2 * An async IO implementation for Linux
3 * Written by Benjamin LaHaise <bcrl@kvack.org>
5 * Implements an efficient asynchronous io interface.
7 * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved.
9 * See ../COPYING for licensing terms.
11 #include <linux/kernel.h>
12 #include <linux/init.h>
13 #include <linux/errno.h>
14 #include <linux/time.h>
15 #include <linux/aio_abi.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/backing-dev.h>
19 #include <linux/uio.h>
23 #include <linux/sched.h>
25 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/mmu_context.h>
29 #include <linux/slab.h>
30 #include <linux/timer.h>
31 #include <linux/aio.h>
32 #include <linux/highmem.h>
33 #include <linux/workqueue.h>
34 #include <linux/security.h>
35 #include <linux/eventfd.h>
36 #include <linux/blkdev.h>
37 #include <linux/compat.h>
39 #include <asm/kmap_types.h>
40 #include <asm/uaccess.h>
43 #define dprintk printk
45 #define dprintk(x...) do { ; } while (0)
48 /*------ sysctl variables----*/
49 static DEFINE_SPINLOCK(aio_nr_lock);
50 unsigned long aio_nr; /* current system wide number of aio requests */
51 unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */
52 /*----end sysctl variables---*/
54 static struct kmem_cache *kiocb_cachep;
55 static struct kmem_cache *kioctx_cachep;
57 static struct workqueue_struct *aio_wq;
59 static void aio_kick_handler(struct work_struct *);
60 static void aio_queue_work(struct kioctx *);
63 * Creates the slab caches used by the aio routines, panic on
64 * failure as this is done early during the boot sequence.
66 static int __init aio_setup(void)
68 kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
69 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
71 aio_wq = alloc_workqueue("aio", 0, 1); /* used to limit concurrency */
74 pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page));
78 __initcall(aio_setup);
80 static void aio_free_ring(struct kioctx *ctx)
82 struct aio_ring_info *info = &ctx->ring_info;
85 for (i=0; i<info->nr_pages; i++)
86 put_page(info->ring_pages[i]);
88 if (info->mmap_size) {
89 BUG_ON(ctx->mm != current->mm);
90 vm_munmap(info->mmap_base, info->mmap_size);
93 if (info->ring_pages && info->ring_pages != info->internal_pages)
94 kfree(info->ring_pages);
95 info->ring_pages = NULL;
99 static int aio_setup_ring(struct kioctx *ctx)
101 struct aio_ring *ring;
102 struct aio_ring_info *info = &ctx->ring_info;
103 unsigned nr_events = ctx->max_reqs;
104 unsigned long size, populate;
107 /* Compensate for the ring buffer's head/tail overlap entry */
108 nr_events += 2; /* 1 is required, 2 for good luck */
110 size = sizeof(struct aio_ring);
111 size += sizeof(struct io_event) * nr_events;
112 nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT;
117 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event);
120 info->ring_pages = info->internal_pages;
121 if (nr_pages > AIO_RING_PAGES) {
122 info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
123 if (!info->ring_pages)
127 info->mmap_size = nr_pages * PAGE_SIZE;
128 dprintk("attempting mmap of %lu bytes\n", info->mmap_size);
129 down_write(&ctx->mm->mmap_sem);
130 info->mmap_base = do_mmap_pgoff(NULL, 0, info->mmap_size,
131 PROT_READ|PROT_WRITE,
132 MAP_ANONYMOUS|MAP_PRIVATE, 0,
134 if (IS_ERR((void *)info->mmap_base)) {
135 up_write(&ctx->mm->mmap_sem);
141 dprintk("mmap address: 0x%08lx\n", info->mmap_base);
142 info->nr_pages = get_user_pages(current, ctx->mm,
143 info->mmap_base, nr_pages,
144 1, 0, info->ring_pages, NULL);
145 up_write(&ctx->mm->mmap_sem);
147 if (unlikely(info->nr_pages != nr_pages)) {
152 mm_populate(info->mmap_base, populate);
154 ctx->user_id = info->mmap_base;
156 info->nr = nr_events; /* trusted copy */
158 ring = kmap_atomic(info->ring_pages[0]);
159 ring->nr = nr_events; /* user copy */
160 ring->id = ctx->user_id;
161 ring->head = ring->tail = 0;
162 ring->magic = AIO_RING_MAGIC;
163 ring->compat_features = AIO_RING_COMPAT_FEATURES;
164 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
165 ring->header_length = sizeof(struct aio_ring);
172 /* aio_ring_event: returns a pointer to the event at the given index from
173 * kmap_atomic(). Release the pointer with put_aio_ring_event();
175 #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
176 #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
177 #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
179 #define aio_ring_event(info, nr) ({ \
180 unsigned pos = (nr) + AIO_EVENTS_OFFSET; \
181 struct io_event *__event; \
182 __event = kmap_atomic( \
183 (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE]); \
184 __event += pos % AIO_EVENTS_PER_PAGE; \
188 #define put_aio_ring_event(event) do { \
189 struct io_event *__event = (event); \
191 kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK)); \
194 static void ctx_rcu_free(struct rcu_head *head)
196 struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
197 kmem_cache_free(kioctx_cachep, ctx);
201 * Called when the last user of an aio context has gone away,
202 * and the struct needs to be freed.
