1 // SPDX-License-Identifier: GPL-2.0
3 * Shared application/kernel submission and completion ring pairs, for
4 * supporting fast/efficient IO.
6 * A note on the read/write ordering memory barriers that are matched between
7 * the application and kernel side.
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
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
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
30 * Also see the examples in the liburing library:
32 * git://git.kernel.dk/liburing
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.
39 * Copyright (C) 2018-2019 Jens Axboe
40 * Copyright (c) 2018-2019 Christoph Hellwig
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>
50 #include <linux/sched/signal.h>
52 #include <linux/file.h>
53 #include <linux/fdtable.h>
55 #include <linux/mman.h>
56 #include <linux/mmu_context.h>
57 #include <linux/percpu.h>
58 #include <linux/slab.h>
59 #include <linux/kthread.h>
60 #include <linux/blkdev.h>
61 #include <linux/bvec.h>
62 #include <linux/net.h>
64 #include <net/af_unix.h>
66 #include <linux/anon_inodes.h>
67 #include <linux/sched/mm.h>
68 #include <linux/uaccess.h>
69 #include <linux/nospec.h>
70 #include <linux/sizes.h>
71 #include <linux/hugetlb.h>
72 #include <linux/highmem.h>
73 #include <linux/namei.h>
74 #include <linux/fsnotify.h>
75 #include <linux/fadvise.h>
77 #define CREATE_TRACE_POINTS
78 #include <trace/events/io_uring.h>
80 #include <uapi/linux/io_uring.h>
85 #define IORING_MAX_ENTRIES 32768
86 #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
89 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
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)
97 u32 head ____cacheline_aligned_in_smp;
98 u32 tail ____cacheline_aligned_in_smp;
102 * This data is shared with the application through the mmap at offsets
103 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
105 * The offsets to the member fields are published through struct
106 * io_sqring_offsets when calling io_uring_setup.
110 * Head and tail offsets into the ring; the offsets need to be
111 * masked to get valid indices.
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
117 struct io_uring sq, cq;
119 * Bitmasks to apply to head and tail offsets (constant, equals
122 u32 sq_ring_mask, cq_ring_mask;
123 /* Ring sizes (constant, power of 2) */
124 u32 sq_ring_entries, cq_ring_entries;
126 * Number of invalid entries dropped by the kernel due to
127 * invalid index stored in array
129 * Written by the kernel, shouldn't be modified by the
130 * application (i.e. get number of "new events" by comparing to
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).
141 * Written by the kernel, shouldn't be modified by the
144 * The application needs a full memory barrier before checking
145 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
149 * Number of completion events lost because the queue was full;
150 * this should be avoided by the application by making sure
151 * there are not more requests pending than there is space in
152 * the completion queue.
154 * Written by the kernel, shouldn't be modified by the
155 * application (i.e. get number of "new events" by comparing to
158 * As completion events come in out of order this counter is not
159 * ordered with any other data.
163 * Ring buffer of completion events.
165 * The kernel writes completion events fresh every time they are
166 * produced, so the application is allowed to modify pending
169 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
172 struct io_mapped_ubuf {
175 struct bio_vec *bvec;
176 unsigned int nr_bvecs;
179 struct fixed_file_table {
187 struct fixed_file_data {
188 struct fixed_file_table *table;
189 struct io_ring_ctx *ctx;
191 struct percpu_ref refs;
192 struct llist_head put_llist;
194 struct work_struct ref_work;
195 struct completion done;
200 struct percpu_ref refs;
201 } ____cacheline_aligned_in_smp;
207 bool cq_overflow_flushed;
211 * Ring buffer of indices into array of io_uring_sqe, which is
212 * mmapped by the application using the IORING_OFF_SQES offset.
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.
218 * The kernel modifies neither the indices array nor the entries
222 unsigned cached_sq_head;
225 unsigned sq_thread_idle;
226 unsigned cached_sq_dropped;
227 atomic_t cached_cq_overflow;
228 unsigned long sq_check_overflow;
230 struct list_head defer_list;
231 struct list_head timeout_list;
232 struct list_head cq_overflow_list;
234 wait_queue_head_t inflight_wait;
235 struct io_uring_sqe *sq_sqes;
236 } ____cacheline_aligned_in_smp;
238 struct io_rings *rings;
242 struct task_struct *sqo_thread; /* if using sq thread polling */
243 struct mm_struct *sqo_mm;
244 wait_queue_head_t sqo_wait;
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).
251 struct fixed_file_data *file_data;
252 unsigned nr_user_files;
254 /* if used, fixed mapped user buffers */
255 unsigned nr_user_bufs;
256 struct io_mapped_ubuf *user_bufs;
258 struct user_struct *user;
260 const struct cred *creds;
262 /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */
263 struct completion *completions;
265 /* if all else fails... */
266 struct io_kiocb *fallback_req;
268 #if defined(CONFIG_UNIX)
269 struct socket *ring_sock;
273 unsigned cached_cq_tail;
276 atomic_t cq_timeouts;
277 unsigned long cq_check_overflow;
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;
284 struct mutex uring_lock;
285 wait_queue_head_t wait;
286 } ____cacheline_aligned_in_smp;
289 spinlock_t completion_lock;
290 struct llist_head poll_llist;
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.
298 struct list_head poll_list;
299 struct hlist_head *cancel_hash;
300 unsigned cancel_hash_bits;
301 bool poll_multi_file;
303 spinlock_t inflight_lock;
304 struct list_head inflight_list;
305 } ____cacheline_aligned_in_smp;
309 * First field must be the file pointer in all the
310 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
312 struct io_poll_iocb {
315 struct wait_queue_head *head;
321 struct wait_queue_entry wait;
326 struct file *put_file;
330 struct io_timeout_data {
331 struct io_kiocb *req;
332 struct hrtimer timer;
333 struct timespec64 ts;
334 enum hrtimer_mode mode;
340 struct sockaddr __user *addr;
341 int __user *addr_len;
366 /* NOTE: kiocb has the file as the first member, so don't do it here */
374 struct sockaddr __user *addr;
380 struct user_msghdr __user *msg;
391 const char __user *fname;
392 struct filename *filename;
393 struct statx __user *buffer;
397 struct io_files_update {
418 struct io_async_connect {
419 struct sockaddr_storage address;
422 struct io_async_msghdr {
423 struct iovec fast_iov[UIO_FASTIOV];
425 struct sockaddr __user *uaddr;
430 struct iovec fast_iov[UIO_FASTIOV];
436 struct io_async_open {
437 struct filename *filename;
440 struct io_async_ctx {
442 struct io_async_rw rw;
443 struct io_async_msghdr msg;
444 struct io_async_connect connect;
445 struct io_timeout_data timeout;
446 struct io_async_open open;
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.
460 struct io_poll_iocb poll;
461 struct io_accept accept;
463 struct io_cancel cancel;
464 struct io_timeout timeout;
465 struct io_connect connect;
466 struct io_sr_msg sr_msg;
468 struct io_close close;
469 struct io_files_update files_update;
470 struct io_fadvise fadvise;
471 struct io_madvise madvise;
474 struct io_async_ctx *io;
477 * ring_file is only used in the submission path, and
478 * llist_node is only used for poll deferred completions
480 struct file *ring_file;
481 struct llist_node llist_node;
486 bool needs_fixed_file;
489 struct io_ring_ctx *ctx;
491 struct list_head list;
492 struct hlist_node hash_node;
494 struct list_head link_list;
497 #define REQ_F_NOWAIT 1 /* must not punt to workers */
498 #define REQ_F_IOPOLL_COMPLETED 2 /* polled IO has completed */
499 #define REQ_F_FIXED_FILE 4 /* ctx owns file */
500 #define REQ_F_LINK_NEXT 8 /* already grabbed next link */
501 #define REQ_F_IO_DRAIN 16 /* drain existing IO first */
502 #define REQ_F_IO_DRAINED 32 /* drain done */
503 #define REQ_F_LINK 64 /* linked sqes */
504 #define REQ_F_LINK_TIMEOUT 128 /* has linked timeout */
505 #define REQ_F_FAIL_LINK 256 /* fail rest of links */
506 #define REQ_F_DRAIN_LINK 512 /* link should be fully drained */
507 #define REQ_F_TIMEOUT 1024 /* timeout request */
508 #define REQ_F_ISREG 2048 /* regular file */
509 #define REQ_F_MUST_PUNT 4096 /* must be punted even for NONBLOCK */
510 #define REQ_F_TIMEOUT_NOSEQ 8192 /* no timeout sequence */
511 #define REQ_F_INFLIGHT 16384 /* on inflight list */
512 #define REQ_F_COMP_LOCKED 32768 /* completion under lock */
513 #define REQ_F_HARDLINK 65536 /* doesn't sever on completion < 0 */
514 #define REQ_F_FORCE_ASYNC 131072 /* IOSQE_ASYNC */
515 #define REQ_F_CUR_POS 262144 /* read/write uses file position */
520 struct list_head inflight_entry;
522 struct io_wq_work work;
525 #define IO_PLUG_THRESHOLD 2
526 #define IO_IOPOLL_BATCH 8
528 struct io_submit_state {
529 struct blk_plug plug;
532 * io_kiocb alloc cache
534 void *reqs[IO_IOPOLL_BATCH];
535 unsigned int free_reqs;
536 unsigned int cur_req;
539 * File reference cache
543 unsigned int has_refs;
544 unsigned int used_refs;
545 unsigned int ios_left;
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;
563 static const struct io_op_def io_op_defs[] = {
568 /* IORING_OP_READV */
572 .unbound_nonreg_file = 1,
575 /* IORING_OP_WRITEV */
580 .unbound_nonreg_file = 1,
583 /* IORING_OP_FSYNC */
587 /* IORING_OP_READ_FIXED */
589 .unbound_nonreg_file = 1,
592 /* IORING_OP_WRITE_FIXED */
595 .unbound_nonreg_file = 1,
598 /* IORING_OP_POLL_ADD */
600 .unbound_nonreg_file = 1,
603 /* IORING_OP_POLL_REMOVE */
606 /* IORING_OP_SYNC_FILE_RANGE */
610 /* IORING_OP_SENDMSG */
614 .unbound_nonreg_file = 1,
617 /* IORING_OP_RECVMSG */
621 .unbound_nonreg_file = 1,
624 /* IORING_OP_TIMEOUT */
629 /* IORING_OP_TIMEOUT_REMOVE */
632 /* IORING_OP_ACCEPT */
635 .unbound_nonreg_file = 1,
638 /* IORING_OP_ASYNC_CANCEL */
641 /* IORING_OP_LINK_TIMEOUT */
646 /* IORING_OP_CONNECT */
650 .unbound_nonreg_file = 1,
653 /* IORING_OP_FALLOCATE */
657 /* IORING_OP_OPENAT */
662 /* IORING_OP_CLOSE */
666 /* IORING_OP_FILES_UPDATE */
670 /* IORING_OP_STATX */
679 .unbound_nonreg_file = 1,
682 /* IORING_OP_WRITE */
685 .unbound_nonreg_file = 1,
688 /* IORING_OP_FADVISE */
692 /* IORING_OP_MADVISE */
697 static void io_wq_submit_work(struct io_wq_work **workptr);
698 static void io_cqring_fill_event(struct io_kiocb *req, long res);
699 static void io_put_req(struct io_kiocb *req);
700 static void __io_double_put_req(struct io_kiocb *req);
701 static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
702 static void io_queue_linked_timeout(struct io_kiocb *req);
703 static int __io_sqe_files_update(struct io_ring_ctx *ctx,
704 struct io_uring_files_update *ip,
707 static struct kmem_cache *req_cachep;
709 static const struct file_operations io_uring_fops;
711 struct sock *io_uring_get_socket(struct file *file)
713 #if defined(CONFIG_UNIX)
714 if (file->f_op == &io_uring_fops) {
715 struct io_ring_ctx *ctx = file->private_data;
717 return ctx->ring_sock->sk;
722 EXPORT_SYMBOL(io_uring_get_socket);
724 static void io_ring_ctx_ref_free(struct percpu_ref *ref)
726 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
728 complete(&ctx->completions[0]);
731 static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
733 struct io_ring_ctx *ctx;
736 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
740 ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL);
741 if (!ctx->fallback_req)
744 ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL);
745 if (!ctx->completions)
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.
