file is only used in the submission path, and
478 * llist_node is only used for poll deferred completions
479 */
480 struct file *ring_file;
481 struct llist_node llist_node;
482 };
cf6fd4bd
PB		 483 int ring_fd;
484 bool has_user;
485 bool in_async;
486 bool needs_fixed_file;
d625c6ee 487 u8 opcode;
2b188cc1
JA		 488
489 struct io_ring_ctx *ctx;
eac406c6
JA		 490 union {
491 struct list_head list;
78076bb6 492 struct hlist_node hash_node;
eac406c6 493 };
9e645e11 494 struct list_head link_list;
2b188cc1 495 unsigned int flags;
c16361c1 496 refcount_t refs;
8449eeda 497#define REQ_F_NOWAIT 1 /* must not punt to workers */
def596e9 498#define REQ_F_IOPOLL_COMPLETED 2 /* polled IO has completed */
6b06314c 499#define REQ_F_FIXED_FILE 4 /* ctx owns file */
4d7dd462 500#define REQ_F_LINK_NEXT 8 /* already grabbed next link */
e2033e33
SB		 501#define REQ_F_IO_DRAIN 16 /* drain existing IO first */
502#define REQ_F_IO_DRAINED 32 /* drain done */
9e645e11 503#define REQ_F_LINK 64 /* linked sqes */
2665abfd 504#define REQ_F_LINK_TIMEOUT 128 /* has linked timeout */
f7b76ac9 505#define REQ_F_FAIL_LINK 256 /* fail rest of links */
1b4a51b6 506#define REQ_F_DRAIN_LINK 512 /* link should be fully drained */
5262f567 507#define REQ_F_TIMEOUT 1024 /* timeout request */
491381ce
JA		 508#define REQ_F_ISREG 2048 /* regular file */
509#define REQ_F_MUST_PUNT 4096 /* must be punted even for NONBLOCK */
93bd25bb 510#define REQ_F_TIMEOUT_NOSEQ 8192 /* no timeout sequence */
fb4b3d3f
LT		 511#define REQ_F_INFLIGHT 16384 /* on inflight list */
512#define REQ_F_COMP_LOCKED 32768 /* completion under lock */
4e88d6e7 513#define REQ_F_HARDLINK 65536 /* doesn't sever on completion < 0 */
ce35a47a 514#define REQ_F_FORCE_ASYNC 131072 /* IOSQE_ASYNC */
ba04291e 515#define REQ_F_CUR_POS 262144 /* read/write uses file position */
2b188cc1 516 u64 user_data;
9e645e11 517 u32 result;
de0617e4 518 u32 sequence;
2b188cc1 519
fcb323cc
JA		 520 struct list_head inflight_entry;
521
561fb04a 522 struct io_wq_work work;
2b188cc1
JA		 523};
524
525#define IO_PLUG_THRESHOLD 2
def596e9 526#define IO_IOPOLL_BATCH 8
2b188cc1 527
9a56a232
JA		 528struct io_submit_state {
529 struct blk_plug plug;
530
2579f913
JA		 531 /*
532 * io_kiocb alloc cache
533 */
534 void *reqs[IO_IOPOLL_BATCH];
535 unsigned int free_reqs;
536 unsigned int cur_req;
537
9a56a232
JA		 538 /*
539 * File reference cache
540 */
541 struct file *file;
542 unsigned int fd;
543 unsigned int has_refs;
544 unsigned int used_refs;
545 unsigned int ios_left;
546};
547
d3656344
JA		 548struct io_op_def {
549 /* needs req->io allocated for deferral/async */
550 unsigned async_ctx : 1;
551 /* needs current->mm setup, does mm access */
552 unsigned needs_mm : 1;
553 /* needs req->file assigned */
554 unsigned needs_file : 1;
555 /* needs req->file assigned IFF fd is >= 0 */
556 unsigned fd_non_neg : 1;
557 /* hash wq insertion if file is a regular file */
558 unsigned hash_reg_file : 1;
559 /* unbound wq insertion if file is a non-regular file */
560 unsigned unbound_nonreg_file : 1;
561};
562
563static const struct io_op_def io_op_defs[] = {
564 {
565 /* IORING_OP_NOP */
566 },
567 {
568 /* IORING_OP_READV */
569 .async_ctx = 1,
570 .needs_mm = 1,
571 .needs_file = 1,
572 .unbound_nonreg_file = 1,
573 },
574 {
575 /* IORING_OP_WRITEV */
576 .async_ctx = 1,
577 .needs_mm = 1,
578 .needs_file = 1,
579 .hash_reg_file = 1,
580 .unbound_nonreg_file = 1,
581 },
582 {
583 /* IORING_OP_FSYNC */
584 .needs_file = 1,
585 },
586 {
587 /* IORING_OP_READ_FIXED */
588 .needs_file = 1,
589 .unbound_nonreg_file = 1,
590 },
591 {
592 /* IORING_OP_WRITE_FIXED */
593 .needs_file = 1,
594 .hash_reg_file = 1,
595 .unbound_nonreg_file = 1,
596 },
597 {
598 /* IORING_OP_POLL_ADD */
599 .needs_file = 1,
600 .unbound_nonreg_file = 1,
601 },
602 {
603 /* IORING_OP_POLL_REMOVE */
604 },
605 {
606 /* IORING_OP_SYNC_FILE_RANGE */
607 .needs_file = 1,
608 },
609 {
610 /* IORING_OP_SENDMSG */
611 .async_ctx = 1,
612 .needs_mm = 1,
613 .needs_file = 1,
614 .unbound_nonreg_file = 1,
615 },
616 {
617 /* IORING_OP_RECVMSG */
618 .async_ctx = 1,
619 .needs_mm = 1,
620 .needs_file = 1,
621 .unbound_nonreg_file = 1,
622 },
623 {
624 /* IORING_OP_TIMEOUT */
625 .async_ctx = 1,
626 .needs_mm = 1,
627 },
628 {
629 /* IORING_OP_TIMEOUT_REMOVE */
630 },
631 {
632 /* IORING_OP_ACCEPT */
633 .needs_mm = 1,
634 .needs_file = 1,
635 .unbound_nonreg_file = 1,
636 },
637 {
638 /* IORING_OP_ASYNC_CANCEL */
639 },
640 {
641 /* IORING_OP_LINK_TIMEOUT */
642 .async_ctx = 1,
643 .needs_mm = 1,
644 },
645 {
646 /* IORING_OP_CONNECT */
647 .async_ctx = 1,
648 .needs_mm = 1,
649 .needs_file = 1,
650 .unbound_nonreg_file = 1,
651 },
652 {
653 /* IORING_OP_FALLOCATE */
654 .needs_file = 1,
655 },
656 {
657 /* IORING_OP_OPENAT */
658 .needs_file = 1,
659 .fd_non_neg = 1,
660 },
661 {
662 /* IORING_OP_CLOSE */
663 .needs_file = 1,
664 },
665 {
666 /* IORING_OP_FILES_UPDATE */
667 .needs_mm = 1,
668 },
669 {
670 /* IORING_OP_STATX */
671 .needs_mm = 1,
672 .needs_file = 1,
673 .fd_non_neg = 1,
674 },
3a6820f2
JA		 675 {
676 /* IORING_OP_READ */
677 .needs_mm = 1,
678 .needs_file = 1,
679 .unbound_nonreg_file = 1,
680 },
681 {
682 /* IORING_OP_WRITE */
683 .needs_mm = 1,
684 .needs_file = 1,
685 .unbound_nonreg_file = 1,
686 },
4840e418
JA		 687 {
688 /* IORING_OP_FADVISE */
689 .needs_file = 1,
690 },
c1ca757b
JA		 691 {
692 /* IORING_OP_MADVISE */
693 .needs_mm = 1,
694 },
d3656344
JA		 695};
696
561fb04a 697static void io_wq_submit_work(struct io_wq_work **workptr);
78e19bbe 698static void io_cqring_fill_event(struct io_kiocb *req, long res);
ec9c02ad 699static void io_put_req(struct io_kiocb *req);
978db57e 700static void __io_double_put_req(struct io_kiocb *req);
94ae5e77
JA		 701static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
702static void io_queue_linked_timeout(struct io_kiocb *req);
05f3fb3c
JA		 703static int __io_sqe_files_update(struct io_ring_ctx *ctx,
704 struct io_uring_files_update *ip,
705 unsigned nr_args);
de0617e4 706
2b188cc1
JA		 707static struct kmem_cache *req_cachep;
708
709static const struct file_operations io_uring_fops;
710
711struct sock *io_uring_get_socket(struct file *file)
712{
713#if defined(CONFIG_UNIX)
714 if (file->f_op == &io_uring_fops) {
715 struct io_ring_ctx *ctx = file->private_data;
716
717 return ctx->ring_sock->sk;
718 }
719#endif
720 return NULL;
721}
722EXPORT_SYMBOL(io_uring_get_socket);
723
724static void io_ring_ctx_ref_free(struct percpu_ref *ref)
725{
726 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
727
206aefde 728 complete(&ctx->completions[0]);
2b188cc1
JA		 729}
730
731static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
732{
733 struct io_ring_ctx *ctx;
78076bb6 734 int hash_bits;
2b188cc1
JA		 735
736 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
737 if (!ctx)
738 return NULL;
739
0ddf92e8
JA		 740 ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL);
741 if (!ctx->fallback_req)
742 goto err;
743
206aefde
JA		 744 ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL);
745 if (!ctx->completions)
746 goto err;
747
78076bb6
JA		 748 /*
749 * Use 5 bits less than the max cq entries, that should give us around
750 * 32 entries per hash list if totally full and uniformly spread.
751 */
752 hash_bits = ilog2(p->cq_entries);
753 hash_bits -= 5;
754 if (hash_bits <= 0)
755 hash_bits = 1;
756 ctx->cancel_hash_bits = hash_bits;
757 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
758 GFP_KERNEL);
759 if (!ctx->cancel_hash)
760 goto err;
761 __hash_init(ctx->cancel_hash, 1U << hash_bits);
762
21482896 763 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
206aefde
JA		 764 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
765 goto err;
2b188cc1
JA		 766
767 ctx->flags = p->flags;
768 init_waitqueue_head(&ctx->cq_wait);
1d7bb1d5 769 INIT_LIST_HEAD(&ctx->cq_overflow_list);
206aefde
JA		 770 init_completion(&ctx->completions[0]);
771 init_completion(&ctx->completions[1]);
2b188cc1
JA		 772 mutex_init(&ctx->uring_lock);
773 init_waitqueue_head(&ctx->wait);
774 spin_lock_init(&ctx->completion_lock);
e94f141b 775 init_llist_head(&ctx->poll_llist);
def596e9 776 INIT_LIST_HEAD(&ctx->poll_list);
de0617e4 777 INIT_LIST_HEAD(&ctx->defer_list);
5262f567 778 INIT_LIST_HEAD(&ctx->timeout_list);
fcb323cc
JA		 779 init_waitqueue_head(&ctx->inflight_wait);
780 spin_lock_init(&ctx->inflight_lock);
781 INIT_LIST_HEAD(&ctx->inflight_list);
2b188cc1 782 return ctx;
206aefde 783err:
0ddf92e8
JA		 784 if (ctx->fallback_req)
785 kmem_cache_free(req_cachep, ctx->fallback_req);
206aefde 786 kfree(ctx->completions);
78076bb6 787 kfree(ctx->cancel_hash);
206aefde
JA		 788 kfree(ctx);
789 return NULL;
2b188cc1
JA		 790}
791
9d858b21 792static inline bool __req_need_defer(struct io_kiocb *req)
7adf4eaf 793{
a197f664
JL		 794 struct io_ring_ctx *ctx = req->ctx;
795
498ccd9e
JA		 796 return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped
797 + atomic_read(&ctx->cached_cq_overflow);
7adf4eaf
JA		 798}
799
9d858b21 800static inline bool req_need_defer(struct io_kiocb *req)
de0617e4 801{
9d858b21
BL		 802 if ((req->flags & (REQ_F_IO_DRAIN|REQ_F_IO_DRAINED)) == REQ_F_IO_DRAIN)
803 return __req_need_defer(req);
de0617e4 804
9d858b21 805 return false;
de0617e4
JA		 806}
807
7adf4eaf 808static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx)
de0617e4
JA		 809{
810 struct io_kiocb *req;
811
7adf4eaf 812 req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list);
9d858b21 813 if (req && !req_need_defer(req)) {
de0617e4
JA		 814 list_del_init(&req->list);
815 return req;
816 }
817
818 return NULL;
819}
820
5262f567
JA		 821static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx)
822{
7adf4eaf
JA		 823 struct io_kiocb *req;
824
825 req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list);
93bd25bb
JA		 826 if (req) {
827 if (req->flags & REQ_F_TIMEOUT_NOSEQ)
828 return NULL;
fb4b3d3f 829 if (!__req_need_defer(req)) {
93bd25bb
JA		 830 list_del_init(&req->list);
831 return req;
832 }
7adf4eaf
JA		 833 }
834
835 return NULL;
5262f567
JA		 836}
837
de0617e4 838static void __io_commit_cqring(struct io_ring_ctx *ctx)
2b188cc1 839{
75b28aff 840 struct io_rings *rings = ctx->rings;
2b188cc1 841
75b28aff 842 if (ctx->cached_cq_tail != READ_ONCE(rings->cq.tail)) {
2b188cc1 843 /* order cqe stores with ring update */
75b28aff 844 smp_store_release(&rings->cq.tail, ctx->cached_cq_tail);
2b188cc1 845
2b188cc1
JA		 846 if (wq_has_sleeper(&ctx->cq_wait)) {
847 wake_up_interruptible(&ctx->cq_wait);
848 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
849 }
850 }
851}
852
94ae5e77
JA		 853static inline bool io_prep_async_work(struct io_kiocb *req,
854 struct io_kiocb **link)
18d9be1a 855{
d3656344 856 const struct io_op_def *def = &io_op_defs[req->opcode];
561fb04a 857 bool do_hashed = false;
54a91f3b 858
d3656344
JA		 859 if (req->flags & REQ_F_ISREG) {
860 if (def->hash_reg_file)
3529d8c2 861 do_hashed = true;
d3656344
JA		 862 } else {
863 if (def->unbound_nonreg_file)
3529d8c2 864 req->work.flags |= IO_WQ_WORK_UNBOUND;
54a91f3b 865 }
d3656344 866 if (def->needs_mm)
3529d8c2 867 req->work.flags |= IO_WQ_WORK_NEEDS_USER;
54a91f3b 868
94ae5e77 869 *link = io_prep_linked_timeout(req);
561fb04a
JA		 870 return do_hashed;
871}
872
a197f664 873static inline void io_queue_async_work(struct io_kiocb *req)
561fb04a 874{
a197f664 875 struct io_ring_ctx *ctx = req->ctx;
94ae5e77
JA		 876 struct io_kiocb *link;
877 bool do_hashed;
878
879 do_hashed = io_prep_async_work(req, &link);
561fb04a
JA		 880
881 trace_io_uring_queue_async_work(ctx, do_hashed, req, &req->work,
882 req->flags);
883 if (!do_hashed) {
884 io_wq_enqueue(ctx->io_wq, &req->work);
885 } else {
886 io_wq_enqueue_hashed(ctx->io_wq, &req->work,
887 file_inode(req->file));
888 }
94ae5e77
JA		 889
890 if (link)
891 io_queue_linked_timeout(link);
18d9be1a
JA		 892}
893
5262f567
JA		 894static void io_kill_timeout(struct io_kiocb *req)
895{
896 int ret;
897
2d28390a 898 ret = hrtimer_try_to_cancel(&req->io->timeout.timer);
5262f567
JA		 899 if (ret != -1) {
900 atomic_inc(&req->ctx->cq_timeouts);
842f9612 901 list_del_init(&req->list);
78e19bbe 902 io_cqring_fill_event(req, 0);
ec9c02ad 903 io_put_req(req);
5262f567
JA		 904 }
905}
906
907static void io_kill_timeouts(struct io_ring_ctx *ctx)
908{
909 struct io_kiocb *req, *tmp;
910
911 spin_lock_irq(&ctx->completion_lock);
912 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list)
913 io_kill_timeout(req);
914 spin_unlock_irq(&ctx->completion_lock);
915}
916
de0617e4
JA		 917static void io_commit_cqring(struct io_ring_ctx *ctx)
918{
919 struct io_kiocb *req;
920
5262f567
JA		 921 while ((req = io_get_timeout_req(ctx)) != NULL)
922 io_kill_timeout(req);
923
de0617e4
JA		 924 __io_commit_cqring(ctx);
925
926 while ((req = io_get_deferred_req(ctx)) != NULL) {
927 req->flags |= REQ_F_IO_DRAINED;
a197f664 928 io_queue_async_work(req);
de0617e4
JA		 929 }
930}
931
2b188cc1
JA		 932static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
933{
75b28aff 934 struct io_rings *rings = ctx->rings;
2b188cc1
JA		 935 unsigned tail;
936
937 tail = ctx->cached_cq_tail;
115e12e5
SB		 938 /*
939 * writes to the cq entry need to come after reading head; the
940 * control dependency is enough as we're using WRITE_ONCE to
941 * fill the cq entry
942 */
75b28aff 943 if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries)
2b188cc1
JA		 944 return NULL;
945
946 ctx->cached_cq_tail++;
75b28aff 947 return &rings->cqes[tail & ctx->cq_mask];
2b188cc1
JA		 948}
949
1d7bb1d5
JA		 950static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
951{
952 if (waitqueue_active(&ctx->wait))
953 wake_up(&ctx->wait);
954 if (waitqueue_active(&ctx->sqo_wait))
955 wake_up(&ctx->sqo_wait);
956 if (ctx->cq_ev_fd)
957 eventfd_signal(ctx->cq_ev_fd, 1);
958}
959
c4a2ed72
JA		 960/* Returns true if there are no backlogged entries after the flush */
961static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
1d7bb1d5
JA		 962{
963 struct io_rings *rings = ctx->rings;
964 struct io_uring_cqe *cqe;
965 struct io_kiocb *req;
966 unsigned long flags;
967 LIST_HEAD(list);
968
969 if (!force) {
970 if (list_empty_careful(&ctx->cq_overflow_list))
c4a2ed72 971 return true;
1d7bb1d5
JA		 972 if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) ==
973 rings->cq_ring_entries))
c4a2ed72 974 return false;
1d7bb1d5
JA		 975 }
976
977 spin_lock_irqsave(&ctx->completion_lock, flags);
978
979 /* if force is set, the ring is going away. always drop after that */
980 if (force)
981 ctx->cq_overflow_flushed = true;
982
c4a2ed72 983 cqe = NULL;
1d7bb1d5
JA		 984 while (!list_empty(&ctx->cq_overflow_list)) {
985 cqe = io_get_cqring(ctx);
986 if (!cqe && !force)
987 break;
988
989 req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb,
990 list);
991 list_move(&req->list, &list);
992 if (cqe) {
993 WRITE_ONCE(cqe->user_data, req->user_data);
994 WRITE_ONCE(cqe->res, req->result);
995 WRITE_ONCE(cqe->flags, 0);
996 } else {
997 WRITE_ONCE(ctx->rings->cq_overflow,
998 atomic_inc_return(&ctx->cached_cq_overflow));
999 }
1000 }
1001
1002 io_commit_cqring(ctx);
ad3eb2c8
JA		 1003 if (cqe) {
1004 clear_bit(0, &ctx->sq_check_overflow);
1005 clear_bit(0, &ctx->cq_check_overflow);
1006 }
1d7bb1d5
JA		 1007 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1008 io_cqring_ev_posted(ctx);
1009
1010 while (!list_empty(&list)) {
1011 req = list_first_entry(&list, struct io_kiocb, list);
1012 list_del(&req->list);
ec9c02ad 1013 io_put_req(req);
1d7bb1d5 1014 }
c4a2ed72
JA		 1015
1016 return cqe != NULL;
1d7bb1d5
JA		 1017}
1018
78e19bbe 1019static void io_cqring_fill_event(struct io_kiocb *req, long res)
2b188cc1 1020{
78e19bbe 1021 struct io_ring_ctx *ctx = req->ctx;
2b188cc1
JA		 1022 struct io_uring_cqe *cqe;
1023
78e19bbe 1024 trace_io_uring_complete(ctx, req->user_data, res);
51c3ff62 1025
2b188cc1
JA		 1026 /*
1027 * If we can't get a cq entry, userspace overflowed the
1028 * submission (by quite a lot). Increment the overflow count in
1029 * the ring.
1030 */
1031 cqe = io_get_cqring(ctx);
1d7bb1d5 1032 if (likely(cqe)) {
78e19bbe 1033 WRITE_ONCE(cqe->user_data, req->user_data);
2b188cc1 1034 WRITE_ONCE(cqe->res, res);
c71ffb67 1035 WRITE_ONCE(cqe->flags, 0);
1d7bb1d5 1036 } else if (ctx->cq_overflow_flushed) {
498ccd9e
JA		 1037 WRITE_ONCE(ctx->rings->cq_overflow,
1038 atomic_inc_return(&ctx->cached_cq_overflow));
1d7bb1d5 1039 } else {
ad3eb2c8
JA		 1040 if (list_empty(&ctx->cq_overflow_list)) {
1041 set_bit(0, &ctx->sq_check_overflow);
1042 set_bit(0, &ctx->cq_check_overflow);
1043 }
1d7bb1d5
JA		 1044 refcount_inc(&req->refs);
1045 req->result = res;
1046 list_add_tail(&req->list, &ctx->cq_overflow_list);
2b188cc1
JA		 1047 }
1048}
1049
78e19bbe 1050static void io_cqring_add_event(struct io_kiocb *req, long res)
2b188cc1 1051{
78e19bbe 1052 struct io_ring_ctx *ctx = req->ctx;
2b188cc1
JA		 1053 unsigned long flags;
1054
1055 spin_lock_irqsave(&ctx->completion_lock, flags);
78e19bbe 1056 io_cqring_fill_event(req, res);
2b188cc1
JA		 1057 io_commit_cqring(ctx);
1058 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1059
8c838788 1060 io_cqring_ev_posted(ctx);
2b188cc1
JA		 1061}
1062
0ddf92e8
JA		 1063static inline bool io_is_fallback_req(struct io_kiocb *req)
1064{
1065 return req == (struct io_kiocb *)
1066 ((unsigned long) req->ctx->fallback_req & ~1UL);
1067}
1068
1069static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx)
1070{
1071 struct io_kiocb *req;
1072
1073 req = ctx->fallback_req;
1074 if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req))
1075 return req;
1076
1077 return NULL;
1078}
1079
2579f913
JA		 1080static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
1081 struct io_submit_state *state)
2b188cc1 1082{
fd6fab2c 1083 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
2b188cc1
JA		 1084 struct io_kiocb *req;
1085
2579f913 1086 if (!state) {
fd6fab2c 1087 req = kmem_cache_alloc(req_cachep, gfp);
2579f913 1088 if (unlikely(!req))
0ddf92e8 1089 goto fallback;
2579f913
JA		 1090 } else if (!state->free_reqs) {
1091 size_t sz;
1092 int ret;
1093
1094 sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs));
fd6fab2c
JA		 1095 ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs);
1096
1097 /*
1098 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1099 * retry single alloc to be on the safe side.
1100 */
1101 if (unlikely(ret <= 0)) {
1102 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1103 if (!state->reqs[0])
0ddf92e8 1104 goto fallback;
fd6fab2c
JA		 1105 ret = 1;
1106 }
2579f913
JA		 1107 state->free_reqs = ret - 1;
1108 state->cur_req = 1;
1109 req = state->reqs[0];
1110 } else {
1111 req = state->reqs[state->cur_req];
1112 state->free_reqs--;
1113 state->cur_req++;
2b188cc1
JA		 1114 }
1115
0ddf92e8 1116got_it:
1a6b74fc 1117 req->io = NULL;
cf6fd4bd 1118 req->ring_file = NULL;
60c112b0 1119 req->file = NULL;
2579f913
JA		 1120 req->ctx = ctx;
1121 req->flags = 0;
e65ef56d
JA		 1122 /* one is dropped after submission, the other at completion */
1123 refcount_set(&req->refs, 2);
9e645e11 1124 req->result = 0;
561fb04a 1125 INIT_IO_WORK(&req->work, io_wq_submit_work);
2579f913 1126 return req;
0ddf92e8
JA		 1127fallback:
1128 req = io_get_fallback_req(ctx);
1129 if (req)
1130 goto got_it;
6805b32e 1131 percpu_ref_put(&ctx->refs);
2b188cc1
JA		 1132 return NULL;
1133}
1134
2b85edfc
PB		 1135static void __io_req_do_free(struct io_kiocb *req)
1136{
1137 if (likely(!io_is_fallback_req(req)))
1138 kmem_cache_free(req_cachep, req);
1139 else
1140 clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req);
1141}
1142
c6ca97b3 1143static void __io_req_aux_free(struct io_kiocb *req)
2b188cc1 1144{
fcb323cc
JA		 1145 struct io_ring_ctx *ctx = req->ctx;
1146
1a6b74fc
JA		 1147 if (req->io)
1148 kfree(req->io);
05f3fb3c
JA		 1149 if (req->file) {
1150 if (req->flags & REQ_F_FIXED_FILE)
1151 percpu_ref_put(&ctx->file_data->refs);
1152 else
1153 fput(req->file);
1154 }
c6ca97b3
JA		 1155}
1156
1157static void __io_free_req(struct io_kiocb *req)
1158{
1159 __io_req_aux_free(req);
1160
fcb323cc 1161 if (req->flags & REQ_F_INFLIGHT) {
c6ca97b3 1162 struct io_ring_ctx *ctx = req->ctx;
fcb323cc
JA		 1163 unsigned long flags;
1164
1165 spin_lock_irqsave(&ctx->inflight_lock, flags);
1166 list_del(&req->inflight_entry);
1167 if (waitqueue_active(&ctx->inflight_wait))
1168 wake_up(&ctx->inflight_wait);
1169 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1170 }
2b85edfc
PB		 1171
1172 percpu_ref_put(&req->ctx->refs);
1173 __io_req_do_free(req);
e65ef56d
JA		 1174}
1175
c6ca97b3
JA		 1176struct req_batch {
1177 void *reqs[IO_IOPOLL_BATCH];
1178 int to_free;
1179 int need_iter;
1180};
1181
1182static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb)
1183{
1184 if (!rb->to_free)
1185 return;
1186 if (rb->need_iter) {
1187 int i, inflight = 0;
1188 unsigned long flags;
1189
1190 for (i = 0; i < rb->to_free; i++) {
1191 struct io_kiocb *req = rb->reqs[i];
1192
1193 if (req->flags & REQ_F_FIXED_FILE)
1194 req->file = NULL;
1195 if (req->flags & REQ_F_INFLIGHT)
1196 inflight++;
1197 else
1198 rb->reqs[i] = NULL;
1199 __io_req_aux_free(req);
1200 }
1201 if (!inflight)
1202 goto do_free;
1203
1204 spin_lock_irqsave(&ctx->inflight_lock, flags);
1205 for (i = 0; i < rb->to_free; i++) {
1206 struct io_kiocb *req = rb->reqs[i];
1207
1208 if (req) {
1209 list_del(&req->inflight_entry);
1210 if (!--inflight)
1211 break;
1212 }
1213 }
1214 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1215
1216 if (waitqueue_active(&ctx->inflight_wait))
1217 wake_up(&ctx->inflight_wait);
1218 }
1219do_free:
1220 kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs);
1221 percpu_ref_put_many(&ctx->refs, rb->to_free);
1222 percpu_ref_put_many(&ctx->file_data->refs, rb->to_free);
1223 rb->to_free = rb->need_iter = 0;
1224}
1225
a197f664 1226static bool io_link_cancel_timeout(struct io_kiocb *req)
2665abfd 1227{
a197f664 1228 struct io_ring_ctx *ctx = req->ctx;
2665abfd
JA		 1229 int ret;
1230
2d28390a 1231 ret = hrtimer_try_to_cancel(&req->io->timeout.timer);
2665abfd 1232 if (ret != -1) {
78e19bbe 1233 io_cqring_fill_event(req, -ECANCELED);
2665abfd
JA		 1234 io_commit_cqring(ctx);
1235 req->flags &= ~REQ_F_LINK;
ec9c02ad 1236 io_put_req(req);
2665abfd
JA		 1237 return true;
1238 }
1239
1240 return false;
e65ef56d
JA		 1241}
1242
ba816ad6 1243static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
9e645e11 1244{
2665abfd 1245 struct io_ring_ctx *ctx = req->ctx;
2665abfd 1246 bool wake_ev = false;
9e645e11 1247
4d7dd462
JA		 1248 /* Already got next link */
1249 if (req->flags & REQ_F_LINK_NEXT)
1250 return;
1251
9e645e11
JA		 1252 /*
1253 * The list should never be empty when we are called here. But could
1254 * potentially happen if the chain is messed up, check to be on the
1255 * safe side.