204 static void __put_ioctx(struct kioctx *ctx)
206 unsigned nr_events = ctx->max_reqs;
207 BUG_ON(ctx->reqs_active);
209 cancel_delayed_work_sync(&ctx->wq);
214 spin_lock(&aio_nr_lock);
215 BUG_ON(aio_nr - nr_events > aio_nr);
217 spin_unlock(&aio_nr_lock);
219 pr_debug("__put_ioctx: freeing %p\n", ctx);
220 call_rcu(&ctx->rcu_head, ctx_rcu_free);
223 static inline int try_get_ioctx(struct kioctx *kioctx)
225 return atomic_inc_not_zero(&kioctx->users);
228 static inline void put_ioctx(struct kioctx *kioctx)
230 BUG_ON(atomic_read(&kioctx->users) <= 0);
231 if (unlikely(atomic_dec_and_test(&kioctx->users)))
236 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
238 static struct kioctx *ioctx_alloc(unsigned nr_events)
240 struct mm_struct *mm;
244 /* Prevent overflows */
245 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
246 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
247 pr_debug("ENOMEM: nr_events too high\n");
248 return ERR_PTR(-EINVAL);
251 if (!nr_events || (unsigned long)nr_events > aio_max_nr)
252 return ERR_PTR(-EAGAIN);
254 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
256 return ERR_PTR(-ENOMEM);
258 ctx->max_reqs = nr_events;
259 mm = ctx->mm = current->mm;
260 atomic_inc(&mm->mm_count);
262 atomic_set(&ctx->users, 2);
263 spin_lock_init(&ctx->ctx_lock);
264 spin_lock_init(&ctx->ring_info.ring_lock);
265 init_waitqueue_head(&ctx->wait);
267 INIT_LIST_HEAD(&ctx->active_reqs);
268 INIT_LIST_HEAD(&ctx->run_list);
269 INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler);
271 if (aio_setup_ring(ctx) < 0)
274 /* limit the number of system wide aios */
275 spin_lock(&aio_nr_lock);
276 if (aio_nr + nr_events > aio_max_nr ||
277 aio_nr + nr_events < aio_nr) {
278 spin_unlock(&aio_nr_lock);
281 aio_nr += ctx->max_reqs;
282 spin_unlock(&aio_nr_lock);
284 /* now link into global list. */
285 spin_lock(&mm->ioctx_lock);
286 hlist_add_head_rcu(&ctx->list, &mm->ioctx_list);
287 spin_unlock(&mm->ioctx_lock);
289 dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
290 ctx, ctx->user_id, current->mm, ctx->ring_info.nr);
298 kmem_cache_free(kioctx_cachep, ctx);
299 dprintk("aio: error allocating ioctx %d\n", err);
304 * Cancels all outstanding aio requests on an aio context. Used
305 * when the processes owning a context have all exited to encourage
306 * the rapid destruction of the kioctx.
308 static void kill_ctx(struct kioctx *ctx)
310 int (*cancel)(struct kiocb *, struct io_event *);
311 struct task_struct *tsk = current;
312 DECLARE_WAITQUEUE(wait, tsk);
315 spin_lock_irq(&ctx->ctx_lock);
317 while (!list_empty(&ctx->active_reqs)) {
318 struct list_head *pos = ctx->active_reqs.next;
319 struct kiocb *iocb = list_kiocb(pos);
320 list_del_init(&iocb->ki_list);
321 cancel = iocb->ki_cancel;
322 kiocbSetCancelled(iocb);
325 spin_unlock_irq(&ctx->ctx_lock);
327 spin_lock_irq(&ctx->ctx_lock);
331 if (!ctx->reqs_active)
334 add_wait_queue(&ctx->wait, &wait);
335 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
336 while (ctx->reqs_active) {
337 spin_unlock_irq(&ctx->ctx_lock);
339 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
340 spin_lock_irq(&ctx->ctx_lock);
342 __set_task_state(tsk, TASK_RUNNING);
343 remove_wait_queue(&ctx->wait, &wait);
346 spin_unlock_irq(&ctx->ctx_lock);
349 /* wait_on_sync_kiocb:
350 * Waits on the given sync kiocb to complete.
352 ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
354 while (iocb->ki_users) {
355 set_current_state(TASK_UNINTERRUPTIBLE);
360 __set_current_state(TASK_RUNNING);
361 return iocb->ki_user_data;
363 EXPORT_SYMBOL(wait_on_sync_kiocb);
365 /* exit_aio: called when the last user of mm goes away. At this point,
366 * there is no way for any new requests to be submited or any of the
367 * io_* syscalls to be called on the context. However, there may be
368 * outstanding requests which hold references to the context; as they
369 * go away, they will call put_ioctx and release any pinned memory
370 * associated with the request (held via struct page * references).
372 void exit_aio(struct mm_struct *mm)
376 while (!hlist_empty(&mm->ioctx_list)) {
377 ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list);
378 hlist_del_rcu(&ctx->list);
382 if (1 != atomic_read(&ctx->users))
384 "exit_aio:ioctx still alive: %d %d %d\n",
385 atomic_read(&ctx->users), ctx->dead,
388 * We don't need to bother with munmap() here -
389 * exit_mmap(mm) is coming and it'll unmap everything.
390 * Since aio_free_ring() uses non-zero ->mmap_size
391 * as indicator that it needs to unmap the area,
392 * just set it to 0; aio_free_ring() is the only
393 * place that uses ->mmap_size, so it's safe.
394 * That way we get all munmap done to current->mm -
395 * all other callers have ctx->mm == current->mm.
397 ctx->ring_info.mmap_size = 0;
403 * Allocate a slot for an aio request. Increments the users count
404 * of the kioctx so that the kioctx stays around until all requests are
405 * complete. Returns NULL if no requests are free.
407 * Returns with kiocb->users set to 2. The io submit code path holds
408 * an extra reference while submitting the i/o.
409 * This prevents races between the aio code path referencing the
410 * req (after submitting it) and aio_complete() freeing the req.