752 hash_bits = ilog2(p->cq_entries);
756 ctx->cancel_hash_bits = hash_bits;
757 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
759 if (!ctx->cancel_hash)
761 __hash_init(ctx->cancel_hash, 1U << hash_bits);
763 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
764 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
767 ctx->flags = p->flags;
768 init_waitqueue_head(&ctx->cq_wait);
769 INIT_LIST_HEAD(&ctx->cq_overflow_list);
770 init_completion(&ctx->completions[0]);
771 init_completion(&ctx->completions[1]);
772 mutex_init(&ctx->uring_lock);
773 init_waitqueue_head(&ctx->wait);
774 spin_lock_init(&ctx->completion_lock);
775 init_llist_head(&ctx->poll_llist);
776 INIT_LIST_HEAD(&ctx->poll_list);
777 INIT_LIST_HEAD(&ctx->defer_list);
778 INIT_LIST_HEAD(&ctx->timeout_list);
779 init_waitqueue_head(&ctx->inflight_wait);
780 spin_lock_init(&ctx->inflight_lock);
781 INIT_LIST_HEAD(&ctx->inflight_list);
784 if (ctx->fallback_req)
785 kmem_cache_free(req_cachep, ctx->fallback_req);
786 kfree(ctx->completions);
787 kfree(ctx->cancel_hash);
792 static inline bool __req_need_defer(struct io_kiocb *req)
794 struct io_ring_ctx *ctx = req->ctx;
796 return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped
797 + atomic_read(&ctx->cached_cq_overflow);
800 static inline bool req_need_defer(struct io_kiocb *req)
802 if ((req->flags & (REQ_F_IO_DRAIN|REQ_F_IO_DRAINED)) == REQ_F_IO_DRAIN)
803 return __req_need_defer(req);
808 static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx)
810 struct io_kiocb *req;
812 req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list);
813 if (req && !req_need_defer(req)) {
814 list_del_init(&req->list);
821 static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx)
823 struct io_kiocb *req;
825 req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list);
827 if (req->flags & REQ_F_TIMEOUT_NOSEQ)
829 if (!__req_need_defer(req)) {
830 list_del_init(&req->list);
838 static void __io_commit_cqring(struct io_ring_ctx *ctx)
840 struct io_rings *rings = ctx->rings;
842 if (ctx->cached_cq_tail != READ_ONCE(rings->cq.tail)) {
843 /* order cqe stores with ring update */
844 smp_store_release(&rings->cq.tail, ctx->cached_cq_tail);
846 if (wq_has_sleeper(&ctx->cq_wait)) {
847 wake_up_interruptible(&ctx->cq_wait);
848 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
853 static inline bool io_prep_async_work(struct io_kiocb *req,
854 struct io_kiocb **link)
856 const struct io_op_def *def = &io_op_defs[req->opcode];
857 bool do_hashed = false;
859 if (req->flags & REQ_F_ISREG) {
860 if (def->hash_reg_file)
863 if (def->unbound_nonreg_file)
864 req->work.flags |= IO_WQ_WORK_UNBOUND;
867 req->work.flags |= IO_WQ_WORK_NEEDS_USER;
869 *link = io_prep_linked_timeout(req);
873 static inline void io_queue_async_work(struct io_kiocb *req)
875 struct io_ring_ctx *ctx = req->ctx;
876 struct io_kiocb *link;
879 do_hashed = io_prep_async_work(req, &link);
881 trace_io_uring_queue_async_work(ctx, do_hashed, req, &req->work,
884 io_wq_enqueue(ctx->io_wq, &req->work);
886 io_wq_enqueue_hashed(ctx->io_wq, &req->work,
887 file_inode(req->file));
891 io_queue_linked_timeout(link);
894 static void io_kill_timeout(struct io_kiocb *req)
898 ret = hrtimer_try_to_cancel(&req->io->timeout.timer);
900 atomic_inc(&req->ctx->cq_timeouts);
901 list_del_init(&req->list);
902 io_cqring_fill_event(req, 0);
907 static void io_kill_timeouts(struct io_ring_ctx *ctx)
909 struct io_kiocb *req, *tmp;
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);
917 static void io_commit_cqring(struct io_ring_ctx *ctx)
919 struct io_kiocb *req;
921 while ((req = io_get_timeout_req(ctx)) != NULL)
922 io_kill_timeout(req);
924 __io_commit_cqring(ctx);
926 while ((req = io_get_deferred_req(ctx)) != NULL) {
927 req->flags |= REQ_F_IO_DRAINED;
928 io_queue_async_work(req);
932 static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
934 struct io_rings *rings = ctx->rings;
937 tail = ctx->cached_cq_tail;
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
943 if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries)
946 ctx->cached_cq_tail++;
947 return &rings->cqes[tail & ctx->cq_mask];
950 static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
952 if (waitqueue_active(&ctx->wait))
954 if (waitqueue_active(&ctx->sqo_wait))
955 wake_up(&ctx->sqo_wait);
957 eventfd_signal(ctx->cq_ev_fd, 1);
960 /* Returns true if there are no backlogged entries after the flush */
961 static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
963 struct io_rings *rings = ctx->rings;
964 struct io_uring_cqe *cqe;
965 struct io_kiocb *req;
970 if (list_empty_careful(&ctx->cq_overflow_list))
972 if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) ==
973 rings->cq_ring_entries))
977 spin_lock_irqsave(&ctx->completion_lock, flags);
979 /* if force is set, the ring is going away. always drop after that */
981 ctx->cq_overflow_flushed = true;
984 while (!list_empty(&ctx->cq_overflow_list)) {
985 cqe = io_get_cqring(ctx);
989 req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb,
991 list_move(&req->list, &list);
993 WRITE_ONCE(cqe->user_data, req->user_data);
994 WRITE_ONCE(cqe->res, req->result);
995 WRITE_ONCE(cqe->flags, 0);
997 WRITE_ONCE(ctx->rings->cq_overflow,
998 atomic_inc_return(&ctx->cached_cq_overflow));
1002 io_commit_cqring(ctx);
1004 clear_bit(0, &ctx->sq_check_overflow);
1005 clear_bit(0, &ctx->cq_check_overflow);
1007 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1008 io_cqring_ev_posted(ctx);
1010 while (!list_empty(&list)) {
1011 req = list_first_entry(&list, struct io_kiocb, list);
1012 list_del(&req->list);
1019 static void io_cqring_fill_event(struct io_kiocb *req, long res)
1021 struct io_ring_ctx *ctx = req->ctx;
1022 struct io_uring_cqe *cqe;
1024 trace_io_uring_complete(ctx, req->user_data, res);
1027 * If we can't get a cq entry, userspace overflowed the
1028 * submission (by quite a lot). Increment the overflow count in
1031 cqe = io_get_cqring(ctx);
1033 WRITE_ONCE(cqe->user_data, req->user_data);
1034 WRITE_ONCE(cqe->res, res);
1035 WRITE_ONCE(cqe->flags, 0);
1036 } else if (ctx->cq_overflow_flushed) {
1037 WRITE_ONCE(ctx->rings->cq_overflow,
1038 atomic_inc_return(&ctx->cached_cq_overflow));
1040 if (list_empty(&ctx->cq_overflow_list)) {
1041 set_bit(0, &ctx->sq_check_overflow);
1042 set_bit(0, &ctx->cq_check_overflow);
1044 refcount_inc(&req->refs);
1046 list_add_tail(&req->list, &ctx->cq_overflow_list);
1050 static void io_cqring_add_event(struct io_kiocb *req, long res)
1052 struct io_ring_ctx *ctx = req->ctx;
1053 unsigned long flags;
1055 spin_lock_irqsave(&ctx->completion_lock, flags);
1056 io_cqring_fill_event(req, res);
1057 io_commit_cqring(ctx);
1058 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1060 io_cqring_ev_posted(ctx);
1063 static inline bool io_is_fallback_req(struct io_kiocb *req)
1065 return req == (struct io_kiocb *)
1066 ((unsigned long) req->ctx->fallback_req & ~1UL);
1069 static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx)
1071 struct io_kiocb *req;
1073 req = ctx->fallback_req;
1074 if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req))
1080 static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
1081 struct io_submit_state *state)
1083 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
1084 struct io_kiocb *req;
1087 req = kmem_cache_alloc(req_cachep, gfp);
1090 } else if (!state->free_reqs) {
1094 sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs));
1095 ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs);
1098 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1099 * retry single alloc to be on the safe side.
1101 if (unlikely(ret <= 0)) {
1102 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1103 if (!state->reqs[0])
1107 state->free_reqs = ret - 1;
1109 req = state->reqs[0];
1111 req = state->reqs[state->cur_req];
1118 req->ring_file = NULL;
1122 /* one is dropped after submission, the other at completion */
1123 refcount_set(&req->refs, 2);
1125 INIT_IO_WORK(&req->work, io_wq_submit_work);
1128 req = io_get_fallback_req(ctx);
1131 percpu_ref_put(&ctx->refs);
1135 static void __io_req_do_free(struct io_kiocb *req)
1137 if (likely(!io_is_fallback_req(req)))
1138 kmem_cache_free(req_cachep, req);
1140 clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req);
1143 static void __io_req_aux_free(struct io_kiocb *req)
1145 struct io_ring_ctx *ctx = req->ctx;
1150 if (req->flags & REQ_F_FIXED_FILE)
1151 percpu_ref_put(&ctx->file_data->refs);
1157 static void __io_free_req(struct io_kiocb *req)
1159 __io_req_aux_free(req);
1161 if (req->flags & REQ_F_INFLIGHT) {
1162 struct io_ring_ctx *ctx = req->ctx;
1163 unsigned long flags;
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);
1172 percpu_ref_put(&req->ctx->refs);
1173 __io_req_do_free(req);
1177 void *reqs[IO_IOPOLL_BATCH];
1182 static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb)
1186 if (rb->need_iter) {
1187 int i, inflight = 0;
1188 unsigned long flags;
1190 for (i = 0; i < rb->to_free; i++) {
1191 struct io_kiocb *req = rb->reqs[i];
1193 if (req->flags & REQ_F_FIXED_FILE)
1195 if (req->flags & REQ_F_INFLIGHT)
1199 __io_req_aux_free(req);
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];
1209 list_del(&req->inflight_entry);
1214 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1216 if (waitqueue_active(&ctx->inflight_wait))
1217 wake_up(&ctx->inflight_wait);
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;
1226 static bool io_link_cancel_timeout(struct io_kiocb *req)
1228 struct io_ring_ctx *ctx = req->ctx;
1231 ret = hrtimer_try_to_cancel(&req->io->timeout.timer);
1233 io_cqring_fill_event(req, -ECANCELED);
1234 io_commit_cqring(ctx);
1235 req->flags &= ~REQ_F_LINK;
1243 static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
1245 struct io_ring_ctx *ctx = req->ctx;
1246 bool wake_ev = false;
1248 /* Already got next link */
1249 if (req->flags & REQ_F_LINK_NEXT)
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
1257 while (!list_empty(&req->link_list)) {
1258 struct io_kiocb *nxt = list_first_entry(&req->link_list,
1259 struct io_kiocb, link_list);
1261 if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) &&
1262 (nxt->flags & REQ_F_TIMEOUT))) {
1263 list_del_init(&nxt->link_list);
1264 wake_ev |= io_link_cancel_timeout(nxt);
1265 req->flags &= ~REQ_F_LINK_TIMEOUT;
1269 list_del_init(&req->link_list);
1270 if (!list_empty(&nxt->link_list))
1271 nxt->flags |= REQ_F_LINK;
1276 req->flags |= REQ_F_LINK_NEXT;
1278 io_cqring_ev_posted(ctx);
1282 * Called if REQ_F_LINK is set, and we fail the head request
1284 static void io_fail_links(struct io_kiocb *req)
1286 struct io_ring_ctx *ctx = req->ctx;
1287 unsigned long flags;
1289 spin_lock_irqsave(&ctx->completion_lock, flags);
1291 while (!list_empty(&req->link_list)) {
1292 struct io_kiocb *link = list_first_entry(&req->link_list,
1293 struct io_kiocb, link_list);
1295 list_del_init(&link->link_list);
1296 trace_io_uring_fail_link(req, link);
1298 if ((req->flags & REQ_F_LINK_TIMEOUT) &&
1299 link->opcode == IORING_OP_LINK_TIMEOUT) {
1300 io_link_cancel_timeout(link);
1302 io_cqring_fill_event(link, -ECANCELED);
1303 __io_double_put_req(link);
1305 req->flags &= ~REQ_F_LINK_TIMEOUT;
1308 io_commit_cqring(ctx);
1309 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1310 io_cqring_ev_posted(ctx);
1313 static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt)
1315 if (likely(!(req->flags & REQ_F_LINK)))
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
1324 if (req->flags & REQ_F_FAIL_LINK) {
1326 } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) ==
1327 REQ_F_LINK_TIMEOUT) {
1328 struct io_ring_ctx *ctx = req->ctx;
1329 unsigned long flags;
1332 * If this is a timeout link, we could be racing with the
1333 * timeout timer. Grab the completion lock for this case to
1334 * protect against that.
1336 spin_lock_irqsave(&ctx->completion_lock, flags);
1337 io_req_link_next(req, nxt);
1338 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1340 io_req_link_next(req, nxt);
1344 static void io_free_req(struct io_kiocb *req)
1346 struct io_kiocb *nxt = NULL;
1348 io_req_find_next(req, &nxt);
1352 io_queue_async_work(nxt);
1356 * Drop reference to request, return next in chain (if there is one) if this
1357 * was the last reference to this request.
1359 __attribute__((nonnull))
1360 static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
1362 io_req_find_next(req, nxtptr);
1364 if (refcount_dec_and_test(&req->refs))
1368 static void io_put_req(struct io_kiocb *req)
1370 if (refcount_dec_and_test(&req->refs))
1375 * Must only be used if we don't need to care about links, usually from
1376 * within the completion handling itself.
1378 static void __io_double_put_req(struct io_kiocb *req)
1380 /* drop both submit and complete references */
1381 if (refcount_sub_and_test(2, &req->refs))
1385 static void io_double_put_req(struct io_kiocb *req)
1387 /* drop both submit and complete references */
1388 if (refcount_sub_and_test(2, &req->refs))
1392 static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush)
1394 struct io_rings *rings = ctx->rings;
1396 if (test_bit(0, &ctx->cq_check_overflow)) {
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.
1402 if (noflush && !list_empty(&ctx->cq_overflow_list))
1405 io_cqring_overflow_flush(ctx, false);
1408 /* See comment at the top of this file */
1410 return ctx->cached_cq_tail - READ_ONCE(rings->cq.head);
1413 static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
1415 struct io_rings *rings = ctx->rings;
1417 /* make sure SQ entry isn't read before tail */
1418 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
1421 static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req)
1423 if ((req->flags & REQ_F_LINK) || io_is_fallback_req(req))
1426 if (!(req->flags & REQ_F_FIXED_FILE) || req->io)
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);
1436 * Find and free completed poll iocbs
1438 static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
1439 struct list_head *done)
1441 struct req_batch rb;
1442 struct io_kiocb *req;
1444 rb.to_free = rb.need_iter = 0;
1445 while (!list_empty(done)) {
1446 req = list_first_entry(done, struct io_kiocb, list);
1447 list_del(&req->list);
1449 io_cqring_fill_event(req, req->result);
1452 if (refcount_dec_and_test(&req->refs) &&
1453 !io_req_multi_free(&rb, req))
1457 io_commit_cqring(ctx);
1458 io_free_req_many(ctx, &rb);
1461 static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
1464 struct io_kiocb *req, *tmp;
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.
1473 spin = !ctx->poll_multi_file && *nr_events < min;
1476 list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) {
1477 struct kiocb *kiocb = &req->rw.kiocb;
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.
1484 if (req->flags & REQ_F_IOPOLL_COMPLETED) {
1485 list_move_tail(&req->list, &done);
1488 if (!list_empty(&done))
1491 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
1500 if (!list_empty(&done))
1501 io_iopoll_complete(ctx, nr_events, &done);
1507 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
1508 * non-spinning poll check - we'll still enter the driver poll loop, but only
1509 * as a non-spinning completion check.
1511 static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
1514 while (!list_empty(&ctx->poll_list) && !need_resched()) {
1517 ret = io_do_iopoll(ctx, nr_events, min);
1520 if (!min || *nr_events >= min)
1528 * We can't just wait for polled events to come to us, we have to actively
1529 * find and complete them.
1531 static void io_iopoll_reap_events(struct io_ring_ctx *ctx)
1533 if (!(ctx->flags & IORING_SETUP_IOPOLL))
1536 mutex_lock(&ctx->uring_lock);
1537 while (!list_empty(&ctx->poll_list)) {
1538 unsigned int nr_events = 0;
1540 io_iopoll_getevents(ctx, &nr_events, 1);
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.
1548 mutex_unlock(&ctx->uring_lock);
1551 static int __io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events,
1554 int iters = 0, ret = 0;
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).
1564 if (io_cqring_events(ctx, false))
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
1577 if (!(++iters & 7)) {
1578 mutex_unlock(&ctx->uring_lock);
1579 mutex_lock(&ctx->uring_lock);
1582 if (*nr_events < min)
1583 tmin = min - *nr_events;
1585 ret = io_iopoll_getevents(ctx, nr_events, tmin);
1589 } while (min && !*nr_events && !need_resched());
1594 static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events,
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.
1604 mutex_lock(&ctx->uring_lock);
1605 ret = __io_iopoll_check(ctx, nr_events, min);
1606 mutex_unlock(&ctx->uring_lock);
1610 static void kiocb_end_write(struct io_kiocb *req)
1613 * Tell lockdep we inherited freeze protection from submission
1616 if (req->flags & REQ_F_ISREG) {
1617 struct inode *inode = file_inode(req->file);
1619 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
1621 file_end_write(req->file);
1624 static inline void req_set_fail_links(struct io_kiocb *req)
1626 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1627 req->flags |= REQ_F_FAIL_LINK;
1630 static void io_complete_rw_common(struct kiocb *kiocb, long res)
1632 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
1634 if (kiocb->ki_flags & IOCB_WRITE)
1635 kiocb_end_write(req);
1637 if (res != req->result)
1638 req_set_fail_links(req);
1639 io_cqring_add_event(req, res);
1642 static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
1644 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
1646 io_complete_rw_common(kiocb, res);
1650 static struct io_kiocb *__io_complete_rw(struct kiocb *kiocb, long res)
1652 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
1653 struct io_kiocb *nxt = NULL;
1655 io_complete_rw_common(kiocb, res);
1656 io_put_req_find_next(req, &nxt);
1661 static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
1663 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
1665 if (kiocb->ki_flags & IOCB_WRITE)
1666 kiocb_end_write(req);
1668 if (res != req->result)
1669 req_set_fail_links(req);
1672 req->flags |= REQ_F_IOPOLL_COMPLETED;
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.
1681 static void io_iopoll_req_issued(struct io_kiocb *req)
1683 struct io_ring_ctx *ctx = req->ctx;
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.
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;
1695 list_req = list_first_entry(&ctx->poll_list, struct io_kiocb,
1697 if (list_req->file != req->file)
1698 ctx->poll_multi_file = true;
1702 * For fast devices, IO may have already completed. If it has, add
1703 * it to the front so we find it first.
1705 if (req->flags & REQ_F_IOPOLL_COMPLETED)
1706 list_add(&req->list, &ctx->poll_list);
1708 list_add_tail(&req->list, &ctx->poll_list);
1711 static void io_file_put(struct io_submit_state *state)
1714 int diff = state->has_refs - state->used_refs;
1717 fput_many(state->file, diff);
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.