1256 */
4493233e
PB		 1257 while (!list_empty(&req->link_list)) {
1258 struct io_kiocb *nxt = list_first_entry(&req->link_list,
1259 struct io_kiocb, link_list);
94ae5e77 1260
4493233e
PB		 1261 if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) &&
1262 (nxt->flags & REQ_F_TIMEOUT))) {
1263 list_del_init(&nxt->link_list);
94ae5e77 1264 wake_ev |= io_link_cancel_timeout(nxt);
94ae5e77
JA		 1265 req->flags &= ~REQ_F_LINK_TIMEOUT;
1266 continue;
1267 }
9e645e11 1268
4493233e
PB		 1269 list_del_init(&req->link_list);
1270 if (!list_empty(&nxt->link_list))
1271 nxt->flags |= REQ_F_LINK;
b18fdf71 1272 *nxtptr = nxt;
94ae5e77 1273 break;
9e645e11 1274 }
2665abfd 1275
4d7dd462 1276 req->flags |= REQ_F_LINK_NEXT;
2665abfd
JA		 1277 if (wake_ev)
1278 io_cqring_ev_posted(ctx);
9e645e11
JA		 1279}
1280
1281/*
1282 * Called if REQ_F_LINK is set, and we fail the head request
1283 */
1284static void io_fail_links(struct io_kiocb *req)
1285{
2665abfd 1286 struct io_ring_ctx *ctx = req->ctx;
2665abfd
JA		 1287 unsigned long flags;
1288
1289 spin_lock_irqsave(&ctx->completion_lock, flags);
9e645e11
JA		 1290
1291 while (!list_empty(&req->link_list)) {
4493233e
PB		 1292 struct io_kiocb *link = list_first_entry(&req->link_list,
1293 struct io_kiocb, link_list);
9e645e11 1294
4493233e 1295 list_del_init(&link->link_list);
c826bd7a 1296 trace_io_uring_fail_link(req, link);
2665abfd
JA		 1297
1298 if ((req->flags & REQ_F_LINK_TIMEOUT) &&
d625c6ee 1299 link->opcode == IORING_OP_LINK_TIMEOUT) {
a197f664 1300 io_link_cancel_timeout(link);
2665abfd 1301 } else {
78e19bbe 1302 io_cqring_fill_event(link, -ECANCELED);
978db57e 1303 __io_double_put_req(link);
2665abfd 1304 }
5d960724 1305 req->flags &= ~REQ_F_LINK_TIMEOUT;
9e645e11 1306 }
2665abfd
JA		 1307
1308 io_commit_cqring(ctx);
1309 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1310 io_cqring_ev_posted(ctx);
9e645e11
JA		 1311}
1312
4d7dd462 1313static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt)
9e645e11 1314{
4d7dd462 1315 if (likely(!(req->flags & REQ_F_LINK)))
2665abfd 1316 return;
2665abfd 1317
9e645e11
JA		 1318 /*
1319 * If LINK is set, we have dependent requests in this chain. If we
1320 * didn't fail this request, queue the first one up, moving any other
1321 * dependencies to the next request. In case of failure, fail the rest
1322 * of the chain.
1323 */
2665abfd
JA		 1324 if (req->flags & REQ_F_FAIL_LINK) {
1325 io_fail_links(req);
7c9e7f0f
JA		 1326 } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) ==
1327 REQ_F_LINK_TIMEOUT) {
2665abfd
JA		 1328 struct io_ring_ctx *ctx = req->ctx;
1329 unsigned long flags;
1330
1331 /*
1332 * If this is a timeout link, we could be racing with the
1333 * timeout timer. Grab the completion lock for this case to
7c9e7f0f 1334 * protect against that.
2665abfd
JA		 1335 */
1336 spin_lock_irqsave(&ctx->completion_lock, flags);
1337 io_req_link_next(req, nxt);
1338 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1339 } else {
1340 io_req_link_next(req, nxt);
9e645e11 1341 }
4d7dd462 1342}
9e645e11 1343
c69f8dbe
JL		 1344static void io_free_req(struct io_kiocb *req)
1345{
944e58bf
PB		 1346 struct io_kiocb *nxt = NULL;
1347
1348 io_req_find_next(req, &nxt);
70cf9f32 1349 __io_free_req(req);
944e58bf
PB		 1350
1351 if (nxt)
1352 io_queue_async_work(nxt);
c69f8dbe
JL		 1353}
1354
ba816ad6
JA		 1355/*
1356 * Drop reference to request, return next in chain (if there is one) if this
1357 * was the last reference to this request.
1358 */
f9bd67f6 1359__attribute__((nonnull))
ec9c02ad 1360static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
e65ef56d 1361{
f9bd67f6 1362 io_req_find_next(req, nxtptr);
4d7dd462 1363
e65ef56d 1364 if (refcount_dec_and_test(&req->refs))
4d7dd462 1365 __io_free_req(req);
2b188cc1
JA		 1366}
1367
e65ef56d
JA		 1368static void io_put_req(struct io_kiocb *req)
1369{
1370 if (refcount_dec_and_test(&req->refs))
1371 io_free_req(req);
2b188cc1
JA		 1372}
1373
978db57e
JA		 1374/*
1375 * Must only be used if we don't need to care about links, usually from
1376 * within the completion handling itself.
1377 */
1378static void __io_double_put_req(struct io_kiocb *req)
78e19bbe
JA		 1379{
1380 /* drop both submit and complete references */
1381 if (refcount_sub_and_test(2, &req->refs))
1382 __io_free_req(req);
1383}
1384
978db57e
JA		 1385static void io_double_put_req(struct io_kiocb *req)
1386{
1387 /* drop both submit and complete references */
1388 if (refcount_sub_and_test(2, &req->refs))
1389 io_free_req(req);
1390}
1391
1d7bb1d5 1392static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush)
a3a0e43f 1393{
84f97dc2
JA		 1394 struct io_rings *rings = ctx->rings;
1395
ad3eb2c8
JA		 1396 if (test_bit(0, &ctx->cq_check_overflow)) {
1397 /*
1398 * noflush == true is from the waitqueue handler, just ensure
1399 * we wake up the task, and the next invocation will flush the
1400 * entries. We cannot safely to it from here.
1401 */
1402 if (noflush && !list_empty(&ctx->cq_overflow_list))
1403 return -1U;
1d7bb1d5 1404
ad3eb2c8
JA		 1405 io_cqring_overflow_flush(ctx, false);
1406 }
1d7bb1d5 1407
a3a0e43f
JA		 1408 /* See comment at the top of this file */
1409 smp_rmb();
ad3eb2c8 1410 return ctx->cached_cq_tail - READ_ONCE(rings->cq.head);
a3a0e43f
JA		 1411}
1412
fb5ccc98
PB		 1413static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
1414{
1415 struct io_rings *rings = ctx->rings;
1416
1417 /* make sure SQ entry isn't read before tail */
1418 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
1419}
1420
8237e045 1421static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req)
e94f141b 1422{
c6ca97b3
JA		 1423 if ((req->flags & REQ_F_LINK) || io_is_fallback_req(req))
1424 return false;
e94f141b 1425
c6ca97b3
JA		 1426 if (!(req->flags & REQ_F_FIXED_FILE) || req->io)
1427 rb->need_iter++;
1428
1429 rb->reqs[rb->to_free++] = req;
1430 if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs)))
1431 io_free_req_many(req->ctx, rb);
1432 return true;
e94f141b
JA		 1433}
1434
def596e9
JA		 1435/*
1436 * Find and free completed poll iocbs
1437 */
1438static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
1439 struct list_head *done)
1440{
8237e045 1441 struct req_batch rb;
def596e9 1442 struct io_kiocb *req;
def596e9 1443
c6ca97b3 1444 rb.to_free = rb.need_iter = 0;
def596e9
JA		 1445 while (!list_empty(done)) {
1446 req = list_first_entry(done, struct io_kiocb, list);
1447 list_del(&req->list);
1448
78e19bbe 1449 io_cqring_fill_event(req, req->result);
def596e9
JA		 1450 (*nr_events)++;
1451
8237e045
JA		 1452 if (refcount_dec_and_test(&req->refs) &&
1453 !io_req_multi_free(&rb, req))
1454 io_free_req(req);
def596e9 1455 }
def596e9 1456
09bb8394 1457 io_commit_cqring(ctx);
8237e045 1458 io_free_req_many(ctx, &rb);
def596e9
JA		 1459}
1460
1461static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
1462 long min)
1463{
1464 struct io_kiocb *req, *tmp;
1465 LIST_HEAD(done);
1466 bool spin;
1467 int ret;
1468
1469 /*
1470 * Only spin for completions if we don't have multiple devices hanging
1471 * off our complete list, and we're under the requested amount.
1472 */
1473 spin = !ctx->poll_multi_file && *nr_events < min;
1474
1475 ret = 0;
1476 list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) {
9adbd45d 1477 struct kiocb *kiocb = &req->rw.kiocb;
def596e9
JA		 1478
1479 /*
1480 * Move completed entries to our local list. If we find a
1481 * request that requires polling, break out and complete
1482 * the done list first, if we have entries there.
1483 */
1484 if (req->flags & REQ_F_IOPOLL_COMPLETED) {
1485 list_move_tail(&req->list, &done);
1486 continue;
1487 }
1488 if (!list_empty(&done))
1489 break;
1490
1491 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
1492 if (ret < 0)
1493 break;
1494
1495 if (ret && spin)
1496 spin = false;
1497 ret = 0;
1498 }
1499
1500 if (!list_empty(&done))
1501 io_iopoll_complete(ctx, nr_events, &done);
1502
1503 return ret;
1504}
1505
1506/*
d195a66e 1507 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
def596e9
JA		 1508 * non-spinning poll check - we'll still enter the driver poll loop, but only
1509 * as a non-spinning completion check.
1510 */
1511static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
1512 long min)
1513{
08f5439f 1514 while (!list_empty(&ctx->poll_list) && !need_resched()) {
def596e9
JA		 1515 int ret;
1516
1517 ret = io_do_iopoll(ctx, nr_events, min);
1518 if (ret < 0)
1519 return ret;
1520 if (!min || *nr_events >= min)
1521 return 0;
1522 }
1523
1524 return 1;
1525}
1526
1527/*
1528 * We can't just wait for polled events to come to us, we have to actively
1529 * find and complete them.
1530 */
1531static void io_iopoll_reap_events(struct io_ring_ctx *ctx)
1532{
1533 if (!(ctx->flags & IORING_SETUP_IOPOLL))
1534 return;
1535
1536 mutex_lock(&ctx->uring_lock);
1537 while (!list_empty(&ctx->poll_list)) {
1538 unsigned int nr_events = 0;
1539
1540 io_iopoll_getevents(ctx, &nr_events, 1);
08f5439f
JA		 1541
1542 /*
1543 * Ensure we allow local-to-the-cpu processing to take place,
1544 * in this case we need to ensure that we reap all events.
1545 */
1546 cond_resched();
def596e9
JA		 1547 }
1548 mutex_unlock(&ctx->uring_lock);
1549}
1550
2b2ed975
JA		 1551static int __io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events,
1552 long min)
def596e9 1553{
2b2ed975 1554 int iters = 0, ret = 0;
500f9fba 1555
def596e9
JA		 1556 do {
1557 int tmin = 0;
1558
a3a0e43f
JA		 1559 /*
1560 * Don't enter poll loop if we already have events pending.
1561 * If we do, we can potentially be spinning for commands that
1562 * already triggered a CQE (eg in error).
1563 */
1d7bb1d5 1564 if (io_cqring_events(ctx, false))
a3a0e43f
JA		 1565 break;
1566
500f9fba
JA		 1567 /*
1568 * If a submit got punted to a workqueue, we can have the
1569 * application entering polling for a command before it gets
1570 * issued. That app will hold the uring_lock for the duration
1571 * of the poll right here, so we need to take a breather every
1572 * now and then to ensure that the issue has a chance to add
1573 * the poll to the issued list. Otherwise we can spin here
1574 * forever, while the workqueue is stuck trying to acquire the
1575 * very same mutex.
1576 */
1577 if (!(++iters & 7)) {
1578 mutex_unlock(&ctx->uring_lock);
1579 mutex_lock(&ctx->uring_lock);
1580 }
1581
def596e9
JA		 1582 if (*nr_events < min)
1583 tmin = min - *nr_events;
1584
1585 ret = io_iopoll_getevents(ctx, nr_events, tmin);
1586 if (ret <= 0)
1587 break;
1588 ret = 0;
1589 } while (min && !*nr_events && !need_resched());
1590
2b2ed975
JA		 1591 return ret;
1592}
1593
1594static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events,
1595 long min)
1596{
1597 int ret;
1598
1599 /*
1600 * We disallow the app entering submit/complete with polling, but we
1601 * still need to lock the ring to prevent racing with polled issue
1602 * that got punted to a workqueue.
1603 */
1604 mutex_lock(&ctx->uring_lock);
1605 ret = __io_iopoll_check(ctx, nr_events, min);
500f9fba 1606 mutex_unlock(&ctx->uring_lock);
def596e9
JA		 1607 return ret;
1608}
1609
491381ce 1610static void kiocb_end_write(struct io_kiocb *req)
2b188cc1 1611{
491381ce
JA		 1612 /*
1613 * Tell lockdep we inherited freeze protection from submission
1614 * thread.
1615 */
1616 if (req->flags & REQ_F_ISREG) {
1617 struct inode *inode = file_inode(req->file);
2b188cc1 1618
491381ce 1619 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
2b188cc1 1620 }
491381ce 1621 file_end_write(req->file);
2b188cc1
JA		 1622}
1623
4e88d6e7
JA		 1624static inline void req_set_fail_links(struct io_kiocb *req)
1625{
1626 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1627 req->flags |= REQ_F_FAIL_LINK;
1628}
1629
ba816ad6 1630static void io_complete_rw_common(struct kiocb *kiocb, long res)
2b188cc1 1631{
9adbd45d 1632 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
2b188cc1 1633
491381ce
JA		 1634 if (kiocb->ki_flags & IOCB_WRITE)
1635 kiocb_end_write(req);
2b188cc1 1636
4e88d6e7
JA		 1637 if (res != req->result)
1638 req_set_fail_links(req);
78e19bbe 1639 io_cqring_add_event(req, res);
ba816ad6
JA		 1640}
1641
1642static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
1643{
9adbd45d 1644 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
ba816ad6
JA		 1645
1646 io_complete_rw_common(kiocb, res);
e65ef56d 1647 io_put_req(req);
2b188cc1
JA		 1648}
1649
ba816ad6
JA		 1650static struct io_kiocb *__io_complete_rw(struct kiocb *kiocb, long res)
1651{
9adbd45d 1652 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
ec9c02ad 1653 struct io_kiocb *nxt = NULL;
ba816ad6
JA		 1654
1655 io_complete_rw_common(kiocb, res);
ec9c02ad
JL		 1656 io_put_req_find_next(req, &nxt);
1657
1658 return nxt;
2b188cc1
JA		 1659}
1660
def596e9
JA		 1661static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
1662{
9adbd45d 1663 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
def596e9 1664
491381ce
JA		 1665 if (kiocb->ki_flags & IOCB_WRITE)
1666 kiocb_end_write(req);
def596e9 1667
4e88d6e7
JA		 1668 if (res != req->result)
1669 req_set_fail_links(req);
9e645e11 1670 req->result = res;
def596e9
JA		 1671 if (res != -EAGAIN)
1672 req->flags |= REQ_F_IOPOLL_COMPLETED;
1673}
1674
1675/*
1676 * After the iocb has been issued, it's safe to be found on the poll list.
1677 * Adding the kiocb to the list AFTER submission ensures that we don't
1678 * find it from a io_iopoll_getevents() thread before the issuer is done
1679 * accessing the kiocb cookie.
1680 */
1681static void io_iopoll_req_issued(struct io_kiocb *req)
1682{
1683 struct io_ring_ctx *ctx = req->ctx;
1684
1685 /*
1686 * Track whether we have multiple files in our lists. This will impact
1687 * how we do polling eventually, not spinning if we're on potentially
1688 * different devices.
1689 */
1690 if (list_empty(&ctx->poll_list)) {
1691 ctx->poll_multi_file = false;
1692 } else if (!ctx->poll_multi_file) {
1693 struct io_kiocb *list_req;
1694
1695 list_req = list_first_entry(&ctx->poll_list, struct io_kiocb,
1696 list);
9adbd45d 1697 if (list_req->file != req->file)
def596e9
JA		 1698 ctx->poll_multi_file = true;
1699 }
1700
1701 /*
1702 * For fast devices, IO may have already completed. If it has, add
1703 * it to the front so we find it first.
1704 */
1705 if (req->flags & REQ_F_IOPOLL_COMPLETED)
1706 list_add(&req->list, &ctx->poll_list);
1707 else
1708 list_add_tail(&req->list, &ctx->poll_list);
1709}
1710
3d6770fb 1711static void io_file_put(struct io_submit_state *state)
9a56a232 1712{
3d6770fb 1713 if (state->file) {
9a56a232
JA		 1714 int diff = state->has_refs - state->used_refs;
1715
1716 if (diff)
1717 fput_many(state->file, diff);
1718 state->file = NULL;
1719 }
1720}
1721
1722/*
1723 * Get as many references to a file as we have IOs left in this submission,
1724 * assuming most submissions are for one file, or at least that each file
1725 * has more than one submission.
1726 */
1727static struct file *io_file_get(struct io_submit_state *state, int fd)
1728{
1729 if (!state)
1730 return fget(fd);
1731
1732 if (state->file) {
1733 if (state->fd == fd) {
1734 state->used_refs++;
1735 state->ios_left--;
1736 return state->file;
1737 }
3d6770fb 1738 io_file_put(state);
9a56a232
JA		 1739 }
1740 state->file = fget_many(fd, state->ios_left);
1741 if (!state->file)
1742 return NULL;
1743
1744 state->fd = fd;
1745 state->has_refs = state->ios_left;
1746 state->used_refs = 1;
1747 state->ios_left--;
1748 return state->file;
1749}
1750
2b188cc1
JA		 1751/*
1752 * If we tracked the file through the SCM inflight mechanism, we could support
1753 * any file. For now, just ensure that anything potentially problematic is done
1754 * inline.
1755 */
1756static bool io_file_supports_async(struct file *file)
1757{
1758 umode_t mode = file_inode(file)->i_mode;
1759
10d59345 1760 if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode))
2b188cc1
JA		 1761 return true;
1762 if (S_ISREG(mode) && file->f_op != &io_uring_fops)
1763 return true;
1764
1765 return false;
1766}
1767
3529d8c2
JA		 1768static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe,
1769 bool force_nonblock)
2b188cc1 1770{
def596e9 1771 struct io_ring_ctx *ctx = req->ctx;
9adbd45d 1772 struct kiocb *kiocb = &req->rw.kiocb;
09bb8394
JA		 1773 unsigned ioprio;
1774 int ret;
2b188cc1 1775
09bb8394
JA		 1776 if (!req->file)
1777 return -EBADF;
2b188cc1 1778
491381ce
JA		 1779 if (S_ISREG(file_inode(req->file)->i_mode))
1780 req->flags |= REQ_F_ISREG;
1781
2b188cc1 1782 kiocb->ki_pos = READ_ONCE(sqe->off);
ba04291e
JA		 1783 if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) {
1784 req->flags |= REQ_F_CUR_POS;
1785 kiocb->ki_pos = req->file->f_pos;
1786 }
2b188cc1
JA		 1787 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
1788 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
1789
1790 ioprio = READ_ONCE(sqe->ioprio);
1791 if (ioprio) {
1792 ret = ioprio_check_cap(ioprio);
1793 if (ret)
09bb8394 1794 return ret;
2b188cc1
JA		 1795
1796 kiocb->ki_ioprio = ioprio;
1797 } else
1798 kiocb->ki_ioprio = get_current_ioprio();
1799
1800 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
1801 if (unlikely(ret))
09bb8394 1802 return ret;
8449eeda
SB		 1803
1804 /* don't allow async punt if RWF_NOWAIT was requested */
491381ce
JA		 1805 if ((kiocb->ki_flags & IOCB_NOWAIT) ||
1806 (req->file->f_flags & O_NONBLOCK))
8449eeda
SB		 1807 req->flags |= REQ_F_NOWAIT;
1808
1809 if (force_nonblock)
2b188cc1 1810 kiocb->ki_flags |= IOCB_NOWAIT;
8449eeda 1811
def596e9 1812 if (ctx->flags & IORING_SETUP_IOPOLL) {
def596e9
JA		 1813 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
1814 !kiocb->ki_filp->f_op->iopoll)
09bb8394 1815 return -EOPNOTSUPP;
2b188cc1 1816
def596e9
JA		 1817 kiocb->ki_flags |= IOCB_HIPRI;
1818 kiocb->ki_complete = io_complete_rw_iopoll;
6873e0bd 1819 req->result = 0;
def596e9 1820 } else {
09bb8394
JA		 1821 if (kiocb->ki_flags & IOCB_HIPRI)
1822 return -EINVAL;
def596e9
JA		 1823 kiocb->ki_complete = io_complete_rw;
1824 }
9adbd45d 1825
3529d8c2
JA		 1826 req->rw.addr = READ_ONCE(sqe->addr);
1827 req->rw.len = READ_ONCE(sqe->len);
9adbd45d
JA		 1828 /* we own ->private, reuse it for the buffer index */
1829 req->rw.kiocb.private = (void *) (unsigned long)
3529d8c2 1830 READ_ONCE(sqe->buf_index);
2b188cc1 1831 return 0;
2b188cc1
JA		 1832}
1833
1834static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
1835{
1836 switch (ret) {
1837 case -EIOCBQUEUED:
1838 break;
1839 case -ERESTARTSYS:
1840 case -ERESTARTNOINTR:
1841 case -ERESTARTNOHAND:
1842 case -ERESTART_RESTARTBLOCK:
1843 /*
1844 * We can't just restart the syscall, since previously
1845 * submitted sqes may already be in progress. Just fail this
1846 * IO with EINTR.
1847 */
1848 ret = -EINTR;
1849 /* fall through */
1850 default:
1851 kiocb->ki_complete(kiocb, ret, 0);
1852 }
1853}
1854
ba816ad6
JA		 1855static void kiocb_done(struct kiocb *kiocb, ssize_t ret, struct io_kiocb **nxt,
1856 bool in_async)
1857{
ba04291e
JA		 1858 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
1859
1860 if (req->flags & REQ_F_CUR_POS)
1861 req->file->f_pos = kiocb->ki_pos;
f9bd67f6 1862 if (in_async && ret >= 0 && kiocb->ki_complete == io_complete_rw)
ba816ad6
JA		 1863 *nxt = __io_complete_rw(kiocb, ret);
1864 else
1865 io_rw_done(kiocb, ret);
1866}
1867
9adbd45d 1868static ssize_t io_import_fixed(struct io_kiocb *req, int rw,
7d009165 1869 struct iov_iter *iter)
edafccee 1870{
9adbd45d
JA		 1871 struct io_ring_ctx *ctx = req->ctx;
1872 size_t len = req->rw.len;
edafccee
JA		 1873 struct io_mapped_ubuf *imu;
1874 unsigned index, buf_index;
1875 size_t offset;
1876 u64 buf_addr;
1877
1878 /* attempt to use fixed buffers without having provided iovecs */
1879 if (unlikely(!ctx->user_bufs))
1880 return -EFAULT;
1881
9adbd45d 1882 buf_index = (unsigned long) req->rw.kiocb.private;
edafccee
JA		 1883 if (unlikely(buf_index >= ctx->nr_user_bufs))
1884 return -EFAULT;
1885
1886 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
1887 imu = &ctx->user_bufs[index];
9adbd45d 1888 buf_addr = req->rw.addr;
edafccee
JA		 1889
1890 /* overflow */
1891 if (buf_addr + len < buf_addr)
1892 return -EFAULT;
1893 /* not inside the mapped region */
1894 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
1895 return -EFAULT;
1896
1897 /*
1898 * May not be a start of buffer, set size appropriately
1899 * and advance us to the beginning.
1900 */
1901 offset = buf_addr - imu->ubuf;
1902 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
bd11b3a3
JA		 1903
1904 if (offset) {
1905 /*
1906 * Don't use iov_iter_advance() here, as it's really slow for
1907 * using the latter parts of a big fixed buffer - it iterates
1908 * over each segment manually. We can cheat a bit here, because
1909 * we know that:
1910 *
1911 * 1) it's a BVEC iter, we set it up
1912 * 2) all bvecs are PAGE_SIZE in size, except potentially the
1913 * first and last bvec
1914 *
1915 * So just find our index, and adjust the iterator afterwards.
1916 * If the offset is within the first bvec (or the whole first
1917 * bvec, just use iov_iter_advance(). This makes it easier
1918 * since we can just skip the first segment, which may not
1919 * be PAGE_SIZE aligned.
1920 */
1921 const struct bio_vec *bvec = imu->bvec;
1922
1923 if (offset <= bvec->bv_len) {
1924 iov_iter_advance(iter, offset);
1925 } else {
1926 unsigned long seg_skip;
1927
1928 /* skip first vec */
1929 offset -= bvec->bv_len;
1930 seg_skip = 1 + (offset >> PAGE_SHIFT);
1931
1932 iter->bvec = bvec + seg_skip;
1933 iter->nr_segs -= seg_skip;
99c79f66 1934 iter->count -= bvec->bv_len + offset;
bd11b3a3 1935 iter->iov_offset = offset & ~PAGE_MASK;
bd11b3a3
JA		 1936 }
1937 }
1938
5e559561 1939 return len;
edafccee
JA		 1940}
1941
cf6fd4bd
PB		 1942static ssize_t io_import_iovec(int rw, struct io_kiocb *req,
1943 struct iovec **iovec, struct iov_iter *iter)
2b188cc1 1944{
9adbd45d
JA		 1945 void __user *buf = u64_to_user_ptr(req->rw.addr);
1946 size_t sqe_len = req->rw.len;
edafccee
JA		 1947 u8 opcode;
1948
d625c6ee 1949 opcode = req->opcode;
7d009165 1950 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
edafccee 1951 *iovec = NULL;
9adbd45d 1952 return io_import_fixed(req, rw, iter);
edafccee 1953 }
2b188cc1 1954
9adbd45d
JA		 1955 /* buffer index only valid with fixed read/write */
1956 if (req->rw.kiocb.private)
1957 return -EINVAL;
1958
3a6820f2
JA		 1959 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
1960 ssize_t ret;
1961 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
1962 *iovec = NULL;
1963 return ret;
1964 }
1965
f67676d1
JA		 1966 if (req->io) {
1967 struct io_async_rw *iorw = &req->io->rw;
1968
1969 *iovec = iorw->iov;
1970 iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size);
1971 if (iorw->iov == iorw->fast_iov)
1972 *iovec = NULL;
1973 return iorw->size;
1974 }
1975
cf6fd4bd 1976 if (!req->has_user)
2b188cc1
JA		 1977 return -EFAULT;
1978
1979#ifdef CONFIG_COMPAT
cf6fd4bd 1980 if (req->ctx->compat)
2b188cc1
JA		 1981 return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV,
1982 iovec, iter);
1983#endif
1984
1985 return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter);
1986}
1987
31b51510 1988/*
32960613
JA		 1989 * For files that don't have ->read_iter() and ->write_iter(), handle them
1990 * by looping over ->read() or ->write() manually.
31b51510 1991 */
32960613
JA		 1992static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb,
1993 struct iov_iter *iter)
1994{
1995 ssize_t ret = 0;
1996
1997 /*
1998 * Don't support polled IO through this interface, and we can't
1999 * support non-blocking either. For the latter, this just causes
2000 * the kiocb to be handled from an async context.