412 static struct kiocb *__aio_get_req(struct kioctx *ctx)
414 struct kiocb *req = NULL;
416 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL);
424 req->ki_cancel = NULL;
425 req->ki_retry = NULL;
428 req->ki_iovec = NULL;
429 INIT_LIST_HEAD(&req->ki_run_list);
430 req->ki_eventfd = NULL;
436 * struct kiocb's are allocated in batches to reduce the number of
437 * times the ctx lock is acquired and released.
439 #define KIOCB_BATCH_SIZE 32L
441 struct list_head head;
442 long count; /* number of requests left to allocate */
445 static void kiocb_batch_init(struct kiocb_batch *batch, long total)
447 INIT_LIST_HEAD(&batch->head);
448 batch->count = total;
451 static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch)
453 struct kiocb *req, *n;
455 if (list_empty(&batch->head))
458 spin_lock_irq(&ctx->ctx_lock);
459 list_for_each_entry_safe(req, n, &batch->head, ki_batch) {
460 list_del(&req->ki_batch);
461 list_del(&req->ki_list);
462 kmem_cache_free(kiocb_cachep, req);
465 if (unlikely(!ctx->reqs_active && ctx->dead))
466 wake_up_all(&ctx->wait);
467 spin_unlock_irq(&ctx->ctx_lock);
471 * Allocate a batch of kiocbs. This avoids taking and dropping the
472 * context lock a lot during setup.
474 static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch)
476 unsigned short allocated, to_alloc;
478 struct kiocb *req, *n;
479 struct aio_ring *ring;
481 to_alloc = min(batch->count, KIOCB_BATCH_SIZE);
482 for (allocated = 0; allocated < to_alloc; allocated++) {
483 req = __aio_get_req(ctx);
485 /* allocation failed, go with what we've got */
487 list_add(&req->ki_batch, &batch->head);
493 spin_lock_irq(&ctx->ctx_lock);
494 ring = kmap_atomic(ctx->ring_info.ring_pages[0]);
496 avail = aio_ring_avail(&ctx->ring_info, ring) - ctx->reqs_active;
498 if (avail < allocated) {
499 /* Trim back the number of requests. */
500 list_for_each_entry_safe(req, n, &batch->head, ki_batch) {
501 list_del(&req->ki_batch);
502 kmem_cache_free(kiocb_cachep, req);
503 if (--allocated <= avail)
508 batch->count -= allocated;
509 list_for_each_entry(req, &batch->head, ki_batch) {
510 list_add(&req->ki_list, &ctx->active_reqs);
515 spin_unlock_irq(&ctx->ctx_lock);
521 static inline struct kiocb *aio_get_req(struct kioctx *ctx,
522 struct kiocb_batch *batch)
526 if (list_empty(&batch->head))
527 if (kiocb_batch_refill(ctx, batch) == 0)
529 req = list_first_entry(&batch->head, struct kiocb, ki_batch);
530 list_del(&req->ki_batch);
534 static inline void really_put_req(struct kioctx *ctx, struct kiocb *req)
536 assert_spin_locked(&ctx->ctx_lock);
538 if (req->ki_eventfd != NULL)
539 eventfd_ctx_put(req->ki_eventfd);
542 if (req->ki_iovec != &req->ki_inline_vec)
543 kfree(req->ki_iovec);
544 kmem_cache_free(kiocb_cachep, req);
547 if (unlikely(!ctx->reqs_active && ctx->dead))
548 wake_up_all(&ctx->wait);
552 * Returns true if this put was the last user of the request.
554 static int __aio_put_req(struct kioctx *ctx, struct kiocb *req)
556 dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n",
557 req, atomic_long_read(&req->ki_filp->f_count));
559 assert_spin_locked(&ctx->ctx_lock);
562 BUG_ON(req->ki_users < 0);
563 if (likely(req->ki_users))
565 list_del(&req->ki_list); /* remove from active_reqs */
566 req->ki_cancel = NULL;
567 req->ki_retry = NULL;
571 really_put_req(ctx, req);
576 * Returns true if this put was the last user of the kiocb,
577 * false if the request is still in use.
579 int aio_put_req(struct kiocb *req)
581 struct kioctx *ctx = req->ki_ctx;
583 spin_lock_irq(&ctx->ctx_lock);
584 ret = __aio_put_req(ctx, req);
585 spin_unlock_irq(&ctx->ctx_lock);
588 EXPORT_SYMBOL(aio_put_req);
590 static struct kioctx *lookup_ioctx(unsigned long ctx_id)
592 struct mm_struct *mm = current->mm;
593 struct kioctx *ctx, *ret = NULL;
594 struct hlist_node *n;
598 hlist_for_each_entry_rcu(ctx, n, &mm->ioctx_list, list) {
600 * RCU protects us against accessing freed memory but
601 * we have to be careful not to get a reference when the
602 * reference count already dropped to 0 (ctx->dead test
603 * is unreliable because of races).
605 if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){
616 * Queue up a kiocb to be retried. Assumes that the kiocb
617 * has already been marked as kicked, and places it on
618 * the retry run list for the corresponding ioctx, if it
619 * isn't already queued. Returns 1 if it actually queued
620 * the kiocb (to tell the caller to activate the work
621 * queue to process it), or 0, if it found that it was
624 static inline int __queue_kicked_iocb(struct kiocb *iocb)
626 struct kioctx *ctx = iocb->ki_ctx;
628 assert_spin_locked(&ctx->ctx_lock);
630 if (list_empty(&iocb->ki_run_list)) {
631 list_add_tail(&iocb->ki_run_list,
639 * This is the core aio execution routine. It is
640 * invoked both for initial i/o submission and
641 * subsequent retries via the aio_kick_handler.