1727 static struct file *io_file_get(struct io_submit_state *state, int fd)
1733 if (state->fd == fd) {
1740 state->file = fget_many(fd, state->ios_left);
1745 state->has_refs = state->ios_left;
1746 state->used_refs = 1;
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
1756 static bool io_file_supports_async(struct file *file)
1758 umode_t mode = file_inode(file)->i_mode;
1760 if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode))
1762 if (S_ISREG(mode) && file->f_op != &io_uring_fops)
1768 static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe,
1769 bool force_nonblock)
1771 struct io_ring_ctx *ctx = req->ctx;
1772 struct kiocb *kiocb = &req->rw.kiocb;
1779 if (S_ISREG(file_inode(req->file)->i_mode))
1780 req->flags |= REQ_F_ISREG;
1782 kiocb->ki_pos = READ_ONCE(sqe->off);
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;
1787 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
1788 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
1790 ioprio = READ_ONCE(sqe->ioprio);
1792 ret = ioprio_check_cap(ioprio);
1796 kiocb->ki_ioprio = ioprio;
1798 kiocb->ki_ioprio = get_current_ioprio();
1800 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
1804 /* don't allow async punt if RWF_NOWAIT was requested */
1805 if ((kiocb->ki_flags & IOCB_NOWAIT) ||
1806 (req->file->f_flags & O_NONBLOCK))
1807 req->flags |= REQ_F_NOWAIT;
1810 kiocb->ki_flags |= IOCB_NOWAIT;
1812 if (ctx->flags & IORING_SETUP_IOPOLL) {
1813 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
1814 !kiocb->ki_filp->f_op->iopoll)
1817 kiocb->ki_flags |= IOCB_HIPRI;
1818 kiocb->ki_complete = io_complete_rw_iopoll;
1821 if (kiocb->ki_flags & IOCB_HIPRI)
1823 kiocb->ki_complete = io_complete_rw;
1826 req->rw.addr = READ_ONCE(sqe->addr);
1827 req->rw.len = READ_ONCE(sqe->len);
1828 /* we own ->private, reuse it for the buffer index */
1829 req->rw.kiocb.private = (void *) (unsigned long)
1830 READ_ONCE(sqe->buf_index);
1834 static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
1840 case -ERESTARTNOINTR:
1841 case -ERESTARTNOHAND:
1842 case -ERESTART_RESTARTBLOCK:
1844 * We can't just restart the syscall, since previously
1845 * submitted sqes may already be in progress. Just fail this
1851 kiocb->ki_complete(kiocb, ret, 0);
1855 static void kiocb_done(struct kiocb *kiocb, ssize_t ret, struct io_kiocb **nxt,
1858 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
1860 if (req->flags & REQ_F_CUR_POS)
1861 req->file->f_pos = kiocb->ki_pos;
1862 if (in_async && ret >= 0 && kiocb->ki_complete == io_complete_rw)
1863 *nxt = __io_complete_rw(kiocb, ret);
1865 io_rw_done(kiocb, ret);
1868 static ssize_t io_import_fixed(struct io_kiocb *req, int rw,
1869 struct iov_iter *iter)
1871 struct io_ring_ctx *ctx = req->ctx;
1872 size_t len = req->rw.len;
1873 struct io_mapped_ubuf *imu;
1874 unsigned index, buf_index;
1878 /* attempt to use fixed buffers without having provided iovecs */
1879 if (unlikely(!ctx->user_bufs))
1882 buf_index = (unsigned long) req->rw.kiocb.private;
1883 if (unlikely(buf_index >= ctx->nr_user_bufs))
1886 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
1887 imu = &ctx->user_bufs[index];
1888 buf_addr = req->rw.addr;
1891 if (buf_addr + len < buf_addr)
1893 /* not inside the mapped region */
1894 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
1898 * May not be a start of buffer, set size appropriately
1899 * and advance us to the beginning.
1901 offset = buf_addr - imu->ubuf;
1902 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
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
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
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.
1921 const struct bio_vec *bvec = imu->bvec;
1923 if (offset <= bvec->bv_len) {
1924 iov_iter_advance(iter, offset);
1926 unsigned long seg_skip;
1928 /* skip first vec */
1929 offset -= bvec->bv_len;
1930 seg_skip = 1 + (offset >> PAGE_SHIFT);
1932 iter->bvec = bvec + seg_skip;
1933 iter->nr_segs -= seg_skip;
1934 iter->count -= bvec->bv_len + offset;
1935 iter->iov_offset = offset & ~PAGE_MASK;
1942 static ssize_t io_import_iovec(int rw, struct io_kiocb *req,
1943 struct iovec **iovec, struct iov_iter *iter)
1945 void __user *buf = u64_to_user_ptr(req->rw.addr);
1946 size_t sqe_len = req->rw.len;
1949 opcode = req->opcode;
1950 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
1952 return io_import_fixed(req, rw, iter);
1955 /* buffer index only valid with fixed read/write */
1956 if (req->rw.kiocb.private)
1959 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
1961 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
1967 struct io_async_rw *iorw = &req->io->rw;
1970 iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size);
1971 if (iorw->iov == iorw->fast_iov)
1979 #ifdef CONFIG_COMPAT
1980 if (req->ctx->compat)
1981 return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV,
1985 return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter);
1989 * For files that don't have ->read_iter() and ->write_iter(), handle them
1990 * by looping over ->read() or ->write() manually.
1992 static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb,
1993 struct iov_iter *iter)
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.
2002 if (kiocb->ki_flags & IOCB_HIPRI)
2004 if (kiocb->ki_flags & IOCB_NOWAIT)
2007 while (iov_iter_count(iter)) {
2011 if (!iov_iter_is_bvec(iter)) {
2012 iovec = iov_iter_iovec(iter);
2014 /* fixed buffers import bvec */
2015 iovec.iov_base = kmap(iter->bvec->bv_page)
2017 iovec.iov_len = min(iter->count,
2018 iter->bvec->bv_len - iter->iov_offset);
2022 nr = file->f_op->read(file, iovec.iov_base,
2023 iovec.iov_len, &kiocb->ki_pos);
2025 nr = file->f_op->write(file, iovec.iov_base,
2026 iovec.iov_len, &kiocb->ki_pos);
2029 if (iov_iter_is_bvec(iter))
2030 kunmap(iter->bvec->bv_page);
2038 if (nr != iovec.iov_len)
2040 iov_iter_advance(iter, nr);
2046 static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size,
2047 struct iovec *iovec, struct iovec *fast_iov,
2048 struct iov_iter *iter)
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);
2060 static int io_alloc_async_ctx(struct io_kiocb *req)
2062 if (!io_op_defs[req->opcode].async_ctx)
2064 req->io = kmalloc(sizeof(*req->io), GFP_KERNEL);
2065 return req->io == NULL;
2068 static void io_rw_async(struct io_wq_work **workptr)
2070 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2071 struct iovec *iov = NULL;
2073 if (req->io->rw.iov != req->io->rw.fast_iov)
2074 iov = req->io->rw.iov;
2075 io_wq_submit_work(workptr);
2079 static 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)
2083 if (req->opcode == IORING_OP_READ_FIXED ||
2084 req->opcode == IORING_OP_WRITE_FIXED)
2086 if (!req->io && io_alloc_async_ctx(req))
2089 io_req_map_rw(req, io_size, iovec, fast_iov, iter);
2090 req->work.func = io_rw_async;
2094 static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
2095 bool force_nonblock)
2097 struct io_async_ctx *io;
2098 struct iov_iter iter;
2101 ret = io_prep_rw(req, sqe, force_nonblock);
2105 if (unlikely(!(req->file->f_mode & FMODE_READ)))
2112 io->rw.iov = io->rw.fast_iov;
2114 ret = io_import_iovec(READ, req, &io->rw.iov, &iter);
2119 io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter);
2123 static int io_read(struct io_kiocb *req, struct io_kiocb **nxt,
2124 bool force_nonblock)
2126 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
2127 struct kiocb *kiocb = &req->rw.kiocb;
2128 struct iov_iter iter;
2130 ssize_t io_size, ret;
2132 ret = io_import_iovec(READ, req, &iovec, &iter);
2136 /* Ensure we clear previously set non-block flag */
2137 if (!force_nonblock)
2138 req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT;
2142 if (req->flags & REQ_F_LINK)
2143 req->result = io_size;
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
2149 if (force_nonblock && !io_file_supports_async(req->file)) {
2150 req->flags |= REQ_F_MUST_PUNT;
2154 iov_count = iov_iter_count(&iter);
2155 ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count);
2159 if (req->file->f_op->read_iter)
2160 ret2 = call_read_iter(req->file, kiocb, &iter);
2162 ret2 = loop_rw_iter(READ, req->file, kiocb, &iter);
2164 /* Catch -EAGAIN return for forced non-blocking submission */
2165 if (!force_nonblock || ret2 != -EAGAIN) {
2166 kiocb_done(kiocb, ret2, nxt, req->in_async);
2169 ret = io_setup_async_rw(req, io_size, iovec,
2170 inline_vecs, &iter);
2177 if (!io_wq_current_is_worker())
2182 static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
2183 bool force_nonblock)
2185 struct io_async_ctx *io;
2186 struct iov_iter iter;
2189 ret = io_prep_rw(req, sqe, force_nonblock);
2193 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
2200 io->rw.iov = io->rw.fast_iov;
2202 ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter);
2207 io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter);
2211 static int io_write(struct io_kiocb *req, struct io_kiocb **nxt,
2212 bool force_nonblock)
2214 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
2215 struct kiocb *kiocb = &req->rw.kiocb;
2216 struct iov_iter iter;
2218 ssize_t ret, io_size;
2220 ret = io_import_iovec(WRITE, req, &iovec, &iter);
2224 /* Ensure we clear previously set non-block flag */
2225 if (!force_nonblock)
2226 req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT;
2230 if (req->flags & REQ_F_LINK)
2231 req->result = io_size;
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
2237 if (force_nonblock && !io_file_supports_async(req->file)) {
2238 req->flags |= REQ_F_MUST_PUNT;
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))
2247 iov_count = iov_iter_count(&iter);
2248 ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count);
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.
2259 if (req->flags & REQ_F_ISREG) {
2260 __sb_start_write(file_inode(req->file)->i_sb,
2261 SB_FREEZE_WRITE, true);
2262 __sb_writers_release(file_inode(req->file)->i_sb,
2265 kiocb->ki_flags |= IOCB_WRITE;
2267 if (req->file->f_op->write_iter)
2268 ret2 = call_write_iter(req->file, kiocb, &iter);
2270 ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter);
2271 if (!force_nonblock || ret2 != -EAGAIN) {
2272 kiocb_done(kiocb, ret2, nxt, req->in_async);
2275 ret = io_setup_async_rw(req, io_size, iovec,
2276 inline_vecs, &iter);
2283 if (!io_wq_current_is_worker())
2289 * IORING_OP_NOP just posts a completion event, nothing else.
2291 static int io_nop(struct io_kiocb *req)
2293 struct io_ring_ctx *ctx = req->ctx;
2295 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
2298 io_cqring_add_event(req, 0);
2303 static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2305 struct io_ring_ctx *ctx = req->ctx;
2310 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
2312 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
2315 req->sync.flags = READ_ONCE(sqe->fsync_flags);
2316 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
2319 req->sync.off = READ_ONCE(sqe->off);
2320 req->sync.len = READ_ONCE(sqe->len);
2324 static bool io_req_cancelled(struct io_kiocb *req)
2326 if (req->work.flags & IO_WQ_WORK_CANCEL) {
2327 req_set_fail_links(req);
2328 io_cqring_add_event(req, -ECANCELED);
2336 static void io_link_work_cb(struct io_wq_work **workptr)
2338 struct io_wq_work *work = *workptr;
2339 struct io_kiocb *link = work->data;
2341 io_queue_linked_timeout(link);
2342 work->func = io_wq_submit_work;
2345 static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt)
2347 struct io_kiocb *link;
2349 io_prep_async_work(nxt, &link);
2350 *workptr = &nxt->work;
2352 nxt->work.flags |= IO_WQ_WORK_CB;
2353 nxt->work.func = io_link_work_cb;
2354 nxt->work.data = link;
2358 static void io_fsync_finish(struct io_wq_work **workptr)
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;
2365 if (io_req_cancelled(req))
2368 ret = vfs_fsync_range(req->file, req->sync.off,
2369 end > 0 ? end : LLONG_MAX,
2370 req->sync.flags & IORING_FSYNC_DATASYNC);
2372 req_set_fail_links(req);
2373 io_cqring_add_event(req, ret);
2374 io_put_req_find_next(req, &nxt);
2376 io_wq_assign_next(workptr, nxt);
2379 static int io_fsync(struct io_kiocb *req, struct io_kiocb **nxt,
2380 bool force_nonblock)
2382 struct io_wq_work *work, *old_work;
2384 /* fsync always requires a blocking context */
2385 if (force_nonblock) {
2387 req->work.func = io_fsync_finish;
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);
2398 static void io_fallocate_finish(struct io_wq_work **workptr)
2400 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2401 struct io_kiocb *nxt = NULL;
2404 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
2407 req_set_fail_links(req);
2408 io_cqring_add_event(req, ret);
2409 io_put_req_find_next(req, &nxt);
2411 io_wq_assign_next(workptr, nxt);
2414 static int io_fallocate_prep(struct io_kiocb *req,
2415 const struct io_uring_sqe *sqe)
2417 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
2420 req->sync.off = READ_ONCE(sqe->off);
2421 req->sync.len = READ_ONCE(sqe->addr);
2422 req->sync.mode = READ_ONCE(sqe->len);
2426 static int io_fallocate(struct io_kiocb *req, struct io_kiocb **nxt,
2427 bool force_nonblock)
2429 struct io_wq_work *work, *old_work;
2431 /* fallocate always requiring blocking context */
2432 if (force_nonblock) {
2434 req->work.func = io_fallocate_finish;
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);
2446 static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2450 if (sqe->ioprio || sqe->buf_index)
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);
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;
2468 static int io_openat(struct io_kiocb *req, struct io_kiocb **nxt,
2469 bool force_nonblock)
2471 struct open_flags op;
2472 struct open_how how;
2476 if (force_nonblock) {
2477 req->work.flags |= IO_WQ_WORK_NEEDS_FILES;
2481 how = build_open_how(req->open.flags, req->open.mode);
2482 ret = build_open_flags(&how, &op);
2486 ret = get_unused_fd_flags(how.flags);
2490 file = do_filp_open(req->open.dfd, req->open.filename, &op);
2493 ret = PTR_ERR(file);
2495 fsnotify_open(file);
2496 fd_install(ret, file);
2499 putname(req->open.filename);
2501 req_set_fail_links(req);
2502 io_cqring_add_event(req, ret);
2503 io_put_req_find_next(req, nxt);
2507 static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2509 #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
2510 if (sqe->ioprio || sqe->buf_index || sqe->off)
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);
2522 static int io_madvise(struct io_kiocb *req, struct io_kiocb **nxt,
2523 bool force_nonblock)
2525 #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
2526 struct io_madvise *ma = &req->madvise;
2532 ret = do_madvise(ma->addr, ma->len, ma->advice);
2534 req_set_fail_links(req);
2535 io_cqring_add_event(req, ret);
2536 io_put_req_find_next(req, nxt);
2543 static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2545 if (sqe->ioprio || sqe->buf_index || sqe->addr)
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);
2554 static int io_fadvise(struct io_kiocb *req, struct io_kiocb **nxt,
2555 bool force_nonblock)
2557 struct io_fadvise *fa = &req->fadvise;
2560 /* DONTNEED may block, others _should_ not */
2561 if (fa->advice == POSIX_FADV_DONTNEED && force_nonblock)
2564 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
2566 req_set_fail_links(req);
2567 io_cqring_add_event(req, ret);
2568 io_put_req_find_next(req, nxt);
2572 static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2574 unsigned lookup_flags;
2577 if (sqe->ioprio || sqe->buf_index)
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);
2586 if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.flags))
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;
2599 static int io_statx(struct io_kiocb *req, struct io_kiocb **nxt,
2600 bool force_nonblock)
2602 struct io_open *ctx = &req->open;
2603 unsigned lookup_flags;
2611 if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->flags))
2615 /* filename_lookup() drops it, keep a reference */
2616 ctx->filename->refcnt++;
2618 ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path,
2623 ret = vfs_getattr(&path, &stat, ctx->mask, ctx->flags);
2625 if (retry_estale(ret, lookup_flags)) {
2626 lookup_flags |= LOOKUP_REVAL;
2630 ret = cp_statx(&stat, ctx->buffer);
2632 putname(ctx->filename);
2634 req_set_fail_links(req);
2635 io_cqring_add_event(req, ret);
2636 io_put_req_find_next(req, nxt);
2640 static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2643 * If we queue this for async, it must not be cancellable. That would
2644 * leave the 'file' in an undeterminate state.