2001 */
2002 if (kiocb->ki_flags & IOCB_HIPRI)
2003 return -EOPNOTSUPP;
2004 if (kiocb->ki_flags & IOCB_NOWAIT)
2005 return -EAGAIN;
2006
2007 while (iov_iter_count(iter)) {
311ae9e1 2008 struct iovec iovec;
32960613
JA		 2009 ssize_t nr;
2010
311ae9e1
PB		 2011 if (!iov_iter_is_bvec(iter)) {
2012 iovec = iov_iter_iovec(iter);
2013 } else {
2014 /* fixed buffers import bvec */
2015 iovec.iov_base = kmap(iter->bvec->bv_page)
2016 + iter->iov_offset;
2017 iovec.iov_len = min(iter->count,
2018 iter->bvec->bv_len - iter->iov_offset);
2019 }
2020
32960613
JA		 2021 if (rw == READ) {
2022 nr = file->f_op->read(file, iovec.iov_base,
2023 iovec.iov_len, &kiocb->ki_pos);
2024 } else {
2025 nr = file->f_op->write(file, iovec.iov_base,
2026 iovec.iov_len, &kiocb->ki_pos);
2027 }
2028
311ae9e1
PB		 2029 if (iov_iter_is_bvec(iter))
2030 kunmap(iter->bvec->bv_page);
2031
32960613
JA		 2032 if (nr < 0) {
2033 if (!ret)
2034 ret = nr;
2035 break;
2036 }
2037 ret += nr;
2038 if (nr != iovec.iov_len)
2039 break;
2040 iov_iter_advance(iter, nr);
2041 }
2042
2043 return ret;
2044}
2045
b7bb4f7d 2046static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size,
f67676d1
JA		 2047 struct iovec *iovec, struct iovec *fast_iov,
2048 struct iov_iter *iter)
2049{
2050 req->io->rw.nr_segs = iter->nr_segs;
2051 req->io->rw.size = io_size;
2052 req->io->rw.iov = iovec;
2053 if (!req->io->rw.iov) {
2054 req->io->rw.iov = req->io->rw.fast_iov;
2055 memcpy(req->io->rw.iov, fast_iov,
2056 sizeof(struct iovec) * iter->nr_segs);
2057 }
2058}
2059
b7bb4f7d 2060static int io_alloc_async_ctx(struct io_kiocb *req)
f67676d1 2061{
d3656344
JA		 2062 if (!io_op_defs[req->opcode].async_ctx)
2063 return 0;
f67676d1 2064 req->io = kmalloc(sizeof(*req->io), GFP_KERNEL);
06b76d44 2065 return req->io == NULL;
b7bb4f7d
JA		 2066}
2067
2068static void io_rw_async(struct io_wq_work **workptr)
2069{
2070 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2071 struct iovec *iov = NULL;
2072
2073 if (req->io->rw.iov != req->io->rw.fast_iov)
2074 iov = req->io->rw.iov;
2075 io_wq_submit_work(workptr);
2076 kfree(iov);
2077}
2078
2079static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size,
2080 struct iovec *iovec, struct iovec *fast_iov,
2081 struct iov_iter *iter)
2082{
74566df3
JA		 2083 if (req->opcode == IORING_OP_READ_FIXED ||
2084 req->opcode == IORING_OP_WRITE_FIXED)
2085 return 0;
b7bb4f7d
JA		 2086 if (!req->io && io_alloc_async_ctx(req))
2087 return -ENOMEM;
2088
2089 io_req_map_rw(req, io_size, iovec, fast_iov, iter);
2090 req->work.func = io_rw_async;
2091 return 0;
f67676d1
JA		 2092}
2093
3529d8c2
JA		 2094static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
2095 bool force_nonblock)
f67676d1 2096{
3529d8c2
JA		 2097 struct io_async_ctx *io;
2098 struct iov_iter iter;
f67676d1
JA		 2099 ssize_t ret;
2100
3529d8c2
JA		 2101 ret = io_prep_rw(req, sqe, force_nonblock);
2102 if (ret)
2103 return ret;
f67676d1 2104
3529d8c2
JA		 2105 if (unlikely(!(req->file->f_mode & FMODE_READ)))
2106 return -EBADF;
f67676d1 2107
3529d8c2
JA		 2108 if (!req->io)
2109 return 0;
2110
2111 io = req->io;
2112 io->rw.iov = io->rw.fast_iov;
2113 req->io = NULL;
2114 ret = io_import_iovec(READ, req, &io->rw.iov, &iter);
2115 req->io = io;
2116 if (ret < 0)
2117 return ret;
2118
2119 io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter);
2120 return 0;
f67676d1
JA		 2121}
2122
267bc904 2123static int io_read(struct io_kiocb *req, struct io_kiocb **nxt,
8358e3a8 2124 bool force_nonblock)
2b188cc1
JA		 2125{
2126 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
9adbd45d 2127 struct kiocb *kiocb = &req->rw.kiocb;
2b188cc1 2128 struct iov_iter iter;
31b51510 2129 size_t iov_count;
f67676d1 2130 ssize_t io_size, ret;
2b188cc1 2131
3529d8c2 2132 ret = io_import_iovec(READ, req, &iovec, &iter);
06b76d44
JA		 2133 if (ret < 0)
2134 return ret;
2b188cc1 2135
fd6c2e4c
JA		 2136 /* Ensure we clear previously set non-block flag */
2137 if (!force_nonblock)
9adbd45d 2138 req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT;
fd6c2e4c 2139
797f3f53 2140 req->result = 0;
f67676d1 2141 io_size = ret;
9e645e11 2142 if (req->flags & REQ_F_LINK)
f67676d1
JA		 2143 req->result = io_size;
2144
2145 /*
2146 * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so
2147 * we know to async punt it even if it was opened O_NONBLOCK
2148 */
9adbd45d 2149 if (force_nonblock && !io_file_supports_async(req->file)) {
f67676d1
JA		 2150 req->flags |= REQ_F_MUST_PUNT;
2151 goto copy_iov;
2152 }
9e645e11 2153
31b51510 2154 iov_count = iov_iter_count(&iter);
9adbd45d 2155 ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count);
2b188cc1
JA		 2156 if (!ret) {
2157 ssize_t ret2;
2158
9adbd45d
JA		 2159 if (req->file->f_op->read_iter)
2160 ret2 = call_read_iter(req->file, kiocb, &iter);
32960613 2161 else
9adbd45d 2162 ret2 = loop_rw_iter(READ, req->file, kiocb, &iter);
32960613 2163
9d93a3f5 2164 /* Catch -EAGAIN return for forced non-blocking submission */
f67676d1 2165 if (!force_nonblock || ret2 != -EAGAIN) {
cf6fd4bd 2166 kiocb_done(kiocb, ret2, nxt, req->in_async);
f67676d1
JA		 2167 } else {
2168copy_iov:
b7bb4f7d 2169 ret = io_setup_async_rw(req, io_size, iovec,
f67676d1
JA		 2170 inline_vecs, &iter);
2171 if (ret)
2172 goto out_free;
2173 return -EAGAIN;
2174 }
2b188cc1 2175 }
f67676d1 2176out_free:
b7bb4f7d
JA		 2177 if (!io_wq_current_is_worker())
2178 kfree(iovec);
2b188cc1
JA		 2179 return ret;
2180}
2181
3529d8c2
JA		 2182static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
2183 bool force_nonblock)
f67676d1 2184{
3529d8c2
JA		 2185 struct io_async_ctx *io;
2186 struct iov_iter iter;
f67676d1
JA		 2187 ssize_t ret;
2188
3529d8c2
JA		 2189 ret = io_prep_rw(req, sqe, force_nonblock);
2190 if (ret)
2191 return ret;
f67676d1 2192
3529d8c2
JA		 2193 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
2194 return -EBADF;
f67676d1 2195
3529d8c2
JA		 2196 if (!req->io)
2197 return 0;
2198
2199 io = req->io;
2200 io->rw.iov = io->rw.fast_iov;
2201 req->io = NULL;
2202 ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter);
2203 req->io = io;
2204 if (ret < 0)
2205 return ret;
2206
2207 io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter);
2208 return 0;
f67676d1
JA		 2209}
2210
267bc904 2211static int io_write(struct io_kiocb *req, struct io_kiocb **nxt,
8358e3a8 2212 bool force_nonblock)
2b188cc1
JA		 2213{
2214 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
9adbd45d 2215 struct kiocb *kiocb = &req->rw.kiocb;
2b188cc1 2216 struct iov_iter iter;
31b51510 2217 size_t iov_count;
f67676d1 2218 ssize_t ret, io_size;
2b188cc1 2219
3529d8c2 2220 ret = io_import_iovec(WRITE, req, &iovec, &iter);
06b76d44
JA		 2221 if (ret < 0)
2222 return ret;
2b188cc1 2223
fd6c2e4c
JA		 2224 /* Ensure we clear previously set non-block flag */
2225 if (!force_nonblock)
9adbd45d 2226 req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT;
fd6c2e4c 2227
797f3f53 2228 req->result = 0;
f67676d1 2229 io_size = ret;
9e645e11 2230 if (req->flags & REQ_F_LINK)
f67676d1 2231 req->result = io_size;
9e645e11 2232
f67676d1
JA		 2233 /*
2234 * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so
2235 * we know to async punt it even if it was opened O_NONBLOCK
2236 */
2237 if (force_nonblock && !io_file_supports_async(req->file)) {
2238 req->flags |= REQ_F_MUST_PUNT;
2239 goto copy_iov;
2240 }
31b51510 2241
10d59345
JA		 2242 /* file path doesn't support NOWAIT for non-direct_IO */
2243 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
2244 (req->flags & REQ_F_ISREG))
f67676d1 2245 goto copy_iov;
31b51510 2246
f67676d1 2247 iov_count = iov_iter_count(&iter);
9adbd45d 2248 ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count);
2b188cc1 2249 if (!ret) {
9bf7933f
RP		 2250 ssize_t ret2;
2251
2b188cc1
JA		 2252 /*
2253 * Open-code file_start_write here to grab freeze protection,
2254 * which will be released by another thread in
2255 * io_complete_rw(). Fool lockdep by telling it the lock got
2256 * released so that it doesn't complain about the held lock when
2257 * we return to userspace.
2258 */
491381ce 2259 if (req->flags & REQ_F_ISREG) {
9adbd45d 2260 __sb_start_write(file_inode(req->file)->i_sb,
2b188cc1 2261 SB_FREEZE_WRITE, true);
9adbd45d 2262 __sb_writers_release(file_inode(req->file)->i_sb,
2b188cc1
JA		 2263 SB_FREEZE_WRITE);
2264 }
2265 kiocb->ki_flags |= IOCB_WRITE;
9bf7933f 2266
9adbd45d
JA		 2267 if (req->file->f_op->write_iter)
2268 ret2 = call_write_iter(req->file, kiocb, &iter);
32960613 2269 else
9adbd45d 2270 ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter);
f67676d1 2271 if (!force_nonblock || ret2 != -EAGAIN) {
cf6fd4bd 2272 kiocb_done(kiocb, ret2, nxt, req->in_async);
f67676d1
JA		 2273 } else {
2274copy_iov:
b7bb4f7d 2275 ret = io_setup_async_rw(req, io_size, iovec,
f67676d1
JA		 2276 inline_vecs, &iter);
2277 if (ret)
2278 goto out_free;
2279 return -EAGAIN;
2280 }
2b188cc1 2281 }
31b51510 2282out_free:
b7bb4f7d
JA		 2283 if (!io_wq_current_is_worker())
2284 kfree(iovec);
2b188cc1
JA		 2285 return ret;
2286}
2287
2288/*
2289 * IORING_OP_NOP just posts a completion event, nothing else.
2290 */
78e19bbe 2291static int io_nop(struct io_kiocb *req)
2b188cc1
JA		 2292{
2293 struct io_ring_ctx *ctx = req->ctx;
2b188cc1 2294
def596e9
JA		 2295 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
2296 return -EINVAL;
2297
78e19bbe 2298 io_cqring_add_event(req, 0);
e65ef56d 2299 io_put_req(req);
2b188cc1
JA		 2300 return 0;
2301}
2302
3529d8c2 2303static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe)
c992fe29 2304{
6b06314c 2305 struct io_ring_ctx *ctx = req->ctx;
c992fe29 2306
09bb8394
JA		 2307 if (!req->file)
2308 return -EBADF;
c992fe29 2309
6b06314c 2310 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
def596e9 2311 return -EINVAL;
edafccee 2312 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
c992fe29
CH		 2313 return -EINVAL;
2314
8ed8d3c3
JA		 2315 req->sync.flags = READ_ONCE(sqe->fsync_flags);
2316 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
2317 return -EINVAL;
2318
2319 req->sync.off = READ_ONCE(sqe->off);
2320 req->sync.len = READ_ONCE(sqe->len);
c992fe29
CH		 2321 return 0;
2322}
2323
8ed8d3c3
JA		 2324static bool io_req_cancelled(struct io_kiocb *req)
2325{
2326 if (req->work.flags & IO_WQ_WORK_CANCEL) {
2327 req_set_fail_links(req);
2328 io_cqring_add_event(req, -ECANCELED);
2329 io_put_req(req);
2330 return true;
2331 }
2332
2333 return false;
2334}
2335
78912934
JA		 2336static void io_link_work_cb(struct io_wq_work **workptr)
2337{
2338 struct io_wq_work *work = *workptr;
2339 struct io_kiocb *link = work->data;
2340
2341 io_queue_linked_timeout(link);
2342 work->func = io_wq_submit_work;
2343}
2344
2345static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt)
2346{
2347 struct io_kiocb *link;
2348
2349 io_prep_async_work(nxt, &link);
2350 *workptr = &nxt->work;
2351 if (link) {
2352 nxt->work.flags |= IO_WQ_WORK_CB;
2353 nxt->work.func = io_link_work_cb;
2354 nxt->work.data = link;
2355 }
2356}
2357
8ed8d3c3
JA		 2358static void io_fsync_finish(struct io_wq_work **workptr)
2359{
2360 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2361 loff_t end = req->sync.off + req->sync.len;
2362 struct io_kiocb *nxt = NULL;
2363 int ret;
2364
2365 if (io_req_cancelled(req))
2366 return;
2367
9adbd45d 2368 ret = vfs_fsync_range(req->file, req->sync.off,
8ed8d3c3
JA		 2369 end > 0 ? end : LLONG_MAX,
2370 req->sync.flags & IORING_FSYNC_DATASYNC);
2371 if (ret < 0)
2372 req_set_fail_links(req);
2373 io_cqring_add_event(req, ret);
2374 io_put_req_find_next(req, &nxt);
2375 if (nxt)
78912934 2376 io_wq_assign_next(workptr, nxt);
8ed8d3c3
JA		 2377}
2378
fc4df999
JA		 2379static int io_fsync(struct io_kiocb *req, struct io_kiocb **nxt,
2380 bool force_nonblock)
c992fe29 2381{
8ed8d3c3 2382 struct io_wq_work *work, *old_work;
c992fe29
CH		 2383
2384 /* fsync always requires a blocking context */
8ed8d3c3
JA		 2385 if (force_nonblock) {
2386 io_put_req(req);
2387 req->work.func = io_fsync_finish;
c992fe29 2388 return -EAGAIN;
8ed8d3c3 2389 }
c992fe29 2390
8ed8d3c3
JA		 2391 work = old_work = &req->work;
2392 io_fsync_finish(&work);
2393 if (work && work != old_work)
2394 *nxt = container_of(work, struct io_kiocb, work);
c992fe29
CH		 2395 return 0;
2396}
2397
d63d1b5e
JA		 2398static void io_fallocate_finish(struct io_wq_work **workptr)
2399{
2400 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2401 struct io_kiocb *nxt = NULL;
2402 int ret;
2403
2404 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
2405 req->sync.len);
2406 if (ret < 0)
2407 req_set_fail_links(req);
2408 io_cqring_add_event(req, ret);
2409 io_put_req_find_next(req, &nxt);
2410 if (nxt)
2411 io_wq_assign_next(workptr, nxt);
2412}
2413
2414static int io_fallocate_prep(struct io_kiocb *req,
2415 const struct io_uring_sqe *sqe)
2416{
2417 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
2418 return -EINVAL;
2419
2420 req->sync.off = READ_ONCE(sqe->off);
2421 req->sync.len = READ_ONCE(sqe->addr);
2422 req->sync.mode = READ_ONCE(sqe->len);
2423 return 0;
2424}
2425
2426static int io_fallocate(struct io_kiocb *req, struct io_kiocb **nxt,
2427 bool force_nonblock)
2428{
2429 struct io_wq_work *work, *old_work;
2430
2431 /* fallocate always requiring blocking context */
2432 if (force_nonblock) {
2433 io_put_req(req);
2434 req->work.func = io_fallocate_finish;
2435 return -EAGAIN;
2436 }
2437
2438 work = old_work = &req->work;
2439 io_fallocate_finish(&work);
2440 if (work && work != old_work)
2441 *nxt = container_of(work, struct io_kiocb, work);
2442
2443 return 0;
2444}
2445
15b71abe
JA		 2446static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2447{
2448 int ret;
2449
2450 if (sqe->ioprio || sqe->buf_index)
2451 return -EINVAL;
2452
2453 req->open.dfd = READ_ONCE(sqe->fd);
2454 req->open.mode = READ_ONCE(sqe->len);
2455 req->open.fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
2456 req->open.flags = READ_ONCE(sqe->open_flags);
2457
2458 req->open.filename = getname(req->open.fname);
2459 if (IS_ERR(req->open.filename)) {
2460 ret = PTR_ERR(req->open.filename);
2461 req->open.filename = NULL;
2462 return ret;
2463 }
2464
2465 return 0;
2466}
2467
2468static int io_openat(struct io_kiocb *req, struct io_kiocb **nxt,
2469 bool force_nonblock)
2470{
2471 struct open_flags op;
2472 struct open_how how;
2473 struct file *file;
2474 int ret;
2475
2476 if (force_nonblock) {
2477 req->work.flags |= IO_WQ_WORK_NEEDS_FILES;
2478 return -EAGAIN;
2479 }
2480
2481 how = build_open_how(req->open.flags, req->open.mode);
2482 ret = build_open_flags(&how, &op);
2483 if (ret)
2484 goto err;
2485
2486 ret = get_unused_fd_flags(how.flags);
2487 if (ret < 0)
2488 goto err;
2489
2490 file = do_filp_open(req->open.dfd, req->open.filename, &op);
2491 if (IS_ERR(file)) {
2492 put_unused_fd(ret);
2493 ret = PTR_ERR(file);
2494 } else {
2495 fsnotify_open(file);
2496 fd_install(ret, file);
2497 }
2498err:
2499 putname(req->open.filename);
2500 if (ret < 0)
2501 req_set_fail_links(req);
2502 io_cqring_add_event(req, ret);
2503 io_put_req_find_next(req, nxt);
2504 return 0;
2505}
2506
c1ca757b
JA		 2507static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2508{
2509#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
2510 if (sqe->ioprio || sqe->buf_index || sqe->off)
2511 return -EINVAL;
2512
2513 req->madvise.addr = READ_ONCE(sqe->addr);
2514 req->madvise.len = READ_ONCE(sqe->len);
2515 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
2516 return 0;
2517#else
2518 return -EOPNOTSUPP;
2519#endif
2520}
2521
2522static int io_madvise(struct io_kiocb *req, struct io_kiocb **nxt,
2523 bool force_nonblock)
2524{
2525#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
2526 struct io_madvise *ma = &req->madvise;
2527 int ret;
2528
2529 if (force_nonblock)
2530 return -EAGAIN;
2531
2532 ret = do_madvise(ma->addr, ma->len, ma->advice);
2533 if (ret < 0)
2534 req_set_fail_links(req);
2535 io_cqring_add_event(req, ret);
2536 io_put_req_find_next(req, nxt);
2537 return 0;
2538#else
2539 return -EOPNOTSUPP;
2540#endif
2541}
2542
4840e418
JA		 2543static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2544{
2545 if (sqe->ioprio || sqe->buf_index || sqe->addr)
2546 return -EINVAL;
2547
2548 req->fadvise.offset = READ_ONCE(sqe->off);
2549 req->fadvise.len = READ_ONCE(sqe->len);
2550 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
2551 return 0;
2552}
2553
2554static int io_fadvise(struct io_kiocb *req, struct io_kiocb **nxt,
2555 bool force_nonblock)
2556{
2557 struct io_fadvise *fa = &req->fadvise;
2558 int ret;
2559
2560 /* DONTNEED may block, others _should_ not */
2561 if (fa->advice == POSIX_FADV_DONTNEED && force_nonblock)
2562 return -EAGAIN;
2563
2564 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
2565 if (ret < 0)
2566 req_set_fail_links(req);
2567 io_cqring_add_event(req, ret);
2568 io_put_req_find_next(req, nxt);
2569 return 0;
2570}
2571
eddc7ef5
JA		 2572static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2573{
2574 unsigned lookup_flags;
2575 int ret;
2576
2577 if (sqe->ioprio || sqe->buf_index)
2578 return -EINVAL;
2579
2580 req->open.dfd = READ_ONCE(sqe->fd);
2581 req->open.mask = READ_ONCE(sqe->len);
2582 req->open.fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
2583 req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
2584 req->open.flags = READ_ONCE(sqe->statx_flags);
2585
2586 if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.flags))
2587 return -EINVAL;
2588
2589 req->open.filename = getname_flags(req->open.fname, lookup_flags, NULL);
2590 if (IS_ERR(req->open.filename)) {
2591 ret = PTR_ERR(req->open.filename);
2592 req->open.filename = NULL;
2593 return ret;
2594 }
2595
2596 return 0;
2597}
2598
2599static int io_statx(struct io_kiocb *req, struct io_kiocb **nxt,
2600 bool force_nonblock)
2601{
2602 struct io_open *ctx = &req->open;
2603 unsigned lookup_flags;
2604 struct path path;
2605 struct kstat stat;
2606 int ret;
2607
2608 if (force_nonblock)
2609 return -EAGAIN;
2610
2611 if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->flags))
2612 return -EINVAL;
2613
2614retry:
2615 /* filename_lookup() drops it, keep a reference */
2616 ctx->filename->refcnt++;
2617
2618 ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path,
2619 NULL);
2620 if (ret)
2621 goto err;
2622
2623 ret = vfs_getattr(&path, &stat, ctx->mask, ctx->flags);
2624 path_put(&path);
2625 if (retry_estale(ret, lookup_flags)) {
2626 lookup_flags |= LOOKUP_REVAL;
2627 goto retry;
2628 }
2629 if (!ret)
2630 ret = cp_statx(&stat, ctx->buffer);
2631err:
2632 putname(ctx->filename);
2633 if (ret < 0)
2634 req_set_fail_links(req);
2635 io_cqring_add_event(req, ret);
2636 io_put_req_find_next(req, nxt);
2637 return 0;
2638}
2639
b5dba59e
JA		 2640static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2641{
2642 /*
2643 * If we queue this for async, it must not be cancellable. That would
2644 * leave the 'file' in an undeterminate state.
2645 */
2646 req->work.flags |= IO_WQ_WORK_NO_CANCEL;
2647
2648 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
2649 sqe->rw_flags || sqe->buf_index)
2650 return -EINVAL;
2651 if (sqe->flags & IOSQE_FIXED_FILE)
2652 return -EINVAL;
2653
2654 req->close.fd = READ_ONCE(sqe->fd);
2655 if (req->file->f_op == &io_uring_fops ||
2656 req->close.fd == req->ring_fd)
2657 return -EBADF;
2658
2659 return 0;
2660}
2661
2662static void io_close_finish(struct io_wq_work **workptr)
2663{
2664 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2665 struct io_kiocb *nxt = NULL;
2666
2667 /* Invoked with files, we need to do the close */
2668 if (req->work.files) {
2669 int ret;
2670
2671 ret = filp_close(req->close.put_file, req->work.files);
2672 if (ret < 0) {
2673 req_set_fail_links(req);
2674 }
2675 io_cqring_add_event(req, ret);
2676 }
2677
2678 fput(req->close.put_file);
2679
2680 /* we bypassed the re-issue, drop the submission reference */
2681 io_put_req(req);
2682 io_put_req_find_next(req, &nxt);
2683 if (nxt)
2684 io_wq_assign_next(workptr, nxt);
2685}
2686
2687static int io_close(struct io_kiocb *req, struct io_kiocb **nxt,
2688 bool force_nonblock)
2689{
2690 int ret;
2691
2692 req->close.put_file = NULL;
2693 ret = __close_fd_get_file(req->close.fd, &req->close.put_file);
2694 if (ret < 0)
2695 return ret;
2696
2697 /* if the file has a flush method, be safe and punt to async */
2698 if (req->close.put_file->f_op->flush && !io_wq_current_is_worker()) {
2699 req->work.flags |= IO_WQ_WORK_NEEDS_FILES;
2700 goto eagain;
2701 }
2702
2703 /*
2704 * No ->flush(), safely close from here and just punt the
2705 * fput() to async context.