642 * Expects to be invoked with iocb->ki_ctx->lock
643 * already held. The lock is released and reacquired
644 * as needed during processing.
646 * Calls the iocb retry method (already setup for the
647 * iocb on initial submission) for operation specific
648 * handling, but takes care of most of common retry
649 * execution details for a given iocb. The retry method
650 * needs to be non-blocking as far as possible, to avoid
651 * holding up other iocbs waiting to be serviced by the
652 * retry kernel thread.
654 * The trickier parts in this code have to do with
655 * ensuring that only one retry instance is in progress
656 * for a given iocb at any time. Providing that guarantee
657 * simplifies the coding of individual aio operations as
658 * it avoids various potential races.
660 static ssize_t aio_run_iocb(struct kiocb *iocb)
662 struct kioctx *ctx = iocb->ki_ctx;
663 ssize_t (*retry)(struct kiocb *);
666 if (!(retry = iocb->ki_retry)) {
667 printk("aio_run_iocb: iocb->ki_retry = NULL\n");
672 * We don't want the next retry iteration for this
673 * operation to start until this one has returned and
674 * updated the iocb state. However, wait_queue functions
675 * can trigger a kick_iocb from interrupt context in the
676 * meantime, indicating that data is available for the next
677 * iteration. We want to remember that and enable the
678 * next retry iteration _after_ we are through with
681 * So, in order to be able to register a "kick", but
682 * prevent it from being queued now, we clear the kick
683 * flag, but make the kick code *think* that the iocb is
684 * still on the run list until we are actually done.
685 * When we are done with this iteration, we check if
686 * the iocb was kicked in the meantime and if so, queue
690 kiocbClearKicked(iocb);
693 * This is so that aio_complete knows it doesn't need to
694 * pull the iocb off the run list (We can't just call
695 * INIT_LIST_HEAD because we don't want a kick_iocb to
696 * queue this on the run list yet)
698 iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL;
699 spin_unlock_irq(&ctx->ctx_lock);
701 /* Quit retrying if the i/o has been cancelled */
702 if (kiocbIsCancelled(iocb)) {
704 aio_complete(iocb, ret, 0);
705 /* must not access the iocb after this */
710 * Now we are all set to call the retry method in async
715 if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) {
717 * There's no easy way to restart the syscall since other AIO's
718 * may be already running. Just fail this IO with EINTR.
720 if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
721 ret == -ERESTARTNOHAND || ret == -ERESTART_RESTARTBLOCK))
723 aio_complete(iocb, ret, 0);
726 spin_lock_irq(&ctx->ctx_lock);
728 if (-EIOCBRETRY == ret) {
730 * OK, now that we are done with this iteration
731 * and know that there is more left to go,
732 * this is where we let go so that a subsequent
733 * "kick" can start the next iteration
736 /* will make __queue_kicked_iocb succeed from here on */
737 INIT_LIST_HEAD(&iocb->ki_run_list);
738 /* we must queue the next iteration ourselves, if it
739 * has already been kicked */
740 if (kiocbIsKicked(iocb)) {
741 __queue_kicked_iocb(iocb);
744 * __queue_kicked_iocb will always return 1 here, because
745 * iocb->ki_run_list is empty at this point so it should
746 * be safe to unconditionally queue the context into the
757 * Process all pending retries queued on the ioctx
759 * Assumes it is operating within the aio issuer's mm
762 static int __aio_run_iocbs(struct kioctx *ctx)
765 struct list_head run_list;
767 assert_spin_locked(&ctx->ctx_lock);
769 list_replace_init(&ctx->run_list, &run_list);
770 while (!list_empty(&run_list)) {
771 iocb = list_entry(run_list.next, struct kiocb,
773 list_del(&iocb->ki_run_list);
775 * Hold an extra reference while retrying i/o.
777 iocb->ki_users++; /* grab extra reference */
779 __aio_put_req(ctx, iocb);
781 if (!list_empty(&ctx->run_list))
786 static void aio_queue_work(struct kioctx * ctx)
788 unsigned long timeout;
790 * if someone is waiting, get the work started right
791 * away, otherwise, use a longer delay
794 if (waitqueue_active(&ctx->wait))
798 queue_delayed_work(aio_wq, &ctx->wq, timeout);
803 * Process all pending retries queued on the ioctx
804 * run list, and keep running them until the list
806 * Assumes it is operating within the aio issuer's mm context.
808 static inline void aio_run_all_iocbs(struct kioctx *ctx)
810 spin_lock_irq(&ctx->ctx_lock);
811 while (__aio_run_iocbs(ctx))
813 spin_unlock_irq(&ctx->ctx_lock);
818 * Work queue handler triggered to process pending
819 * retries on an ioctx. Takes on the aio issuer's
820 * mm context before running the iocbs, so that
821 * copy_xxx_user operates on the issuer's address
823 * Run on aiod's context.
825 static void aio_kick_handler(struct work_struct *work)
827 struct kioctx *ctx = container_of(work, struct kioctx, wq.work);
828 mm_segment_t oldfs = get_fs();
829 struct mm_struct *mm;
834 spin_lock_irq(&ctx->ctx_lock);
835 requeue =__aio_run_iocbs(ctx);
837 spin_unlock_irq(&ctx->ctx_lock);
841 * we're in a worker thread already; no point using non-zero delay
844 queue_delayed_work(aio_wq, &ctx->wq, 0);
849 * Called by kick_iocb to queue the kiocb for retry
850 * and if required activate the aio work queue to process
853 static void try_queue_kicked_iocb(struct kiocb *iocb)
855 struct kioctx *ctx = iocb->ki_ctx;
859 spin_lock_irqsave(&ctx->ctx_lock, flags);
860 /* set this inside the lock so that we can't race with aio_run_iocb()
861 * testing it and putting the iocb on the run list under the lock */
862 if (!kiocbTryKick(iocb))
863 run = __queue_kicked_iocb(iocb);
864 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
871 * Called typically from a wait queue callback context
872 * to trigger a retry of the iocb.