2646 req->work.flags |= IO_WQ_WORK_NO_CANCEL;
2648 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
2649 sqe->rw_flags || sqe->buf_index)
2651 if (sqe->flags & IOSQE_FIXED_FILE)
2654 req->close.fd = READ_ONCE(sqe->fd);
2655 if (req->file->f_op == &io_uring_fops ||
2656 req->close.fd == req->ring_fd)
2662 static void io_close_finish(struct io_wq_work **workptr)
2664 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2665 struct io_kiocb *nxt = NULL;
2667 /* Invoked with files, we need to do the close */
2668 if (req->work.files) {
2671 ret = filp_close(req->close.put_file, req->work.files);
2673 req_set_fail_links(req);
2675 io_cqring_add_event(req, ret);
2678 fput(req->close.put_file);
2680 /* we bypassed the re-issue, drop the submission reference */
2682 io_put_req_find_next(req, &nxt);
2684 io_wq_assign_next(workptr, nxt);
2687 static int io_close(struct io_kiocb *req, struct io_kiocb **nxt,
2688 bool force_nonblock)
2692 req->close.put_file = NULL;
2693 ret = __close_fd_get_file(req->close.fd, &req->close.put_file);
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;
2704 * No ->flush(), safely close from here and just punt the
2705 * fput() to async context.
2707 ret = filp_close(req->close.put_file, current->files);
2710 req_set_fail_links(req);
2711 io_cqring_add_event(req, ret);
2713 if (io_wq_current_is_worker()) {
2714 struct io_wq_work *old_work, *work;
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);
2724 req->work.func = io_close_finish;
2728 static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2730 struct io_ring_ctx *ctx = req->ctx;
2735 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
2737 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
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);
2746 static void io_sync_file_range_finish(struct io_wq_work **workptr)
2748 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2749 struct io_kiocb *nxt = NULL;
2752 if (io_req_cancelled(req))
2755 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
2758 req_set_fail_links(req);
2759 io_cqring_add_event(req, ret);
2760 io_put_req_find_next(req, &nxt);
2762 io_wq_assign_next(workptr, nxt);
2765 static int io_sync_file_range(struct io_kiocb *req, struct io_kiocb **nxt,
2766 bool force_nonblock)
2768 struct io_wq_work *work, *old_work;
2770 /* sync_file_range always requires a blocking context */
2771 if (force_nonblock) {
2773 req->work.func = io_sync_file_range_finish;
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);
2784 #if defined(CONFIG_NET)
2785 static void io_sendrecv_async(struct io_wq_work **workptr)
2787 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
2788 struct iovec *iov = NULL;
2790 if (req->io->rw.iov != req->io->rw.fast_iov)
2791 iov = req->io->msg.iov;
2792 io_wq_submit_work(workptr);
2797 static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2799 #if defined(CONFIG_NET)
2800 struct io_sr_msg *sr = &req->sr_msg;
2801 struct io_async_ctx *io = req->io;
2803 sr->msg_flags = READ_ONCE(sqe->msg_flags);
2804 sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr));
2809 io->msg.iov = io->msg.fast_iov;
2810 return sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags,
2817 static int io_sendmsg(struct io_kiocb *req, struct io_kiocb **nxt,
2818 bool force_nonblock)
2820 #if defined(CONFIG_NET)
2821 struct io_async_msghdr *kmsg = NULL;
2822 struct socket *sock;
2825 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
2828 sock = sock_from_file(req->file, &ret);
2830 struct io_async_ctx io;
2831 struct sockaddr_storage addr;
2835 kmsg = &req->io->msg;
2836 kmsg->msg.msg_name = &addr;
2837 /* if iov is set, it's allocated already */
2839 kmsg->iov = kmsg->fast_iov;
2840 kmsg->msg.msg_iter.iov = kmsg->iov;
2842 struct io_sr_msg *sr = &req->sr_msg;
2845 kmsg->msg.msg_name = &addr;
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);
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;
2860 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
2861 if (force_nonblock && ret == -EAGAIN) {
2864 if (io_alloc_async_ctx(req))
2866 memcpy(&req->io->msg, &io.msg, sizeof(io.msg));
2867 req->work.func = io_sendrecv_async;
2870 if (ret == -ERESTARTSYS)
2874 if (!io_wq_current_is_worker() && kmsg && kmsg->iov != kmsg->fast_iov)
2876 io_cqring_add_event(req, ret);
2878 req_set_fail_links(req);
2879 io_put_req_find_next(req, nxt);
2886 static int io_recvmsg_prep(struct io_kiocb *req,
2887 const struct io_uring_sqe *sqe)
2889 #if defined(CONFIG_NET)
2890 struct io_sr_msg *sr = &req->sr_msg;
2891 struct io_async_ctx *io = req->io;
2893 sr->msg_flags = READ_ONCE(sqe->msg_flags);
2894 sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr));
2899 io->msg.iov = io->msg.fast_iov;
2900 return recvmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags,
2901 &io->msg.uaddr, &io->msg.iov);
2907 static int io_recvmsg(struct io_kiocb *req, struct io_kiocb **nxt,
2908 bool force_nonblock)
2910 #if defined(CONFIG_NET)
2911 struct io_async_msghdr *kmsg = NULL;
2912 struct socket *sock;
2915 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
2918 sock = sock_from_file(req->file, &ret);
2920 struct io_async_ctx io;
2921 struct sockaddr_storage addr;
2925 kmsg = &req->io->msg;
2926 kmsg->msg.msg_name = &addr;
2927 /* if iov is set, it's allocated already */
2929 kmsg->iov = kmsg->fast_iov;
2930 kmsg->msg.msg_iter.iov = kmsg->iov;
2932 struct io_sr_msg *sr = &req->sr_msg;
2935 kmsg->msg.msg_name = &addr;
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,
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;
2951 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg,
2952 kmsg->uaddr, flags);
2953 if (force_nonblock && ret == -EAGAIN) {
2956 if (io_alloc_async_ctx(req))
2958 memcpy(&req->io->msg, &io.msg, sizeof(io.msg));
2959 req->work.func = io_sendrecv_async;
2962 if (ret == -ERESTARTSYS)
2966 if (!io_wq_current_is_worker() && kmsg && kmsg->iov != kmsg->fast_iov)
2968 io_cqring_add_event(req, ret);
2970 req_set_fail_links(req);
2971 io_put_req_find_next(req, nxt);
2978 static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
2980 #if defined(CONFIG_NET)
2981 struct io_accept *accept = &req->accept;
2983 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
2985 if (sqe->ioprio || sqe->len || sqe->buf_index)
2988 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
2989 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
2990 accept->flags = READ_ONCE(sqe->accept_flags);
2997 #if defined(CONFIG_NET)
2998 static int __io_accept(struct io_kiocb *req, struct io_kiocb **nxt,
2999 bool force_nonblock)
3001 struct io_accept *accept = &req->accept;
3002 unsigned file_flags;
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)
3010 if (ret == -ERESTARTSYS)
3013 req_set_fail_links(req);
3014 io_cqring_add_event(req, ret);
3015 io_put_req_find_next(req, nxt);
3019 static void io_accept_finish(struct io_wq_work **workptr)
3021 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
3022 struct io_kiocb *nxt = NULL;
3024 if (io_req_cancelled(req))
3026 __io_accept(req, &nxt, false);
3028 io_wq_assign_next(workptr, nxt);
3032 static int io_accept(struct io_kiocb *req, struct io_kiocb **nxt,
3033 bool force_nonblock)
3035 #if defined(CONFIG_NET)
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;
3051 static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3053 #if defined(CONFIG_NET)
3054 struct io_connect *conn = &req->connect;
3055 struct io_async_ctx *io = req->io;
3057 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
3059 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
3062 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
3063 conn->addr_len = READ_ONCE(sqe->addr2);
3068 return move_addr_to_kernel(conn->addr, conn->addr_len,
3069 &io->connect.address);
3075 static int io_connect(struct io_kiocb *req, struct io_kiocb **nxt,
3076 bool force_nonblock)
3078 #if defined(CONFIG_NET)
3079 struct io_async_ctx __io, *io;
3080 unsigned file_flags;
3086 ret = move_addr_to_kernel(req->connect.addr,
3087 req->connect.addr_len,
3088 &__io.connect.address);
3094 file_flags = force_nonblock ? O_NONBLOCK : 0;
3096 ret = __sys_connect_file(req->file, &io->connect.address,
3097 req->connect.addr_len, file_flags);
3098 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
3101 if (io_alloc_async_ctx(req)) {
3105 memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect));
3108 if (ret == -ERESTARTSYS)
3112 req_set_fail_links(req);
3113 io_cqring_add_event(req, ret);
3114 io_put_req_find_next(req, nxt);
3121 static void io_poll_remove_one(struct io_kiocb *req)
3123 struct io_poll_iocb *poll = &req->poll;
3125 spin_lock(&poll->head->lock);
3126 WRITE_ONCE(poll->canceled, true);
3127 if (!list_empty(&poll->wait.entry)) {
3128 list_del_init(&poll->wait.entry);
3129 io_queue_async_work(req);
3131 spin_unlock(&poll->head->lock);
3132 hash_del(&req->hash_node);
3135 static void io_poll_remove_all(struct io_ring_ctx *ctx)
3137 struct hlist_node *tmp;
3138 struct io_kiocb *req;
3141 spin_lock_irq(&ctx->completion_lock);
3142 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
3143 struct hlist_head *list;
3145 list = &ctx->cancel_hash[i];
3146 hlist_for_each_entry_safe(req, tmp, list, hash_node)
3147 io_poll_remove_one(req);
3149 spin_unlock_irq(&ctx->completion_lock);
3152 static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
3154 struct hlist_head *list;
3155 struct io_kiocb *req;
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) {
3160 io_poll_remove_one(req);
3168 static int io_poll_remove_prep(struct io_kiocb *req,
3169 const struct io_uring_sqe *sqe)
3171 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3173 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
3177 req->poll.addr = READ_ONCE(sqe->addr);
3182 * Find a running poll command that matches one specified in sqe->addr,
3183 * and remove it if found.
3185 static int io_poll_remove(struct io_kiocb *req)
3187 struct io_ring_ctx *ctx = req->ctx;
3191 addr = req->poll.addr;
3192 spin_lock_irq(&ctx->completion_lock);
3193 ret = io_poll_cancel(ctx, addr);
3194 spin_unlock_irq(&ctx->completion_lock);
3196 io_cqring_add_event(req, ret);
3198 req_set_fail_links(req);
3203 static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
3205 struct io_ring_ctx *ctx = req->ctx;
3207 req->poll.done = true;
3209 io_cqring_fill_event(req, error);
3211 io_cqring_fill_event(req, mangle_poll(mask));
3212 io_commit_cqring(ctx);
3215 static void io_poll_complete_work(struct io_wq_work **workptr)
3217 struct io_wq_work *work = *workptr;
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;
3222 struct io_kiocb *nxt = NULL;
3226 if (work->flags & IO_WQ_WORK_CANCEL) {
3227 WRITE_ONCE(poll->canceled, true);
3229 } else if (READ_ONCE(poll->canceled)) {
3233 if (ret != -ECANCELED)
3234 mask = vfs_poll(poll->file, &pt) & poll->events;
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.
3243 spin_lock_irq(&ctx->completion_lock);
3244 if (!mask && ret != -ECANCELED) {
3245 add_wait_queue(poll->head, &poll->wait);
3246 spin_unlock_irq(&ctx->completion_lock);
3249 hash_del(&req->hash_node);
3250 io_poll_complete(req, mask, ret);
3251 spin_unlock_irq(&ctx->completion_lock);
3253 io_cqring_ev_posted(ctx);
3256 req_set_fail_links(req);
3257 io_put_req_find_next(req, &nxt);
3259 io_wq_assign_next(workptr, nxt);
3262 static void __io_poll_flush(struct io_ring_ctx *ctx, struct llist_node *nodes)
3264 struct io_kiocb *req, *tmp;
3265 struct req_batch rb;
3267 rb.to_free = rb.need_iter = 0;
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);
3273 if (refcount_dec_and_test(&req->refs) &&
3274 !io_req_multi_free(&rb, req)) {
3275 req->flags |= REQ_F_COMP_LOCKED;
3279 spin_unlock_irq(&ctx->completion_lock);
3281 io_cqring_ev_posted(ctx);
3282 io_free_req_many(ctx, &rb);
3285 static void io_poll_flush(struct io_wq_work **workptr)
3287 struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work);
3288 struct llist_node *nodes;
3290 nodes = llist_del_all(&req->ctx->poll_llist);
3292 __io_poll_flush(req->ctx, nodes);
3295 static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
3298 struct io_poll_iocb *poll = wait->private;
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);
3303 /* for instances that support it check for an event match first: */
3304 if (mask && !(mask & poll->events))
3307 list_del_init(&poll->wait.entry);
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.
3316 unsigned long flags;
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;
3324 spin_unlock_irqrestore(&ctx->completion_lock, flags);
3326 io_cqring_ev_posted(ctx);
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))
3335 req->work.func = io_poll_flush;
3339 io_queue_async_work(req);
3344 struct io_poll_table {
3345 struct poll_table_struct pt;
3346 struct io_kiocb *req;
3350 static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
3351 struct poll_table_struct *p)
3353 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
3355 if (unlikely(pt->req->poll.head)) {
3356 pt->error = -EINVAL;
3361 pt->req->poll.head = head;
3362 add_wait_queue(head, &pt->req->poll.wait);
3365 static void io_poll_req_insert(struct io_kiocb *req)
3367 struct io_ring_ctx *ctx = req->ctx;
3368 struct hlist_head *list;
3370 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
3371 hlist_add_head(&req->hash_node, list);
3374 static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3376 struct io_poll_iocb *poll = &req->poll;
3379 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3381 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
3386 events = READ_ONCE(sqe->poll_events);
3387 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP;
3391 static int io_poll_add(struct io_kiocb *req, struct io_kiocb **nxt)
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;
3399 INIT_IO_WORK(&req->work, io_poll_complete_work);
3400 INIT_HLIST_NODE(&req->hash_node);
3404 poll->canceled = false;
3406 ipt.pt._qproc = io_poll_queue_proc;
3407 ipt.pt._key = poll->events;
3409 ipt.error = -EINVAL; /* same as no support for IOCB_CMD_POLL */
3411 /* initialized the list so that we can do list_empty checks */
3412 INIT_LIST_HEAD(&poll->wait.entry);
3413 init_waitqueue_func_entry(&poll->wait, io_poll_wake);
3414 poll->wait.private = poll;
3416 INIT_LIST_HEAD(&req->list);
3418 mask = vfs_poll(poll->file, &ipt.pt) & poll->events;
3420 spin_lock_irq(&ctx->completion_lock);
3421 if (likely(poll->head)) {
3422 spin_lock(&poll->head->lock);
3423 if (unlikely(list_empty(&poll->wait.entry))) {
3429 if (mask || ipt.error)
3430 list_del_init(&poll->wait.entry);
3432 WRITE_ONCE(poll->canceled, true);
3433 else if (!poll->done) /* actually waiting for an event */
3434 io_poll_req_insert(req);
3435 spin_unlock(&poll->head->lock);
3437 if (mask) { /* no async, we'd stolen it */
3439 io_poll_complete(req, mask, 0);
3441 spin_unlock_irq(&ctx->completion_lock);
3444 io_cqring_ev_posted(ctx);
3445 io_put_req_find_next(req, nxt);
3450 static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
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;
3456 unsigned long flags;
3458 atomic_inc(&ctx->cq_timeouts);
3460 spin_lock_irqsave(&ctx->completion_lock, flags);
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.
3465 if (!list_empty(&req->list)) {
3466 struct io_kiocb *prev;
3469 * Adjust the reqs sequence before the current one because it
3470 * will consume a slot in the cq_ring and the cq_tail
3471 * pointer will be increased, otherwise other timeout reqs may
3472 * return in advance without waiting for enough wait_nr.