2706 */
2707 ret = filp_close(req->close.put_file, current->files);
2708
2709 if (ret < 0)
2710 req_set_fail_links(req);
2711 io_cqring_add_event(req, ret);
2712
2713 if (io_wq_current_is_worker()) {
2714 struct io_wq_work *old_work, *work;
2715
2716 old_work = work = &req->work;
2717 io_close_finish(&work);
2718 if (work && work != old_work)
2719 *nxt = container_of(work, struct io_kiocb, work);
2720 return 0;
2721 }
2722
2723eagain:
2724 req->work.func = io_close_finish;
2725 return -EAGAIN;
2726}
2727
3529d8c2 2728static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe)
5d17b4a4
JA		 2729{
2730 struct io_ring_ctx *ctx = req->ctx;
5d17b4a4
JA		 2731
2732 if (!req->file)
2733 return -EBADF;
5d17b4a4
JA		 2734
2735 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
2736 return -EINVAL;
2737 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
2738 return -EINVAL;
2739
8ed8d3c3
JA		 2740 req->sync.off = READ_ONCE(sqe->off);
2741 req->sync.len = READ_ONCE(sqe->len);
2742 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
8ed8d3c3
JA		 2743 return 0;
2744}
2745
2746static void io_sync_file_range_finish(struct io_wq_work **workptr)
2747{
2748 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2749 struct io_kiocb *nxt = NULL;
2750 int ret;
2751
2752 if (io_req_cancelled(req))
2753 return;
2754
9adbd45d 2755 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
8ed8d3c3
JA		 2756 req->sync.flags);
2757 if (ret < 0)
2758 req_set_fail_links(req);
2759 io_cqring_add_event(req, ret);
2760 io_put_req_find_next(req, &nxt);
2761 if (nxt)
78912934 2762 io_wq_assign_next(workptr, nxt);
5d17b4a4
JA		 2763}
2764
fc4df999 2765static int io_sync_file_range(struct io_kiocb *req, struct io_kiocb **nxt,
5d17b4a4
JA		 2766 bool force_nonblock)
2767{
8ed8d3c3 2768 struct io_wq_work *work, *old_work;
5d17b4a4
JA		 2769
2770 /* sync_file_range always requires a blocking context */
8ed8d3c3
JA		 2771 if (force_nonblock) {
2772 io_put_req(req);
2773 req->work.func = io_sync_file_range_finish;
5d17b4a4 2774 return -EAGAIN;
8ed8d3c3 2775 }
5d17b4a4 2776
8ed8d3c3
JA		 2777 work = old_work = &req->work;
2778 io_sync_file_range_finish(&work);
2779 if (work && work != old_work)
2780 *nxt = container_of(work, struct io_kiocb, work);
5d17b4a4
JA		 2781 return 0;
2782}
2783
b7bb4f7d
JA		 2784#if defined(CONFIG_NET)
2785static void io_sendrecv_async(struct io_wq_work **workptr)
2786{
2787 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2788 struct iovec *iov = NULL;
2789
2790 if (req->io->rw.iov != req->io->rw.fast_iov)
2791 iov = req->io->msg.iov;
2792 io_wq_submit_work(workptr);
2793 kfree(iov);
2794}
2795#endif
2796
3529d8c2 2797static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
03b1230c 2798{
0fa03c62 2799#if defined(CONFIG_NET)
e47293fd 2800 struct io_sr_msg *sr = &req->sr_msg;
3529d8c2 2801 struct io_async_ctx *io = req->io;
03b1230c 2802
e47293fd
JA		 2803 sr->msg_flags = READ_ONCE(sqe->msg_flags);
2804 sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr));
3529d8c2
JA		 2805
2806 if (!io)
2807 return 0;
2808
d9688565 2809 io->msg.iov = io->msg.fast_iov;
3529d8c2 2810 return sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags,
e47293fd 2811 &io->msg.iov);
03b1230c 2812#else
e47293fd 2813 return -EOPNOTSUPP;
03b1230c
JA		 2814#endif
2815}
2816
fc4df999
JA		 2817static int io_sendmsg(struct io_kiocb *req, struct io_kiocb **nxt,
2818 bool force_nonblock)
aa1fa28f 2819{
03b1230c 2820#if defined(CONFIG_NET)
0b416c3e 2821 struct io_async_msghdr *kmsg = NULL;
0fa03c62
JA		 2822 struct socket *sock;
2823 int ret;
2824
2825 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
2826 return -EINVAL;
2827
2828 sock = sock_from_file(req->file, &ret);
2829 if (sock) {
b7bb4f7d 2830 struct io_async_ctx io;
03b1230c 2831 struct sockaddr_storage addr;
0fa03c62
JA		 2832 unsigned flags;
2833
03b1230c 2834 if (req->io) {
0b416c3e
JA		 2835 kmsg = &req->io->msg;
2836 kmsg->msg.msg_name = &addr;
2837 /* if iov is set, it's allocated already */
2838 if (!kmsg->iov)
2839 kmsg->iov = kmsg->fast_iov;
2840 kmsg->msg.msg_iter.iov = kmsg->iov;
03b1230c 2841 } else {
3529d8c2
JA		 2842 struct io_sr_msg *sr = &req->sr_msg;
2843
0b416c3e
JA		 2844 kmsg = &io.msg;
2845 kmsg->msg.msg_name = &addr;
3529d8c2
JA		 2846
2847 io.msg.iov = io.msg.fast_iov;
2848 ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg,
2849 sr->msg_flags, &io.msg.iov);
03b1230c 2850 if (ret)
3529d8c2 2851 return ret;
03b1230c 2852 }
0fa03c62 2853
e47293fd
JA		 2854 flags = req->sr_msg.msg_flags;
2855 if (flags & MSG_DONTWAIT)
2856 req->flags |= REQ_F_NOWAIT;
2857 else if (force_nonblock)
2858 flags |= MSG_DONTWAIT;
2859
0b416c3e 2860 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
03b1230c 2861 if (force_nonblock && ret == -EAGAIN) {
b7bb4f7d
JA		 2862 if (req->io)
2863 return -EAGAIN;
2864 if (io_alloc_async_ctx(req))
2865 return -ENOMEM;
2866 memcpy(&req->io->msg, &io.msg, sizeof(io.msg));
2867 req->work.func = io_sendrecv_async;
0b416c3e 2868 return -EAGAIN;
03b1230c 2869 }
441cdbd5
JA		 2870 if (ret == -ERESTARTSYS)
2871 ret = -EINTR;
0fa03c62
JA		 2872 }
2873
b7bb4f7d 2874 if (!io_wq_current_is_worker() && kmsg && kmsg->iov != kmsg->fast_iov)
0b416c3e 2875 kfree(kmsg->iov);
78e19bbe 2876 io_cqring_add_event(req, ret);
4e88d6e7
JA		 2877 if (ret < 0)
2878 req_set_fail_links(req);
ec9c02ad 2879 io_put_req_find_next(req, nxt);
5d17b4a4 2880 return 0;
03b1230c
JA		 2881#else
2882 return -EOPNOTSUPP;
aa1fa28f 2883#endif
03b1230c 2884}
aa1fa28f 2885
3529d8c2
JA		 2886static int io_recvmsg_prep(struct io_kiocb *req,
2887 const struct io_uring_sqe *sqe)
aa1fa28f
JA		 2888{
2889#if defined(CONFIG_NET)
e47293fd 2890 struct io_sr_msg *sr = &req->sr_msg;
3529d8c2
JA		 2891 struct io_async_ctx *io = req->io;
2892
2893 sr->msg_flags = READ_ONCE(sqe->msg_flags);
2894 sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr));
06b76d44 2895
3529d8c2 2896 if (!io)
06b76d44 2897 return 0;
03b1230c 2898
d9688565 2899 io->msg.iov = io->msg.fast_iov;
3529d8c2 2900 return recvmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags,
e47293fd 2901 &io->msg.uaddr, &io->msg.iov);
aa1fa28f 2902#else
e47293fd 2903 return -EOPNOTSUPP;
aa1fa28f
JA		 2904#endif
2905}
2906
fc4df999
JA		 2907static int io_recvmsg(struct io_kiocb *req, struct io_kiocb **nxt,
2908 bool force_nonblock)
aa1fa28f
JA		 2909{
2910#if defined(CONFIG_NET)
0b416c3e 2911 struct io_async_msghdr *kmsg = NULL;
03b1230c
JA		 2912 struct socket *sock;
2913 int ret;
2914
2915 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
2916 return -EINVAL;
2917
2918 sock = sock_from_file(req->file, &ret);
2919 if (sock) {
b7bb4f7d 2920 struct io_async_ctx io;
03b1230c 2921 struct sockaddr_storage addr;
03b1230c
JA		 2922 unsigned flags;
2923
03b1230c 2924 if (req->io) {
0b416c3e
JA		 2925 kmsg = &req->io->msg;
2926 kmsg->msg.msg_name = &addr;
2927 /* if iov is set, it's allocated already */
2928 if (!kmsg->iov)
2929 kmsg->iov = kmsg->fast_iov;
2930 kmsg->msg.msg_iter.iov = kmsg->iov;
03b1230c 2931 } else {
3529d8c2
JA		 2932 struct io_sr_msg *sr = &req->sr_msg;
2933
0b416c3e
JA		 2934 kmsg = &io.msg;
2935 kmsg->msg.msg_name = &addr;
3529d8c2
JA		 2936
2937 io.msg.iov = io.msg.fast_iov;
2938 ret = recvmsg_copy_msghdr(&io.msg.msg, sr->msg,
2939 sr->msg_flags, &io.msg.uaddr,
2940 &io.msg.iov);
03b1230c 2941 if (ret)
3529d8c2 2942 return ret;
03b1230c
JA		 2943 }
2944
e47293fd
JA		 2945 flags = req->sr_msg.msg_flags;
2946 if (flags & MSG_DONTWAIT)
2947 req->flags |= REQ_F_NOWAIT;
2948 else if (force_nonblock)
2949 flags |= MSG_DONTWAIT;
2950
2951 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg,
2952 kmsg->uaddr, flags);
03b1230c 2953 if (force_nonblock && ret == -EAGAIN) {
b7bb4f7d
JA		 2954 if (req->io)
2955 return -EAGAIN;
2956 if (io_alloc_async_ctx(req))
2957 return -ENOMEM;
2958 memcpy(&req->io->msg, &io.msg, sizeof(io.msg));
2959 req->work.func = io_sendrecv_async;
0b416c3e 2960 return -EAGAIN;
03b1230c
JA		 2961 }
2962 if (ret == -ERESTARTSYS)
2963 ret = -EINTR;
2964 }
2965
b7bb4f7d 2966 if (!io_wq_current_is_worker() && kmsg && kmsg->iov != kmsg->fast_iov)
0b416c3e 2967 kfree(kmsg->iov);
03b1230c 2968 io_cqring_add_event(req, ret);
4e88d6e7
JA		 2969 if (ret < 0)
2970 req_set_fail_links(req);
03b1230c
JA		 2971 io_put_req_find_next(req, nxt);
2972 return 0;
0fa03c62
JA		 2973#else
2974 return -EOPNOTSUPP;
2975#endif
2976}
5d17b4a4 2977
3529d8c2 2978static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
17f2fe35
JA		 2979{
2980#if defined(CONFIG_NET)
8ed8d3c3
JA		 2981 struct io_accept *accept = &req->accept;
2982
17f2fe35
JA		 2983 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
2984 return -EINVAL;
8042d6ce 2985 if (sqe->ioprio || sqe->len || sqe->buf_index)
17f2fe35
JA		 2986 return -EINVAL;
2987
d55e5f5b
JA		 2988 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
2989 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
8ed8d3c3 2990 accept->flags = READ_ONCE(sqe->accept_flags);
8ed8d3c3
JA		 2991 return 0;
2992#else
2993 return -EOPNOTSUPP;
2994#endif
2995}
17f2fe35 2996
8ed8d3c3
JA		 2997#if defined(CONFIG_NET)
2998static int __io_accept(struct io_kiocb *req, struct io_kiocb **nxt,
2999 bool force_nonblock)
3000{
3001 struct io_accept *accept = &req->accept;
3002 unsigned file_flags;
3003 int ret;
3004
3005 file_flags = force_nonblock ? O_NONBLOCK : 0;
3006 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
3007 accept->addr_len, accept->flags);
3008 if (ret == -EAGAIN && force_nonblock)
17f2fe35 3009 return -EAGAIN;
8e3cca12
JA		 3010 if (ret == -ERESTARTSYS)
3011 ret = -EINTR;
4e88d6e7
JA		 3012 if (ret < 0)
3013 req_set_fail_links(req);
78e19bbe 3014 io_cqring_add_event(req, ret);
ec9c02ad 3015 io_put_req_find_next(req, nxt);
17f2fe35 3016 return 0;
8ed8d3c3
JA		 3017}
3018
3019static void io_accept_finish(struct io_wq_work **workptr)
3020{
3021 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
3022 struct io_kiocb *nxt = NULL;
3023
3024 if (io_req_cancelled(req))
3025 return;
3026 __io_accept(req, &nxt, false);
3027 if (nxt)
78912934 3028 io_wq_assign_next(workptr, nxt);
8ed8d3c3
JA		 3029}
3030#endif
3031
3032static int io_accept(struct io_kiocb *req, struct io_kiocb **nxt,
3033 bool force_nonblock)
3034{
3035#if defined(CONFIG_NET)
3036 int ret;
3037
8ed8d3c3
JA		 3038 ret = __io_accept(req, nxt, force_nonblock);
3039 if (ret == -EAGAIN && force_nonblock) {
3040 req->work.func = io_accept_finish;
3041 req->work.flags |= IO_WQ_WORK_NEEDS_FILES;
3042 io_put_req(req);
3043 return -EAGAIN;
3044 }
3045 return 0;
0fa03c62
JA		 3046#else
3047 return -EOPNOTSUPP;
3048#endif
3049}
5d17b4a4 3050
3529d8c2 3051static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
f499a021
JA		 3052{
3053#if defined(CONFIG_NET)
3529d8c2
JA		 3054 struct io_connect *conn = &req->connect;
3055 struct io_async_ctx *io = req->io;
f499a021 3056
3fbb51c1
JA		 3057 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
3058 return -EINVAL;
3059 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
3060 return -EINVAL;
3061
3529d8c2
JA		 3062 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
3063 conn->addr_len = READ_ONCE(sqe->addr2);
3064
3065 if (!io)
3066 return 0;
3067
3068 return move_addr_to_kernel(conn->addr, conn->addr_len,
3fbb51c1 3069 &io->connect.address);
f499a021 3070#else
3fbb51c1 3071 return -EOPNOTSUPP;
f499a021
JA		 3072#endif
3073}
3074
fc4df999
JA		 3075static int io_connect(struct io_kiocb *req, struct io_kiocb **nxt,
3076 bool force_nonblock)
f8e85cf2
JA		 3077{
3078#if defined(CONFIG_NET)
f499a021 3079 struct io_async_ctx __io, *io;
f8e85cf2 3080 unsigned file_flags;
3fbb51c1 3081 int ret;
f8e85cf2 3082
f499a021
JA		 3083 if (req->io) {
3084 io = req->io;
3085 } else {
3529d8c2
JA		 3086 ret = move_addr_to_kernel(req->connect.addr,
3087 req->connect.addr_len,
3088 &__io.connect.address);
f499a021
JA		 3089 if (ret)
3090 goto out;
3091 io = &__io;
3092 }
3093
3fbb51c1
JA		 3094 file_flags = force_nonblock ? O_NONBLOCK : 0;
3095
3096 ret = __sys_connect_file(req->file, &io->connect.address,
3097 req->connect.addr_len, file_flags);
87f80d62 3098 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
b7bb4f7d
JA		 3099 if (req->io)
3100 return -EAGAIN;
3101 if (io_alloc_async_ctx(req)) {
f499a021
JA		 3102 ret = -ENOMEM;
3103 goto out;
3104 }
b7bb4f7d 3105 memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect));
f8e85cf2 3106 return -EAGAIN;
f499a021 3107 }
f8e85cf2
JA		 3108 if (ret == -ERESTARTSYS)
3109 ret = -EINTR;
f499a021 3110out:
4e88d6e7
JA		 3111 if (ret < 0)
3112 req_set_fail_links(req);
f8e85cf2
JA		 3113 io_cqring_add_event(req, ret);
3114 io_put_req_find_next(req, nxt);
3115 return 0;
3116#else
3117 return -EOPNOTSUPP;
3118#endif
3119}
3120
221c5eb2
JA		 3121static void io_poll_remove_one(struct io_kiocb *req)
3122{
3123 struct io_poll_iocb *poll = &req->poll;
3124
3125 spin_lock(&poll->head->lock);
3126 WRITE_ONCE(poll->canceled, true);
392edb45
JA		 3127 if (!list_empty(&poll->wait.entry)) {
3128 list_del_init(&poll->wait.entry);
a197f664 3129 io_queue_async_work(req);
221c5eb2
JA		 3130 }
3131 spin_unlock(&poll->head->lock);
78076bb6 3132 hash_del(&req->hash_node);
221c5eb2
JA		 3133}
3134
3135static void io_poll_remove_all(struct io_ring_ctx *ctx)
3136{
78076bb6 3137 struct hlist_node *tmp;
221c5eb2 3138 struct io_kiocb *req;
78076bb6 3139 int i;
221c5eb2
JA		 3140
3141 spin_lock_irq(&ctx->completion_lock);
78076bb6
JA		 3142 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
3143 struct hlist_head *list;
3144
3145 list = &ctx->cancel_hash[i];
3146 hlist_for_each_entry_safe(req, tmp, list, hash_node)
3147 io_poll_remove_one(req);
221c5eb2
JA		 3148 }
3149 spin_unlock_irq(&ctx->completion_lock);
3150}
3151
47f46768
JA		 3152static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
3153{
78076bb6 3154 struct hlist_head *list;
47f46768
JA		 3155 struct io_kiocb *req;
3156
78076bb6
JA		 3157 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
3158 hlist_for_each_entry(req, list, hash_node) {
3159 if (sqe_addr == req->user_data) {
eac406c6
JA		 3160 io_poll_remove_one(req);
3161 return 0;
3162 }
47f46768
JA		 3163 }
3164
3165 return -ENOENT;
3166}
3167
3529d8c2
JA		 3168static int io_poll_remove_prep(struct io_kiocb *req,
3169 const struct io_uring_sqe *sqe)
0969e783 3170{
0969e783
JA		 3171 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3172 return -EINVAL;
3173 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
3174 sqe->poll_events)
3175 return -EINVAL;
3176
3177 req->poll.addr = READ_ONCE(sqe->addr);
0969e783
JA		 3178 return 0;
3179}
3180
221c5eb2
JA		 3181/*
3182 * Find a running poll command that matches one specified in sqe->addr,
3183 * and remove it if found.
3184 */
fc4df999 3185static int io_poll_remove(struct io_kiocb *req)
221c5eb2
JA		 3186{
3187 struct io_ring_ctx *ctx = req->ctx;
0969e783 3188 u64 addr;
47f46768 3189 int ret;
221c5eb2 3190
0969e783 3191 addr = req->poll.addr;
221c5eb2 3192 spin_lock_irq(&ctx->completion_lock);
0969e783 3193 ret = io_poll_cancel(ctx, addr);
221c5eb2
JA		 3194 spin_unlock_irq(&ctx->completion_lock);
3195
78e19bbe 3196 io_cqring_add_event(req, ret);
4e88d6e7
JA		 3197 if (ret < 0)
3198 req_set_fail_links(req);
e65ef56d 3199 io_put_req(req);
221c5eb2
JA		 3200 return 0;
3201}
3202
b0dd8a41 3203static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
221c5eb2 3204{
a197f664
JL		 3205 struct io_ring_ctx *ctx = req->ctx;
3206
8c838788 3207 req->poll.done = true;
b0dd8a41
JA		 3208 if (error)
3209 io_cqring_fill_event(req, error);
3210 else
3211 io_cqring_fill_event(req, mangle_poll(mask));
8c838788 3212 io_commit_cqring(ctx);
221c5eb2
JA		 3213}
3214
561fb04a 3215static void io_poll_complete_work(struct io_wq_work **workptr)
221c5eb2 3216{
561fb04a 3217 struct io_wq_work *work = *workptr;
221c5eb2
JA		 3218 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
3219 struct io_poll_iocb *poll = &req->poll;
3220 struct poll_table_struct pt = { ._key = poll->events };
3221 struct io_ring_ctx *ctx = req->ctx;
89723d0b 3222 struct io_kiocb *nxt = NULL;
221c5eb2 3223 __poll_t mask = 0;
b0dd8a41 3224 int ret = 0;
221c5eb2 3225
b0dd8a41 3226 if (work->flags & IO_WQ_WORK_CANCEL) {
561fb04a 3227 WRITE_ONCE(poll->canceled, true);
b0dd8a41
JA		 3228 ret = -ECANCELED;
3229 } else if (READ_ONCE(poll->canceled)) {
3230 ret = -ECANCELED;
3231 }
561fb04a 3232
b0dd8a41 3233 if (ret != -ECANCELED)
221c5eb2
JA		 3234 mask = vfs_poll(poll->file, &pt) & poll->events;
3235
3236 /*
3237 * Note that ->ki_cancel callers also delete iocb from active_reqs after
3238 * calling ->ki_cancel. We need the ctx_lock roundtrip here to
3239 * synchronize with them. In the cancellation case the list_del_init
3240 * itself is not actually needed, but harmless so we keep it in to
3241 * avoid further branches in the fast path.
3242 */
3243 spin_lock_irq(&ctx->completion_lock);
b0dd8a41 3244 if (!mask && ret != -ECANCELED) {
392edb45 3245 add_wait_queue(poll->head, &poll->wait);
221c5eb2
JA		 3246 spin_unlock_irq(&ctx->completion_lock);
3247 return;
3248 }
78076bb6 3249 hash_del(&req->hash_node);
b0dd8a41 3250 io_poll_complete(req, mask, ret);
221c5eb2
JA		 3251 spin_unlock_irq(&ctx->completion_lock);
3252
8c838788 3253 io_cqring_ev_posted(ctx);
89723d0b 3254
4e88d6e7
JA		 3255 if (ret < 0)
3256 req_set_fail_links(req);
ec9c02ad 3257 io_put_req_find_next(req, &nxt);
89723d0b 3258 if (nxt)
78912934 3259 io_wq_assign_next(workptr, nxt);
221c5eb2
JA		 3260}
3261
e94f141b
JA		 3262static void __io_poll_flush(struct io_ring_ctx *ctx, struct llist_node *nodes)
3263{
e94f141b 3264 struct io_kiocb *req, *tmp;
8237e045 3265 struct req_batch rb;
e94f141b 3266
c6ca97b3 3267 rb.to_free = rb.need_iter = 0;
e94f141b
JA		 3268 spin_lock_irq(&ctx->completion_lock);
3269 llist_for_each_entry_safe(req, tmp, nodes, llist_node) {
3270 hash_del(&req->hash_node);
3271 io_poll_complete(req, req->result, 0);
3272
8237e045
JA		 3273 if (refcount_dec_and_test(&req->refs) &&
3274 !io_req_multi_free(&rb, req)) {
3275 req->flags |= REQ_F_COMP_LOCKED;
3276 io_free_req(req);
e94f141b
JA		 3277 }
3278 }
3279 spin_unlock_irq(&ctx->completion_lock);
3280
3281 io_cqring_ev_posted(ctx);
8237e045 3282 io_free_req_many(ctx, &rb);
e94f141b
JA		 3283}
3284
3285static void io_poll_flush(struct io_wq_work **workptr)
3286{
3287 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
3288 struct llist_node *nodes;
3289
3290 nodes = llist_del_all(&req->ctx->poll_llist);
3291 if (nodes)
3292 __io_poll_flush(req->ctx, nodes);
3293}
3294
221c5eb2
JA		 3295static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
3296 void *key)
3297{
e944475e 3298 struct io_poll_iocb *poll = wait->private;
221c5eb2
JA		 3299 struct io_kiocb *req = container_of(poll, struct io_kiocb, poll);
3300 struct io_ring_ctx *ctx = req->ctx;
3301 __poll_t mask = key_to_poll(key);
221c5eb2
JA		 3302
3303 /* for instances that support it check for an event match first: */
8c838788
JA		 3304 if (mask && !(mask & poll->events))
3305 return 0;
221c5eb2 3306
392edb45 3307 list_del_init(&poll->wait.entry);
221c5eb2 3308
7c9e7f0f
JA		 3309 /*
3310 * Run completion inline if we can. We're using trylock here because
3311 * we are violating the completion_lock -> poll wq lock ordering.
3312 * If we have a link timeout we're going to need the completion_lock
3313 * for finalizing the request, mark us as having grabbed that already.
3314 */
e94f141b
JA		 3315 if (mask) {
3316 unsigned long flags;
221c5eb2 3317
e94f141b
JA		 3318 if (llist_empty(&ctx->poll_llist) &&
3319 spin_trylock_irqsave(&ctx->completion_lock, flags)) {
3320 hash_del(&req->hash_node);
3321 io_poll_complete(req, mask, 0);
3322 req->flags |= REQ_F_COMP_LOCKED;
3323 io_put_req(req);
3324 spin_unlock_irqrestore(&ctx->completion_lock, flags);
3325
3326 io_cqring_ev_posted(ctx);
3327 req = NULL;
3328 } else {
3329 req->result = mask;
3330 req->llist_node.next = NULL;
3331 /* if the list wasn't empty, we're done */
3332 if (!llist_add(&req->llist_node, &ctx->poll_llist))
3333 req = NULL;
3334 else
3335 req->work.func = io_poll_flush;
3336 }
221c5eb2 3337 }
e94f141b
JA		 3338 if (req)
3339 io_queue_async_work(req);
221c5eb2 3340
221c5eb2
JA		 3341 return 1;
3342}
3343
3344struct io_poll_table {
3345 struct poll_table_struct pt;
3346 struct io_kiocb *req;
3347 int error;
3348};
3349
3350static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
3351 struct poll_table_struct *p)
3352{
3353 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
3354
3355 if (unlikely(pt->req->poll.head)) {
3356 pt->error = -EINVAL;
3357 return;
3358 }
3359
3360 pt->error = 0;
3361 pt->req->poll.head = head;
392edb45 3362 add_wait_queue(head, &pt->req->poll.wait);
221c5eb2
JA		 3363}
3364
eac406c6
JA		 3365static void io_poll_req_insert(struct io_kiocb *req)
3366{
3367 struct io_ring_ctx *ctx = req->ctx;
78076bb6
JA		 3368 struct hlist_head *list;
3369
3370 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
3371 hlist_add_head(&req->hash_node, list);
eac406c6
JA		 3372}
3373
3529d8c2 3374static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
221c5eb2
JA		 3375{
3376 struct io_poll_iocb *poll = &req->poll;
221c5eb2 3377 u16 events;
221c5eb2
JA		 3378
3379 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3380 return -EINVAL;
3381 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
3382 return -EINVAL;
09bb8394
JA		 3383 if (!poll->file)
3384 return -EBADF;
221c5eb2 3385
221c5eb2
JA		 3386 events = READ_ONCE(sqe->poll_events);
3387 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP;
0969e783
JA		 3388 return 0;
3389}
3390
3391static int io_poll_add(struct io_kiocb *req, struct io_kiocb **nxt)
3392{
3393 struct io_poll_iocb *poll = &req->poll;
3394 struct io_ring_ctx *ctx = req->ctx;
3395 struct io_poll_table ipt;
3396 bool cancel = false;
3397 __poll_t mask;
0969e783
JA		 3398
3399 INIT_IO_WORK(&req->work, io_poll_complete_work);
78076bb6 3400 INIT_HLIST_NODE(&req->hash_node);
221c5eb2 3401
221c5eb2 3402 poll->head = NULL;
8c838788 3403 poll->done = false;
221c5eb2
JA		 3404 poll->canceled = false;
3405
3406 ipt.pt._qproc = io_poll_queue_proc;
3407 ipt.pt._key = poll->events;
3408 ipt.req = req;
3409 ipt.error = -EINVAL; /* same as no support for IOCB_CMD_POLL */
3410
3411 /* initialized the list so that we can do list_empty checks */
392edb45
JA		 3412 INIT_LIST_HEAD(&poll->wait.entry);
3413 init_waitqueue_func_entry(&poll->wait, io_poll_wake);
3414 poll->wait.private = poll;
221c5eb2 3415
36703247
JA		 3416 INIT_LIST_HEAD(&req->list);
3417
221c5eb2 3418 mask = vfs_poll(poll->file, &ipt.pt) & poll->events;
221c5eb2
JA		 3419
3420 spin_lock_irq(&ctx->completion_lock);
8c838788
JA		 3421 if (likely(poll->head)) {
3422 spin_lock(&poll->head->lock);
392edb45 3423 if (unlikely(list_empty(&poll->wait.entry))) {
8c838788
JA		 3424 if (ipt.error)
3425 cancel = true;
3426 ipt.error = 0;
3427 mask = 0;
3428 }
3429 if (mask || ipt.error)
392edb45 3430 list_del_init(&poll->wait.entry);
8c838788
JA		 3431 else if (cancel)
3432 WRITE_ONCE(poll->canceled, true);
3433 else if (!poll->done) /* actually waiting for an event */
eac406c6 3434 io_poll_req_insert(req);
8c838788
JA		 3435 spin_unlock(&poll->head->lock);
3436 }
3437 if (mask) { /* no async, we'd stolen it */
221c5eb2 3438 ipt.error = 0;
b0dd8a41 3439 io_poll_complete(req, mask, 0);
221c5eb2 3440 }
221c5eb2
JA		 3441 spin_unlock_irq(&ctx->completion_lock);
3442
8c838788
JA		 3443 if (mask) {
3444 io_cqring_ev_posted(ctx);
ec9c02ad 3445 io_put_req_find_next(req, nxt);
221c5eb2 3446 }
8c838788 3447 return ipt.error;
221c5eb2
JA		 3448}
3449
5262f567
JA		 3450static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
3451{
ad8a48ac
JA		 3452 struct io_timeout_data *data = container_of(timer,
3453 struct io_timeout_data, timer);
3454 struct io_kiocb *req = data->req;
3455 struct io_ring_ctx *ctx = req->ctx;
5262f567
JA		 3456 unsigned long flags;
3457
5262f567
JA		 3458 atomic_inc(&ctx->cq_timeouts);
3459
3460 spin_lock_irqsave(&ctx->completion_lock, flags);
ef03681a 3461 /*
11365043
JA		 3462 * We could be racing with timeout deletion. If the list is empty,
3463 * then timeout lookup already found it and will be handling it.
ef03681a 3464 */
842f9612 3465 if (!list_empty(&req->list)) {
11365043 3466 struct io_kiocb *prev;
5262f567 3467
11365043
JA		 3468 /*
3469 * Adjust the reqs sequence before the current one because it
d195a66e 3470 * will consume a slot in the cq_ring and the cq_tail
11365043
JA		 3471 * pointer will be increased, otherwise other timeout reqs may
3472 * return in advance without waiting for enough wait_nr.