873 * The retry is usually executed by aio workqueue
874 * threads (See aio_kick_handler).
876 void kick_iocb(struct kiocb *iocb)
878 /* sync iocbs are easy: they can only ever be executing from a
880 if (is_sync_kiocb(iocb)) {
881 kiocbSetKicked(iocb);
882 wake_up_process(iocb->ki_obj.tsk);
886 try_queue_kicked_iocb(iocb);
888 EXPORT_SYMBOL(kick_iocb);
891 * Called when the io request on the given iocb is complete.
892 * Returns true if this is the last user of the request. The
893 * only other user of the request can be the cancellation code.
895 int aio_complete(struct kiocb *iocb, long res, long res2)
897 struct kioctx *ctx = iocb->ki_ctx;
898 struct aio_ring_info *info;
899 struct aio_ring *ring;
900 struct io_event *event;
906 * Special case handling for sync iocbs:
907 * - events go directly into the iocb for fast handling
908 * - the sync task with the iocb in its stack holds the single iocb
909 * ref, no other paths have a way to get another ref
910 * - the sync task helpfully left a reference to itself in the iocb
912 if (is_sync_kiocb(iocb)) {
913 BUG_ON(iocb->ki_users != 1);
914 iocb->ki_user_data = res;
916 wake_up_process(iocb->ki_obj.tsk);
920 info = &ctx->ring_info;
922 /* add a completion event to the ring buffer.
923 * must be done holding ctx->ctx_lock to prevent
924 * other code from messing with the tail
925 * pointer since we might be called from irq
928 spin_lock_irqsave(&ctx->ctx_lock, flags);
930 if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list))
931 list_del_init(&iocb->ki_run_list);
934 * cancelled requests don't get events, userland was given one
935 * when the event got cancelled.
937 if (kiocbIsCancelled(iocb))
940 ring = kmap_atomic(info->ring_pages[0]);
943 event = aio_ring_event(info, tail);
944 if (++tail >= info->nr)
947 event->obj = (u64)(unsigned long)iocb->ki_obj.user;
948 event->data = iocb->ki_user_data;
952 dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n",
953 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
956 /* after flagging the request as done, we
957 * must never even look at it again
959 smp_wmb(); /* make event visible before updating tail */
964 put_aio_ring_event(event);
967 pr_debug("added to ring %p at [%lu]\n", iocb, tail);
970 * Check if the user asked us to deliver the result through an
971 * eventfd. The eventfd_signal() function is safe to be called
974 if (iocb->ki_eventfd != NULL)
975 eventfd_signal(iocb->ki_eventfd, 1);
978 /* everything turned out well, dispose of the aiocb. */
979 ret = __aio_put_req(ctx, iocb);
982 * We have to order our ring_info tail store above and test
983 * of the wait list below outside the wait lock. This is
984 * like in wake_up_bit() where clearing a bit has to be
985 * ordered with the unlocked test.
989 if (waitqueue_active(&ctx->wait))
992 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
995 EXPORT_SYMBOL(aio_complete);
998 * Pull an event off of the ioctx's event ring. Returns the number of
999 * events fetched (0 or 1 ;-)
1000 * FIXME: make this use cmpxchg.
1001 * TODO: make the ringbuffer user mmap()able (requires FIXME).
1003 static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent)
1005 struct aio_ring_info *info = &ioctx->ring_info;
1006 struct aio_ring *ring;
1010 ring = kmap_atomic(info->ring_pages[0]);
1011 dprintk("in aio_read_evt h%lu t%lu m%lu\n",
1012 (unsigned long)ring->head, (unsigned long)ring->tail,
1013 (unsigned long)ring->nr);
1015 if (ring->head == ring->tail)
1018 spin_lock(&info->ring_lock);
1020 head = ring->head % info->nr;
1021 if (head != ring->tail) {
1022 struct io_event *evp = aio_ring_event(info, head);
1024 head = (head + 1) % info->nr;
1025 smp_mb(); /* finish reading the event before updatng the head */
1028 put_aio_ring_event(evp);
1030 spin_unlock(&info->ring_lock);
1033 kunmap_atomic(ring);
1034 dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret,
1035 (unsigned long)ring->head, (unsigned long)ring->tail);
1039 struct aio_timeout {
1040 struct timer_list timer;
1042 struct task_struct *p;
1045 static void timeout_func(unsigned long data)
1047 struct aio_timeout *to = (struct aio_timeout *)data;
1050 wake_up_process(to->p);
1053 static inline void init_timeout(struct aio_timeout *to)
1055 setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to);
1060 static inline void set_timeout(long start_jiffies, struct aio_timeout *to,
1061 const struct timespec *ts)
1063 to->timer.expires = start_jiffies + timespec_to_jiffies(ts);
1064 if (time_after(to->timer.expires, jiffies))
1065 add_timer(&to->timer);
1070 static inline void clear_timeout(struct aio_timeout *to)
1072 del_singleshot_timer_sync(&to->timer);
1075 static int read_events(struct kioctx *ctx,
1076 long min_nr, long nr,
1077 struct io_event __user *event,
1078 struct timespec __user *timeout)
1080 long start_jiffies = jiffies;
1081 struct task_struct *tsk = current;
1082 DECLARE_WAITQUEUE(wait, tsk);
1085 struct io_event ent;
1086 struct aio_timeout to;
1089 /* needed to zero any padding within an entry (there shouldn't be
1090 * any, but C is fun!