3475 list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list)
3477 list_del_init(&req->list);
3480 io_cqring_fill_event(req, -ETIME);
3481 io_commit_cqring(ctx);
3482 spin_unlock_irqrestore(&ctx->completion_lock, flags);
3484 io_cqring_ev_posted(ctx);
3485 req_set_fail_links(req);
3487 return HRTIMER_NORESTART;
3490 static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
3492 struct io_kiocb *req;
3495 list_for_each_entry(req, &ctx->timeout_list, list) {
3496 if (user_data == req->user_data) {
3497 list_del_init(&req->list);
3506 ret = hrtimer_try_to_cancel(&req->io->timeout.timer);
3510 req_set_fail_links(req);
3511 io_cqring_fill_event(req, -ECANCELED);
3516 static int io_timeout_remove_prep(struct io_kiocb *req,
3517 const struct io_uring_sqe *sqe)
3519 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3521 if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len)
3524 req->timeout.addr = READ_ONCE(sqe->addr);
3525 req->timeout.flags = READ_ONCE(sqe->timeout_flags);
3526 if (req->timeout.flags)
3533 * Remove or update an existing timeout command
3535 static int io_timeout_remove(struct io_kiocb *req)
3537 struct io_ring_ctx *ctx = req->ctx;
3540 spin_lock_irq(&ctx->completion_lock);
3541 ret = io_timeout_cancel(ctx, req->timeout.addr);
3543 io_cqring_fill_event(req, ret);
3544 io_commit_cqring(ctx);
3545 spin_unlock_irq(&ctx->completion_lock);
3546 io_cqring_ev_posted(ctx);
3548 req_set_fail_links(req);
3553 static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
3554 bool is_timeout_link)
3556 struct io_timeout_data *data;
3559 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3561 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
3563 if (sqe->off && is_timeout_link)
3565 flags = READ_ONCE(sqe->timeout_flags);
3566 if (flags & ~IORING_TIMEOUT_ABS)
3569 req->timeout.count = READ_ONCE(sqe->off);
3571 if (!req->io && io_alloc_async_ctx(req))
3574 data = &req->io->timeout;
3576 req->flags |= REQ_F_TIMEOUT;
3578 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
3581 if (flags & IORING_TIMEOUT_ABS)
3582 data->mode = HRTIMER_MODE_ABS;
3584 data->mode = HRTIMER_MODE_REL;
3586 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
3590 static int io_timeout(struct io_kiocb *req)
3593 struct io_ring_ctx *ctx = req->ctx;
3594 struct io_timeout_data *data;
3595 struct list_head *entry;
3598 data = &req->io->timeout;
3601 * sqe->off holds how many events that need to occur for this
3602 * timeout event to be satisfied. If it isn't set, then this is
3603 * a pure timeout request, sequence isn't used.
3605 count = req->timeout.count;
3607 req->flags |= REQ_F_TIMEOUT_NOSEQ;
3608 spin_lock_irq(&ctx->completion_lock);
3609 entry = ctx->timeout_list.prev;
3613 req->sequence = ctx->cached_sq_head + count - 1;
3614 data->seq_offset = count;
3617 * Insertion sort, ensuring the first entry in the list is always
3618 * the one we need first.
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);
3623 unsigned nxt_sq_head;
3624 long long tmp, tmp_nxt;
3625 u32 nxt_offset = nxt->io->timeout.seq_offset;
3627 if (nxt->flags & REQ_F_TIMEOUT_NOSEQ)
3631 * Since cached_sq_head + count - 1 can overflow, use type long
3634 tmp = (long long)ctx->cached_sq_head + count - 1;
3635 nxt_sq_head = nxt->sequence - nxt_offset + 1;
3636 tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1;
3639 * cached_sq_head may overflow, and it will never overflow twice
3640 * once there is some timeout req still be valid.
3642 if (ctx->cached_sq_head < nxt_sq_head)
3649 * Sequence of reqs after the insert one and itself should
3650 * be adjusted because each timeout req consumes a slot.
3655 req->sequence -= span;
3657 list_add(&req->list, entry);
3658 data->timer.function = io_timeout_fn;
3659 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
3660 spin_unlock_irq(&ctx->completion_lock);
3664 static bool io_cancel_cb(struct io_wq_work *work, void *data)
3666 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
3668 return req->user_data == (unsigned long) data;
3671 static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr)
3673 enum io_wq_cancel cancel_ret;
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:
3681 case IO_WQ_CANCEL_RUNNING:
3684 case IO_WQ_CANCEL_NOTFOUND:
3692 static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
3693 struct io_kiocb *req, __u64 sqe_addr,
3694 struct io_kiocb **nxt, int success_ret)
3696 unsigned long flags;
3699 ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr);
3700 if (ret != -ENOENT) {
3701 spin_lock_irqsave(&ctx->completion_lock, flags);
3705 spin_lock_irqsave(&ctx->completion_lock, flags);
3706 ret = io_timeout_cancel(ctx, sqe_addr);
3709 ret = io_poll_cancel(ctx, sqe_addr);
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);
3719 req_set_fail_links(req);
3720 io_put_req_find_next(req, nxt);
3723 static int io_async_cancel_prep(struct io_kiocb *req,
3724 const struct io_uring_sqe *sqe)
3726 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3728 if (sqe->flags || sqe->ioprio || sqe->off || sqe->len ||
3732 req->cancel.addr = READ_ONCE(sqe->addr);
3736 static int io_async_cancel(struct io_kiocb *req, struct io_kiocb **nxt)
3738 struct io_ring_ctx *ctx = req->ctx;
3740 io_async_find_and_cancel(ctx, req, req->cancel.addr, nxt, 0);
3744 static int io_files_update_prep(struct io_kiocb *req,
3745 const struct io_uring_sqe *sqe)
3747 if (sqe->flags || sqe->ioprio || sqe->rw_flags)
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)
3754 req->files_update.arg = READ_ONCE(sqe->addr);
3758 static int io_files_update(struct io_kiocb *req, bool force_nonblock)
3760 struct io_ring_ctx *ctx = req->ctx;
3761 struct io_uring_files_update up;
3764 if (force_nonblock) {
3765 req->work.flags |= IO_WQ_WORK_NEEDS_FILES;
3769 up.offset = req->files_update.offset;
3770 up.fds = req->files_update.arg;
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);
3777 req_set_fail_links(req);
3778 io_cqring_add_event(req, ret);
3783 static int io_req_defer_prep(struct io_kiocb *req,
3784 const struct io_uring_sqe *sqe)
3788 switch (req->opcode) {
3791 case IORING_OP_READV:
3792 case IORING_OP_READ_FIXED:
3793 case IORING_OP_READ:
3794 ret = io_read_prep(req, sqe, true);
3796 case IORING_OP_WRITEV:
3797 case IORING_OP_WRITE_FIXED:
3798 case IORING_OP_WRITE:
3799 ret = io_write_prep(req, sqe, true);
3801 case IORING_OP_POLL_ADD:
3802 ret = io_poll_add_prep(req, sqe);
3804 case IORING_OP_POLL_REMOVE:
3805 ret = io_poll_remove_prep(req, sqe);
3807 case IORING_OP_FSYNC:
3808 ret = io_prep_fsync(req, sqe);
3810 case IORING_OP_SYNC_FILE_RANGE:
3811 ret = io_prep_sfr(req, sqe);
3813 case IORING_OP_SENDMSG:
3814 ret = io_sendmsg_prep(req, sqe);
3816 case IORING_OP_RECVMSG:
3817 ret = io_recvmsg_prep(req, sqe);
3819 case IORING_OP_CONNECT:
3820 ret = io_connect_prep(req, sqe);
3822 case IORING_OP_TIMEOUT:
3823 ret = io_timeout_prep(req, sqe, false);
3825 case IORING_OP_TIMEOUT_REMOVE:
3826 ret = io_timeout_remove_prep(req, sqe);
3828 case IORING_OP_ASYNC_CANCEL:
3829 ret = io_async_cancel_prep(req, sqe);
3831 case IORING_OP_LINK_TIMEOUT:
3832 ret = io_timeout_prep(req, sqe, true);
3834 case IORING_OP_ACCEPT:
3835 ret = io_accept_prep(req, sqe);
3837 case IORING_OP_FALLOCATE:
3838 ret = io_fallocate_prep(req, sqe);
3840 case IORING_OP_OPENAT:
3841 ret = io_openat_prep(req, sqe);
3843 case IORING_OP_CLOSE:
3844 ret = io_close_prep(req, sqe);
3846 case IORING_OP_FILES_UPDATE:
3847 ret = io_files_update_prep(req, sqe);
3849 case IORING_OP_STATX:
3850 ret = io_statx_prep(req, sqe);
3852 case IORING_OP_FADVISE:
3853 ret = io_fadvise_prep(req, sqe);
3855 case IORING_OP_MADVISE:
3856 ret = io_madvise_prep(req, sqe);
3859 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
3868 static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3870 struct io_ring_ctx *ctx = req->ctx;
3873 /* Still need defer if there is pending req in defer list. */
3874 if (!req_need_defer(req) && list_empty(&ctx->defer_list))
3877 if (!req->io && io_alloc_async_ctx(req))
3880 ret = io_req_defer_prep(req, sqe);
3884 spin_lock_irq(&ctx->completion_lock);
3885 if (!req_need_defer(req) && list_empty(&ctx->defer_list)) {
3886 spin_unlock_irq(&ctx->completion_lock);
3890 trace_io_uring_defer(ctx, req, req->user_data);
3891 list_add_tail(&req->list, &ctx->defer_list);
3892 spin_unlock_irq(&ctx->completion_lock);
3893 return -EIOCBQUEUED;
3896 static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
3897 struct io_kiocb **nxt, bool force_nonblock)
3899 struct io_ring_ctx *ctx = req->ctx;
3902 switch (req->opcode) {
3906 case IORING_OP_READV:
3907 case IORING_OP_READ_FIXED:
3908 case IORING_OP_READ:
3910 ret = io_read_prep(req, sqe, force_nonblock);
3914 ret = io_read(req, nxt, force_nonblock);
3916 case IORING_OP_WRITEV:
3917 case IORING_OP_WRITE_FIXED:
3918 case IORING_OP_WRITE:
3920 ret = io_write_prep(req, sqe, force_nonblock);
3924 ret = io_write(req, nxt, force_nonblock);
3926 case IORING_OP_FSYNC:
3928 ret = io_prep_fsync(req, sqe);
3932 ret = io_fsync(req, nxt, force_nonblock);
3934 case IORING_OP_POLL_ADD:
3936 ret = io_poll_add_prep(req, sqe);
3940 ret = io_poll_add(req, nxt);
3942 case IORING_OP_POLL_REMOVE:
3944 ret = io_poll_remove_prep(req, sqe);
3948 ret = io_poll_remove(req);
3950 case IORING_OP_SYNC_FILE_RANGE:
3952 ret = io_prep_sfr(req, sqe);
3956 ret = io_sync_file_range(req, nxt, force_nonblock);
3958 case IORING_OP_SENDMSG:
3960 ret = io_sendmsg_prep(req, sqe);
3964 ret = io_sendmsg(req, nxt, force_nonblock);
3966 case IORING_OP_RECVMSG:
3968 ret = io_recvmsg_prep(req, sqe);
3972 ret = io_recvmsg(req, nxt, force_nonblock);
3974 case IORING_OP_TIMEOUT:
3976 ret = io_timeout_prep(req, sqe, false);
3980 ret = io_timeout(req);
3982 case IORING_OP_TIMEOUT_REMOVE:
3984 ret = io_timeout_remove_prep(req, sqe);
3988 ret = io_timeout_remove(req);
3990 case IORING_OP_ACCEPT:
3992 ret = io_accept_prep(req, sqe);
3996 ret = io_accept(req, nxt, force_nonblock);
3998 case IORING_OP_CONNECT:
4000 ret = io_connect_prep(req, sqe);
4004 ret = io_connect(req, nxt, force_nonblock);
4006 case IORING_OP_ASYNC_CANCEL:
4008 ret = io_async_cancel_prep(req, sqe);
4012 ret = io_async_cancel(req, nxt);
4014 case IORING_OP_FALLOCATE:
4016 ret = io_fallocate_prep(req, sqe);
4020 ret = io_fallocate(req, nxt, force_nonblock);
4022 case IORING_OP_OPENAT:
4024 ret = io_openat_prep(req, sqe);
4028 ret = io_openat(req, nxt, force_nonblock);
4030 case IORING_OP_CLOSE:
4032 ret = io_close_prep(req, sqe);
4036 ret = io_close(req, nxt, force_nonblock);
4038 case IORING_OP_FILES_UPDATE:
4040 ret = io_files_update_prep(req, sqe);
4044 ret = io_files_update(req, force_nonblock);
4046 case IORING_OP_STATX:
4048 ret = io_statx_prep(req, sqe);
4052 ret = io_statx(req, nxt, force_nonblock);
4054 case IORING_OP_FADVISE:
4056 ret = io_fadvise_prep(req, sqe);
4060 ret = io_fadvise(req, nxt, force_nonblock);
4062 case IORING_OP_MADVISE:
4064 ret = io_madvise_prep(req, sqe);
4068 ret = io_madvise(req, nxt, force_nonblock);
4078 if (ctx->flags & IORING_SETUP_IOPOLL) {
4079 const bool in_async = io_wq_current_is_worker();
4081 if (req->result == -EAGAIN)
4084 /* workqueue context doesn't hold uring_lock, grab it now */
4086 mutex_lock(&ctx->uring_lock);
4088 io_iopoll_req_issued(req);
4091 mutex_unlock(&ctx->uring_lock);
4097 static void io_wq_submit_work(struct io_wq_work **workptr)
4099 struct io_wq_work *work = *workptr;
4100 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
4101 struct io_kiocb *nxt = NULL;
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) {
4111 req->has_user = (work->flags & IO_WQ_WORK_HAS_MM) != 0;
4112 req->in_async = true;
4114 ret = io_issue_sqe(req, NULL, &nxt, false);
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.
4126 /* drop submission reference */
4130 req_set_fail_links(req);
4131 io_cqring_add_event(req, ret);
4135 /* if a dependent link is ready, pass it back */
4137 io_wq_assign_next(workptr, nxt);
4140 static int io_req_needs_file(struct io_kiocb *req, int fd)
4142 if (!io_op_defs[req->opcode].needs_file)
4144 if (fd == -1 && io_op_defs[req->opcode].fd_non_neg)
4149 static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
4152 struct fixed_file_table *table;
4154 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
4155 return table->files[index & IORING_FILE_TABLE_MASK];;
4158 static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
4159 const struct io_uring_sqe *sqe)
4161 struct io_ring_ctx *ctx = req->ctx;
4165 flags = READ_ONCE(sqe->flags);
4166 fd = READ_ONCE(sqe->fd);
4168 if (flags & IOSQE_IO_DRAIN)
4169 req->flags |= REQ_F_IO_DRAIN;
4171 if (!io_req_needs_file(req, fd))
4174 if (flags & IOSQE_FIXED_FILE) {
4175 if (unlikely(!ctx->file_data ||
4176 (unsigned) fd >= ctx->nr_user_files))
4178 fd = array_index_nospec(fd, ctx->nr_user_files);
4179 req->file = io_file_from_index(ctx, fd);
4182 req->flags |= REQ_F_FIXED_FILE;
4183 percpu_ref_get(&ctx->file_data->refs);
4185 if (req->needs_fixed_file)
4187 trace_io_uring_file_get(ctx, fd);
4188 req->file = io_file_get(state, fd);
4189 if (unlikely(!req->file))
4196 static int io_grab_files(struct io_kiocb *req)
4199 struct io_ring_ctx *ctx = req->ctx;
4201 if (!req->ring_file)
4205 spin_lock_irq(&ctx->inflight_lock);
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.