3473 */
3474 prev = req;
3475 list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list)
3476 prev->sequence++;
11365043 3477 list_del_init(&req->list);
11365043 3478 }
5262f567 3479
78e19bbe 3480 io_cqring_fill_event(req, -ETIME);
5262f567
JA		 3481 io_commit_cqring(ctx);
3482 spin_unlock_irqrestore(&ctx->completion_lock, flags);
3483
3484 io_cqring_ev_posted(ctx);
4e88d6e7 3485 req_set_fail_links(req);
5262f567
JA		 3486 io_put_req(req);
3487 return HRTIMER_NORESTART;
3488}
3489
47f46768
JA		 3490static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
3491{
3492 struct io_kiocb *req;
3493 int ret = -ENOENT;
3494
3495 list_for_each_entry(req, &ctx->timeout_list, list) {
3496 if (user_data == req->user_data) {
3497 list_del_init(&req->list);
3498 ret = 0;
3499 break;
3500 }
3501 }
3502
3503 if (ret == -ENOENT)
3504 return ret;
3505
2d28390a 3506 ret = hrtimer_try_to_cancel(&req->io->timeout.timer);
47f46768
JA		 3507 if (ret == -1)
3508 return -EALREADY;
3509
4e88d6e7 3510 req_set_fail_links(req);
47f46768
JA		 3511 io_cqring_fill_event(req, -ECANCELED);
3512 io_put_req(req);
3513 return 0;
3514}
3515
3529d8c2
JA		 3516static int io_timeout_remove_prep(struct io_kiocb *req,
3517 const struct io_uring_sqe *sqe)
b29472ee 3518{
b29472ee
JA		 3519 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3520 return -EINVAL;
3521 if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len)
3522 return -EINVAL;
3523
3524 req->timeout.addr = READ_ONCE(sqe->addr);
3525 req->timeout.flags = READ_ONCE(sqe->timeout_flags);
3526 if (req->timeout.flags)
3527 return -EINVAL;
3528
b29472ee
JA		 3529 return 0;
3530}
3531
11365043
JA		 3532/*
3533 * Remove or update an existing timeout command
3534 */
fc4df999 3535static int io_timeout_remove(struct io_kiocb *req)
11365043
JA		 3536{
3537 struct io_ring_ctx *ctx = req->ctx;
47f46768 3538 int ret;
11365043 3539
11365043 3540 spin_lock_irq(&ctx->completion_lock);
b29472ee 3541 ret = io_timeout_cancel(ctx, req->timeout.addr);
11365043 3542
47f46768 3543 io_cqring_fill_event(req, ret);
11365043
JA		 3544 io_commit_cqring(ctx);
3545 spin_unlock_irq(&ctx->completion_lock);
5262f567 3546 io_cqring_ev_posted(ctx);
4e88d6e7
JA		 3547 if (ret < 0)
3548 req_set_fail_links(req);
ec9c02ad 3549 io_put_req(req);
11365043 3550 return 0;
5262f567
JA		 3551}
3552
3529d8c2 3553static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
2d28390a 3554 bool is_timeout_link)
5262f567 3555{
ad8a48ac 3556 struct io_timeout_data *data;
a41525ab 3557 unsigned flags;
5262f567 3558
ad8a48ac 3559 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5262f567 3560 return -EINVAL;
ad8a48ac 3561 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
a41525ab 3562 return -EINVAL;
2d28390a
JA		 3563 if (sqe->off && is_timeout_link)
3564 return -EINVAL;
a41525ab
JA		 3565 flags = READ_ONCE(sqe->timeout_flags);
3566 if (flags & ~IORING_TIMEOUT_ABS)
5262f567 3567 return -EINVAL;
bdf20073 3568
26a61679
JA		 3569 req->timeout.count = READ_ONCE(sqe->off);
3570
3529d8c2 3571 if (!req->io && io_alloc_async_ctx(req))
26a61679
JA		 3572 return -ENOMEM;
3573
3574 data = &req->io->timeout;
ad8a48ac 3575 data->req = req;
ad8a48ac
JA		 3576 req->flags |= REQ_F_TIMEOUT;
3577
3578 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
5262f567
JA		 3579 return -EFAULT;
3580
11365043 3581 if (flags & IORING_TIMEOUT_ABS)
ad8a48ac 3582 data->mode = HRTIMER_MODE_ABS;
11365043 3583 else
ad8a48ac 3584 data->mode = HRTIMER_MODE_REL;
11365043 3585
ad8a48ac
JA		 3586 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
3587 return 0;
3588}
3589
fc4df999 3590static int io_timeout(struct io_kiocb *req)
ad8a48ac
JA		 3591{
3592 unsigned count;
3593 struct io_ring_ctx *ctx = req->ctx;
3594 struct io_timeout_data *data;
3595 struct list_head *entry;
3596 unsigned span = 0;
ad8a48ac 3597
2d28390a 3598 data = &req->io->timeout;
93bd25bb 3599
5262f567
JA		 3600 /*
3601 * sqe->off holds how many events that need to occur for this
93bd25bb
JA		 3602 * timeout event to be satisfied. If it isn't set, then this is
3603 * a pure timeout request, sequence isn't used.
5262f567 3604 */
26a61679 3605 count = req->timeout.count;
93bd25bb
JA		 3606 if (!count) {
3607 req->flags |= REQ_F_TIMEOUT_NOSEQ;
3608 spin_lock_irq(&ctx->completion_lock);
3609 entry = ctx->timeout_list.prev;
3610 goto add;
3611 }
5262f567
JA		 3612
3613 req->sequence = ctx->cached_sq_head + count - 1;
2d28390a 3614 data->seq_offset = count;
5262f567
JA		 3615
3616 /*
3617 * Insertion sort, ensuring the first entry in the list is always
3618 * the one we need first.
3619 */
5262f567
JA		 3620 spin_lock_irq(&ctx->completion_lock);
3621 list_for_each_prev(entry, &ctx->timeout_list) {
3622 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list);
5da0fb1a 3623 unsigned nxt_sq_head;
3624 long long tmp, tmp_nxt;
2d28390a 3625 u32 nxt_offset = nxt->io->timeout.seq_offset;
5262f567 3626
93bd25bb
JA		 3627 if (nxt->flags & REQ_F_TIMEOUT_NOSEQ)
3628 continue;
3629
5da0fb1a 3630 /*
3631 * Since cached_sq_head + count - 1 can overflow, use type long
3632 * long to store it.
3633 */
3634 tmp = (long long)ctx->cached_sq_head + count - 1;
cc42e0ac
PB		 3635 nxt_sq_head = nxt->sequence - nxt_offset + 1;
3636 tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1;
5da0fb1a 3637
3638 /*
3639 * cached_sq_head may overflow, and it will never overflow twice
3640 * once there is some timeout req still be valid.
3641 */
3642 if (ctx->cached_sq_head < nxt_sq_head)
8b07a65a 3643 tmp += UINT_MAX;
5da0fb1a 3644
a1f58ba4 3645 if (tmp > tmp_nxt)
5262f567 3646 break;
a1f58ba4 3647
3648 /*
3649 * Sequence of reqs after the insert one and itself should
3650 * be adjusted because each timeout req consumes a slot.
3651 */
3652 span++;
3653 nxt->sequence++;
5262f567 3654 }
a1f58ba4 3655 req->sequence -= span;
93bd25bb 3656add:
5262f567 3657 list_add(&req->list, entry);
ad8a48ac
JA		 3658 data->timer.function = io_timeout_fn;
3659 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
5262f567 3660 spin_unlock_irq(&ctx->completion_lock);
5262f567
JA		 3661 return 0;
3662}
5262f567 3663
62755e35
JA		 3664static bool io_cancel_cb(struct io_wq_work *work, void *data)
3665{
3666 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
3667
3668 return req->user_data == (unsigned long) data;
3669}
3670
e977d6d3 3671static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr)
62755e35 3672{
62755e35 3673 enum io_wq_cancel cancel_ret;
62755e35
JA		 3674 int ret = 0;
3675
62755e35
JA		 3676 cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr);
3677 switch (cancel_ret) {
3678 case IO_WQ_CANCEL_OK:
3679 ret = 0;
3680 break;
3681 case IO_WQ_CANCEL_RUNNING:
3682 ret = -EALREADY;
3683 break;
3684 case IO_WQ_CANCEL_NOTFOUND:
3685 ret = -ENOENT;
3686 break;
3687 }
3688
e977d6d3
JA		 3689 return ret;
3690}
3691
47f46768
JA		 3692static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
3693 struct io_kiocb *req, __u64 sqe_addr,
b0dd8a41 3694 struct io_kiocb **nxt, int success_ret)
47f46768
JA		 3695{
3696 unsigned long flags;
3697 int ret;
3698
3699 ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr);
3700 if (ret != -ENOENT) {
3701 spin_lock_irqsave(&ctx->completion_lock, flags);
3702 goto done;
3703 }
3704
3705 spin_lock_irqsave(&ctx->completion_lock, flags);
3706 ret = io_timeout_cancel(ctx, sqe_addr);
3707 if (ret != -ENOENT)
3708 goto done;
3709 ret = io_poll_cancel(ctx, sqe_addr);
3710done:
b0dd8a41
JA		 3711 if (!ret)
3712 ret = success_ret;
47f46768
JA		 3713 io_cqring_fill_event(req, ret);
3714 io_commit_cqring(ctx);
3715 spin_unlock_irqrestore(&ctx->completion_lock, flags);
3716 io_cqring_ev_posted(ctx);
3717
4e88d6e7
JA		 3718 if (ret < 0)
3719 req_set_fail_links(req);
47f46768
JA		 3720 io_put_req_find_next(req, nxt);
3721}
3722
3529d8c2
JA		 3723static int io_async_cancel_prep(struct io_kiocb *req,
3724 const struct io_uring_sqe *sqe)
e977d6d3 3725{
fbf23849 3726 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
e977d6d3
JA		 3727 return -EINVAL;
3728 if (sqe->flags || sqe->ioprio || sqe->off || sqe->len ||
3729 sqe->cancel_flags)
3730 return -EINVAL;
3731
fbf23849
JA		 3732 req->cancel.addr = READ_ONCE(sqe->addr);
3733 return 0;
3734}
3735
3736static int io_async_cancel(struct io_kiocb *req, struct io_kiocb **nxt)
3737{
3738 struct io_ring_ctx *ctx = req->ctx;
fbf23849
JA		 3739
3740 io_async_find_and_cancel(ctx, req, req->cancel.addr, nxt, 0);
5262f567
JA		 3741 return 0;
3742}
3743
05f3fb3c
JA		 3744static int io_files_update_prep(struct io_kiocb *req,
3745 const struct io_uring_sqe *sqe)
3746{
3747 if (sqe->flags || sqe->ioprio || sqe->rw_flags)
3748 return -EINVAL;
3749
3750 req->files_update.offset = READ_ONCE(sqe->off);
3751 req->files_update.nr_args = READ_ONCE(sqe->len);
3752 if (!req->files_update.nr_args)
3753 return -EINVAL;
3754 req->files_update.arg = READ_ONCE(sqe->addr);
3755 return 0;
3756}
3757
3758static int io_files_update(struct io_kiocb *req, bool force_nonblock)
3759{
3760 struct io_ring_ctx *ctx = req->ctx;
3761 struct io_uring_files_update up;
3762 int ret;
3763
3764 if (force_nonblock) {
3765 req->work.flags |= IO_WQ_WORK_NEEDS_FILES;
3766 return -EAGAIN;
3767 }
3768
3769 up.offset = req->files_update.offset;
3770 up.fds = req->files_update.arg;
3771
3772 mutex_lock(&ctx->uring_lock);
3773 ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args);
3774 mutex_unlock(&ctx->uring_lock);
3775
3776 if (ret < 0)
3777 req_set_fail_links(req);
3778 io_cqring_add_event(req, ret);
3779 io_put_req(req);
3780 return 0;
3781}
3782
3529d8c2
JA		 3783static int io_req_defer_prep(struct io_kiocb *req,
3784 const struct io_uring_sqe *sqe)
f67676d1 3785{
e781573e 3786 ssize_t ret = 0;
f67676d1 3787
d625c6ee 3788 switch (req->opcode) {
e781573e
JA		 3789 case IORING_OP_NOP:
3790 break;
f67676d1
JA		 3791 case IORING_OP_READV:
3792 case IORING_OP_READ_FIXED:
3a6820f2 3793 case IORING_OP_READ:
3529d8c2 3794 ret = io_read_prep(req, sqe, true);
f67676d1
JA		 3795 break;
3796 case IORING_OP_WRITEV:
3797 case IORING_OP_WRITE_FIXED:
3a6820f2 3798 case IORING_OP_WRITE:
3529d8c2 3799 ret = io_write_prep(req, sqe, true);
f67676d1 3800 break;
0969e783 3801 case IORING_OP_POLL_ADD:
3529d8c2 3802 ret = io_poll_add_prep(req, sqe);
0969e783
JA		 3803 break;
3804 case IORING_OP_POLL_REMOVE:
3529d8c2 3805 ret = io_poll_remove_prep(req, sqe);
0969e783 3806 break;
8ed8d3c3 3807 case IORING_OP_FSYNC:
3529d8c2 3808 ret = io_prep_fsync(req, sqe);
8ed8d3c3
JA		 3809 break;
3810 case IORING_OP_SYNC_FILE_RANGE:
3529d8c2 3811 ret = io_prep_sfr(req, sqe);
8ed8d3c3 3812 break;
03b1230c 3813 case IORING_OP_SENDMSG:
3529d8c2 3814 ret = io_sendmsg_prep(req, sqe);
03b1230c
JA		 3815 break;
3816 case IORING_OP_RECVMSG:
3529d8c2 3817 ret = io_recvmsg_prep(req, sqe);
03b1230c 3818 break;
f499a021 3819 case IORING_OP_CONNECT:
3529d8c2 3820 ret = io_connect_prep(req, sqe);
f499a021 3821 break;
2d28390a 3822 case IORING_OP_TIMEOUT:
3529d8c2 3823 ret = io_timeout_prep(req, sqe, false);
b7bb4f7d 3824 break;
b29472ee 3825 case IORING_OP_TIMEOUT_REMOVE:
3529d8c2 3826 ret = io_timeout_remove_prep(req, sqe);
b29472ee 3827 break;
fbf23849 3828 case IORING_OP_ASYNC_CANCEL:
3529d8c2 3829 ret = io_async_cancel_prep(req, sqe);
fbf23849 3830 break;
2d28390a 3831 case IORING_OP_LINK_TIMEOUT:
3529d8c2 3832 ret = io_timeout_prep(req, sqe, true);
b7bb4f7d 3833 break;
8ed8d3c3 3834 case IORING_OP_ACCEPT:
3529d8c2 3835 ret = io_accept_prep(req, sqe);
8ed8d3c3 3836 break;
d63d1b5e
JA		 3837 case IORING_OP_FALLOCATE:
3838 ret = io_fallocate_prep(req, sqe);
3839 break;
15b71abe
JA		 3840 case IORING_OP_OPENAT:
3841 ret = io_openat_prep(req, sqe);
3842 break;
b5dba59e
JA		 3843 case IORING_OP_CLOSE:
3844 ret = io_close_prep(req, sqe);
3845 break;
05f3fb3c
JA		 3846 case IORING_OP_FILES_UPDATE:
3847 ret = io_files_update_prep(req, sqe);
3848 break;
eddc7ef5
JA		 3849 case IORING_OP_STATX:
3850 ret = io_statx_prep(req, sqe);
3851 break;
4840e418
JA		 3852 case IORING_OP_FADVISE:
3853 ret = io_fadvise_prep(req, sqe);
3854 break;
c1ca757b
JA		 3855 case IORING_OP_MADVISE:
3856 ret = io_madvise_prep(req, sqe);
3857 break;
f67676d1 3858 default:
e781573e
JA		 3859 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
3860 req->opcode);
3861 ret = -EINVAL;
b7bb4f7d 3862 break;
f67676d1
JA		 3863 }
3864
b7bb4f7d 3865 return ret;
f67676d1
JA		 3866}
3867
3529d8c2 3868static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe)
de0617e4 3869{
a197f664 3870 struct io_ring_ctx *ctx = req->ctx;
f67676d1 3871 int ret;
de0617e4 3872
9d858b21
BL		 3873 /* Still need defer if there is pending req in defer list. */
3874 if (!req_need_defer(req) && list_empty(&ctx->defer_list))
de0617e4
JA		 3875 return 0;
3876
3529d8c2 3877 if (!req->io && io_alloc_async_ctx(req))
de0617e4
JA		 3878 return -EAGAIN;
3879
3529d8c2 3880 ret = io_req_defer_prep(req, sqe);
b7bb4f7d 3881 if (ret < 0)
2d28390a 3882 return ret;
2d28390a 3883
de0617e4 3884 spin_lock_irq(&ctx->completion_lock);
9d858b21 3885 if (!req_need_defer(req) && list_empty(&ctx->defer_list)) {
de0617e4 3886 spin_unlock_irq(&ctx->completion_lock);
de0617e4
JA		 3887 return 0;
3888 }
3889
915967f6 3890 trace_io_uring_defer(ctx, req, req->user_data);
de0617e4
JA		 3891 list_add_tail(&req->list, &ctx->defer_list);
3892 spin_unlock_irq(&ctx->completion_lock);
3893 return -EIOCBQUEUED;
3894}
3895
3529d8c2
JA		 3896static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
3897 struct io_kiocb **nxt, bool force_nonblock)
2b188cc1 3898{
a197f664 3899 struct io_ring_ctx *ctx = req->ctx;
d625c6ee 3900 int ret;
2b188cc1 3901
d625c6ee 3902 switch (req->opcode) {
2b188cc1 3903 case IORING_OP_NOP:
78e19bbe 3904 ret = io_nop(req);
2b188cc1
JA		 3905 break;
3906 case IORING_OP_READV:
edafccee 3907 case IORING_OP_READ_FIXED:
3a6820f2 3908 case IORING_OP_READ:
3529d8c2
JA		 3909 if (sqe) {
3910 ret = io_read_prep(req, sqe, force_nonblock);
3911 if (ret < 0)
3912 break;
3913 }
267bc904 3914 ret = io_read(req, nxt, force_nonblock);
edafccee 3915 break;
3529d8c2 3916 case IORING_OP_WRITEV:
edafccee 3917 case IORING_OP_WRITE_FIXED:
3a6820f2 3918 case IORING_OP_WRITE:
3529d8c2
JA		 3919 if (sqe) {
3920 ret = io_write_prep(req, sqe, force_nonblock);
3921 if (ret < 0)
3922 break;
3923 }
267bc904 3924 ret = io_write(req, nxt, force_nonblock);
2b188cc1 3925 break;
c992fe29 3926 case IORING_OP_FSYNC:
3529d8c2
JA		 3927 if (sqe) {
3928 ret = io_prep_fsync(req, sqe);
3929 if (ret < 0)
3930 break;
3931 }
fc4df999 3932 ret = io_fsync(req, nxt, force_nonblock);
c992fe29 3933 break;
221c5eb2 3934 case IORING_OP_POLL_ADD:
3529d8c2
JA		 3935 if (sqe) {
3936 ret = io_poll_add_prep(req, sqe);
3937 if (ret)
3938 break;
3939 }
fc4df999 3940 ret = io_poll_add(req, nxt);
221c5eb2
JA		 3941 break;
3942 case IORING_OP_POLL_REMOVE:
3529d8c2
JA		 3943 if (sqe) {
3944 ret = io_poll_remove_prep(req, sqe);
3945 if (ret < 0)
3946 break;
3947 }
fc4df999 3948 ret = io_poll_remove(req);
221c5eb2 3949 break;
5d17b4a4 3950 case IORING_OP_SYNC_FILE_RANGE:
3529d8c2
JA		 3951 if (sqe) {
3952 ret = io_prep_sfr(req, sqe);
3953 if (ret < 0)
3954 break;
3955 }
fc4df999 3956 ret = io_sync_file_range(req, nxt, force_nonblock);
5d17b4a4 3957 break;
0fa03c62 3958 case IORING_OP_SENDMSG:
3529d8c2
JA		 3959 if (sqe) {
3960 ret = io_sendmsg_prep(req, sqe);
3961 if (ret < 0)
3962 break;
3963 }
fc4df999 3964 ret = io_sendmsg(req, nxt, force_nonblock);
0fa03c62 3965 break;
aa1fa28f 3966 case IORING_OP_RECVMSG:
3529d8c2
JA		 3967 if (sqe) {
3968 ret = io_recvmsg_prep(req, sqe);
3969 if (ret)
3970 break;
3971 }
fc4df999 3972 ret = io_recvmsg(req, nxt, force_nonblock);
aa1fa28f 3973 break;
5262f567 3974 case IORING_OP_TIMEOUT:
3529d8c2
JA		 3975 if (sqe) {
3976 ret = io_timeout_prep(req, sqe, false);
3977 if (ret)
3978 break;
3979 }
fc4df999 3980 ret = io_timeout(req);
5262f567 3981 break;
11365043 3982 case IORING_OP_TIMEOUT_REMOVE:
3529d8c2
JA		 3983 if (sqe) {
3984 ret = io_timeout_remove_prep(req, sqe);
3985 if (ret)
3986 break;
3987 }
fc4df999 3988 ret = io_timeout_remove(req);
11365043 3989 break;
17f2fe35 3990 case IORING_OP_ACCEPT:
3529d8c2
JA		 3991 if (sqe) {
3992 ret = io_accept_prep(req, sqe);
3993 if (ret)
3994 break;
3995 }
fc4df999 3996 ret = io_accept(req, nxt, force_nonblock);
17f2fe35 3997 break;
f8e85cf2 3998 case IORING_OP_CONNECT:
3529d8c2
JA		 3999 if (sqe) {
4000 ret = io_connect_prep(req, sqe);
4001 if (ret)
4002 break;
4003 }
fc4df999 4004 ret = io_connect(req, nxt, force_nonblock);
f8e85cf2 4005 break;
62755e35 4006 case IORING_OP_ASYNC_CANCEL:
3529d8c2
JA		 4007 if (sqe) {
4008 ret = io_async_cancel_prep(req, sqe);
4009 if (ret)
4010 break;
4011 }
fc4df999 4012 ret = io_async_cancel(req, nxt);
62755e35 4013 break;
d63d1b5e
JA		 4014 case IORING_OP_FALLOCATE:
4015 if (sqe) {
4016 ret = io_fallocate_prep(req, sqe);
4017 if (ret)
4018 break;
4019 }
4020 ret = io_fallocate(req, nxt, force_nonblock);
4021 break;
15b71abe
JA		 4022 case IORING_OP_OPENAT:
4023 if (sqe) {
4024 ret = io_openat_prep(req, sqe);
4025 if (ret)
4026 break;
4027 }
4028 ret = io_openat(req, nxt, force_nonblock);
4029 break;
b5dba59e
JA		 4030 case IORING_OP_CLOSE:
4031 if (sqe) {
4032 ret = io_close_prep(req, sqe);
4033 if (ret)
4034 break;
4035 }
4036 ret = io_close(req, nxt, force_nonblock);
4037 break;
05f3fb3c
JA		 4038 case IORING_OP_FILES_UPDATE:
4039 if (sqe) {
4040 ret = io_files_update_prep(req, sqe);
4041 if (ret)
4042 break;
4043 }
4044 ret = io_files_update(req, force_nonblock);
4045 break;
eddc7ef5
JA		 4046 case IORING_OP_STATX:
4047 if (sqe) {
4048 ret = io_statx_prep(req, sqe);
4049 if (ret)
4050 break;
4051 }
4052 ret = io_statx(req, nxt, force_nonblock);
4053 break;
4840e418
JA		 4054 case IORING_OP_FADVISE:
4055 if (sqe) {
4056 ret = io_fadvise_prep(req, sqe);
4057 if (ret)
4058 break;
4059 }
4060 ret = io_fadvise(req, nxt, force_nonblock);
4061 break;
c1ca757b
JA		 4062 case IORING_OP_MADVISE:
4063 if (sqe) {
4064 ret = io_madvise_prep(req, sqe);
4065 if (ret)
4066 break;
4067 }
4068 ret = io_madvise(req, nxt, force_nonblock);
4069 break;
2b188cc1
JA		 4070 default:
4071 ret = -EINVAL;
4072 break;
4073 }
4074
def596e9
JA		 4075 if (ret)
4076 return ret;
4077
4078 if (ctx->flags & IORING_SETUP_IOPOLL) {
11ba820b
JA		 4079 const bool in_async = io_wq_current_is_worker();
4080
9e645e11 4081 if (req->result == -EAGAIN)
def596e9
JA		 4082 return -EAGAIN;
4083
11ba820b
JA		 4084 /* workqueue context doesn't hold uring_lock, grab it now */
4085 if (in_async)
4086 mutex_lock(&ctx->uring_lock);
4087
def596e9 4088 io_iopoll_req_issued(req);
11ba820b
JA		 4089
4090 if (in_async)
4091 mutex_unlock(&ctx->uring_lock);
def596e9
JA		 4092 }
4093
4094 return 0;
2b188cc1
JA		 4095}
4096
561fb04a 4097static void io_wq_submit_work(struct io_wq_work **workptr)
2b188cc1 4098{
561fb04a 4099 struct io_wq_work *work = *workptr;
2b188cc1 4100 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
561fb04a
JA		 4101 struct io_kiocb *nxt = NULL;
4102 int ret = 0;
2b188cc1 4103
0c9d5ccd
JA		 4104 /* if NO_CANCEL is set, we must still run the work */
4105 if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) ==
4106 IO_WQ_WORK_CANCEL) {
561fb04a 4107 ret = -ECANCELED;
0c9d5ccd 4108 }
31b51510 4109
561fb04a 4110 if (!ret) {
cf6fd4bd
PB		 4111 req->has_user = (work->flags & IO_WQ_WORK_HAS_MM) != 0;
4112 req->in_async = true;
561fb04a 4113 do {
3529d8c2 4114 ret = io_issue_sqe(req, NULL, &nxt, false);
561fb04a
JA		 4115 /*
4116 * We can get EAGAIN for polled IO even though we're
4117 * forcing a sync submission from here, since we can't
4118 * wait for request slots on the block side.
4119 */
4120 if (ret != -EAGAIN)
4121 break;
4122 cond_resched();
4123 } while (1);
4124 }
31b51510 4125
561fb04a 4126 /* drop submission reference */
ec9c02ad 4127 io_put_req(req);
817869d2 4128
561fb04a 4129 if (ret) {
4e88d6e7 4130 req_set_fail_links(req);
78e19bbe 4131 io_cqring_add_event(req, ret);
817869d2 4132 io_put_req(req);
edafccee 4133 }
2b188cc1 4134
561fb04a 4135 /* if a dependent link is ready, pass it back */
78912934
JA		 4136 if (!ret && nxt)
4137 io_wq_assign_next(workptr, nxt);
2b188cc1
JA		 4138}
4139
15b71abe 4140static int io_req_needs_file(struct io_kiocb *req, int fd)
09bb8394 4141{
d3656344 4142 if (!io_op_defs[req->opcode].needs_file)
9e3aa61a 4143 return 0;
d3656344
JA		 4144 if (fd == -1 && io_op_defs[req->opcode].fd_non_neg)
4145 return 0;
4146 return 1;
09bb8394
JA		 4147}
4148
65e19f54
JA		 4149static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
4150 int index)
4151{
4152 struct fixed_file_table *table;
4153
05f3fb3c
JA		 4154 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
4155 return table->files[index & IORING_FILE_TABLE_MASK];;
65e19f54
JA		 4156}
4157
3529d8c2
JA		 4158static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
4159 const struct io_uring_sqe *sqe)
09bb8394 4160{
a197f664 4161 struct io_ring_ctx *ctx = req->ctx;
09bb8394 4162 unsigned flags;
d3656344 4163 int fd;
09bb8394 4164
3529d8c2
JA		 4165 flags = READ_ONCE(sqe->flags);
4166 fd = READ_ONCE(sqe->fd);
09bb8394 4167
4fe2c963 4168 if (flags & IOSQE_IO_DRAIN)
de0617e4 4169 req->flags |= REQ_F_IO_DRAIN;
de0617e4 4170
d3656344
JA		 4171 if (!io_req_needs_file(req, fd))
4172 return 0;
09bb8394
JA		 4173
4174 if (flags & IOSQE_FIXED_FILE) {
05f3fb3c 4175 if (unlikely(!ctx->file_data ||
09bb8394
JA		 4176 (unsigned) fd >= ctx->nr_user_files))
4177 return -EBADF;
b7620121 4178 fd = array_index_nospec(fd, ctx->nr_user_files);
65e19f54
JA		 4179 req->file = io_file_from_index(ctx, fd);
4180 if (!req->file)
08a45173 4181 return -EBADF;
09bb8394 4182 req->flags |= REQ_F_FIXED_FILE;
05f3fb3c 4183 percpu_ref_get(&ctx->file_data->refs);
09bb8394 4184 } else {
cf6fd4bd 4185 if (req->needs_fixed_file)
09bb8394 4186 return -EBADF;
c826bd7a 4187 trace_io_uring_file_get(ctx, fd);
09bb8394
JA		 4188 req->file = io_file_get(state, fd);
4189 if (unlikely(!req->file))
4190 return -EBADF;
4191 }
4192
4193 return 0;
4194}
4195
a197f664 4196static int io_grab_files(struct io_kiocb *req)
fcb323cc
JA		 4197{
4198 int ret = -EBADF;
a197f664 4199 struct io_ring_ctx *ctx = req->ctx;
fcb323cc 4200
b5dba59e
JA		 4201 if (!req->ring_file)
4202 return -EBADF;
4203
fcb323cc
JA		 4204 rcu_read_lock();
4205 spin_lock_irq(&ctx->inflight_lock);
4206 /*
4207 * We use the f_ops->flush() handler to ensure that we can flush
4208 * out work accessing these files if the fd is closed. Check if
4209 * the fd has changed since we started down this path, and disallow
4210 * this operation if it has.