1092 memset(&ent, 0, sizeof(ent));
1095 while (likely(i < nr)) {
1096 ret = aio_read_evt(ctx, &ent);
1097 if (unlikely(ret <= 0))
1100 dprintk("read event: %Lx %Lx %Lx %Lx\n",
1101 ent.data, ent.obj, ent.res, ent.res2);
1103 /* Could we split the check in two? */
1105 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1106 dprintk("aio: lost an event due to EFAULT.\n");
1111 /* Good, event copied to userland, update counts. */
1123 /* racey check, but it gets redone */
1124 if (!retry && unlikely(!list_empty(&ctx->run_list))) {
1126 aio_run_all_iocbs(ctx);
1134 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1137 set_timeout(start_jiffies, &to, &ts);
1140 while (likely(i < nr)) {
1141 add_wait_queue_exclusive(&ctx->wait, &wait);
1143 set_task_state(tsk, TASK_INTERRUPTIBLE);
1144 ret = aio_read_evt(ctx, &ent);
1149 if (unlikely(ctx->dead)) {
1153 if (to.timed_out) /* Only check after read evt */
1155 /* Try to only show up in io wait if there are ops
1157 if (ctx->reqs_active)
1161 if (signal_pending(tsk)) {
1165 /*ret = aio_read_evt(ctx, &ent);*/
1168 set_task_state(tsk, TASK_RUNNING);
1169 remove_wait_queue(&ctx->wait, &wait);
1171 if (unlikely(ret <= 0))
1175 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1176 dprintk("aio: lost an event due to EFAULT.\n");
1180 /* Good, event copied to userland, update counts. */
1188 destroy_timer_on_stack(&to.timer);
1192 /* Take an ioctx and remove it from the list of ioctx's. Protects
1193 * against races with itself via ->dead.
1195 static void io_destroy(struct kioctx *ioctx)
1197 struct mm_struct *mm = current->mm;
1200 /* delete the entry from the list is someone else hasn't already */
1201 spin_lock(&mm->ioctx_lock);
1202 was_dead = ioctx->dead;
1204 hlist_del_rcu(&ioctx->list);
1205 spin_unlock(&mm->ioctx_lock);
1207 dprintk("aio_release(%p)\n", ioctx);
1208 if (likely(!was_dead))
1209 put_ioctx(ioctx); /* twice for the list */
1214 * Wake up any waiters. The setting of ctx->dead must be seen
1215 * by other CPUs at this point. Right now, we rely on the
1216 * locking done by the above calls to ensure this consistency.
1218 wake_up_all(&ioctx->wait);
1222 * Create an aio_context capable of receiving at least nr_events.
1223 * ctxp must not point to an aio_context that already exists, and
1224 * must be initialized to 0 prior to the call. On successful
1225 * creation of the aio_context, *ctxp is filled in with the resulting
1226 * handle. May fail with -EINVAL if *ctxp is not initialized,
1227 * if the specified nr_events exceeds internal limits. May fail
1228 * with -EAGAIN if the specified nr_events exceeds the user's limit
1229 * of available events. May fail with -ENOMEM if insufficient kernel
1230 * resources are available. May fail with -EFAULT if an invalid
1231 * pointer is passed for ctxp. Will fail with -ENOSYS if not
1234 SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
1236 struct kioctx *ioctx = NULL;
1240 ret = get_user(ctx, ctxp);
1245 if (unlikely(ctx || nr_events == 0)) {
1246 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
1251 ioctx = ioctx_alloc(nr_events);
1252 ret = PTR_ERR(ioctx);
1253 if (!IS_ERR(ioctx)) {
1254 ret = put_user(ioctx->user_id, ctxp);
1265 * Destroy the aio_context specified. May cancel any outstanding
1266 * AIOs and block on completion. Will fail with -ENOSYS if not
1267 * implemented. May fail with -EINVAL if the context pointed to
1270 SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
1272 struct kioctx *ioctx = lookup_ioctx(ctx);
1273 if (likely(NULL != ioctx)) {
1278 pr_debug("EINVAL: io_destroy: invalid context id\n");
1282 static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
1284 struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];
1288 while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
1289 ssize_t this = min((ssize_t)iov->iov_len, ret);
1290 iov->iov_base += this;
1291 iov->iov_len -= this;
1292 iocb->ki_left -= this;
1294 if (iov->iov_len == 0) {
1300 /* the caller should not have done more io than what fit in
1301 * the remaining iovecs */
1302 BUG_ON(ret > 0 && iocb->ki_left == 0);
1305 static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
1307 struct file *file = iocb->ki_filp;
1308 struct address_space *mapping = file->f_mapping;
1309 struct inode *inode = mapping->host;
1310 ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
1311 unsigned long, loff_t);
1313 unsigned short opcode;
1315 if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
1316 (iocb->ki_opcode == IOCB_CMD_PREAD)) {
1317 rw_op = file->f_op->aio_read;
1318 opcode = IOCB_CMD_PREADV;
1320 rw_op = file->f_op->aio_write;
1321 opcode = IOCB_CMD_PWRITEV;
1324 /* This matches the pread()/pwrite() logic */