4212 if (fcheck(req->ring_fd) == req->ring_file) {
4213 list_add(&req->inflight_entry, &ctx->inflight_list);
4214 req->flags |= REQ_F_INFLIGHT;
4215 req->work.files = current->files;
4218 spin_unlock_irq(&ctx->inflight_lock);
4224 static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
4226 struct io_timeout_data *data = container_of(timer,
4227 struct io_timeout_data, timer);
4228 struct io_kiocb *req = data->req;
4229 struct io_ring_ctx *ctx = req->ctx;
4230 struct io_kiocb *prev = NULL;
4231 unsigned long flags;
4233 spin_lock_irqsave(&ctx->completion_lock, flags);
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.
4239 if (!list_empty(&req->link_list)) {
4240 prev = list_entry(req->link_list.prev, struct io_kiocb,
4242 if (refcount_inc_not_zero(&prev->refs)) {
4243 list_del_init(&req->link_list);
4244 prev->flags &= ~REQ_F_LINK_TIMEOUT;
4249 spin_unlock_irqrestore(&ctx->completion_lock, flags);
4252 req_set_fail_links(prev);
4253 io_async_find_and_cancel(ctx, req, prev->user_data, NULL,
4257 io_cqring_add_event(req, -ETIME);
4260 return HRTIMER_NORESTART;
4263 static void io_queue_linked_timeout(struct io_kiocb *req)
4265 struct io_ring_ctx *ctx = req->ctx;
4268 * If the list is now empty, then our linked request finished before
4269 * we got a chance to setup the timer
4271 spin_lock_irq(&ctx->completion_lock);
4272 if (!list_empty(&req->link_list)) {
4273 struct io_timeout_data *data = &req->io->timeout;
4275 data->timer.function = io_link_timeout_fn;
4276 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
4279 spin_unlock_irq(&ctx->completion_lock);
4281 /* drop submission reference */
4285 static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
4287 struct io_kiocb *nxt;
4289 if (!(req->flags & REQ_F_LINK))
4292 nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb,
4294 if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT)
4297 req->flags |= REQ_F_LINK_TIMEOUT;
4301 static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4303 struct io_kiocb *linked_timeout;
4304 struct io_kiocb *nxt = NULL;
4308 linked_timeout = io_prep_linked_timeout(req);
4310 ret = io_issue_sqe(req, sqe, &nxt, true);
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
4316 if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) ||
4317 (req->flags & REQ_F_MUST_PUNT))) {
4318 if (req->work.flags & IO_WQ_WORK_NEEDS_FILES) {
4319 ret = io_grab_files(req);
4325 * Queued up for async execution, worker will release
4326 * submit reference when the iocb is actually submitted.
4328 io_queue_async_work(req);
4333 /* drop submission reference */
4336 if (linked_timeout) {
4338 io_queue_linked_timeout(linked_timeout);
4340 io_put_req(linked_timeout);
4343 /* and drop final reference, if we failed */
4345 io_cqring_add_event(req, ret);
4346 req_set_fail_links(req);
4357 static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4361 if (unlikely(req->ctx->drain_next)) {
4362 req->flags |= REQ_F_IO_DRAIN;
4363 req->ctx->drain_next = false;
4365 req->ctx->drain_next = (req->flags & REQ_F_DRAIN_LINK);
4367 ret = io_req_defer(req, sqe);
4369 if (ret != -EIOCBQUEUED) {
4370 io_cqring_add_event(req, ret);
4371 req_set_fail_links(req);
4372 io_double_put_req(req);
4374 } else if ((req->flags & REQ_F_FORCE_ASYNC) &&
4375 !io_wq_current_is_worker()) {
4377 * Never try inline submit of IOSQE_ASYNC is set, go straight
4378 * to async execution.
4380 req->work.flags |= IO_WQ_WORK_CONCURRENT;
4381 io_queue_async_work(req);
4383 __io_queue_sqe(req, sqe);
4387 static inline void io_queue_link_head(struct io_kiocb *req)
4389 if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
4390 io_cqring_add_event(req, -ECANCELED);
4391 io_double_put_req(req);
4393 io_queue_sqe(req, NULL);
4396 #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
4397 IOSQE_IO_HARDLINK | IOSQE_ASYNC)
4399 static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
4400 struct io_submit_state *state, struct io_kiocb **link)
4402 struct io_ring_ctx *ctx = req->ctx;
4403 unsigned int sqe_flags;
4406 sqe_flags = READ_ONCE(sqe->flags);
4408 /* enforce forwards compatibility on users */
4409 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
4413 if (sqe_flags & IOSQE_ASYNC)
4414 req->flags |= REQ_F_FORCE_ASYNC;
4416 ret = io_req_set_file(state, req, sqe);
4417 if (unlikely(ret)) {
4419 io_cqring_add_event(req, ret);
4420 io_double_put_req(req);
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.
4432 struct io_kiocb *head = *link;
4434 if (sqe_flags & IOSQE_IO_DRAIN)
4435 head->flags |= REQ_F_DRAIN_LINK | REQ_F_IO_DRAIN;
4437 if (sqe_flags & IOSQE_IO_HARDLINK)
4438 req->flags |= REQ_F_HARDLINK;
4440 if (io_alloc_async_ctx(req)) {
4445 ret = io_req_defer_prep(req, sqe);
4447 /* fail even hard links since we don't submit */
4448 head->flags |= REQ_F_FAIL_LINK;
4451 trace_io_uring_link(ctx, req, head);
4452 list_add_tail(&req->link_list, &head->link_list);
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);
4459 } else if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
4460 req->flags |= REQ_F_LINK;
4461 if (sqe_flags & IOSQE_IO_HARDLINK)
4462 req->flags |= REQ_F_HARDLINK;
4464 INIT_LIST_HEAD(&req->link_list);
4465 ret = io_req_defer_prep(req, sqe);
4467 req->flags |= REQ_F_FAIL_LINK;
4470 io_queue_sqe(req, sqe);
4477 * Batched submission is done, ensure local IO is flushed out.
4479 static void io_submit_state_end(struct io_submit_state *state)
4481 blk_finish_plug(&state->plug);
4483 if (state->free_reqs)
4484 kmem_cache_free_bulk(req_cachep, state->free_reqs,
4485 &state->reqs[state->cur_req]);
4489 * Start submission side cache.
4491 static void io_submit_state_start(struct io_submit_state *state,
4492 unsigned int max_ios)
4494 blk_start_plug(&state->plug);
4495 state->free_reqs = 0;
4497 state->ios_left = max_ios;
4500 static void io_commit_sqring(struct io_ring_ctx *ctx)
4502 struct io_rings *rings = ctx->rings;
4504 if (ctx->cached_sq_head != READ_ONCE(rings->sq.head)) {
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.
4510 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
4515 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
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.
4522 static bool io_get_sqring(struct io_ring_ctx *ctx, struct io_kiocb *req,
4523 const struct io_uring_sqe **sqe_ptr)
4525 struct io_rings *rings = ctx->rings;
4526 u32 *sq_array = ctx->sq_array;
4530 * The cached sq head (or cq tail) serves two purposes:
4532 * 1) allows us to batch the cost of updating the user visible
4534 * 2) allows the kernel side to track the head on its own, even
4535 * though the application is the one updating it.
4537 head = ctx->cached_sq_head;
4538 /* make sure SQ entry isn't read before tail */
4539 if (unlikely(head == smp_load_acquire(&rings->sq.tail)))
4542 head = READ_ONCE(sq_array[head & ctx->sq_mask]);
4543 if (likely(head < ctx->sq_entries)) {
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
4549 req->sequence = ctx->cached_sq_head;
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);
4553 ctx->cached_sq_head++;
4557 /* drop invalid entries */
4558 ctx->cached_sq_head++;
4559 ctx->cached_sq_dropped++;
4560 WRITE_ONCE(rings->sq_dropped, ctx->cached_sq_dropped);
4564 static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
4565 struct file *ring_file, int ring_fd,
4566 struct mm_struct **mm, bool async)
4568 struct io_submit_state state, *statep = NULL;
4569 struct io_kiocb *link = NULL;
4570 int i, submitted = 0;
4571 bool mm_fault = false;
4573 /* if we have a backlog and couldn't flush it all, return BUSY */
4574 if (test_bit(0, &ctx->sq_check_overflow)) {
4575 if (!list_empty(&ctx->cq_overflow_list) &&
4576 !io_cqring_overflow_flush(ctx, false))
4580 if (!percpu_ref_tryget_many(&ctx->refs, nr))
4583 if (nr > IO_PLUG_THRESHOLD) {
4584 io_submit_state_start(&state, nr);
4588 for (i = 0; i < nr; i++) {
4589 const struct io_uring_sqe *sqe;
4590 struct io_kiocb *req;
4592 req = io_get_req(ctx, statep);
4593 if (unlikely(!req)) {
4595 submitted = -EAGAIN;
4598 if (!io_get_sqring(ctx, req, &sqe)) {
4599 __io_req_do_free(req);
4603 /* will complete beyond this point, count as submitted */
4606 if (unlikely(req->opcode >= IORING_OP_LAST)) {
4607 io_cqring_add_event(req, -EINVAL);
4608 io_double_put_req(req);
4612 if (io_op_defs[req->opcode].needs_mm && !*mm) {
4613 mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm);
4615 use_mm(ctx->sqo_mm);
4620 req->ring_file = ring_file;
4621 req->ring_fd = ring_fd;
4622 req->has_user = *mm != NULL;
4623 req->in_async = async;
4624 req->needs_fixed_file = async;
4625 trace_io_uring_submit_sqe(ctx, req->user_data, true, async);
4626 if (!io_submit_sqe(req, sqe, statep, &link))
4630 if (submitted != nr)
4631 percpu_ref_put_many(&ctx->refs, nr - submitted);
4633 io_queue_link_head(link);
4635 io_submit_state_end(&state);
4637 /* Commit SQ ring head once we've consumed and submitted all SQEs */
4638 io_commit_sqring(ctx);
4643 static int io_sq_thread(void *data)
4645 struct io_ring_ctx *ctx = data;
4646 struct mm_struct *cur_mm = NULL;
4647 const struct cred *old_cred;
4648 mm_segment_t old_fs;
4651 unsigned long timeout;
4654 complete(&ctx->completions[1]);
4658 old_cred = override_creds(ctx->creds);
4660 ret = timeout = inflight = 0;
4661 while (!kthread_should_park()) {
4662 unsigned int to_submit;
4665 unsigned nr_events = 0;
4667 if (ctx->flags & IORING_SETUP_IOPOLL) {
4669 * inflight is the count of the maximum possible
4670 * entries we submitted, but it can be smaller
4671 * if we dropped some of them. If we don't have
4672 * poll entries available, then we know that we
4673 * have nothing left to poll for. Reset the
4674 * inflight count to zero in that case.
4676 mutex_lock(&ctx->uring_lock);
4677 if (!list_empty(&ctx->poll_list))
4678 __io_iopoll_check(ctx, &nr_events, 0);
4681 mutex_unlock(&ctx->uring_lock);
4684 * Normal IO, just pretend everything completed.
4685 * We don't have to poll completions for that.
4687 nr_events = inflight;
4690 inflight -= nr_events;
4692 timeout = jiffies + ctx->sq_thread_idle;
4695 to_submit = io_sqring_entries(ctx);
4698 * If submit got -EBUSY, flag us as needing the application
4699 * to enter the kernel to reap and flush events.
4701 if (!to_submit || ret == -EBUSY) {
4703 * We're polling. If we're within the defined idle
4704 * period, then let us spin without work before going
4705 * to sleep. The exception is if we got EBUSY doing
4706 * more IO, we should wait for the application to
4707 * reap events and wake us up.
4710 (!time_after(jiffies, timeout) && ret != -EBUSY)) {
4716 * Drop cur_mm before scheduling, we can't hold it for
4717 * long periods (or over schedule()). Do this before
4718 * adding ourselves to the waitqueue, as the unuse/drop
4727 prepare_to_wait(&ctx->sqo_wait, &wait,
4728 TASK_INTERRUPTIBLE);
4730 /* Tell userspace we may need a wakeup call */
4731 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
4732 /* make sure to read SQ tail after writing flags */
4735 to_submit = io_sqring_entries(ctx);
4736 if (!to_submit || ret == -EBUSY) {
4737 if (kthread_should_park()) {
4738 finish_wait(&ctx->sqo_wait, &wait);
4741 if (signal_pending(current))
4742 flush_signals(current);
4744 finish_wait(&ctx->sqo_wait, &wait);
4746 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
4749 finish_wait(&ctx->sqo_wait, &wait);
4751 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
4754 to_submit = min(to_submit, ctx->sq_entries);
4755 mutex_lock(&ctx->uring_lock);
4756 ret = io_submit_sqes(ctx, to_submit, NULL, -1, &cur_mm, true);
4757 mutex_unlock(&ctx->uring_lock);
4767 revert_creds(old_cred);
4774 struct io_wait_queue {
4775 struct wait_queue_entry wq;
4776 struct io_ring_ctx *ctx;
4778 unsigned nr_timeouts;
4781 static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
4783 struct io_ring_ctx *ctx = iowq->ctx;
4786 * Wake up if we have enough events, or if a timeout occurred since we
4787 * started waiting. For timeouts, we always want to return to userspace,
4788 * regardless of event count.
4790 return io_cqring_events(ctx, noflush) >= iowq->to_wait ||
4791 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
4794 static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
4795 int wake_flags, void *key)
4797 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
4800 /* use noflush == true, as we can't safely rely on locking context */
4801 if (!io_should_wake(iowq, true))
4804 return autoremove_wake_function(curr, mode, wake_flags, key);
4808 * Wait until events become available, if we don't already have some. The
4809 * application must reap them itself, as they reside on the shared cq ring.
4811 static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
4812 const sigset_t __user *sig, size_t sigsz)
4814 struct io_wait_queue iowq = {
4817 .func = io_wake_function,
4818 .entry = LIST_HEAD_INIT(iowq.wq.entry),
4821 .to_wait = min_events,
4823 struct io_rings *rings = ctx->rings;
4826 if (io_cqring_events(ctx, false) >= min_events)
4830 #ifdef CONFIG_COMPAT
4831 if (in_compat_syscall())
4832 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
4836 ret = set_user_sigmask(sig, sigsz);
4842 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
4843 trace_io_uring_cqring_wait(ctx, min_events);
4845 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
4846 TASK_INTERRUPTIBLE);
4847 if (io_should_wake(&iowq, false))
4850 if (signal_pending(current)) {
4855 finish_wait(&ctx->wait, &iowq.wq);
4857 restore_saved_sigmask_unless(ret == -EINTR);
4859 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
4862 static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
4864 #if defined(CONFIG_UNIX)
4865 if (ctx->ring_sock) {
4866 struct sock *sock = ctx->ring_sock->sk;
4867 struct sk_buff *skb;
4869 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
4875 for (i = 0; i < ctx->nr_user_files; i++) {
4878 file = io_file_from_index(ctx, i);
4885 static void io_file_ref_kill(struct percpu_ref *ref)
4887 struct fixed_file_data *data;
4889 data = container_of(ref, struct fixed_file_data, refs);
4890 complete(&data->done);
4893 static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
4895 struct fixed_file_data *data = ctx->file_data;
4896 unsigned nr_tables, i;
4901 /* protect against inflight atomic switch, which drops the ref */
4902 flush_work(&data->ref_work);
4903 percpu_ref_get(&data->refs);
4904 percpu_ref_kill_and_confirm(&data->refs, io_file_ref_kill);
4905 wait_for_completion(&data->done);
4906 percpu_ref_put(&data->refs);
4907 percpu_ref_exit(&data->refs);
4909 __io_sqe_files_unregister(ctx);
4910 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
4911 for (i = 0; i < nr_tables; i++)
4912 kfree(data->table[i].files);
4915 ctx->file_data = NULL;
4916 ctx->nr_user_files = 0;
4920 static void io_sq_thread_stop(struct io_ring_ctx *ctx)
4922 if (ctx->sqo_thread) {
4923 wait_for_completion(&ctx->completions[1]);
4925 * The park is a bit of a work-around, without it we get
4926 * warning spews on shutdown with SQPOLL set and affinity
4927 * set to a single CPU.
4929 kthread_park(ctx->sqo_thread);
4930 kthread_stop(ctx->sqo_thread);
4931 ctx->sqo_thread = NULL;
4935 static void io_finish_async(struct io_ring_ctx *ctx)
4937 io_sq_thread_stop(ctx);
4940 io_wq_destroy(ctx->io_wq);
4945 #if defined(CONFIG_UNIX)
4947 * Ensure the UNIX gc is aware of our file set, so we are certain that
4948 * the io_uring can be safely unregistered on process exit, even if we have
4949 * loops in the file referencing.