4211 */
cf6fd4bd 4212 if (fcheck(req->ring_fd) == req->ring_file) {
fcb323cc
JA		 4213 list_add(&req->inflight_entry, &ctx->inflight_list);
4214 req->flags |= REQ_F_INFLIGHT;
4215 req->work.files = current->files;
4216 ret = 0;
4217 }
4218 spin_unlock_irq(&ctx->inflight_lock);
4219 rcu_read_unlock();
4220
4221 return ret;
4222}
4223
2665abfd 4224static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
2b188cc1 4225{
ad8a48ac
JA		 4226 struct io_timeout_data *data = container_of(timer,
4227 struct io_timeout_data, timer);
4228 struct io_kiocb *req = data->req;
2665abfd
JA		 4229 struct io_ring_ctx *ctx = req->ctx;
4230 struct io_kiocb *prev = NULL;
4231 unsigned long flags;
2665abfd
JA		 4232
4233 spin_lock_irqsave(&ctx->completion_lock, flags);
4234
4235 /*
4236 * We don't expect the list to be empty, that will only happen if we
4237 * race with the completion of the linked work.
4238 */
4493233e
PB		 4239 if (!list_empty(&req->link_list)) {
4240 prev = list_entry(req->link_list.prev, struct io_kiocb,
4241 link_list);
5d960724 4242 if (refcount_inc_not_zero(&prev->refs)) {
4493233e 4243 list_del_init(&req->link_list);
5d960724
JA		 4244 prev->flags &= ~REQ_F_LINK_TIMEOUT;
4245 } else
76a46e06 4246 prev = NULL;
2665abfd
JA		 4247 }
4248
4249 spin_unlock_irqrestore(&ctx->completion_lock, flags);
4250
4251 if (prev) {
4e88d6e7 4252 req_set_fail_links(prev);
b0dd8a41
JA		 4253 io_async_find_and_cancel(ctx, req, prev->user_data, NULL,
4254 -ETIME);
76a46e06 4255 io_put_req(prev);
47f46768
JA		 4256 } else {
4257 io_cqring_add_event(req, -ETIME);
4258 io_put_req(req);
2665abfd 4259 }
2665abfd
JA		 4260 return HRTIMER_NORESTART;
4261}
4262
ad8a48ac 4263static void io_queue_linked_timeout(struct io_kiocb *req)
2665abfd 4264{
76a46e06 4265 struct io_ring_ctx *ctx = req->ctx;
2665abfd 4266
76a46e06
JA		 4267 /*
4268 * If the list is now empty, then our linked request finished before
4269 * we got a chance to setup the timer
4270 */
4271 spin_lock_irq(&ctx->completion_lock);
4493233e 4272 if (!list_empty(&req->link_list)) {
2d28390a 4273 struct io_timeout_data *data = &req->io->timeout;
94ae5e77 4274
ad8a48ac
JA		 4275 data->timer.function = io_link_timeout_fn;
4276 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
4277 data->mode);
2665abfd 4278 }
76a46e06 4279 spin_unlock_irq(&ctx->completion_lock);
2665abfd 4280
2665abfd 4281 /* drop submission reference */
76a46e06
JA		 4282 io_put_req(req);
4283}
2665abfd 4284
ad8a48ac 4285static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
2665abfd
JA		 4286{
4287 struct io_kiocb *nxt;
4288
4289 if (!(req->flags & REQ_F_LINK))
4290 return NULL;
4291
4493233e
PB		 4292 nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb,
4293 link_list);
d625c6ee 4294 if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT)
76a46e06 4295 return NULL;
2665abfd 4296
76a46e06 4297 req->flags |= REQ_F_LINK_TIMEOUT;
76a46e06 4298 return nxt;
2665abfd
JA		 4299}
4300
3529d8c2 4301static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2b188cc1 4302{
4a0a7a18 4303 struct io_kiocb *linked_timeout;
f9bd67f6 4304 struct io_kiocb *nxt = NULL;
e0c5c576 4305 int ret;
2b188cc1 4306
4a0a7a18
JA		 4307again:
4308 linked_timeout = io_prep_linked_timeout(req);
4309
3529d8c2 4310 ret = io_issue_sqe(req, sqe, &nxt, true);
491381ce
JA		 4311
4312 /*
4313 * We async punt it if the file wasn't marked NOWAIT, or if the file
4314 * doesn't support non-blocking read/write attempts
4315 */
4316 if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) ||
4317 (req->flags & REQ_F_MUST_PUNT))) {
bbad27b2
PB		 4318 if (req->work.flags & IO_WQ_WORK_NEEDS_FILES) {
4319 ret = io_grab_files(req);
4320 if (ret)
4321 goto err;
2b188cc1 4322 }
bbad27b2
PB		 4323
4324 /*
4325 * Queued up for async execution, worker will release
4326 * submit reference when the iocb is actually submitted.
4327 */
4328 io_queue_async_work(req);
4a0a7a18 4329 goto done_req;
2b188cc1 4330 }
e65ef56d 4331
fcb323cc 4332err:
76a46e06 4333 /* drop submission reference */
ec9c02ad 4334 io_put_req(req);
e65ef56d 4335
f9bd67f6 4336 if (linked_timeout) {
76a46e06 4337 if (!ret)
f9bd67f6 4338 io_queue_linked_timeout(linked_timeout);
76a46e06 4339 else
f9bd67f6 4340 io_put_req(linked_timeout);
76a46e06
JA		 4341 }
4342
e65ef56d 4343 /* and drop final reference, if we failed */
9e645e11 4344 if (ret) {
78e19bbe 4345 io_cqring_add_event(req, ret);
4e88d6e7 4346 req_set_fail_links(req);
e65ef56d 4347 io_put_req(req);
9e645e11 4348 }
4a0a7a18
JA		 4349done_req:
4350 if (nxt) {
4351 req = nxt;
4352 nxt = NULL;
4353 goto again;
4354 }
2b188cc1
JA		 4355}
4356
3529d8c2 4357static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4fe2c963
JL		 4358{
4359 int ret;
4360
1b4a51b6
PB		 4361 if (unlikely(req->ctx->drain_next)) {
4362 req->flags |= REQ_F_IO_DRAIN;
4363 req->ctx->drain_next = false;
4364 }
4365 req->ctx->drain_next = (req->flags & REQ_F_DRAIN_LINK);
4366
3529d8c2 4367 ret = io_req_defer(req, sqe);
4fe2c963
JL		 4368 if (ret) {
4369 if (ret != -EIOCBQUEUED) {
78e19bbe 4370 io_cqring_add_event(req, ret);
4e88d6e7 4371 req_set_fail_links(req);
78e19bbe 4372 io_double_put_req(req);
4fe2c963 4373 }
ce35a47a
JA		 4374 } else if ((req->flags & REQ_F_FORCE_ASYNC) &&
4375 !io_wq_current_is_worker()) {
4376 /*
4377 * Never try inline submit of IOSQE_ASYNC is set, go straight
4378 * to async execution.
4379 */
4380 req->work.flags |= IO_WQ_WORK_CONCURRENT;
4381 io_queue_async_work(req);
4382 } else {
3529d8c2 4383 __io_queue_sqe(req, sqe);
ce35a47a 4384 }
4fe2c963
JL		 4385}
4386
1b4a51b6 4387static inline void io_queue_link_head(struct io_kiocb *req)
4fe2c963 4388{
94ae5e77 4389 if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
1b4a51b6
PB		 4390 io_cqring_add_event(req, -ECANCELED);
4391 io_double_put_req(req);
4392 } else
3529d8c2 4393 io_queue_sqe(req, NULL);
4fe2c963
JL		 4394}
4395
4e88d6e7 4396#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
ce35a47a 4397 IOSQE_IO_HARDLINK | IOSQE_ASYNC)
9e645e11 4398
3529d8c2
JA		 4399static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
4400 struct io_submit_state *state, struct io_kiocb **link)
9e645e11 4401{
a197f664 4402 struct io_ring_ctx *ctx = req->ctx;
32fe525b 4403 unsigned int sqe_flags;
9e645e11
JA		 4404 int ret;
4405
32fe525b
PB		 4406 sqe_flags = READ_ONCE(sqe->flags);
4407
9e645e11 4408 /* enforce forwards compatibility on users */
32fe525b 4409 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
9e645e11 4410 ret = -EINVAL;
196be95c 4411 goto err_req;
9e645e11 4412 }
32fe525b 4413 if (sqe_flags & IOSQE_ASYNC)
ce35a47a 4414 req->flags |= REQ_F_FORCE_ASYNC;
9e645e11 4415
3529d8c2 4416 ret = io_req_set_file(state, req, sqe);
9e645e11
JA		 4417 if (unlikely(ret)) {
4418err_req:
78e19bbe
JA		 4419 io_cqring_add_event(req, ret);
4420 io_double_put_req(req);
2e6e1fde 4421 return false;
9e645e11
JA		 4422 }
4423
9e645e11
JA		 4424 /*
4425 * If we already have a head request, queue this one for async
4426 * submittal once the head completes. If we don't have a head but
4427 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
4428 * submitted sync once the chain is complete. If none of those
4429 * conditions are true (normal request), then just queue it.
4430 */
4431 if (*link) {
9d76377f 4432 struct io_kiocb *head = *link;
9e645e11 4433
32fe525b 4434 if (sqe_flags & IOSQE_IO_DRAIN)
9d76377f 4435 head->flags |= REQ_F_DRAIN_LINK | REQ_F_IO_DRAIN;
1b4a51b6 4436
32fe525b 4437 if (sqe_flags & IOSQE_IO_HARDLINK)
4e88d6e7
JA		 4438 req->flags |= REQ_F_HARDLINK;
4439
b7bb4f7d 4440 if (io_alloc_async_ctx(req)) {
9e645e11
JA		 4441 ret = -EAGAIN;
4442 goto err_req;
4443 }
4444
3529d8c2 4445 ret = io_req_defer_prep(req, sqe);
2d28390a 4446 if (ret) {
4e88d6e7 4447 /* fail even hard links since we don't submit */
9d76377f 4448 head->flags |= REQ_F_FAIL_LINK;
f67676d1 4449 goto err_req;
2d28390a 4450 }
9d76377f
PB		 4451 trace_io_uring_link(ctx, req, head);
4452 list_add_tail(&req->link_list, &head->link_list);
32fe525b
PB		 4453
4454 /* last request of a link, enqueue the link */
4455 if (!(sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK))) {
4456 io_queue_link_head(head);
4457 *link = NULL;
4458 }
4459 } else if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
9e645e11 4460 req->flags |= REQ_F_LINK;
32fe525b 4461 if (sqe_flags & IOSQE_IO_HARDLINK)
4e88d6e7 4462 req->flags |= REQ_F_HARDLINK;
9e645e11 4463
9e645e11 4464 INIT_LIST_HEAD(&req->link_list);
3529d8c2
JA		 4465 ret = io_req_defer_prep(req, sqe);
4466 if (ret)
4467 req->flags |= REQ_F_FAIL_LINK;
9e645e11
JA		 4468 *link = req;
4469 } else {
3529d8c2 4470 io_queue_sqe(req, sqe);
9e645e11 4471 }
2e6e1fde
PB		 4472
4473 return true;
9e645e11
JA		 4474}
4475
9a56a232
JA		 4476/*
4477 * Batched submission is done, ensure local IO is flushed out.
4478 */
4479static void io_submit_state_end(struct io_submit_state *state)
4480{
4481 blk_finish_plug(&state->plug);
3d6770fb 4482 io_file_put(state);
2579f913
JA		 4483 if (state->free_reqs)
4484 kmem_cache_free_bulk(req_cachep, state->free_reqs,
4485 &state->reqs[state->cur_req]);
9a56a232
JA		 4486}
4487
4488/*
4489 * Start submission side cache.
4490 */
4491static void io_submit_state_start(struct io_submit_state *state,
22efde59 4492 unsigned int max_ios)
9a56a232
JA		 4493{
4494 blk_start_plug(&state->plug);
2579f913 4495 state->free_reqs = 0;
9a56a232
JA		 4496 state->file = NULL;
4497 state->ios_left = max_ios;
4498}
4499
2b188cc1
JA		 4500static void io_commit_sqring(struct io_ring_ctx *ctx)
4501{
75b28aff 4502 struct io_rings *rings = ctx->rings;
2b188cc1 4503
75b28aff 4504 if (ctx->cached_sq_head != READ_ONCE(rings->sq.head)) {
2b188cc1
JA		 4505 /*
4506 * Ensure any loads from the SQEs are done at this point,
4507 * since once we write the new head, the application could
4508 * write new data to them.
4509 */
75b28aff 4510 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
2b188cc1
JA		 4511 }
4512}
4513
2b188cc1 4514/*
3529d8c2 4515 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
2b188cc1
JA		 4516 * that is mapped by userspace. This means that care needs to be taken to
4517 * ensure that reads are stable, as we cannot rely on userspace always
4518 * being a good citizen. If members of the sqe are validated and then later
4519 * used, it's important that those reads are done through READ_ONCE() to
4520 * prevent a re-load down the line.
4521 */
3529d8c2
JA		 4522static bool io_get_sqring(struct io_ring_ctx *ctx, struct io_kiocb *req,
4523 const struct io_uring_sqe **sqe_ptr)
2b188cc1 4524{
75b28aff
HV		 4525 struct io_rings *rings = ctx->rings;
4526 u32 *sq_array = ctx->sq_array;
2b188cc1
JA		 4527 unsigned head;
4528
4529 /*
4530 * The cached sq head (or cq tail) serves two purposes:
4531 *
4532 * 1) allows us to batch the cost of updating the user visible
4533 * head updates.
4534 * 2) allows the kernel side to track the head on its own, even
4535 * though the application is the one updating it.
4536 */
4537 head = ctx->cached_sq_head;
e523a29c 4538 /* make sure SQ entry isn't read before tail */
9835d6fa 4539 if (unlikely(head == smp_load_acquire(&rings->sq.tail)))
2b188cc1
JA		 4540 return false;
4541
75b28aff 4542 head = READ_ONCE(sq_array[head & ctx->sq_mask]);
9835d6fa 4543 if (likely(head < ctx->sq_entries)) {
cf6fd4bd
PB		 4544 /*
4545 * All io need record the previous position, if LINK vs DARIN,
4546 * it can be used to mark the position of the first IO in the
4547 * link list.
4548 */
4549 req->sequence = ctx->cached_sq_head;
3529d8c2
JA		 4550 *sqe_ptr = &ctx->sq_sqes[head];
4551 req->opcode = READ_ONCE((*sqe_ptr)->opcode);
4552 req->user_data = READ_ONCE((*sqe_ptr)->user_data);
2b188cc1
JA		 4553 ctx->cached_sq_head++;
4554 return true;
4555 }
4556
4557 /* drop invalid entries */
4558 ctx->cached_sq_head++;
498ccd9e
JA		 4559 ctx->cached_sq_dropped++;
4560 WRITE_ONCE(rings->sq_dropped, ctx->cached_sq_dropped);
2b188cc1
JA		 4561 return false;
4562}
4563
fb5ccc98 4564static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
ae9428ca
PB		 4565 struct file *ring_file, int ring_fd,
4566 struct mm_struct **mm, bool async)
6c271ce2
JA		 4567{
4568 struct io_submit_state state, *statep = NULL;
9e645e11 4569 struct io_kiocb *link = NULL;
9e645e11 4570 int i, submitted = 0;
95a1b3ff 4571 bool mm_fault = false;
6c271ce2 4572
c4a2ed72 4573 /* if we have a backlog and couldn't flush it all, return BUSY */
ad3eb2c8
JA		 4574 if (test_bit(0, &ctx->sq_check_overflow)) {
4575 if (!list_empty(&ctx->cq_overflow_list) &&
4576 !io_cqring_overflow_flush(ctx, false))
4577 return -EBUSY;
4578 }
6c271ce2 4579
9ef4f124
PB		 4580 nr = min(nr, ctx->sq_entries);
4581
2b85edfc
PB		 4582 if (!percpu_ref_tryget_many(&ctx->refs, nr))
4583 return -EAGAIN;
4584
6c271ce2 4585 if (nr > IO_PLUG_THRESHOLD) {
22efde59 4586 io_submit_state_start(&state, nr);
6c271ce2
JA		 4587 statep = &state;
4588 }
4589
4590 for (i = 0; i < nr; i++) {
3529d8c2 4591 const struct io_uring_sqe *sqe;
196be95c 4592 struct io_kiocb *req;
fb5ccc98 4593
196be95c
PB		 4594 req = io_get_req(ctx, statep);
4595 if (unlikely(!req)) {
4596 if (!submitted)
4597 submitted = -EAGAIN;
fb5ccc98 4598 break;
196be95c 4599 }
3529d8c2 4600 if (!io_get_sqring(ctx, req, &sqe)) {
2b85edfc 4601 __io_req_do_free(req);
196be95c
PB		 4602 break;
4603 }
fb5ccc98 4604
d3656344
JA		 4605 /* will complete beyond this point, count as submitted */
4606 submitted++;
4607
4608 if (unlikely(req->opcode >= IORING_OP_LAST)) {
4609 io_cqring_add_event(req, -EINVAL);
4610 io_double_put_req(req);
4611 break;
4612 }
4613
4614 if (io_op_defs[req->opcode].needs_mm && !*mm) {
95a1b3ff
PB		 4615 mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm);
4616 if (!mm_fault) {
4617 use_mm(ctx->sqo_mm);
4618 *mm = ctx->sqo_mm;
4619 }
9e645e11 4620 }
9e645e11 4621
cf6fd4bd
PB		 4622 req->ring_file = ring_file;
4623 req->ring_fd = ring_fd;
4624 req->has_user = *mm != NULL;
4625 req->in_async = async;
4626 req->needs_fixed_file = async;
d625c6ee 4627 trace_io_uring_submit_sqe(ctx, req->user_data, true, async);
3529d8c2 4628 if (!io_submit_sqe(req, sqe, statep, &link))
2e6e1fde 4629 break;
6c271ce2
JA		 4630 }
4631
2b85edfc
PB		 4632 if (submitted != nr)
4633 percpu_ref_put_many(&ctx->refs, nr - submitted);
9e645e11 4634 if (link)
1b4a51b6 4635 io_queue_link_head(link);
6c271ce2
JA		 4636 if (statep)
4637 io_submit_state_end(&state);
4638
ae9428ca
PB		 4639 /* Commit SQ ring head once we've consumed and submitted all SQEs */
4640 io_commit_sqring(ctx);
4641
6c271ce2
JA		 4642 return submitted;
4643}
4644
4645static int io_sq_thread(void *data)
4646{
6c271ce2
JA		 4647 struct io_ring_ctx *ctx = data;
4648 struct mm_struct *cur_mm = NULL;
181e448d 4649 const struct cred *old_cred;
6c271ce2
JA		 4650 mm_segment_t old_fs;
4651 DEFINE_WAIT(wait);
4652 unsigned inflight;
4653 unsigned long timeout;
c1edbf5f 4654 int ret;
6c271ce2 4655
206aefde 4656 complete(&ctx->completions[1]);
a4c0b3de 4657
6c271ce2
JA		 4658 old_fs = get_fs();
4659 set_fs(USER_DS);
181e448d 4660 old_cred = override_creds(ctx->creds);
6c271ce2 4661
c1edbf5f 4662 ret = timeout = inflight = 0;
2bbcd6d3 4663 while (!kthread_should_park()) {
fb5ccc98 4664 unsigned int to_submit;
6c271ce2
JA		 4665
4666 if (inflight) {
4667 unsigned nr_events = 0;
4668
4669 if (ctx->flags & IORING_SETUP_IOPOLL) {
2b2ed975
JA		 4670 /*
4671 * inflight is the count of the maximum possible
4672 * entries we submitted, but it can be smaller
4673 * if we dropped some of them. If we don't have
4674 * poll entries available, then we know that we
4675 * have nothing left to poll for. Reset the
4676 * inflight count to zero in that case.
4677 */
4678 mutex_lock(&ctx->uring_lock);
4679 if (!list_empty(&ctx->poll_list))
4680 __io_iopoll_check(ctx, &nr_events, 0);
4681 else
4682 inflight = 0;
4683 mutex_unlock(&ctx->uring_lock);
6c271ce2
JA		 4684 } else {
4685 /*
4686 * Normal IO, just pretend everything completed.
4687 * We don't have to poll completions for that.
4688 */
4689 nr_events = inflight;
4690 }
4691
4692 inflight -= nr_events;
4693 if (!inflight)
4694 timeout = jiffies + ctx->sq_thread_idle;
4695 }
4696
fb5ccc98 4697 to_submit = io_sqring_entries(ctx);
c1edbf5f
JA		 4698
4699 /*
4700 * If submit got -EBUSY, flag us as needing the application
4701 * to enter the kernel to reap and flush events.
4702 */
4703 if (!to_submit || ret == -EBUSY) {
6c271ce2
JA		 4704 /*
4705 * We're polling. If we're within the defined idle
4706 * period, then let us spin without work before going
c1edbf5f
JA		 4707 * to sleep. The exception is if we got EBUSY doing
4708 * more IO, we should wait for the application to
4709 * reap events and wake us up.
6c271ce2 4710 */
c1edbf5f
JA		 4711 if (inflight ||
4712 (!time_after(jiffies, timeout) && ret != -EBUSY)) {
9831a90c 4713 cond_resched();
6c271ce2
JA		 4714 continue;
4715 }
4716
4717 /*
4718 * Drop cur_mm before scheduling, we can't hold it for
4719 * long periods (or over schedule()). Do this before
4720 * adding ourselves to the waitqueue, as the unuse/drop
4721 * may sleep.
4722 */
4723 if (cur_mm) {
4724 unuse_mm(cur_mm);
4725 mmput(cur_mm);
4726 cur_mm = NULL;
4727 }
4728
4729 prepare_to_wait(&ctx->sqo_wait, &wait,
4730 TASK_INTERRUPTIBLE);
4731
4732 /* Tell userspace we may need a wakeup call */
75b28aff 4733 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
0d7bae69
SB		 4734 /* make sure to read SQ tail after writing flags */
4735 smp_mb();
6c271ce2 4736
fb5ccc98 4737 to_submit = io_sqring_entries(ctx);
c1edbf5f 4738 if (!to_submit || ret == -EBUSY) {
2bbcd6d3 4739 if (kthread_should_park()) {
6c271ce2
JA		 4740 finish_wait(&ctx->sqo_wait, &wait);
4741 break;
4742 }
4743 if (signal_pending(current))
4744 flush_signals(current);
4745 schedule();
4746 finish_wait(&ctx->sqo_wait, &wait);
4747
75b28aff 4748 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6c271ce2
JA		 4749 continue;
4750 }
4751 finish_wait(&ctx->sqo_wait, &wait);
4752
75b28aff 4753 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6c271ce2
JA		 4754 }
4755
8a4955ff 4756 mutex_lock(&ctx->uring_lock);
1d7bb1d5 4757 ret = io_submit_sqes(ctx, to_submit, NULL, -1, &cur_mm, true);
8a4955ff 4758 mutex_unlock(&ctx->uring_lock);
1d7bb1d5
JA		 4759 if (ret > 0)
4760 inflight += ret;
6c271ce2
JA		 4761 }
4762
4763 set_fs(old_fs);
4764 if (cur_mm) {
4765 unuse_mm(cur_mm);
4766 mmput(cur_mm);
4767 }
181e448d 4768 revert_creds(old_cred);
06058632 4769
2bbcd6d3 4770 kthread_parkme();
06058632 4771
6c271ce2
JA		 4772 return 0;
4773}
4774
bda52162
JA		 4775struct io_wait_queue {
4776 struct wait_queue_entry wq;
4777 struct io_ring_ctx *ctx;
4778 unsigned to_wait;
4779 unsigned nr_timeouts;
4780};
4781
1d7bb1d5 4782static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
bda52162
JA		 4783{
4784 struct io_ring_ctx *ctx = iowq->ctx;
4785
4786 /*
d195a66e 4787 * Wake up if we have enough events, or if a timeout occurred since we
bda52162
JA		 4788 * started waiting. For timeouts, we always want to return to userspace,
4789 * regardless of event count.
4790 */
1d7bb1d5 4791 return io_cqring_events(ctx, noflush) >= iowq->to_wait ||
bda52162
JA		 4792 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
4793}
4794
4795static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
4796 int wake_flags, void *key)
4797{
4798 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
4799 wq);
4800
1d7bb1d5
JA		 4801 /* use noflush == true, as we can't safely rely on locking context */
4802 if (!io_should_wake(iowq, true))
bda52162
JA		 4803 return -1;
4804
4805 return autoremove_wake_function(curr, mode, wake_flags, key);
4806}
4807
2b188cc1
JA		 4808/*
4809 * Wait until events become available, if we don't already have some. The
4810 * application must reap them itself, as they reside on the shared cq ring.
4811 */
4812static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
4813 const sigset_t __user *sig, size_t sigsz)
4814{
bda52162
JA		 4815 struct io_wait_queue iowq = {
4816 .wq = {
4817 .private = current,
4818 .func = io_wake_function,
4819 .entry = LIST_HEAD_INIT(iowq.wq.entry),
4820 },
4821 .ctx = ctx,
4822 .to_wait = min_events,
4823 };
75b28aff 4824 struct io_rings *rings = ctx->rings;
e9ffa5c2 4825 int ret = 0;
2b188cc1 4826
1d7bb1d5 4827 if (io_cqring_events(ctx, false) >= min_events)
2b188cc1
JA		 4828 return 0;
4829
4830 if (sig) {
9e75ad5d
AB		 4831#ifdef CONFIG_COMPAT
4832 if (in_compat_syscall())
4833 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
b772434b 4834 sigsz);
9e75ad5d
AB		 4835 else
4836#endif
b772434b 4837 ret = set_user_sigmask(sig, sigsz);
9e75ad5d 4838
2b188cc1
JA		 4839 if (ret)
4840 return ret;
4841 }
4842
bda52162 4843 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
c826bd7a 4844 trace_io_uring_cqring_wait(ctx, min_events);
bda52162
JA		 4845 do {
4846 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
4847 TASK_INTERRUPTIBLE);
1d7bb1d5 4848 if (io_should_wake(&iowq, false))
bda52162
JA		 4849 break;
4850 schedule();
4851 if (signal_pending(current)) {
e9ffa5c2 4852 ret = -EINTR;
bda52162
JA		 4853 break;
4854 }
4855 } while (1);
4856 finish_wait(&ctx->wait, &iowq.wq);
4857
e9ffa5c2 4858 restore_saved_sigmask_unless(ret == -EINTR);
2b188cc1 4859
75b28aff 4860 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
2b188cc1
JA		 4861}
4862
6b06314c
JA		 4863static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
4864{
4865#if defined(CONFIG_UNIX)
4866 if (ctx->ring_sock) {
4867 struct sock *sock = ctx->ring_sock->sk;
4868 struct sk_buff *skb;
4869
4870 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
4871 kfree_skb(skb);
4872 }
4873#else
4874 int i;
4875
65e19f54
JA		 4876 for (i = 0; i < ctx->nr_user_files; i++) {
4877 struct file *file;
4878
4879 file = io_file_from_index(ctx, i);
4880 if (file)
4881 fput(file);
4882 }
6b06314c
JA		 4883#endif
4884}
4885
05f3fb3c
JA		 4886static void io_file_ref_kill(struct percpu_ref *ref)
4887{
4888 struct fixed_file_data *data;
4889
4890 data = container_of(ref, struct fixed_file_data, refs);
4891 complete(&data->done);
4892}
4893
6b06314c
JA		 4894static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
4895{
05f3fb3c 4896 struct fixed_file_data *data = ctx->file_data;
65e19f54
JA		 4897 unsigned nr_tables, i;
4898
05f3fb3c 4899 if (!data)
6b06314c
JA		 4900 return -ENXIO;
4901
05f3fb3c
JA		 4902 /* protect against inflight atomic switch, which drops the ref */
4903 flush_work(&data->ref_work);
4904 percpu_ref_get(&data->refs);
4905 percpu_ref_kill_and_confirm(&data->refs, io_file_ref_kill);
4906 wait_for_completion(&data->done);
4907 percpu_ref_put(&data->refs);
4908 percpu_ref_exit(&data->refs);
4909
6b06314c 4910 __io_sqe_files_unregister(ctx);
65e19f54
JA		 4911 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
4912 for (i = 0; i < nr_tables; i++)
05f3fb3c
JA		 4913 kfree(data->table[i].files);
4914 kfree(data->table);
4915 kfree(data);
4916 ctx->file_data = NULL;
6b06314c
JA		 4917 ctx->nr_user_files = 0;
4918 return 0;
4919}
4920
6c271ce2
JA		 4921static void io_sq_thread_stop(struct io_ring_ctx *ctx)
4922{
4923 if (ctx->sqo_thread) {
206aefde 4924 wait_for_completion(&ctx->completions[1]);
2bbcd6d3
RP		 4925 /*
4926 * The park is a bit of a work-around, without it we get
4927 * warning spews on shutdown with SQPOLL set and affinity
4928 * set to a single CPU.