1325 if (iocb->ki_pos < 0)
1329 ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
1330 iocb->ki_nr_segs - iocb->ki_cur_seg,
1333 aio_advance_iovec(iocb, ret);
1335 /* retry all partial writes. retry partial reads as long as its a
1337 } while (ret > 0 && iocb->ki_left > 0 &&
1338 (opcode == IOCB_CMD_PWRITEV ||
1339 (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
1341 /* This means we must have transferred all that we could */
1342 /* No need to retry anymore */
1343 if ((ret == 0) || (iocb->ki_left == 0))
1344 ret = iocb->ki_nbytes - iocb->ki_left;
1346 /* If we managed to write some out we return that, rather than
1347 * the eventual error. */
1348 if (opcode == IOCB_CMD_PWRITEV
1349 && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY
1350 && iocb->ki_nbytes - iocb->ki_left)
1351 ret = iocb->ki_nbytes - iocb->ki_left;
1356 static ssize_t aio_fdsync(struct kiocb *iocb)
1358 struct file *file = iocb->ki_filp;
1359 ssize_t ret = -EINVAL;
1361 if (file->f_op->aio_fsync)
1362 ret = file->f_op->aio_fsync(iocb, 1);
1366 static ssize_t aio_fsync(struct kiocb *iocb)
1368 struct file *file = iocb->ki_filp;
1369 ssize_t ret = -EINVAL;
1371 if (file->f_op->aio_fsync)
1372 ret = file->f_op->aio_fsync(iocb, 0);
1376 static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat)
1380 #ifdef CONFIG_COMPAT
1382 ret = compat_rw_copy_check_uvector(type,
1383 (struct compat_iovec __user *)kiocb->ki_buf,
1384 kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
1388 ret = rw_copy_check_uvector(type,
1389 (struct iovec __user *)kiocb->ki_buf,
1390 kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
1395 ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret);
1399 kiocb->ki_nr_segs = kiocb->ki_nbytes;
1400 kiocb->ki_cur_seg = 0;
1401 /* ki_nbytes/left now reflect bytes instead of segs */
1402 kiocb->ki_nbytes = ret;
1403 kiocb->ki_left = ret;
1410 static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb)
1414 bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left);
1418 kiocb->ki_iovec = &kiocb->ki_inline_vec;
1419 kiocb->ki_iovec->iov_base = kiocb->ki_buf;
1420 kiocb->ki_iovec->iov_len = bytes;
1421 kiocb->ki_nr_segs = 1;
1422 kiocb->ki_cur_seg = 0;
1428 * Performs the initial checks and aio retry method
1429 * setup for the kiocb at the time of io submission.
1431 static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat)
1433 struct file *file = kiocb->ki_filp;
1436 switch (kiocb->ki_opcode) {
1437 case IOCB_CMD_PREAD:
1439 if (unlikely(!(file->f_mode & FMODE_READ)))
1442 if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
1445 ret = aio_setup_single_vector(READ, file, kiocb);
1449 if (file->f_op->aio_read)
1450 kiocb->ki_retry = aio_rw_vect_retry;
1452 case IOCB_CMD_PWRITE:
1454 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1457 if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
1460 ret = aio_setup_single_vector(WRITE, file, kiocb);
1464 if (file->f_op->aio_write)
1465 kiocb->ki_retry = aio_rw_vect_retry;
1467 case IOCB_CMD_PREADV:
1469 if (unlikely(!(file->f_mode & FMODE_READ)))
1471 ret = aio_setup_vectored_rw(READ, kiocb, compat);
1475 if (file->f_op->aio_read)
1476 kiocb->ki_retry = aio_rw_vect_retry;
1478 case IOCB_CMD_PWRITEV:
1480 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1482 ret = aio_setup_vectored_rw(WRITE, kiocb, compat);
1486 if (file->f_op->aio_write)
1487 kiocb->ki_retry = aio_rw_vect_retry;
1489 case IOCB_CMD_FDSYNC:
1491 if (file->f_op->aio_fsync)
1492 kiocb->ki_retry = aio_fdsync;
1494 case IOCB_CMD_FSYNC:
1496 if (file->f_op->aio_fsync)
1497 kiocb->ki_retry = aio_fsync;
1500 dprintk("EINVAL: io_submit: no operation provided\n");
1504 if (!kiocb->ki_retry)
1510 static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
1511 struct iocb *iocb, struct kiocb_batch *batch,
1518 /* enforce forwards compatibility on users */
1519 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
1520 pr_debug("EINVAL: io_submit: reserve field set\n");
1524 /* prevent overflows */
1526 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1527 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1528 ((ssize_t)iocb->aio_nbytes < 0)
1530 pr_debug("EINVAL: io_submit: overflow check\n");
1534 file = fget(iocb->aio_fildes);
1535 if (unlikely(!file))
1538 req = aio_get_req(ctx, batch); /* returns with 2 references to req */
1539 if (unlikely(!req)) {
1543 req->ki_filp = file;
1544 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1546 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
1547 * instance of the file* now. The file descriptor must be
1548 * an eventfd() fd, and will be signaled for each completed
1549 * event using the eventfd_signal() function.