4951 static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
4953 struct sock *sk = ctx->ring_sock->sk;
4954 struct scm_fp_list *fpl;
4955 struct sk_buff *skb;
4958 if (!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) {
4959 unsigned long inflight = ctx->user->unix_inflight + nr;
4961 if (inflight > task_rlimit(current, RLIMIT_NOFILE))
4965 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
4969 skb = alloc_skb(0, GFP_KERNEL);
4978 fpl->user = get_uid(ctx->user);
4979 for (i = 0; i < nr; i++) {
4980 struct file *file = io_file_from_index(ctx, i + offset);
4984 fpl->fp[nr_files] = get_file(file);
4985 unix_inflight(fpl->user, fpl->fp[nr_files]);
4990 fpl->max = SCM_MAX_FD;
4991 fpl->count = nr_files;
4992 UNIXCB(skb).fp = fpl;
4993 skb->destructor = unix_destruct_scm;
4994 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
4995 skb_queue_head(&sk->sk_receive_queue, skb);
4997 for (i = 0; i < nr_files; i++)
5008 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
5009 * causes regular reference counting to break down. We rely on the UNIX
5010 * garbage collection to take care of this problem for us.
5012 static int io_sqe_files_scm(struct io_ring_ctx *ctx)
5014 unsigned left, total;
5018 left = ctx->nr_user_files;
5020 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
5022 ret = __io_sqe_files_scm(ctx, this_files, total);
5026 total += this_files;
5032 while (total < ctx->nr_user_files) {
5033 struct file *file = io_file_from_index(ctx, total);
5043 static int io_sqe_files_scm(struct io_ring_ctx *ctx)
5049 static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables,
5054 for (i = 0; i < nr_tables; i++) {
5055 struct fixed_file_table *table = &ctx->file_data->table[i];
5056 unsigned this_files;
5058 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
5059 table->files = kcalloc(this_files, sizeof(struct file *),
5063 nr_files -= this_files;
5069 for (i = 0; i < nr_tables; i++) {
5070 struct fixed_file_table *table = &ctx->file_data->table[i];
5071 kfree(table->files);
5076 static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file)
5078 #if defined(CONFIG_UNIX)
5079 struct sock *sock = ctx->ring_sock->sk;
5080 struct sk_buff_head list, *head = &sock->sk_receive_queue;
5081 struct sk_buff *skb;
5084 __skb_queue_head_init(&list);
5087 * Find the skb that holds this file in its SCM_RIGHTS. When found,
5088 * remove this entry and rearrange the file array.
5090 skb = skb_dequeue(head);
5092 struct scm_fp_list *fp;
5094 fp = UNIXCB(skb).fp;
5095 for (i = 0; i < fp->count; i++) {
5098 if (fp->fp[i] != file)
5101 unix_notinflight(fp->user, fp->fp[i]);
5102 left = fp->count - 1 - i;
5104 memmove(&fp->fp[i], &fp->fp[i + 1],
5105 left * sizeof(struct file *));
5112 __skb_queue_tail(&list, skb);
5122 __skb_queue_tail(&list, skb);
5124 skb = skb_dequeue(head);
5127 if (skb_peek(&list)) {
5128 spin_lock_irq(&head->lock);
5129 while ((skb = __skb_dequeue(&list)) != NULL)
5130 __skb_queue_tail(head, skb);
5131 spin_unlock_irq(&head->lock);
5138 struct io_file_put {
5139 struct llist_node llist;
5141 struct completion *done;
5144 static void io_ring_file_ref_switch(struct work_struct *work)
5146 struct io_file_put *pfile, *tmp;
5147 struct fixed_file_data *data;
5148 struct llist_node *node;
5150 data = container_of(work, struct fixed_file_data, ref_work);
5152 while ((node = llist_del_all(&data->put_llist)) != NULL) {
5153 llist_for_each_entry_safe(pfile, tmp, node, llist) {
5154 io_ring_file_put(data->ctx, pfile->file);
5156 complete(pfile->done);
5162 percpu_ref_get(&data->refs);
5163 percpu_ref_switch_to_percpu(&data->refs);
5166 static void io_file_data_ref_zero(struct percpu_ref *ref)
5168 struct fixed_file_data *data;
5170 data = container_of(ref, struct fixed_file_data, refs);
5172 /* we can't safely switch from inside this context, punt to wq */
5173 queue_work(system_wq, &data->ref_work);
5176 static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
5179 __s32 __user *fds = (__s32 __user *) arg;
5189 if (nr_args > IORING_MAX_FIXED_FILES)
5192 ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL);
5193 if (!ctx->file_data)
5195 ctx->file_data->ctx = ctx;
5196 init_completion(&ctx->file_data->done);
5198 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
5199 ctx->file_data->table = kcalloc(nr_tables,
5200 sizeof(struct fixed_file_table),
5202 if (!ctx->file_data->table) {
5203 kfree(ctx->file_data);
5204 ctx->file_data = NULL;
5208 if (percpu_ref_init(&ctx->file_data->refs, io_file_data_ref_zero,
5209 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
5210 kfree(ctx->file_data->table);
5211 kfree(ctx->file_data);
5212 ctx->file_data = NULL;
5215 ctx->file_data->put_llist.first = NULL;
5216 INIT_WORK(&ctx->file_data->ref_work, io_ring_file_ref_switch);
5218 if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) {
5219 percpu_ref_exit(&ctx->file_data->refs);
5220 kfree(ctx->file_data->table);
5221 kfree(ctx->file_data);
5222 ctx->file_data = NULL;
5226 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
5227 struct fixed_file_table *table;
5231 if (copy_from_user(&fd, &fds[i], sizeof(fd)))
5233 /* allow sparse sets */
5239 table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT];
5240 index = i & IORING_FILE_TABLE_MASK;
5248 * Don't allow io_uring instances to be registered. If UNIX
5249 * isn't enabled, then this causes a reference cycle and this
5250 * instance can never get freed. If UNIX is enabled we'll
5251 * handle it just fine, but there's still no point in allowing
5252 * a ring fd as it doesn't support regular read/write anyway.
5254 if (file->f_op == &io_uring_fops) {
5259 table->files[index] = file;
5263 for (i = 0; i < ctx->nr_user_files; i++) {
5264 file = io_file_from_index(ctx, i);
5268 for (i = 0; i < nr_tables; i++)
5269 kfree(ctx->file_data->table[i].files);
5271 kfree(ctx->file_data->table);
5272 kfree(ctx->file_data);
5273 ctx->file_data = NULL;
5274 ctx->nr_user_files = 0;
5278 ret = io_sqe_files_scm(ctx);
5280 io_sqe_files_unregister(ctx);
5285 static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
5288 #if defined(CONFIG_UNIX)
5289 struct sock *sock = ctx->ring_sock->sk;
5290 struct sk_buff_head *head = &sock->sk_receive_queue;
5291 struct sk_buff *skb;
5294 * See if we can merge this file into an existing skb SCM_RIGHTS
5295 * file set. If there's no room, fall back to allocating a new skb
5296 * and filling it in.
5298 spin_lock_irq(&head->lock);
5299 skb = skb_peek(head);
5301 struct scm_fp_list *fpl = UNIXCB(skb).fp;
5303 if (fpl->count < SCM_MAX_FD) {
5304 __skb_unlink(skb, head);
5305 spin_unlock_irq(&head->lock);
5306 fpl->fp[fpl->count] = get_file(file);
5307 unix_inflight(fpl->user, fpl->fp[fpl->count]);
5309 spin_lock_irq(&head->lock);
5310 __skb_queue_head(head, skb);
5315 spin_unlock_irq(&head->lock);
5322 return __io_sqe_files_scm(ctx, 1, index);
5328 static void io_atomic_switch(struct percpu_ref *ref)
5330 struct fixed_file_data *data;
5332 data = container_of(ref, struct fixed_file_data, refs);
5333 clear_bit(FFD_F_ATOMIC, &data->state);
5336 static bool io_queue_file_removal(struct fixed_file_data *data,
5339 struct io_file_put *pfile, pfile_stack;
5340 DECLARE_COMPLETION_ONSTACK(done);
5343 * If we fail allocating the struct we need for doing async reomval
5344 * of this file, just punt to sync and wait for it.
5346 pfile = kzalloc(sizeof(*pfile), GFP_KERNEL);
5348 pfile = &pfile_stack;
5349 pfile->done = &done;
5353 llist_add(&pfile->llist, &data->put_llist);
5355 if (pfile == &pfile_stack) {
5356 if (!test_and_set_bit(FFD_F_ATOMIC, &data->state)) {
5357 percpu_ref_put(&data->refs);
5358 percpu_ref_switch_to_atomic(&data->refs,
5361 wait_for_completion(&done);
5362 flush_work(&data->ref_work);
5369 static int __io_sqe_files_update(struct io_ring_ctx *ctx,
5370 struct io_uring_files_update *up,
5373 struct fixed_file_data *data = ctx->file_data;
5374 bool ref_switch = false;
5380 if (check_add_overflow(up->offset, nr_args, &done))
5382 if (done > ctx->nr_user_files)
5386 fds = u64_to_user_ptr(up->fds);
5388 struct fixed_file_table *table;
5392 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
5396 i = array_index_nospec(up->offset, ctx->nr_user_files);
5397 table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT];
5398 index = i & IORING_FILE_TABLE_MASK;
5399 if (table->files[index]) {
5400 file = io_file_from_index(ctx, index);
5401 table->files[index] = NULL;
5402 if (io_queue_file_removal(data, file))
5412 * Don't allow io_uring instances to be registered. If
5413 * UNIX isn't enabled, then this causes a reference
5414 * cycle and this instance can never get freed. If UNIX
5415 * is enabled we'll handle it just fine, but there's
5416 * still no point in allowing a ring fd as it doesn't
5417 * support regular read/write anyway.
5419 if (file->f_op == &io_uring_fops) {
5424 table->files[index] = file;
5425 err = io_sqe_file_register(ctx, file, i);
5434 if (ref_switch && !test_and_set_bit(FFD_F_ATOMIC, &data->state)) {
5435 percpu_ref_put(&data->refs);
5436 percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch);
5439 return done ? done : err;
5441 static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
5444 struct io_uring_files_update up;
5446 if (!ctx->file_data)
5450 if (copy_from_user(&up, arg, sizeof(up)))
5455 return __io_sqe_files_update(ctx, &up, nr_args);
5458 static void io_put_work(struct io_wq_work *work)
5460 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
5465 static void io_get_work(struct io_wq_work *work)
5467 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
5469 refcount_inc(&req->refs);
5472 static int io_sq_offload_start(struct io_ring_ctx *ctx,
5473 struct io_uring_params *p)
5475 struct io_wq_data data;
5476 unsigned concurrency;
5479 init_waitqueue_head(&ctx->sqo_wait);
5480 mmgrab(current->mm);
5481 ctx->sqo_mm = current->mm;
5483 if (ctx->flags & IORING_SETUP_SQPOLL) {
5485 if (!capable(CAP_SYS_ADMIN))
5488 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
5489 if (!ctx->sq_thread_idle)
5490 ctx->sq_thread_idle = HZ;
5492 if (p->flags & IORING_SETUP_SQ_AFF) {
5493 int cpu = p->sq_thread_cpu;
5496 if (cpu >= nr_cpu_ids)
5498 if (!cpu_online(cpu))
5501 ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread,
5505 ctx->sqo_thread = kthread_create(io_sq_thread, ctx,
5508 if (IS_ERR(ctx->sqo_thread)) {
5509 ret = PTR_ERR(ctx->sqo_thread);
5510 ctx->sqo_thread = NULL;
5513 wake_up_process(ctx->sqo_thread);
5514 } else if (p->flags & IORING_SETUP_SQ_AFF) {
5515 /* Can't have SQ_AFF without SQPOLL */
5520 data.mm = ctx->sqo_mm;
5521 data.user = ctx->user;
5522 data.creds = ctx->creds;
5523 data.get_work = io_get_work;
5524 data.put_work = io_put_work;
5526 /* Do QD, or 4 * CPUS, whatever is smallest */
5527 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
5528 ctx->io_wq = io_wq_create(concurrency, &data);
5529 if (IS_ERR(ctx->io_wq)) {
5530 ret = PTR_ERR(ctx->io_wq);
5537 io_finish_async(ctx);
5538 mmdrop(ctx->sqo_mm);
5543 static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages)
5545 atomic_long_sub(nr_pages, &user->locked_vm);
5548 static int io_account_mem(struct user_struct *user, unsigned long nr_pages)
5550 unsigned long page_limit, cur_pages, new_pages;
5552 /* Don't allow more pages than we can safely lock */
5553 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
5556 cur_pages = atomic_long_read(&user->locked_vm);
5557 new_pages = cur_pages + nr_pages;
5558 if (new_pages > page_limit)
5560 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
5561 new_pages) != cur_pages);
5566 static void io_mem_free(void *ptr)
5573 page = virt_to_head_page(ptr);
5574 if (put_page_testzero(page))
5575 free_compound_page(page);
5578 static void *io_mem_alloc(size_t size)
5580 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
5583 return (void *) __get_free_pages(gfp_flags, get_order(size));
5586 static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
5589 struct io_rings *rings;
5590 size_t off, sq_array_size;
5592 off = struct_size(rings, cqes, cq_entries);
5593 if (off == SIZE_MAX)
5597 off = ALIGN(off, SMP_CACHE_BYTES);
5602 sq_array_size = array_size(sizeof(u32), sq_entries);
5603 if (sq_array_size == SIZE_MAX)
5606 if (check_add_overflow(off, sq_array_size, &off))
5615 static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries)
5619 pages = (size_t)1 << get_order(
5620 rings_size(sq_entries, cq_entries, NULL));
5621 pages += (size_t)1 << get_order(
5622 array_size(sizeof(struct io_uring_sqe), sq_entries));
5627 static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx)
5631 if (!ctx->user_bufs)
5634 for (i = 0; i < ctx->nr_user_bufs; i++) {
5635 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
5637 for (j = 0; j < imu->nr_bvecs; j++)
5638 put_user_page(imu->bvec[j].bv_page);
5640 if (ctx->account_mem)
5641 io_unaccount_mem(ctx->user, imu->nr_bvecs);
5646 kfree(ctx->user_bufs);
5647 ctx->user_bufs = NULL;
5648 ctx->nr_user_bufs = 0;
5652 static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
5653 void __user *arg, unsigned index)
5655 struct iovec __user *src;
5657 #ifdef CONFIG_COMPAT
5659 struct compat_iovec __user *ciovs;
5660 struct compat_iovec ciov;
5662 ciovs = (struct compat_iovec __user *) arg;
5663 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
5666 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
5667 dst->iov_len = ciov.iov_len;
5671 src = (struct iovec __user *) arg;
5672 if (copy_from_user(dst, &src[index], sizeof(*dst)))
5677 static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg,
5680 struct vm_area_struct **vmas = NULL;
5681 struct page **pages = NULL;
5682 int i, j, got_pages = 0;
5687 if (!nr_args || nr_args > UIO_MAXIOV)
5690 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
5692 if (!ctx->user_bufs)
5695 for (i = 0; i < nr_args; i++) {
5696 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
5697 unsigned long off, start, end, ubuf;
5702 ret = io_copy_iov(ctx, &iov, arg, i);
5707 * Don't impose further limits on the size and buffer
5708 * constraints here, we'll -EINVAL later when IO is
5709 * submitted if they are wrong.
5712 if (!iov.iov_base || !iov.iov_len)
5715 /* arbitrary limit, but we need something */
5716 if (iov.iov_len > SZ_1G)
5719 ubuf = (unsigned long) iov.iov_base;
5720 end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
5721 start = ubuf >> PAGE_SHIFT;
5722 nr_pages = end - start;
5724 if (ctx->account_mem) {
5725 ret = io_account_mem(ctx->user, nr_pages);
5731 if (!pages || nr_pages > got_pages) {
5734 pages = kvmalloc_array(nr_pages, sizeof(struct page *),
5736 vmas = kvmalloc_array(nr_pages,
5737 sizeof(struct vm_area_struct *),
5739 if (!pages || !vmas) {
5741 if (ctx->account_mem)
5742 io_unaccount_mem(ctx->user, nr_pages);
5745 got_pages = nr_pages;
5748 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
5752 if (ctx->account_mem)
5753 io_unaccount_mem(ctx->user, nr_pages);
5758 down_read(¤t->mm->mmap_sem);
5759 pret = get_user_pages(ubuf, nr_pages,
5760 FOLL_WRITE | FOLL_LONGTERM,
5762 if (pret == nr_pages) {
5763 /* don't support file backed memory */
5764 for (j = 0; j < nr_pages; j++) {
5765 struct vm_area_struct *vma = vmas[j];
5768 !is_file_hugepages(vma->vm_file)) {
5774 ret = pret < 0 ? pret : -EFAULT;
5776 up_read(¤t->mm->mmap_sem);
5779 * if we did partial map, or found file backed vmas,
5780 * release any pages we did get
5783 put_user_pages(pages, pret);
5784 if (ctx->account_mem)
5785 io_unaccount_mem(ctx->user, nr_pages);
5790 off = ubuf & ~PAGE_MASK;
5792 for (j = 0; j < nr_pages; j++) {
5795 vec_len = min_t(size_t, size, PAGE_SIZE - off);
5796 imu->bvec[j].bv_page = pages[j];
5797 imu->bvec[j].bv_len = vec_len;
5798 imu->bvec[j].bv_offset = off;
5802 /* store original address for later verification */
5804 imu->len = iov.iov_len;
5805 imu->nr_bvecs = nr_pages;
5807 ctx->nr_user_bufs++;
5815 io_sqe_buffer_unregister(ctx);
5819 static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
5821 __s32 __user *fds = arg;
5827 if (copy_from_user(&fd, fds, sizeof(*fds)))
5830 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
5831 if (IS_ERR(ctx->cq_ev_fd)) {
5832 int ret = PTR_ERR(ctx->cq_ev_fd);
5833 ctx->cq_ev_fd = NULL;
5840 static int io_eventfd_unregister(struct io_ring_ctx *ctx)
5842 if (ctx->cq_ev_fd) {
5843 eventfd_ctx_put(ctx->cq_ev_fd);
5844 ctx->cq_ev_fd = NULL;
5851 static void io_ring_ctx_free(struct io_ring_ctx *ctx)
5853 io_finish_async(ctx);
5855 mmdrop(ctx->sqo_mm);
5857 io_iopoll_reap_events(ctx);
5858 io_sqe_buffer_unregister(ctx);
5859 io_sqe_files_unregister(ctx);
5860 io_eventfd_unregister(ctx);
5862 #if defined(CONFIG_UNIX)
5863 if (ctx->ring_sock) {
5864 ctx->ring_sock->file = NULL; /* so that iput() is called */
5865 sock_release(ctx->ring_sock);
5869 io_mem_free(ctx->rings);
5870 io_mem_free(ctx->sq_sqes);
5872 percpu_ref_exit(&ctx->refs);
5873 if (ctx->account_mem)
5874 io_unaccount_mem(ctx->user,
5875 ring_pages(ctx->sq_entries, ctx->cq_entries));
5876 free_uid(ctx->user);
5877 put_cred(ctx->creds);
5878 kfree(ctx->completions);
5879 kfree(ctx->cancel_hash);
5880 kmem_cache_free(req_cachep, ctx->fallback_req);
5884 static __poll_t io_uring_poll(struct file *file, poll_table *wait)
5886 struct io_ring_ctx *ctx = file->private_data;
5889 poll_wait(file, &ctx->cq_wait, wait);
5891 * synchronizes with barrier from wq_has_sleeper call in
5895 if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head !=
5896 ctx->rings->sq_ring_entries)
5897 mask |= EPOLLOUT | EPOLLWRNORM;
5898 if (READ_ONCE(ctx->rings->cq.head) != ctx->cached_cq_tail)
5899 mask |= EPOLLIN | EPOLLRDNORM;
5904 static int io_uring_fasync(int fd, struct file *file, int on)
5906 struct io_ring_ctx *ctx = file->private_data;
5908 return fasync_helper(fd, file, on, &ctx->cq_fasync);
5911 static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
5913 mutex_lock(&ctx->uring_lock);
5914 percpu_ref_kill(&ctx->refs);
5915 mutex_unlock(&ctx->uring_lock);
5917 io_kill_timeouts(ctx);
5918 io_poll_remove_all(ctx);
5921 io_wq_cancel_all(ctx->io_wq);
5923 io_iopoll_reap_events(ctx);
5924 /* if we failed setting up the ctx, we might not have any rings */
5926 io_cqring_overflow_flush(ctx, true);
5927 wait_for_completion(&ctx->completions[0]);
5928 io_ring_ctx_free(ctx);
5931 static int io_uring_release(struct inode *inode, struct file *file)
5933 struct io_ring_ctx *ctx = file->private_data;
5935 file->private_data = NULL;
5936 io_ring_ctx_wait_and_kill(ctx);
5940 static void io_uring_cancel_files(struct io_ring_ctx *ctx,
5941 struct files_struct *files)
5943 struct io_kiocb *req;
5946 while (!list_empty_careful(&ctx->inflight_list)) {
5947 struct io_kiocb *cancel_req = NULL;
5949 spin_lock_irq(&ctx->inflight_lock);
5950 list_for_each_entry(req, &ctx->inflight_list, inflight_entry) {
5951 if (req->work.files != files)
5953 /* req is being completed, ignore */
5954 if (!refcount_inc_not_zero(&req->refs))
5960 prepare_to_wait(&ctx->inflight_wait, &wait,
5961 TASK_UNINTERRUPTIBLE);
5962 spin_unlock_irq(&ctx->inflight_lock);
5964 /* We need to keep going until we don't find a matching req */
5968 io_wq_cancel_work(ctx->io_wq, &cancel_req->work);
5969 io_put_req(cancel_req);
5972 finish_wait(&ctx->inflight_wait, &wait);
5975 static int io_uring_flush(struct file *file, void *data)
5977 struct io_ring_ctx *ctx = file->private_data;
5979 io_uring_cancel_files(ctx, data);
5980 if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) {
5981 io_cqring_overflow_flush(ctx, true);
5982 io_wq_cancel_all(ctx->io_wq);
5987 static void *io_uring_validate_mmap_request(struct file *file,
5988 loff_t pgoff, size_t sz)
5990 struct io_ring_ctx *ctx = file->private_data;
5991 loff_t offset = pgoff << PAGE_SHIFT;
5996 case IORING_OFF_SQ_RING:
5997 case IORING_OFF_CQ_RING:
6000 case IORING_OFF_SQES:
6004 return ERR_PTR(-EINVAL);
6007 page = virt_to_head_page(ptr);
6008 if (sz > page_size(page))
6009 return ERR_PTR(-EINVAL);
6016 static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
6018 size_t sz = vma->vm_end - vma->vm_start;
6022 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
6024 return PTR_ERR(ptr);
6026 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
6027 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
6030 #else /* !CONFIG_MMU */
6032 static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
6034 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
6037 static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
6039 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
6042 static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
6043 unsigned long addr, unsigned long len,
6044 unsigned long pgoff, unsigned long flags)
6048 ptr = io_uring_validate_mmap_request(file, pgoff, len);
6050 return PTR_ERR(ptr);
6052 return (unsigned long) ptr;
6055 #endif /* !CONFIG_MMU */
6057 SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
6058 u32, min_complete, u32, flags, const sigset_t __user *, sig,
6061 struct io_ring_ctx *ctx;
6066 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP))
6074 if (f.file->f_op != &io_uring_fops)
6078 ctx = f.file->private_data;
6079 if (!percpu_ref_tryget(&ctx->refs))
6083 * For SQ polling, the thread will do all submissions and completions.
6084 * Just return the requested submit count, and wake the thread if
6088 if (ctx->flags & IORING_SETUP_SQPOLL) {
6089 if (!list_empty_careful(&ctx->cq_overflow_list))
6090 io_cqring_overflow_flush(ctx, false);
6091 if (flags & IORING_ENTER_SQ_WAKEUP)
6092 wake_up(&ctx->sqo_wait);
6093 submitted = to_submit;
6094 } else if (to_submit) {
6095 struct mm_struct *cur_mm;
6097 if (current->mm != ctx->sqo_mm ||
6098 current_cred() != ctx->creds) {
6103 to_submit = min(to_submit, ctx->sq_entries);
6104 mutex_lock(&ctx->uring_lock);
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,
6109 mutex_unlock(&ctx->uring_lock);
6111 if (submitted != to_submit)
6114 if (flags & IORING_ENTER_GETEVENTS) {
6115 unsigned nr_events = 0;
6117 min_complete = min(min_complete, ctx->cq_entries);
6119 if (ctx->flags & IORING_SETUP_IOPOLL) {
6120 ret = io_iopoll_check(ctx, &nr_events, min_complete);
6122 ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
6127 percpu_ref_put(&ctx->refs);
6130 return submitted ? submitted : ret;
6133 static const struct file_operations io_uring_fops = {
6134 .release = io_uring_release,
6135 .flush = io_uring_flush,
6136 .mmap = io_uring_mmap,
6138 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
6139 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
6141 .poll = io_uring_poll,
6142 .fasync = io_uring_fasync,
6145 static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
6146 struct io_uring_params *p)
6148 struct io_rings *rings;
6149 size_t size, sq_array_offset;
6151 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
6152 if (size == SIZE_MAX)
6155 rings = io_mem_alloc(size);
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;
6170 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
6171 if (size == SIZE_MAX) {
6172 io_mem_free(ctx->rings);
6177 ctx->sq_sqes = io_mem_alloc(size);
6178 if (!ctx->sq_sqes) {
6179 io_mem_free(ctx->rings);
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.
6193 static int io_uring_get_fd(struct io_ring_ctx *ctx)
6198 #if defined(CONFIG_UNIX)
6199 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
6205 ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
6209 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
6210 O_RDWR | O_CLOEXEC);
6213 ret = PTR_ERR(file);
6217 #if defined(CONFIG_UNIX)
6218 ctx->ring_sock->file = file;
6220 fd_install(ret, file);
6223 #if defined(CONFIG_UNIX)
6224 sock_release(ctx->ring_sock);
6225 ctx->ring_sock = NULL;
6230 static int io_uring_create(unsigned entries, struct io_uring_params *p)
6232 struct user_struct *user = NULL;
6233 struct io_ring_ctx *ctx;
6237 if (!entries || entries > IORING_MAX_ENTRIES)
6241 * Use twice as many entries for the CQ ring. It's possible for the
6242 * application to drive a higher depth than the size of the SQ ring,
6243 * since the sqes are only used at submission time. This allows for
6244 * some flexibility in overcommitting a bit. If the application has
6245 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
6246 * of CQ ring entries manually.
6248 p->sq_entries = roundup_pow_of_two(entries);
6249 if (p->flags & IORING_SETUP_CQSIZE) {
6251 * If IORING_SETUP_CQSIZE is set, we do the same roundup
6252 * to a power-of-two, if it isn't already. We do NOT impose
6253 * any cq vs sq ring sizing.
6255 if (p->cq_entries < p->sq_entries || p->cq_entries > IORING_MAX_CQ_ENTRIES)
6257 p->cq_entries = roundup_pow_of_two(p->cq_entries);
6259 p->cq_entries = 2 * p->sq_entries;
6262 user = get_uid(current_user());
6263 account_mem = !capable(CAP_IPC_LOCK);
6266 ret = io_account_mem(user,
6267 ring_pages(p->sq_entries, p->cq_entries));
6274 ctx = io_ring_ctx_alloc(p);
6277 io_unaccount_mem(user, ring_pages(p->sq_entries,
6282 ctx->compat = in_compat_syscall();
6283 ctx->account_mem = account_mem;
6285 ctx->creds = get_current_cred();
6287 ret = io_allocate_scq_urings(ctx, p);
6291 ret = io_sq_offload_start(ctx, p);
6295 memset(&p->sq_off, 0, sizeof(p->sq_off));
6296 p->sq_off.head = offsetof(struct io_rings, sq.head);
6297 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
6298 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
6299 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
6300 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
6301 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
6302 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
6304 memset(&p->cq_off, 0, sizeof(p->cq_off));
6305 p->cq_off.head = offsetof(struct io_rings, cq.head);
6306 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
6307 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
6308 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
6309 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
6310 p->cq_off.cqes = offsetof(struct io_rings, cqes);
6313 * Install ring fd as the very last thing, so we don't risk someone
6314 * having closed it before we finish setup
6316 ret = io_uring_get_fd(ctx);
6320 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
6321 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS;
6322 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
6325 io_ring_ctx_wait_and_kill(ctx);
6330 * Sets up an aio uring context, and returns the fd. Applications asks for a
6331 * ring size, we return the actual sq/cq ring sizes (among other things) in the
6332 * params structure passed in.
6334 static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
6336 struct io_uring_params p;
6340 if (copy_from_user(&p, params, sizeof(p)))
6342 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
6347 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
6348 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE))
6351 ret = io_uring_create(entries, &p);
6355 if (copy_to_user(params, &p, sizeof(p)))
6361 SYSCALL_DEFINE2(io_uring_setup, u32, entries,
6362 struct io_uring_params __user *, params)
6364 return io_uring_setup(entries, params);
6367 static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
6368 void __user *arg, unsigned nr_args)
6369 __releases(ctx->uring_lock)
6370 __acquires(ctx->uring_lock)
6375 * We're inside the ring mutex, if the ref is already dying, then
6376 * someone else killed the ctx or is already going through
6377 * io_uring_register().
6379 if (percpu_ref_is_dying(&ctx->refs))
6382 if (opcode != IORING_UNREGISTER_FILES &&
6383 opcode != IORING_REGISTER_FILES_UPDATE) {
6384 percpu_ref_kill(&ctx->refs);
6387 * Drop uring mutex before waiting for references to exit. If
6388 * another thread is currently inside io_uring_enter() it might
6389 * need to grab the uring_lock to make progress. If we hold it
6390 * here across the drain wait, then we can deadlock. It's safe
6391 * to drop the mutex here, since no new references will come in
6392 * after we've killed the percpu ref.
6394 mutex_unlock(&ctx->uring_lock);
6395 wait_for_completion(&ctx->completions[0]);
6396 mutex_lock(&ctx->uring_lock);
6400 case IORING_REGISTER_BUFFERS:
6401 ret = io_sqe_buffer_register(ctx, arg, nr_args);
6403 case IORING_UNREGISTER_BUFFERS:
6407 ret = io_sqe_buffer_unregister(ctx);
6409 case IORING_REGISTER_FILES:
6410 ret = io_sqe_files_register(ctx, arg, nr_args);
6412 case IORING_UNREGISTER_FILES:
6416 ret = io_sqe_files_unregister(ctx);
6418 case IORING_REGISTER_FILES_UPDATE:
6419 ret = io_sqe_files_update(ctx, arg, nr_args);
6421 case IORING_REGISTER_EVENTFD:
6425 ret = io_eventfd_register(ctx, arg);
6427 case IORING_UNREGISTER_EVENTFD:
6431 ret = io_eventfd_unregister(ctx);
6439 if (opcode != IORING_UNREGISTER_FILES &&
6440 opcode != IORING_REGISTER_FILES_UPDATE) {
6441 /* bring the ctx back to life */
6442 reinit_completion(&ctx->completions[0]);
6443 percpu_ref_reinit(&ctx->refs);
6448 SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
6449 void __user *, arg, unsigned int, nr_args)
6451 struct io_ring_ctx *ctx;
6460 if (f.file->f_op != &io_uring_fops)
6463 ctx = f.file->private_data;
6465 mutex_lock(&ctx->uring_lock);
6466 ret = __io_uring_register(ctx, opcode, arg, nr_args);
6467 mutex_unlock(&ctx->uring_lock);
6468 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
6469 ctx->cq_ev_fd != NULL, ret);
6475 static int __init io_uring_init(void)
6477 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
6478 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
6481 __initcall(io_uring_init);