4929 */
06058632 4930 kthread_park(ctx->sqo_thread);
6c271ce2
JA		 4931 kthread_stop(ctx->sqo_thread);
4932 ctx->sqo_thread = NULL;
4933 }
4934}
4935
6b06314c
JA		 4936static void io_finish_async(struct io_ring_ctx *ctx)
4937{
6c271ce2
JA		 4938 io_sq_thread_stop(ctx);
4939
561fb04a
JA		 4940 if (ctx->io_wq) {
4941 io_wq_destroy(ctx->io_wq);
4942 ctx->io_wq = NULL;
6b06314c
JA		 4943 }
4944}
4945
4946#if defined(CONFIG_UNIX)
6b06314c
JA		 4947/*
4948 * Ensure the UNIX gc is aware of our file set, so we are certain that
4949 * the io_uring can be safely unregistered on process exit, even if we have
4950 * loops in the file referencing.
4951 */
4952static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
4953{
4954 struct sock *sk = ctx->ring_sock->sk;
4955 struct scm_fp_list *fpl;
4956 struct sk_buff *skb;
08a45173 4957 int i, nr_files;
6b06314c
JA		 4958
4959 if (!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) {
4960 unsigned long inflight = ctx->user->unix_inflight + nr;
4961
4962 if (inflight > task_rlimit(current, RLIMIT_NOFILE))
4963 return -EMFILE;
4964 }
4965
4966 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
4967 if (!fpl)
4968 return -ENOMEM;
4969
4970 skb = alloc_skb(0, GFP_KERNEL);
4971 if (!skb) {
4972 kfree(fpl);
4973 return -ENOMEM;
4974 }
4975
4976 skb->sk = sk;
6b06314c 4977
08a45173 4978 nr_files = 0;
6b06314c
JA		 4979 fpl->user = get_uid(ctx->user);
4980 for (i = 0; i < nr; i++) {
65e19f54
JA		 4981 struct file *file = io_file_from_index(ctx, i + offset);
4982
4983 if (!file)
08a45173 4984 continue;
65e19f54 4985 fpl->fp[nr_files] = get_file(file);
08a45173
JA		 4986 unix_inflight(fpl->user, fpl->fp[nr_files]);
4987 nr_files++;
6b06314c
JA		 4988 }
4989
08a45173
JA		 4990 if (nr_files) {
4991 fpl->max = SCM_MAX_FD;
4992 fpl->count = nr_files;
4993 UNIXCB(skb).fp = fpl;
05f3fb3c 4994 skb->destructor = unix_destruct_scm;
08a45173
JA		 4995 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
4996 skb_queue_head(&sk->sk_receive_queue, skb);
6b06314c 4997
08a45173
JA		 4998 for (i = 0; i < nr_files; i++)
4999 fput(fpl->fp[i]);
5000 } else {
5001 kfree_skb(skb);
5002 kfree(fpl);
5003 }
6b06314c
JA		 5004
5005 return 0;
5006}
5007
5008/*
5009 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
5010 * causes regular reference counting to break down. We rely on the UNIX
5011 * garbage collection to take care of this problem for us.
5012 */
5013static int io_sqe_files_scm(struct io_ring_ctx *ctx)
5014{
5015 unsigned left, total;
5016 int ret = 0;
5017
5018 total = 0;
5019 left = ctx->nr_user_files;
5020 while (left) {
5021 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
6b06314c
JA		 5022
5023 ret = __io_sqe_files_scm(ctx, this_files, total);
5024 if (ret)
5025 break;
5026 left -= this_files;
5027 total += this_files;
5028 }
5029
5030 if (!ret)
5031 return 0;
5032
5033 while (total < ctx->nr_user_files) {
65e19f54
JA		 5034 struct file *file = io_file_from_index(ctx, total);
5035
5036 if (file)
5037 fput(file);
6b06314c
JA		 5038 total++;
5039 }
5040
5041 return ret;
5042}
5043#else
5044static int io_sqe_files_scm(struct io_ring_ctx *ctx)
5045{
5046 return 0;
5047}
5048#endif
5049
65e19f54
JA		 5050static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables,
5051 unsigned nr_files)
5052{
5053 int i;
5054
5055 for (i = 0; i < nr_tables; i++) {
05f3fb3c 5056 struct fixed_file_table *table = &ctx->file_data->table[i];
65e19f54
JA		 5057 unsigned this_files;
5058
5059 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
5060 table->files = kcalloc(this_files, sizeof(struct file *),
5061 GFP_KERNEL);
5062 if (!table->files)
5063 break;
5064 nr_files -= this_files;
5065 }
5066
5067 if (i == nr_tables)
5068 return 0;
5069
5070 for (i = 0; i < nr_tables; i++) {
05f3fb3c 5071 struct fixed_file_table *table = &ctx->file_data->table[i];
65e19f54
JA		 5072 kfree(table->files);
5073 }
5074 return 1;
5075}
5076
05f3fb3c
JA		 5077static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file)
5078{
5079#if defined(CONFIG_UNIX)
5080 struct sock *sock = ctx->ring_sock->sk;
5081 struct sk_buff_head list, *head = &sock->sk_receive_queue;
5082 struct sk_buff *skb;
5083 int i;
5084
5085 __skb_queue_head_init(&list);
5086
5087 /*
5088 * Find the skb that holds this file in its SCM_RIGHTS. When found,
5089 * remove this entry and rearrange the file array.
5090 */
5091 skb = skb_dequeue(head);
5092 while (skb) {
5093 struct scm_fp_list *fp;
5094
5095 fp = UNIXCB(skb).fp;
5096 for (i = 0; i < fp->count; i++) {
5097 int left;
5098
5099 if (fp->fp[i] != file)
5100 continue;
5101
5102 unix_notinflight(fp->user, fp->fp[i]);
5103 left = fp->count - 1 - i;
5104 if (left) {
5105 memmove(&fp->fp[i], &fp->fp[i + 1],
5106 left * sizeof(struct file *));
5107 }
5108 fp->count--;
5109 if (!fp->count) {
5110 kfree_skb(skb);
5111 skb = NULL;
5112 } else {
5113 __skb_queue_tail(&list, skb);
5114 }
5115 fput(file);
5116 file = NULL;
5117 break;
5118 }
5119
5120 if (!file)
5121 break;
5122
5123 __skb_queue_tail(&list, skb);
5124
5125 skb = skb_dequeue(head);
5126 }
5127
5128 if (skb_peek(&list)) {
5129 spin_lock_irq(&head->lock);
5130 while ((skb = __skb_dequeue(&list)) != NULL)
5131 __skb_queue_tail(head, skb);
5132 spin_unlock_irq(&head->lock);
5133 }
5134#else
5135 fput(file);
5136#endif
5137}
5138
5139struct io_file_put {
5140 struct llist_node llist;
5141 struct file *file;
5142 struct completion *done;
5143};
5144
5145static void io_ring_file_ref_switch(struct work_struct *work)
5146{
5147 struct io_file_put *pfile, *tmp;
5148 struct fixed_file_data *data;
5149 struct llist_node *node;
5150
5151 data = container_of(work, struct fixed_file_data, ref_work);
5152
5153 while ((node = llist_del_all(&data->put_llist)) != NULL) {
5154 llist_for_each_entry_safe(pfile, tmp, node, llist) {
5155 io_ring_file_put(data->ctx, pfile->file);
5156 if (pfile->done)
5157 complete(pfile->done);
5158 else
5159 kfree(pfile);
5160 }
5161 }
5162
5163 percpu_ref_get(&data->refs);
5164 percpu_ref_switch_to_percpu(&data->refs);
5165}
5166
5167static void io_file_data_ref_zero(struct percpu_ref *ref)
5168{
5169 struct fixed_file_data *data;
5170
5171 data = container_of(ref, struct fixed_file_data, refs);
5172
5173 /* we can't safely switch from inside this context, punt to wq */
5174 queue_work(system_wq, &data->ref_work);
5175}
5176
6b06314c
JA		 5177static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
5178 unsigned nr_args)
5179{
5180 __s32 __user *fds = (__s32 __user *) arg;
65e19f54 5181 unsigned nr_tables;
05f3fb3c 5182 struct file *file;
6b06314c
JA		 5183 int fd, ret = 0;
5184 unsigned i;
5185
05f3fb3c 5186 if (ctx->file_data)
6b06314c
JA		 5187 return -EBUSY;
5188 if (!nr_args)
5189 return -EINVAL;
5190 if (nr_args > IORING_MAX_FIXED_FILES)
5191 return -EMFILE;
5192
05f3fb3c
JA		 5193 ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL);
5194 if (!ctx->file_data)
5195 return -ENOMEM;
5196 ctx->file_data->ctx = ctx;
5197 init_completion(&ctx->file_data->done);
5198
65e19f54 5199 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
05f3fb3c
JA		 5200 ctx->file_data->table = kcalloc(nr_tables,
5201 sizeof(struct fixed_file_table),
65e19f54 5202 GFP_KERNEL);
05f3fb3c
JA		 5203 if (!ctx->file_data->table) {
5204 kfree(ctx->file_data);
5205 ctx->file_data = NULL;
6b06314c 5206 return -ENOMEM;
05f3fb3c
JA		 5207 }
5208
5209 if (percpu_ref_init(&ctx->file_data->refs, io_file_data_ref_zero,
5210 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
5211 kfree(ctx->file_data->table);
5212 kfree(ctx->file_data);
5213 ctx->file_data = NULL;
5214 return -ENOMEM;
5215 }
5216 ctx->file_data->put_llist.first = NULL;
5217 INIT_WORK(&ctx->file_data->ref_work, io_ring_file_ref_switch);
6b06314c 5218
65e19f54 5219 if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) {
05f3fb3c
JA		 5220 percpu_ref_exit(&ctx->file_data->refs);
5221 kfree(ctx->file_data->table);
5222 kfree(ctx->file_data);
5223 ctx->file_data = NULL;
65e19f54
JA		 5224 return -ENOMEM;
5225 }
5226
08a45173 5227 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
65e19f54
JA		 5228 struct fixed_file_table *table;
5229 unsigned index;
5230
6b06314c
JA		 5231 ret = -EFAULT;
5232 if (copy_from_user(&fd, &fds[i], sizeof(fd)))
5233 break;
08a45173
JA		 5234 /* allow sparse sets */
5235 if (fd == -1) {
5236 ret = 0;
5237 continue;
5238 }
6b06314c 5239
05f3fb3c 5240 table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT];
65e19f54 5241 index = i & IORING_FILE_TABLE_MASK;
05f3fb3c 5242 file = fget(fd);
6b06314c
JA		 5243
5244 ret = -EBADF;
05f3fb3c 5245 if (!file)
6b06314c 5246 break;
05f3fb3c 5247
6b06314c
JA		 5248 /*
5249 * Don't allow io_uring instances to be registered. If UNIX
5250 * isn't enabled, then this causes a reference cycle and this
5251 * instance can never get freed. If UNIX is enabled we'll
5252 * handle it just fine, but there's still no point in allowing
5253 * a ring fd as it doesn't support regular read/write anyway.
5254 */
05f3fb3c
JA		 5255 if (file->f_op == &io_uring_fops) {
5256 fput(file);
6b06314c
JA		 5257 break;
5258 }
6b06314c 5259 ret = 0;
05f3fb3c 5260 table->files[index] = file;
6b06314c
JA		 5261 }
5262
5263 if (ret) {
65e19f54 5264 for (i = 0; i < ctx->nr_user_files; i++) {
65e19f54
JA		 5265 file = io_file_from_index(ctx, i);
5266 if (file)
5267 fput(file);
5268 }
5269 for (i = 0; i < nr_tables; i++)
05f3fb3c 5270 kfree(ctx->file_data->table[i].files);
6b06314c 5271
05f3fb3c
JA		 5272 kfree(ctx->file_data->table);
5273 kfree(ctx->file_data);
5274 ctx->file_data = NULL;
6b06314c
JA		 5275 ctx->nr_user_files = 0;
5276 return ret;
5277 }
5278
5279 ret = io_sqe_files_scm(ctx);
5280 if (ret)
5281 io_sqe_files_unregister(ctx);
5282
5283 return ret;
5284}
5285
c3a31e60
JA		 5286static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
5287 int index)
5288{
5289#if defined(CONFIG_UNIX)
5290 struct sock *sock = ctx->ring_sock->sk;
5291 struct sk_buff_head *head = &sock->sk_receive_queue;
5292 struct sk_buff *skb;
5293
5294 /*
5295 * See if we can merge this file into an existing skb SCM_RIGHTS
5296 * file set. If there's no room, fall back to allocating a new skb
5297 * and filling it in.
5298 */
5299 spin_lock_irq(&head->lock);
5300 skb = skb_peek(head);
5301 if (skb) {
5302 struct scm_fp_list *fpl = UNIXCB(skb).fp;
5303
5304 if (fpl->count < SCM_MAX_FD) {
5305 __skb_unlink(skb, head);
5306 spin_unlock_irq(&head->lock);
5307 fpl->fp[fpl->count] = get_file(file);
5308 unix_inflight(fpl->user, fpl->fp[fpl->count]);
5309 fpl->count++;
5310 spin_lock_irq(&head->lock);
5311 __skb_queue_head(head, skb);
5312 } else {
5313 skb = NULL;
5314 }
5315 }
5316 spin_unlock_irq(&head->lock);
5317
5318 if (skb) {
5319 fput(file);
5320 return 0;
5321 }
5322
5323 return __io_sqe_files_scm(ctx, 1, index);
5324#else
5325 return 0;
5326#endif
5327}
5328
05f3fb3c 5329static void io_atomic_switch(struct percpu_ref *ref)
c3a31e60 5330{
05f3fb3c
JA		 5331 struct fixed_file_data *data;
5332
5333 data = container_of(ref, struct fixed_file_data, refs);
5334 clear_bit(FFD_F_ATOMIC, &data->state);
5335}
5336
5337static bool io_queue_file_removal(struct fixed_file_data *data,
5338 struct file *file)
5339{
5340 struct io_file_put *pfile, pfile_stack;
5341 DECLARE_COMPLETION_ONSTACK(done);
5342
5343 /*
5344 * If we fail allocating the struct we need for doing async reomval
5345 * of this file, just punt to sync and wait for it.
5346 */
5347 pfile = kzalloc(sizeof(*pfile), GFP_KERNEL);
5348 if (!pfile) {
5349 pfile = &pfile_stack;
5350 pfile->done = &done;
5351 }
5352
5353 pfile->file = file;
5354 llist_add(&pfile->llist, &data->put_llist);
5355
5356 if (pfile == &pfile_stack) {
5357 if (!test_and_set_bit(FFD_F_ATOMIC, &data->state)) {
5358 percpu_ref_put(&data->refs);
5359 percpu_ref_switch_to_atomic(&data->refs,
5360 io_atomic_switch);
5361 }
5362 wait_for_completion(&done);
5363 flush_work(&data->ref_work);
5364 return false;
5365 }
5366
5367 return true;
5368}
5369
5370static int __io_sqe_files_update(struct io_ring_ctx *ctx,
5371 struct io_uring_files_update *up,
5372 unsigned nr_args)
5373{
5374 struct fixed_file_data *data = ctx->file_data;
5375 bool ref_switch = false;
5376 struct file *file;
c3a31e60
JA		 5377 __s32 __user *fds;
5378 int fd, i, err;
5379 __u32 done;
5380
05f3fb3c 5381 if (check_add_overflow(up->offset, nr_args, &done))
c3a31e60
JA		 5382 return -EOVERFLOW;
5383 if (done > ctx->nr_user_files)
5384 return -EINVAL;
5385
5386 done = 0;
05f3fb3c 5387 fds = u64_to_user_ptr(up->fds);
c3a31e60 5388 while (nr_args) {
65e19f54
JA		 5389 struct fixed_file_table *table;
5390 unsigned index;
5391
c3a31e60
JA		 5392 err = 0;
5393 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
5394 err = -EFAULT;
5395 break;
5396 }
05f3fb3c
JA		 5397 i = array_index_nospec(up->offset, ctx->nr_user_files);
5398 table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT];
65e19f54
JA		 5399 index = i & IORING_FILE_TABLE_MASK;
5400 if (table->files[index]) {
05f3fb3c 5401 file = io_file_from_index(ctx, index);
65e19f54 5402 table->files[index] = NULL;
05f3fb3c
JA		 5403 if (io_queue_file_removal(data, file))
5404 ref_switch = true;
c3a31e60
JA		 5405 }
5406 if (fd != -1) {
c3a31e60
JA		 5407 file = fget(fd);
5408 if (!file) {
5409 err = -EBADF;
5410 break;
5411 }
5412 /*
5413 * Don't allow io_uring instances to be registered. If
5414 * UNIX isn't enabled, then this causes a reference
5415 * cycle and this instance can never get freed. If UNIX
5416 * is enabled we'll handle it just fine, but there's
5417 * still no point in allowing a ring fd as it doesn't
5418 * support regular read/write anyway.
5419 */
5420 if (file->f_op == &io_uring_fops) {
5421 fput(file);
5422 err = -EBADF;
5423 break;
5424 }
65e19f54 5425 table->files[index] = file;
c3a31e60
JA		 5426 err = io_sqe_file_register(ctx, file, i);
5427 if (err)
5428 break;
5429 }
5430 nr_args--;
5431 done++;
05f3fb3c
JA		 5432 up->offset++;
5433 }
5434
5435 if (ref_switch && !test_and_set_bit(FFD_F_ATOMIC, &data->state)) {
5436 percpu_ref_put(&data->refs);
5437 percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch);
c3a31e60
JA		 5438 }
5439
5440 return done ? done : err;
5441}
05f3fb3c
JA		 5442static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
5443 unsigned nr_args)
5444{
5445 struct io_uring_files_update up;
5446
5447 if (!ctx->file_data)
5448 return -ENXIO;
5449 if (!nr_args)
5450 return -EINVAL;
5451 if (copy_from_user(&up, arg, sizeof(up)))
5452 return -EFAULT;
5453 if (up.resv)
5454 return -EINVAL;
5455
5456 return __io_sqe_files_update(ctx, &up, nr_args);
5457}
c3a31e60 5458
7d723065
JA		 5459static void io_put_work(struct io_wq_work *work)
5460{
5461 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
5462
5463 io_put_req(req);
5464}
5465
5466static void io_get_work(struct io_wq_work *work)
5467{
5468 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
5469
5470 refcount_inc(&req->refs);
5471}
5472
6c271ce2
JA		 5473static int io_sq_offload_start(struct io_ring_ctx *ctx,
5474 struct io_uring_params *p)
2b188cc1 5475{
576a347b 5476 struct io_wq_data data;
561fb04a 5477 unsigned concurrency;
2b188cc1
JA		 5478 int ret;
5479
6c271ce2 5480 init_waitqueue_head(&ctx->sqo_wait);
2b188cc1
JA		 5481 mmgrab(current->mm);
5482 ctx->sqo_mm = current->mm;
5483
6c271ce2 5484 if (ctx->flags & IORING_SETUP_SQPOLL) {
3ec482d1
JA		 5485 ret = -EPERM;
5486 if (!capable(CAP_SYS_ADMIN))
5487 goto err;
5488
917257da
JA		 5489 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
5490 if (!ctx->sq_thread_idle)
5491 ctx->sq_thread_idle = HZ;
5492
6c271ce2 5493 if (p->flags & IORING_SETUP_SQ_AFF) {
44a9bd18 5494 int cpu = p->sq_thread_cpu;
6c271ce2 5495
917257da 5496 ret = -EINVAL;
44a9bd18
JA		 5497 if (cpu >= nr_cpu_ids)
5498 goto err;
7889f44d 5499 if (!cpu_online(cpu))
917257da
JA		 5500 goto err;
5501
6c271ce2
JA		 5502 ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread,
5503 ctx, cpu,
5504 "io_uring-sq");
5505 } else {
5506 ctx->sqo_thread = kthread_create(io_sq_thread, ctx,
5507 "io_uring-sq");
5508 }
5509 if (IS_ERR(ctx->sqo_thread)) {
5510 ret = PTR_ERR(ctx->sqo_thread);
5511 ctx->sqo_thread = NULL;
5512 goto err;
5513 }
5514 wake_up_process(ctx->sqo_thread);
5515 } else if (p->flags & IORING_SETUP_SQ_AFF) {
5516 /* Can't have SQ_AFF without SQPOLL */
5517 ret = -EINVAL;
5518 goto err;
5519 }
5520
576a347b
JA		 5521 data.mm = ctx->sqo_mm;
5522 data.user = ctx->user;
181e448d 5523 data.creds = ctx->creds;
576a347b
JA		 5524 data.get_work = io_get_work;
5525 data.put_work = io_put_work;
5526
561fb04a
JA		 5527 /* Do QD, or 4 * CPUS, whatever is smallest */
5528 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
576a347b 5529 ctx->io_wq = io_wq_create(concurrency, &data);
975c99a5
JA		 5530 if (IS_ERR(ctx->io_wq)) {
5531 ret = PTR_ERR(ctx->io_wq);
5532 ctx->io_wq = NULL;
2b188cc1
JA		 5533 goto err;
5534 }
5535
5536 return 0;
5537err:
54a91f3b 5538 io_finish_async(ctx);
2b188cc1
JA		 5539 mmdrop(ctx->sqo_mm);
5540 ctx->sqo_mm = NULL;
5541 return ret;
5542}
5543
5544static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages)
5545{
5546 atomic_long_sub(nr_pages, &user->locked_vm);
5547}
5548
5549static int io_account_mem(struct user_struct *user, unsigned long nr_pages)
5550{
5551 unsigned long page_limit, cur_pages, new_pages;
5552
5553 /* Don't allow more pages than we can safely lock */
5554 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
5555
5556 do {
5557 cur_pages = atomic_long_read(&user->locked_vm);
5558 new_pages = cur_pages + nr_pages;
5559 if (new_pages > page_limit)
5560 return -ENOMEM;
5561 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
5562 new_pages) != cur_pages);
5563
5564 return 0;
5565}
5566
5567static void io_mem_free(void *ptr)
5568{
52e04ef4
MR		 5569 struct page *page;
5570
5571 if (!ptr)
5572 return;
2b188cc1 5573
52e04ef4 5574 page = virt_to_head_page(ptr);
2b188cc1
JA		 5575 if (put_page_testzero(page))
5576 free_compound_page(page);
5577}
5578
5579static void *io_mem_alloc(size_t size)
5580{
5581 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
5582 __GFP_NORETRY;
5583
5584 return (void *) __get_free_pages(gfp_flags, get_order(size));
5585}
5586
75b28aff
HV		 5587static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
5588 size_t *sq_offset)
5589{
5590 struct io_rings *rings;
5591 size_t off, sq_array_size;
5592
5593 off = struct_size(rings, cqes, cq_entries);
5594 if (off == SIZE_MAX)
5595 return SIZE_MAX;
5596
5597#ifdef CONFIG_SMP
5598 off = ALIGN(off, SMP_CACHE_BYTES);
5599 if (off == 0)
5600 return SIZE_MAX;
5601#endif
5602
5603 sq_array_size = array_size(sizeof(u32), sq_entries);
5604 if (sq_array_size == SIZE_MAX)
5605 return SIZE_MAX;
5606
5607 if (check_add_overflow(off, sq_array_size, &off))
5608 return SIZE_MAX;
5609
5610 if (sq_offset)
5611 *sq_offset = off;
5612
5613 return off;
5614}
5615
2b188cc1
JA		 5616static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries)
5617{
75b28aff 5618 size_t pages;
2b188cc1 5619
75b28aff
HV		 5620 pages = (size_t)1 << get_order(
5621 rings_size(sq_entries, cq_entries, NULL));
5622 pages += (size_t)1 << get_order(
5623 array_size(sizeof(struct io_uring_sqe), sq_entries));
2b188cc1 5624
75b28aff 5625 return pages;
2b188cc1
JA		 5626}
5627
edafccee
JA		 5628static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx)
5629{
5630 int i, j;
5631
5632 if (!ctx->user_bufs)
5633 return -ENXIO;
5634
5635 for (i = 0; i < ctx->nr_user_bufs; i++) {
5636 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
5637
5638 for (j = 0; j < imu->nr_bvecs; j++)
27c4d3a3 5639 put_user_page(imu->bvec[j].bv_page);
edafccee
JA		 5640
5641 if (ctx->account_mem)
5642 io_unaccount_mem(ctx->user, imu->nr_bvecs);
d4ef6475 5643 kvfree(imu->bvec);
edafccee
JA		 5644 imu->nr_bvecs = 0;
5645 }
5646
5647 kfree(ctx->user_bufs);
5648 ctx->user_bufs = NULL;
5649 ctx->nr_user_bufs = 0;
5650 return 0;
5651}
5652
5653static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
5654 void __user *arg, unsigned index)
5655{
5656 struct iovec __user *src;
5657
5658#ifdef CONFIG_COMPAT
5659 if (ctx->compat) {
5660 struct compat_iovec __user *ciovs;
5661 struct compat_iovec ciov;
5662
5663 ciovs = (struct compat_iovec __user *) arg;
5664 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
5665 return -EFAULT;
5666
d55e5f5b 5667 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
edafccee
JA		 5668 dst->iov_len = ciov.iov_len;
5669 return 0;
5670 }
5671#endif
5672 src = (struct iovec __user *) arg;
5673 if (copy_from_user(dst, &src[index], sizeof(*dst)))
5674 return -EFAULT;
5675 return 0;
5676}
5677
5678static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg,
5679 unsigned nr_args)
5680{
5681 struct vm_area_struct **vmas = NULL;
5682 struct page **pages = NULL;
5683 int i, j, got_pages = 0;
5684 int ret = -EINVAL;
5685
5686 if (ctx->user_bufs)
5687 return -EBUSY;
5688 if (!nr_args || nr_args > UIO_MAXIOV)
5689 return -EINVAL;
5690
5691 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
5692 GFP_KERNEL);
5693 if (!ctx->user_bufs)
5694 return -ENOMEM;
5695
5696 for (i = 0; i < nr_args; i++) {
5697 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
5698 unsigned long off, start, end, ubuf;
5699 int pret, nr_pages;
5700 struct iovec iov;
5701 size_t size;
5702
5703 ret = io_copy_iov(ctx, &iov, arg, i);
5704 if (ret)
a278682d 5705 goto err;
edafccee
JA		 5706
5707 /*
5708 * Don't impose further limits on the size and buffer
5709 * constraints here, we'll -EINVAL later when IO is
5710 * submitted if they are wrong.
5711 */
5712 ret = -EFAULT;
5713 if (!iov.iov_base || !iov.iov_len)
5714 goto err;
5715
5716 /* arbitrary limit, but we need something */
5717 if (iov.iov_len > SZ_1G)
5718 goto err;
5719
5720 ubuf = (unsigned long) iov.iov_base;
5721 end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
5722 start = ubuf >> PAGE_SHIFT;
5723 nr_pages = end - start;
5724
5725 if (ctx->account_mem) {
5726 ret = io_account_mem(ctx->user, nr_pages);
5727 if (ret)
5728 goto err;
5729 }
5730
5731 ret = 0;
5732 if (!pages || nr_pages > got_pages) {
5733 kfree(vmas);
5734 kfree(pages);
d4ef6475 5735 pages = kvmalloc_array(nr_pages, sizeof(struct page *),
edafccee 5736 GFP_KERNEL);
d4ef6475 5737 vmas = kvmalloc_array(nr_pages,
edafccee
JA		 5738 sizeof(struct vm_area_struct *),
5739 GFP_KERNEL);
5740 if (!pages || !vmas) {
5741 ret = -ENOMEM;
5742 if (ctx->account_mem)
5743 io_unaccount_mem(ctx->user, nr_pages);
5744 goto err;
5745 }
5746 got_pages = nr_pages;
5747 }
5748
d4ef6475 5749 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
edafccee
JA		 5750 GFP_KERNEL);
5751 ret = -ENOMEM;
5752 if (!imu->bvec) {
5753 if (ctx->account_mem)
5754 io_unaccount_mem(ctx->user, nr_pages);
5755 goto err;
5756 }
5757
5758 ret = 0;
5759 down_read(&current->mm->mmap_sem);
932f4a63
IW		 5760 pret = get_user_pages(ubuf, nr_pages,
5761 FOLL_WRITE | FOLL_LONGTERM,
5762 pages, vmas);
edafccee
JA		 5763 if (pret == nr_pages) {
5764 /* don't support file backed memory */
5765 for (j = 0; j < nr_pages; j++) {
5766 struct vm_area_struct *vma = vmas[j];
5767
5768 if (vma->vm_file &&
5769 !is_file_hugepages(vma->vm_file)) {
5770 ret = -EOPNOTSUPP;
5771 break;
5772 }
5773 }
5774 } else {
5775 ret = pret < 0 ? pret : -EFAULT;
5776 }
5777 up_read(&current->mm->mmap_sem);
5778 if (ret) {
5779 /*
5780 * if we did partial map, or found file backed vmas,
5781 * release any pages we did get
5782 */
27c4d3a3
JH		 5783 if (pret > 0)
5784 put_user_pages(pages, pret);
edafccee
JA		 5785 if (ctx->account_mem)
5786 io_unaccount_mem(ctx->user, nr_pages);
d4ef6475 5787 kvfree(imu->bvec);
edafccee
JA		 5788 goto err;
5789 }
5790
5791 off = ubuf & ~PAGE_MASK;
5792 size = iov.iov_len;
5793 for (j = 0; j < nr_pages; j++) {
5794 size_t vec_len;
5795
5796 vec_len = min_t(size_t, size, PAGE_SIZE - off);
5797 imu->bvec[j].bv_page = pages[j];
5798 imu->bvec[j].bv_len = vec_len;
5799 imu->bvec[j].bv_offset = off;
5800 off = 0;
5801 size -= vec_len;
5802 }
5803 /* store original address for later verification */
5804 imu->ubuf = ubuf;
5805 imu->len = iov.iov_len;
5806 imu->nr_bvecs = nr_pages;
5807
5808 ctx->nr_user_bufs++;
5809 }
d4ef6475
MR		 5810 kvfree(pages);
5811 kvfree(vmas);
edafccee
JA		 5812 return 0;
5813err:
d4ef6475
MR		 5814 kvfree(pages);
5815 kvfree(vmas);
edafccee
JA		 5816 io_sqe_buffer_unregister(ctx);
5817 return ret;
5818}
5819
9b402849
JA		 5820static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
5821{
5822 __s32 __user *fds = arg;
5823 int fd;
5824
5825 if (ctx->cq_ev_fd)
5826 return -EBUSY;
5827
5828 if (copy_from_user(&fd, fds, sizeof(*fds)))
5829 return -EFAULT;
5830
5831 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
5832 if (IS_ERR(ctx->cq_ev_fd)) {
5833 int ret = PTR_ERR(ctx->cq_ev_fd);
5834 ctx->cq_ev_fd = NULL;
5835 return ret;
5836 }
5837
5838 return 0;
5839}
5840
5841static int io_eventfd_unregister(struct io_ring_ctx *ctx)
5842{
5843 if (ctx->cq_ev_fd) {
5844 eventfd_ctx_put(ctx->cq_ev_fd);
5845 ctx->cq_ev_fd = NULL;
5846 return 0;
5847 }
5848
5849 return -ENXIO;
5850}
5851
2b188cc1
JA		 5852static void io_ring_ctx_free(struct io_ring_ctx *ctx)
5853{
6b06314c 5854 io_finish_async(ctx);
2b188cc1
JA		 5855 if (ctx->sqo_mm)
5856 mmdrop(ctx->sqo_mm);
def596e9
JA		 5857
5858 io_iopoll_reap_events(ctx);
edafccee 5859 io_sqe_buffer_unregister(ctx);
6b06314c 5860 io_sqe_files_unregister(ctx);
9b402849 5861 io_eventfd_unregister(ctx);
def596e9 5862
2b188cc1 5863#if defined(CONFIG_UNIX)
355e8d26
EB		 5864 if (ctx->ring_sock) {
5865 ctx->ring_sock->file = NULL; /* so that iput() is called */
2b188cc1 5866 sock_release(ctx->ring_sock);
355e8d26 5867 }
2b188cc1
JA		 5868#endif
5869
75b28aff 5870 io_mem_free(ctx->rings);
2b188cc1 5871 io_mem_free(ctx->sq_sqes);
2b188cc1
JA		 5872
5873 percpu_ref_exit(&ctx->refs);
5874 if (ctx->account_mem)
5875 io_unaccount_mem(ctx->user,
5876 ring_pages(ctx->sq_entries, ctx->cq_entries));
5877 free_uid(ctx->user);
181e448d 5878 put_cred(ctx->creds);
206aefde 5879 kfree(ctx->completions);
78076bb6 5880 kfree(ctx->cancel_hash);
0ddf92e8 5881 kmem_cache_free(req_cachep, ctx->fallback_req);
2b188cc1
JA		 5882 kfree(ctx);
5883}
5884
5885static __poll_t io_uring_poll(struct file *file, poll_table *wait)
5886{
5887 struct io_ring_ctx *ctx = file->private_data;
5888 __poll_t mask = 0;
5889
5890 poll_wait(file, &ctx->cq_wait, wait);
4f7067c3
SB		 5891 /*
5892 * synchronizes with barrier from wq_has_sleeper call in
5893 * io_commit_cqring
5894 */
2b188cc1 5895 smp_rmb();
75b28aff
HV		 5896 if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head !=
5897 ctx->rings->sq_ring_entries)
2b188cc1 5898 mask |= EPOLLOUT | EPOLLWRNORM;
daa5de54 5899 if (READ_ONCE(ctx->rings->cq.head) != ctx->cached_cq_tail)
2b188cc1
JA		 5900 mask |= EPOLLIN | EPOLLRDNORM;
5901
5902 return mask;
5903}
5904
5905static int io_uring_fasync(int fd, struct file *file, int on)
5906{
5907 struct io_ring_ctx *ctx = file->private_data;
5908
5909 return fasync_helper(fd, file, on, &ctx->cq_fasync);
5910}
5911
5912static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
5913{
5914 mutex_lock(&ctx->uring_lock);
5915 percpu_ref_kill(&ctx->refs);
5916 mutex_unlock(&ctx->uring_lock);
5917
5262f567 5918 io_kill_timeouts(ctx);
221c5eb2 5919 io_poll_remove_all(ctx);
561fb04a
JA		 5920
5921 if (ctx->io_wq)
5922 io_wq_cancel_all(ctx->io_wq);
5923
def596e9 5924 io_iopoll_reap_events(ctx);
15dff286
JA		 5925 /* if we failed setting up the ctx, we might not have any rings */
5926 if (ctx->rings)
5927 io_cqring_overflow_flush(ctx, true);
206aefde 5928 wait_for_completion(&ctx->completions[0]);
2b188cc1
JA		 5929 io_ring_ctx_free(ctx);
5930}
5931
5932static int io_uring_release(struct inode *inode, struct file *file)
5933{
5934 struct io_ring_ctx *ctx = file->private_data;
5935
5936 file->private_data = NULL;
5937 io_ring_ctx_wait_and_kill(ctx);
5938 return 0;
5939}
5940
fcb323cc
JA		 5941static void io_uring_cancel_files(struct io_ring_ctx *ctx,
5942 struct files_struct *files)
5943{
5944 struct io_kiocb *req;
5945 DEFINE_WAIT(wait);
5946
5947 while (!list_empty_careful(&ctx->inflight_list)) {
768134d4 5948 struct io_kiocb *cancel_req = NULL;
fcb323cc
JA		 5949
5950 spin_lock_irq(&ctx->inflight_lock);
5951 list_for_each_entry(req, &ctx->inflight_list, inflight_entry) {
768134d4
JA		 5952 if (req->work.files != files)
5953 continue;
5954 /* req is being completed, ignore */
5955 if (!refcount_inc_not_zero(&req->refs))
5956 continue;
5957 cancel_req = req;
5958 break;
fcb323cc 5959 }
768134d4 5960 if (cancel_req)
fcb323cc 5961 prepare_to_wait(&ctx->inflight_wait, &wait,
768134d4 5962 TASK_UNINTERRUPTIBLE);
fcb323cc
JA		 5963 spin_unlock_irq(&ctx->inflight_lock);
5964
768134d4
JA		 5965 /* We need to keep going until we don't find a matching req */
5966 if (!cancel_req)
fcb323cc 5967 break;
2f6d9b9d
BL		 5968
5969 io_wq_cancel_work(ctx->io_wq, &cancel_req->work);
5970 io_put_req(cancel_req);
fcb323cc
JA		 5971 schedule();
5972 }
768134d4 5973 finish_wait(&ctx->inflight_wait, &wait);
fcb323cc
JA		 5974}
5975
5976static int io_uring_flush(struct file *file, void *data)
5977{
5978 struct io_ring_ctx *ctx = file->private_data;
5979
5980 io_uring_cancel_files(ctx, data);
1d7bb1d5
JA		 5981 if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) {
5982 io_cqring_overflow_flush(ctx, true);
fcb323cc 5983 io_wq_cancel_all(ctx->io_wq);
1d7bb1d5 5984 }
fcb323cc
JA		 5985 return 0;
5986}
5987
6c5c240e
RP		 5988static void *io_uring_validate_mmap_request(struct file *file,
5989 loff_t pgoff, size_t sz)
2b188cc1 5990{
2b188cc1 5991 struct io_ring_ctx *ctx = file->private_data;
6c5c240e 5992 loff_t offset = pgoff << PAGE_SHIFT;
2b188cc1
JA		 5993 struct page *page;
5994 void *ptr;
5995
5996 switch (offset) {
5997 case IORING_OFF_SQ_RING:
75b28aff
HV		 5998 case IORING_OFF_CQ_RING:
5999 ptr = ctx->rings;
2b188cc1
JA		 6000 break;
6001 case IORING_OFF_SQES:
6002 ptr = ctx->sq_sqes;
6003 break;
2b188cc1 6004 default:
6c5c240e 6005 return ERR_PTR(-EINVAL);
2b188cc1
JA		 6006 }
6007
6008 page = virt_to_head_page(ptr);
a50b854e 6009 if (sz > page_size(page))
6c5c240e
RP		 6010 return ERR_PTR(-EINVAL);
6011
6012 return ptr;
6013}
6014
6015#ifdef CONFIG_MMU
6016
6017static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
6018{
6019 size_t sz = vma->vm_end - vma->vm_start;
6020 unsigned long pfn;
6021 void *ptr;
6022
6023 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
6024 if (IS_ERR(ptr))
6025 return PTR_ERR(ptr);
2b188cc1
JA		 6026
6027 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
6028 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
6029}
6030
6c5c240e
RP		 6031#else /* !CONFIG_MMU */
6032
6033static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
6034{
6035 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
6036}
6037
6038static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
6039{
6040 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
6041}
6042
6043static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
6044 unsigned long addr, unsigned long len,
6045 unsigned long pgoff, unsigned long flags)
6046{
6047 void *ptr;
6048
6049 ptr = io_uring_validate_mmap_request(file, pgoff, len);
6050 if (IS_ERR(ptr))
6051 return PTR_ERR(ptr);
6052
6053 return (unsigned long) ptr;
6054}
6055
6056#endif /* !CONFIG_MMU */
6057
2b188cc1
JA		 6058SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
6059 u32, min_complete, u32, flags, const sigset_t __user *, sig,
6060 size_t, sigsz)
6061{
6062 struct io_ring_ctx *ctx;
6063 long ret = -EBADF;
6064 int submitted = 0;
6065 struct fd f;
6066
6c271ce2 6067 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP))
2b188cc1
JA		 6068 return -EINVAL;
6069
6070 f = fdget(fd);
6071 if (!f.file)
6072 return -EBADF;
6073
6074 ret = -EOPNOTSUPP;
6075 if (f.file->f_op != &io_uring_fops)
6076 goto out_fput;
6077
6078 ret = -ENXIO;
6079 ctx = f.file->private_data;
6080 if (!percpu_ref_tryget(&ctx->refs))
6081 goto out_fput;
6082
6c271ce2
JA		 6083 /*
6084 * For SQ polling, the thread will do all submissions and completions.
6085 * Just return the requested submit count, and wake the thread if
6086 * we were asked to.
6087 */
b2a9eada 6088 ret = 0;
6c271ce2 6089 if (ctx->flags & IORING_SETUP_SQPOLL) {
c1edbf5f
JA		 6090 if (!list_empty_careful(&ctx->cq_overflow_list))
6091 io_cqring_overflow_flush(ctx, false);
6c271ce2
JA		 6092 if (flags & IORING_ENTER_SQ_WAKEUP)
6093 wake_up(&ctx->sqo_wait);
6094 submitted = to_submit;
b2a9eada 6095 } else if (to_submit) {
ae9428ca 6096 struct mm_struct *cur_mm;
2b188cc1 6097
44d28279
JA		 6098 if (current->mm != ctx->sqo_mm ||
6099 current_cred() != ctx->creds) {
6100 ret = -EPERM;
6101 goto out;
6102 }
6103
2b188cc1 6104 mutex_lock(&ctx->uring_lock);
ae9428ca
PB		 6105 /* already have mm, so io_submit_sqes() won't try to grab it */
6106 cur_mm = ctx->sqo_mm;
6107 submitted = io_submit_sqes(ctx, to_submit, f.file, fd,
6108 &cur_mm, false);
2b188cc1 6109 mutex_unlock(&ctx->uring_lock);
7c504e65
PB		 6110
6111 if (submitted != to_submit)
6112 goto out;
2b188cc1
JA		 6113 }
6114 if (flags & IORING_ENTER_GETEVENTS) {
def596e9
JA		 6115 unsigned nr_events = 0;
6116
2b188cc1
JA		 6117 min_complete = min(min_complete, ctx->cq_entries);
6118
def596e9 6119 if (ctx->flags & IORING_SETUP_IOPOLL) {
def596e9 6120 ret = io_iopoll_check(ctx, &nr_events, min_complete);
def596e9
JA		 6121 } else {
6122 ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
6123 }
2b188cc1
JA		 6124 }
6125
7c504e65 6126out:
6805b32e 6127 percpu_ref_put(&ctx->refs);
2b188cc1
JA		 6128out_fput:
6129 fdput(f);
6130 return submitted ? submitted : ret;
6131}
6132
6133static const struct file_operations io_uring_fops = {
6134 .release = io_uring_release,
fcb323cc 6135 .flush = io_uring_flush,
2b188cc1 6136 .mmap = io_uring_mmap,
6c5c240e
RP		 6137#ifndef CONFIG_MMU
6138 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
6139 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
6140#endif
2b188cc1
JA		 6141 .poll = io_uring_poll,
6142 .fasync = io_uring_fasync,
6143};
6144
6145static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
6146 struct io_uring_params *p)
6147{
75b28aff
HV		 6148 struct io_rings *rings;
6149 size_t size, sq_array_offset;
2b188cc1 6150
75b28aff
HV		 6151 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
6152 if (size == SIZE_MAX)
6153 return -EOVERFLOW;
6154
6155 rings = io_mem_alloc(size);
6156 if (!rings)
2b188cc1
JA		 6157 return -ENOMEM;
6158
75b28aff
HV		 6159 ctx->rings = rings;
6160 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
6161 rings->sq_ring_mask = p->sq_entries - 1;
6162 rings->cq_ring_mask = p->cq_entries - 1;
6163 rings->sq_ring_entries = p->sq_entries;
6164 rings->cq_ring_entries = p->cq_entries;
6165 ctx->sq_mask = rings->sq_ring_mask;
6166 ctx->cq_mask = rings->cq_ring_mask;
6167 ctx->sq_entries = rings->sq_ring_entries;
6168 ctx->cq_entries = rings->cq_ring_entries;
2b188cc1
JA		 6169
6170 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
eb065d30
JA		 6171 if (size == SIZE_MAX) {
6172 io_mem_free(ctx->rings);
6173 ctx->rings = NULL;
2b188cc1 6174 return -EOVERFLOW;
eb065d30 6175 }
2b188cc1
JA		 6176
6177 ctx->sq_sqes = io_mem_alloc(size);
eb065d30
JA		 6178 if (!ctx->sq_sqes) {
6179 io_mem_free(ctx->rings);
6180 ctx->rings = NULL;
2b188cc1 6181 return -ENOMEM;
eb065d30 6182 }
2b188cc1 6183
2b188cc1
JA		 6184 return 0;
6185}
6186
6187/*
6188 * Allocate an anonymous fd, this is what constitutes the application
6189 * visible backing of an io_uring instance. The application mmaps this
6190 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
6191 * we have to tie this fd to a socket for file garbage collection purposes.
6192 */
6193static int io_uring_get_fd(struct io_ring_ctx *ctx)
6194{
6195 struct file *file;
6196 int ret;
6197
6198#if defined(CONFIG_UNIX)
6199 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
6200 &ctx->ring_sock);
6201 if (ret)
6202 return ret;
6203#endif
6204
6205 ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
6206 if (ret < 0)
6207 goto err;
6208
6209 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
6210 O_RDWR | O_CLOEXEC);
6211 if (IS_ERR(file)) {
6212 put_unused_fd(ret);
6213 ret = PTR_ERR(file);
6214 goto err;
6215 }
6216
6217#if defined(CONFIG_UNIX)
6218 ctx->ring_sock->file = file;
6219#endif
6220 fd_install(ret, file);
6221 return ret;
6222err:
6223#if defined(CONFIG_UNIX)
6224 sock_release(ctx->ring_sock);
6225 ctx->ring_sock = NULL;
6226#endif
6227 return ret;
6228}
6229
6230static int io_uring_create(unsigned entries, struct io_uring_params *p)
6231{
6232 struct user_struct *user = NULL;
6233 struct io_ring_ctx *ctx;
6234 bool account_mem;
6235 int ret;
6236
8110c1a6 6237 if (!entries)
2b188cc1 6238 return -EINVAL;
8110c1a6
JA		 6239 if (entries > IORING_MAX_ENTRIES) {
6240 if (!(p->flags & IORING_SETUP_CLAMP))
6241 return -EINVAL;
6242 entries = IORING_MAX_ENTRIES;
6243 }
2b188cc1
JA		 6244
6245 /*
6246 * Use twice as many entries for the CQ ring. It's possible for the
6247 * application to drive a higher depth than the size of the SQ ring,
6248 * since the sqes are only used at submission time. This allows for
33a107f0
JA		 6249 * some flexibility in overcommitting a bit. If the application has
6250 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
6251 * of CQ ring entries manually.
2b188cc1
JA		 6252 */
6253 p->sq_entries = roundup_pow_of_two(entries);
33a107f0
JA		 6254 if (p->flags & IORING_SETUP_CQSIZE) {
6255 /*
6256 * If IORING_SETUP_CQSIZE is set, we do the same roundup
6257 * to a power-of-two, if it isn't already. We do NOT impose
6258 * any cq vs sq ring sizing.
6259 */
8110c1a6 6260 if (p->cq_entries < p->sq_entries)
33a107f0 6261 return -EINVAL;
8110c1a6
JA		 6262 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
6263 if (!(p->flags & IORING_SETUP_CLAMP))
6264 return -EINVAL;
6265 p->cq_entries = IORING_MAX_CQ_ENTRIES;
6266 }
33a107f0
JA		 6267 p->cq_entries = roundup_pow_of_two(p->cq_entries);
6268 } else {
6269 p->cq_entries = 2 * p->sq_entries;
6270 }
2b188cc1
JA		 6271
6272 user = get_uid(current_user());
6273 account_mem = !capable(CAP_IPC_LOCK);
6274
6275 if (account_mem) {
6276 ret = io_account_mem(user,
6277 ring_pages(p->sq_entries, p->cq_entries));
6278 if (ret) {
6279 free_uid(user);
6280 return ret;
6281 }
6282 }
6283
6284 ctx = io_ring_ctx_alloc(p);
6285 if (!ctx) {
6286 if (account_mem)
6287 io_unaccount_mem(user, ring_pages(p->sq_entries,
6288 p->cq_entries));
6289 free_uid(user);
6290 return -ENOMEM;
6291 }
6292 ctx->compat = in_compat_syscall();
6293 ctx->account_mem = account_mem;
6294 ctx->user = user;
0b8c0ec7 6295 ctx->creds = get_current_cred();
2b188cc1
JA		 6296
6297 ret = io_allocate_scq_urings(ctx, p);
6298 if (ret)
6299 goto err;
6300
6c271ce2 6301 ret = io_sq_offload_start(ctx, p);
2b188cc1
JA		 6302 if (ret)
6303 goto err;
6304
2b188cc1 6305 memset(&p->sq_off, 0, sizeof(p->sq_off));
75b28aff
HV		 6306 p->sq_off.head = offsetof(struct io_rings, sq.head);
6307 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
6308 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
6309 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
6310 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
6311 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
6312 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
2b188cc1
JA		 6313
6314 memset(&p->cq_off, 0, sizeof(p->cq_off));
75b28aff
HV		 6315 p->cq_off.head = offsetof(struct io_rings, cq.head);
6316 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
6317 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
6318 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
6319 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
6320 p->cq_off.cqes = offsetof(struct io_rings, cqes);
ac90f249 6321
044c1ab3
JA		 6322 /*
6323 * Install ring fd as the very last thing, so we don't risk someone
6324 * having closed it before we finish setup
6325 */
6326 ret = io_uring_get_fd(ctx);
6327 if (ret < 0)
6328 goto err;
6329
da8c9690 6330 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
ba04291e 6331 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS;
c826bd7a 6332 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
2b188cc1
JA		 6333 return ret;
6334err:
6335 io_ring_ctx_wait_and_kill(ctx);
6336 return ret;
6337}
6338
6339/*
6340 * Sets up an aio uring context, and returns the fd. Applications asks for a
6341 * ring size, we return the actual sq/cq ring sizes (among other things) in the
6342 * params structure passed in.
6343 */
6344static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
6345{
6346 struct io_uring_params p;
6347 long ret;
6348 int i;
6349
6350 if (copy_from_user(&p, params, sizeof(p)))
6351 return -EFAULT;
6352 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
6353 if (p.resv[i])
6354 return -EINVAL;
6355 }
6356
6c271ce2 6357 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
8110c1a6
JA		 6358 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
6359 IORING_SETUP_CLAMP))
2b188cc1
JA		 6360 return -EINVAL;
6361
6362 ret = io_uring_create(entries, &p);
6363 if (ret < 0)
6364 return ret;
6365
6366 if (copy_to_user(params, &p, sizeof(p)))
6367 return -EFAULT;
6368
6369 return ret;
6370}
6371
6372SYSCALL_DEFINE2(io_uring_setup, u32, entries,
6373 struct io_uring_params __user *, params)
6374{
6375 return io_uring_setup(entries, params);
6376}
6377
edafccee
JA		 6378static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
6379 void __user *arg, unsigned nr_args)
b19062a5
JA		 6380 __releases(ctx->uring_lock)
6381 __acquires(ctx->uring_lock)
edafccee
JA		 6382{
6383 int ret;
6384
35fa71a0
JA		 6385 /*
6386 * We're inside the ring mutex, if the ref is already dying, then
6387 * someone else killed the ctx or is already going through
6388 * io_uring_register().
6389 */
6390 if (percpu_ref_is_dying(&ctx->refs))
6391 return -ENXIO;
6392
05f3fb3c
JA		 6393 if (opcode != IORING_UNREGISTER_FILES &&
6394 opcode != IORING_REGISTER_FILES_UPDATE) {
6395 percpu_ref_kill(&ctx->refs);
b19062a5 6396
05f3fb3c
JA		 6397 /*
6398 * Drop uring mutex before waiting for references to exit. If
6399 * another thread is currently inside io_uring_enter() it might
6400 * need to grab the uring_lock to make progress. If we hold it
6401 * here across the drain wait, then we can deadlock. It's safe
6402 * to drop the mutex here, since no new references will come in
6403 * after we've killed the percpu ref.
6404 */
6405 mutex_unlock(&ctx->uring_lock);
6406 wait_for_completion(&ctx->completions[0]);
6407 mutex_lock(&ctx->uring_lock);
6408 }
edafccee
JA		 6409
6410 switch (opcode) {
6411 case IORING_REGISTER_BUFFERS:
6412 ret = io_sqe_buffer_register(ctx, arg, nr_args);
6413 break;
6414 case IORING_UNREGISTER_BUFFERS:
6415 ret = -EINVAL;
6416 if (arg || nr_args)
6417 break;
6418 ret = io_sqe_buffer_unregister(ctx);
6419 break;
6b06314c
JA		 6420 case IORING_REGISTER_FILES:
6421 ret = io_sqe_files_register(ctx, arg, nr_args);
6422 break;
6423 case IORING_UNREGISTER_FILES:
6424 ret = -EINVAL;
6425 if (arg || nr_args)
6426 break;
6427 ret = io_sqe_files_unregister(ctx);
6428 break;
c3a31e60
JA		 6429 case IORING_REGISTER_FILES_UPDATE:
6430 ret = io_sqe_files_update(ctx, arg, nr_args);
6431 break;
9b402849
JA		 6432 case IORING_REGISTER_EVENTFD:
6433 ret = -EINVAL;
6434 if (nr_args != 1)
6435 break;
6436 ret = io_eventfd_register(ctx, arg);
6437 break;
6438 case IORING_UNREGISTER_EVENTFD:
6439 ret = -EINVAL;
6440 if (arg || nr_args)
6441 break;
6442 ret = io_eventfd_unregister(ctx);
6443 break;
edafccee
JA		 6444 default:
6445 ret = -EINVAL;
6446 break;
6447 }
6448
05f3fb3c
JA		 6449
6450 if (opcode != IORING_UNREGISTER_FILES &&
6451 opcode != IORING_REGISTER_FILES_UPDATE) {
6452 /* bring the ctx back to life */
6453 reinit_completion(&ctx->completions[0]);
6454 percpu_ref_reinit(&ctx->refs);
6455 }
edafccee
JA		 6456 return ret;
6457}
6458
6459SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
6460 void __user *, arg, unsigned int, nr_args)
6461{
6462 struct io_ring_ctx *ctx;
6463 long ret = -EBADF;
6464 struct fd f;
6465
6466 f = fdget(fd);
6467 if (!f.file)
6468 return -EBADF;
6469
6470 ret = -EOPNOTSUPP;
6471 if (f.file->f_op != &io_uring_fops)
6472 goto out_fput;
6473
6474 ctx = f.file->private_data;
6475
6476 mutex_lock(&ctx->uring_lock);
6477 ret = __io_uring_register(ctx, opcode, arg, nr_args);
6478 mutex_unlock(&ctx->uring_lock);
c826bd7a
DD		 6479 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
6480 ctx->cq_ev_fd != NULL, ret);
edafccee
JA		 6481out_fput:
6482 fdput(f);
6483 return ret;
6484}
6485
2b188cc1
JA		 6486static int __init io_uring_init(void)
6487{
d3656344 6488 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
2b188cc1
JA		 6489 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
6490 return 0;
6491};
6492__initcall(io_uring_init);
Linux 6.x block layer and io_uring tree(s)