1551 req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd);
1552 if (IS_ERR(req->ki_eventfd)) {
1553 ret = PTR_ERR(req->ki_eventfd);
1554 req->ki_eventfd = NULL;
1559 ret = put_user(req->ki_key, &user_iocb->aio_key);
1560 if (unlikely(ret)) {
1561 dprintk("EFAULT: aio_key\n");
1565 req->ki_obj.user = user_iocb;
1566 req->ki_user_data = iocb->aio_data;
1567 req->ki_pos = iocb->aio_offset;
1569 req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf;
1570 req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
1571 req->ki_opcode = iocb->aio_lio_opcode;
1573 ret = aio_setup_iocb(req, compat);
1578 spin_lock_irq(&ctx->ctx_lock);
1580 * We could have raced with io_destroy() and are currently holding a
1581 * reference to ctx which should be destroyed. We cannot submit IO
1582 * since ctx gets freed as soon as io_submit() puts its reference. The
1583 * check here is reliable: io_destroy() sets ctx->dead before waiting
1584 * for outstanding IO and the barrier between these two is realized by
1585 * unlock of mm->ioctx_lock and lock of ctx->ctx_lock. Analogously we
1586 * increment ctx->reqs_active before checking for ctx->dead and the
1587 * barrier is realized by unlock and lock of ctx->ctx_lock. Thus if we
1588 * don't see ctx->dead set here, io_destroy() waits for our IO to
1592 spin_unlock_irq(&ctx->ctx_lock);
1597 if (!list_empty(&ctx->run_list)) {
1598 /* drain the run list */
1599 while (__aio_run_iocbs(ctx))
1602 spin_unlock_irq(&ctx->ctx_lock);
1604 aio_put_req(req); /* drop extra ref to req */
1608 aio_put_req(req); /* drop extra ref to req */
1609 aio_put_req(req); /* drop i/o ref to req */
1613 long do_io_submit(aio_context_t ctx_id, long nr,
1614 struct iocb __user *__user *iocbpp, bool compat)
1619 struct blk_plug plug;
1620 struct kiocb_batch batch;
1622 if (unlikely(nr < 0))
1625 if (unlikely(nr > LONG_MAX/sizeof(*iocbpp)))
1626 nr = LONG_MAX/sizeof(*iocbpp);
1628 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1631 ctx = lookup_ioctx(ctx_id);
1632 if (unlikely(!ctx)) {
1633 pr_debug("EINVAL: io_submit: invalid context id\n");
1637 kiocb_batch_init(&batch, nr);
1639 blk_start_plug(&plug);
1642 * AKPM: should this return a partial result if some of the IOs were
1643 * successfully submitted?
1645 for (i=0; i<nr; i++) {
1646 struct iocb __user *user_iocb;
1649 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1654 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1659 ret = io_submit_one(ctx, user_iocb, &tmp, &batch, compat);
1663 blk_finish_plug(&plug);
1665 kiocb_batch_free(ctx, &batch);
1671 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1672 * the number of iocbs queued. May return -EINVAL if the aio_context
1673 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1674 * *iocbpp[0] is not properly initialized, if the operation specified
1675 * is invalid for the file descriptor in the iocb. May fail with
1676 * -EFAULT if any of the data structures point to invalid data. May
1677 * fail with -EBADF if the file descriptor specified in the first
1678 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1679 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1680 * fail with -ENOSYS if not implemented.
1682 SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr,
1683 struct iocb __user * __user *, iocbpp)
1685 return do_io_submit(ctx_id, nr, iocbpp, 0);
1689 * Finds a given iocb for cancellation.
1691 static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
1694 struct list_head *pos;
1696 assert_spin_locked(&ctx->ctx_lock);
1698 /* TODO: use a hash or array, this sucks. */
1699 list_for_each(pos, &ctx->active_reqs) {
1700 struct kiocb *kiocb = list_kiocb(pos);
1701 if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key)
1708 * Attempts to cancel an iocb previously passed to io_submit. If
1709 * the operation is successfully cancelled, the resulting event is
1710 * copied into the memory pointed to by result without being placed
1711 * into the completion queue and 0 is returned. May fail with
1712 * -EFAULT if any of the data structures pointed to are invalid.
1713 * May fail with -EINVAL if aio_context specified by ctx_id is
1714 * invalid. May fail with -EAGAIN if the iocb specified was not
1715 * cancelled. Will fail with -ENOSYS if not implemented.
1717 SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb,
1718 struct io_event __user *, result)
1720 int (*cancel)(struct kiocb *iocb, struct io_event *res);
1722 struct kiocb *kiocb;
1726 ret = get_user(key, &iocb->aio_key);
1730 ctx = lookup_ioctx(ctx_id);
1734 spin_lock_irq(&ctx->ctx_lock);
1736 kiocb = lookup_kiocb(ctx, iocb, key);
1737 if (kiocb && kiocb->ki_cancel) {
1738 cancel = kiocb->ki_cancel;
1740 kiocbSetCancelled(kiocb);
1743 spin_unlock_irq(&ctx->ctx_lock);
1745 if (NULL != cancel) {
1746 struct io_event tmp;
1747 pr_debug("calling cancel\n");
1748 memset(&tmp, 0, sizeof(tmp));
1749 tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user;
1750 tmp.data = kiocb->ki_user_data;
1751 ret = cancel(kiocb, &tmp);
1753 /* Cancellation succeeded -- copy the result
1754 * into the user's buffer.
1756 if (copy_to_user(result, &tmp, sizeof(tmp)))
1768 * Attempts to read at least min_nr events and up to nr events from
1769 * the completion queue for the aio_context specified by ctx_id. If
1770 * it succeeds, the number of read events is returned. May fail with
1771 * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is
1772 * out of range, if timeout is out of range. May fail with -EFAULT
1773 * if any of the memory specified is invalid. May return 0 or
1774 * < min_nr if the timeout specified by timeout has elapsed
1775 * before sufficient events are available, where timeout == NULL
1776 * specifies an infinite timeout. Note that the timeout pointed to by
1777 * timeout is relative and will be updated if not NULL and the
1778 * operation blocks. Will fail with -ENOSYS if not implemented.
1780 SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id,
1783 struct io_event __user *, events,
1784 struct timespec __user *, timeout)
1786 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1789 if (likely(ioctx)) {
1790 if (likely(min_nr <= nr && min_nr >= 0))
1791 ret = read_events(ioctx, min_nr, nr, events, timeout);
1795 asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout);