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/module.h>
17 #include <linux/syscalls.h>
18 #include <linux/uio.h>
22 #include <linux/sched.h>
24 #include <linux/file.h>
26 #include <linux/mman.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/aio.h>
30 #include <linux/highmem.h>
31 #include <linux/workqueue.h>
32 #include <linux/security.h>
33 #include <linux/eventfd.h>
35 #include <asm/kmap_types.h>
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
40 #define dprintk printk
42 #define dprintk(x...) do { ; } while (0)
45 /*------ sysctl variables----*/
46 static DEFINE_SPINLOCK(aio_nr_lock);
47 unsigned long aio_nr; /* current system wide number of aio requests */
48 unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */
49 /*----end sysctl variables---*/
51 static struct kmem_cache *kiocb_cachep;
52 static struct kmem_cache *kioctx_cachep;
54 static struct workqueue_struct *aio_wq;
56 /* Used for rare fput completion. */
57 static void aio_fput_routine(struct work_struct *);
58 static DECLARE_WORK(fput_work, aio_fput_routine);
60 static DEFINE_SPINLOCK(fput_lock);
61 static LIST_HEAD(fput_head);
63 static void aio_kick_handler(struct work_struct *);
64 static void aio_queue_work(struct kioctx *);
67 * Creates the slab caches used by the aio routines, panic on
68 * failure as this is done early during the boot sequence.
70 static int __init aio_setup(void)
72 kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
73 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
75 aio_wq = create_workqueue("aio");
77 pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page));
82 static void aio_free_ring(struct kioctx *ctx)
84 struct aio_ring_info *info = &ctx->ring_info;
87 for (i=0; i<info->nr_pages; i++)
88 put_page(info->ring_pages[i]);
90 if (info->mmap_size) {
91 down_write(&ctx->mm->mmap_sem);
92 do_munmap(ctx->mm, info->mmap_base, info->mmap_size);
93 up_write(&ctx->mm->mmap_sem);
96 if (info->ring_pages && info->ring_pages != info->internal_pages)
97 kfree(info->ring_pages);
98 info->ring_pages = NULL;
102 static int aio_setup_ring(struct kioctx *ctx)
104 struct aio_ring *ring;
105 struct aio_ring_info *info = &ctx->ring_info;
106 unsigned nr_events = ctx->max_reqs;
110 /* Compensate for the ring buffer's head/tail overlap entry */
111 nr_events += 2; /* 1 is required, 2 for good luck */
113 size = sizeof(struct aio_ring);
114 size += sizeof(struct io_event) * nr_events;
115 nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT;
120 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event);
123 info->ring_pages = info->internal_pages;
124 if (nr_pages > AIO_RING_PAGES) {
125 info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
126 if (!info->ring_pages)
130 info->mmap_size = nr_pages * PAGE_SIZE;
131 dprintk("attempting mmap of %lu bytes\n", info->mmap_size);
132 down_write(&ctx->mm->mmap_sem);
133 info->mmap_base = do_mmap(NULL, 0, info->mmap_size,
134 PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE,
136 if (IS_ERR((void *)info->mmap_base)) {
137 up_write(&ctx->mm->mmap_sem);
143 dprintk("mmap address: 0x%08lx\n", info->mmap_base);
144 info->nr_pages = get_user_pages(current, ctx->mm,
145 info->mmap_base, nr_pages,
146 1, 0, info->ring_pages, NULL);
147 up_write(&ctx->mm->mmap_sem);
149 if (unlikely(info->nr_pages != nr_pages)) {
154 ctx->user_id = info->mmap_base;
156 info->nr = nr_events; /* trusted copy */
158 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
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);
166 kunmap_atomic(ring, KM_USER0);
172 /* aio_ring_event: returns a pointer to the event at the given index from
173 * kmap_atomic(, km). 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, km) ({ \
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], km); \
184 __event += pos % AIO_EVENTS_PER_PAGE; \
188 #define put_aio_ring_event(event, km) do { \
189 struct io_event *__event = (event); \
191 kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \
195 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
197 static struct kioctx *ioctx_alloc(unsigned nr_events)
199 struct mm_struct *mm;
202 /* Prevent overflows */
203 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
204 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
205 pr_debug("ENOMEM: nr_events too high\n");
206 return ERR_PTR(-EINVAL);
209 if ((unsigned long)nr_events > aio_max_nr)
210 return ERR_PTR(-EAGAIN);
212 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
214 return ERR_PTR(-ENOMEM);
216 ctx->max_reqs = nr_events;
217 mm = ctx->mm = current->mm;
218 atomic_inc(&mm->mm_count);
220 atomic_set(&ctx->users, 1);
221 spin_lock_init(&ctx->ctx_lock);
222 spin_lock_init(&ctx->ring_info.ring_lock);
223 init_waitqueue_head(&ctx->wait);
225 INIT_LIST_HEAD(&ctx->active_reqs);
226 INIT_LIST_HEAD(&ctx->run_list);
227 INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler);
229 if (aio_setup_ring(ctx) < 0)
232 /* limit the number of system wide aios */
233 spin_lock(&aio_nr_lock);
234 if (aio_nr + ctx->max_reqs > aio_max_nr ||
235 aio_nr + ctx->max_reqs < aio_nr)
238 aio_nr += ctx->max_reqs;
239 spin_unlock(&aio_nr_lock);
240 if (ctx->max_reqs == 0)
243 /* now link into global list. kludge. FIXME */
244 write_lock(&mm->ioctx_list_lock);
245 ctx->next = mm->ioctx_list;
246 mm->ioctx_list = ctx;
247 write_unlock(&mm->ioctx_list_lock);
249 dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
250 ctx, ctx->user_id, current->mm, ctx->ring_info.nr);
255 return ERR_PTR(-EAGAIN);
259 kmem_cache_free(kioctx_cachep, ctx);
260 ctx = ERR_PTR(-ENOMEM);
262 dprintk("aio: error allocating ioctx %p\n", ctx);
267 * Cancels all outstanding aio requests on an aio context. Used
268 * when the processes owning a context have all exited to encourage
269 * the rapid destruction of the kioctx.
271 static void aio_cancel_all(struct kioctx *ctx)
273 int (*cancel)(struct kiocb *, struct io_event *);
275 spin_lock_irq(&ctx->ctx_lock);
277 while (!list_empty(&ctx->active_reqs)) {
278 struct list_head *pos = ctx->active_reqs.next;
279 struct kiocb *iocb = list_kiocb(pos);
280 list_del_init(&iocb->ki_list);
281 cancel = iocb->ki_cancel;
282 kiocbSetCancelled(iocb);
285 spin_unlock_irq(&ctx->ctx_lock);
287 spin_lock_irq(&ctx->ctx_lock);
290 spin_unlock_irq(&ctx->ctx_lock);
293 static void wait_for_all_aios(struct kioctx *ctx)
295 struct task_struct *tsk = current;
296 DECLARE_WAITQUEUE(wait, tsk);
298 spin_lock_irq(&ctx->ctx_lock);
299 if (!ctx->reqs_active)
302 add_wait_queue(&ctx->wait, &wait);
303 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
304 while (ctx->reqs_active) {
305 spin_unlock_irq(&ctx->ctx_lock);
307 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
308 spin_lock_irq(&ctx->ctx_lock);
310 __set_task_state(tsk, TASK_RUNNING);
311 remove_wait_queue(&ctx->wait, &wait);
314 spin_unlock_irq(&ctx->ctx_lock);
317 /* wait_on_sync_kiocb:
318 * Waits on the given sync kiocb to complete.
320 ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
322 while (iocb->ki_users) {
323 set_current_state(TASK_UNINTERRUPTIBLE);
328 __set_current_state(TASK_RUNNING);
329 return iocb->ki_user_data;
332 /* exit_aio: called when the last user of mm goes away. At this point,
333 * there is no way for any new requests to be submited or any of the
334 * io_* syscalls to be called on the context. However, there may be
335 * outstanding requests which hold references to the context; as they
336 * go away, they will call put_ioctx and release any pinned memory
337 * associated with the request (held via struct page * references).
339 void exit_aio(struct mm_struct *mm)
341 struct kioctx *ctx = mm->ioctx_list;
342 mm->ioctx_list = NULL;
344 struct kioctx *next = ctx->next;
348 wait_for_all_aios(ctx);
350 * Ensure we don't leave the ctx on the aio_wq
352 cancel_work_sync(&ctx->wq.work);
354 if (1 != atomic_read(&ctx->users))
356 "exit_aio:ioctx still alive: %d %d %d\n",
357 atomic_read(&ctx->users), ctx->dead,
365 * Called when the last user of an aio context has gone away,
366 * and the struct needs to be freed.
368 void __put_ioctx(struct kioctx *ctx)
370 unsigned nr_events = ctx->max_reqs;
372 BUG_ON(ctx->reqs_active);
374 cancel_delayed_work(&ctx->wq);
375 cancel_work_sync(&ctx->wq.work);
379 pr_debug("__put_ioctx: freeing %p\n", ctx);
380 kmem_cache_free(kioctx_cachep, ctx);
383 spin_lock(&aio_nr_lock);
384 BUG_ON(aio_nr - nr_events > aio_nr);
386 spin_unlock(&aio_nr_lock);
391 * Allocate a slot for an aio request. Increments the users count
392 * of the kioctx so that the kioctx stays around until all requests are
393 * complete. Returns NULL if no requests are free.
395 * Returns with kiocb->users set to 2. The io submit code path holds
396 * an extra reference while submitting the i/o.
397 * This prevents races between the aio code path referencing the
398 * req (after submitting it) and aio_complete() freeing the req.
400 static struct kiocb *__aio_get_req(struct kioctx *ctx)
402 struct kiocb *req = NULL;
403 struct aio_ring *ring;
406 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL);
414 req->ki_cancel = NULL;
415 req->ki_retry = NULL;
418 req->ki_iovec = NULL;
419 INIT_LIST_HEAD(&req->ki_run_list);
420 req->ki_eventfd = ERR_PTR(-EINVAL);
422 /* Check if the completion queue has enough free space to
423 * accept an event from this io.
425 spin_lock_irq(&ctx->ctx_lock);
426 ring = kmap_atomic(ctx->ring_info.ring_pages[0], KM_USER0);
427 if (ctx->reqs_active < aio_ring_avail(&ctx->ring_info, ring)) {
428 list_add(&req->ki_list, &ctx->active_reqs);
432 kunmap_atomic(ring, KM_USER0);
433 spin_unlock_irq(&ctx->ctx_lock);
436 kmem_cache_free(kiocb_cachep, req);
443 static inline struct kiocb *aio_get_req(struct kioctx *ctx)
446 /* Handle a potential starvation case -- should be exceedingly rare as
447 * requests will be stuck on fput_head only if the aio_fput_routine is
448 * delayed and the requests were the last user of the struct file.
450 req = __aio_get_req(ctx);
451 if (unlikely(NULL == req)) {
452 aio_fput_routine(NULL);
453 req = __aio_get_req(ctx);
458 static inline void really_put_req(struct kioctx *ctx, struct kiocb *req)
460 assert_spin_locked(&ctx->ctx_lock);
462 if (!IS_ERR(req->ki_eventfd))
463 fput(req->ki_eventfd);
466 if (req->ki_iovec != &req->ki_inline_vec)
467 kfree(req->ki_iovec);
468 kmem_cache_free(kiocb_cachep, req);
471 if (unlikely(!ctx->reqs_active && ctx->dead))
475 static void aio_fput_routine(struct work_struct *data)
477 spin_lock_irq(&fput_lock);
478 while (likely(!list_empty(&fput_head))) {
479 struct kiocb *req = list_kiocb(fput_head.next);
480 struct kioctx *ctx = req->ki_ctx;
482 list_del(&req->ki_list);
483 spin_unlock_irq(&fput_lock);
485 /* Complete the fput */
486 __fput(req->ki_filp);
488 /* Link the iocb into the context's free list */
489 spin_lock_irq(&ctx->ctx_lock);
490 really_put_req(ctx, req);
491 spin_unlock_irq(&ctx->ctx_lock);
494 spin_lock_irq(&fput_lock);
496 spin_unlock_irq(&fput_lock);
500 * Returns true if this put was the last user of the request.
502 static int __aio_put_req(struct kioctx *ctx, struct kiocb *req)
504 dprintk(KERN_DEBUG "aio_put(%p): f_count=%d\n",
505 req, atomic_read(&req->ki_filp->f_count));
507 assert_spin_locked(&ctx->ctx_lock);
510 BUG_ON(req->ki_users < 0);
511 if (likely(req->ki_users))
513 list_del(&req->ki_list); /* remove from active_reqs */
514 req->ki_cancel = NULL;
515 req->ki_retry = NULL;
517 /* Must be done under the lock to serialise against cancellation.
518 * Call this aio_fput as it duplicates fput via the fput_work.
520 if (unlikely(atomic_dec_and_test(&req->ki_filp->f_count))) {
522 spin_lock(&fput_lock);
523 list_add(&req->ki_list, &fput_head);
524 spin_unlock(&fput_lock);
525 queue_work(aio_wq, &fput_work);
527 really_put_req(ctx, req);
532 * Returns true if this put was the last user of the kiocb,
533 * false if the request is still in use.
535 int aio_put_req(struct kiocb *req)
537 struct kioctx *ctx = req->ki_ctx;
539 spin_lock_irq(&ctx->ctx_lock);
540 ret = __aio_put_req(ctx, req);
541 spin_unlock_irq(&ctx->ctx_lock);
545 /* Lookup an ioctx id. ioctx_list is lockless for reads.
546 * FIXME: this is O(n) and is only suitable for development.
548 struct kioctx *lookup_ioctx(unsigned long ctx_id)
550 struct kioctx *ioctx;
551 struct mm_struct *mm;
554 read_lock(&mm->ioctx_list_lock);
555 for (ioctx = mm->ioctx_list; ioctx; ioctx = ioctx->next)
556 if (likely(ioctx->user_id == ctx_id && !ioctx->dead)) {
560 read_unlock(&mm->ioctx_list_lock);
567 * Makes the calling kernel thread take on the specified
569 * Called by the retry thread execute retries within the
570 * iocb issuer's mm context, so that copy_from/to_user
571 * operations work seamlessly for aio.
572 * (Note: this routine is intended to be called only
573 * from a kernel thread context)
575 static void use_mm(struct mm_struct *mm)
577 struct mm_struct *active_mm;
578 struct task_struct *tsk = current;
581 tsk->flags |= PF_BORROWED_MM;
582 active_mm = tsk->active_mm;
583 atomic_inc(&mm->mm_count);
587 * Note that on UML this *requires* PF_BORROWED_MM to be set, otherwise
588 * it won't work. Update it accordingly if you change it here
590 switch_mm(active_mm, mm, tsk);
598 * Reverses the effect of use_mm, i.e. releases the
599 * specified mm context which was earlier taken on
600 * by the calling kernel thread
601 * (Note: this routine is intended to be called only
602 * from a kernel thread context)
604 static void unuse_mm(struct mm_struct *mm)
606 struct task_struct *tsk = current;
609 tsk->flags &= ~PF_BORROWED_MM;
611 /* active_mm is still 'mm' */
612 enter_lazy_tlb(mm, tsk);
617 * Queue up a kiocb to be retried. Assumes that the kiocb
618 * has already been marked as kicked, and places it on
619 * the retry run list for the corresponding ioctx, if it
620 * isn't already queued. Returns 1 if it actually queued
621 * the kiocb (to tell the caller to activate the work
622 * queue to process it), or 0, if it found that it was
625 static inline int __queue_kicked_iocb(struct kiocb *iocb)
627 struct kioctx *ctx = iocb->ki_ctx;
629 assert_spin_locked(&ctx->ctx_lock);
631 if (list_empty(&iocb->ki_run_list)) {
632 list_add_tail(&iocb->ki_run_list,
640 * This is the core aio execution routine. It is
641 * invoked both for initial i/o submission and
642 * subsequent retries via the aio_kick_handler.
643 * Expects to be invoked with iocb->ki_ctx->lock
644 * already held. The lock is released and reacquired
645 * as needed during processing.
647 * Calls the iocb retry method (already setup for the
648 * iocb on initial submission) for operation specific
649 * handling, but takes care of most of common retry
650 * execution details for a given iocb. The retry method
651 * needs to be non-blocking as far as possible, to avoid
652 * holding up other iocbs waiting to be serviced by the
653 * retry kernel thread.
655 * The trickier parts in this code have to do with
656 * ensuring that only one retry instance is in progress
657 * for a given iocb at any time. Providing that guarantee
658 * simplifies the coding of individual aio operations as
659 * it avoids various potential races.
661 static ssize_t aio_run_iocb(struct kiocb *iocb)
663 struct kioctx *ctx = iocb->ki_ctx;
664 ssize_t (*retry)(struct kiocb *);
667 if (!(retry = iocb->ki_retry)) {
668 printk("aio_run_iocb: iocb->ki_retry = NULL\n");
673 * We don't want the next retry iteration for this
674 * operation to start until this one has returned and
675 * updated the iocb state. However, wait_queue functions
676 * can trigger a kick_iocb from interrupt context in the
677 * meantime, indicating that data is available for the next
678 * iteration. We want to remember that and enable the
679 * next retry iteration _after_ we are through with
682 * So, in order to be able to register a "kick", but
683 * prevent it from being queued now, we clear the kick
684 * flag, but make the kick code *think* that the iocb is
685 * still on the run list until we are actually done.
686 * When we are done with this iteration, we check if
687 * the iocb was kicked in the meantime and if so, queue
691 kiocbClearKicked(iocb);
694 * This is so that aio_complete knows it doesn't need to
695 * pull the iocb off the run list (We can't just call
696 * INIT_LIST_HEAD because we don't want a kick_iocb to
697 * queue this on the run list yet)
699 iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL;
700 spin_unlock_irq(&ctx->ctx_lock);
702 /* Quit retrying if the i/o has been cancelled */
703 if (kiocbIsCancelled(iocb)) {
705 aio_complete(iocb, ret, 0);
706 /* must not access the iocb after this */
711 * Now we are all set to call the retry method in async
716 if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) {
717 BUG_ON(!list_empty(&iocb->ki_wait.task_list));
718 aio_complete(iocb, ret, 0);
721 spin_lock_irq(&ctx->ctx_lock);
723 if (-EIOCBRETRY == ret) {
725 * OK, now that we are done with this iteration
726 * and know that there is more left to go,
727 * this is where we let go so that a subsequent
728 * "kick" can start the next iteration
731 /* will make __queue_kicked_iocb succeed from here on */
732 INIT_LIST_HEAD(&iocb->ki_run_list);
733 /* we must queue the next iteration ourselves, if it
734 * has already been kicked */
735 if (kiocbIsKicked(iocb)) {
736 __queue_kicked_iocb(iocb);
739 * __queue_kicked_iocb will always return 1 here, because
740 * iocb->ki_run_list is empty at this point so it should
741 * be safe to unconditionally queue the context into the
752 * Process all pending retries queued on the ioctx
754 * Assumes it is operating within the aio issuer's mm
757 static int __aio_run_iocbs(struct kioctx *ctx)
760 struct list_head run_list;
762 assert_spin_locked(&ctx->ctx_lock);
764 list_replace_init(&ctx->run_list, &run_list);
765 while (!list_empty(&run_list)) {
766 iocb = list_entry(run_list.next, struct kiocb,
768 list_del(&iocb->ki_run_list);
770 * Hold an extra reference while retrying i/o.
772 iocb->ki_users++; /* grab extra reference */
774 __aio_put_req(ctx, iocb);
776 if (!list_empty(&ctx->run_list))
781 static void aio_queue_work(struct kioctx * ctx)
783 unsigned long timeout;
785 * if someone is waiting, get the work started right
786 * away, otherwise, use a longer delay
789 if (waitqueue_active(&ctx->wait))
793 queue_delayed_work(aio_wq, &ctx->wq, timeout);
799 * Process all pending retries queued on the ioctx
801 * Assumes it is operating within the aio issuer's mm
804 static inline void aio_run_iocbs(struct kioctx *ctx)
808 spin_lock_irq(&ctx->ctx_lock);
810 requeue = __aio_run_iocbs(ctx);
811 spin_unlock_irq(&ctx->ctx_lock);
817 * just like aio_run_iocbs, but keeps running them until
818 * the list stays empty
820 static inline void aio_run_all_iocbs(struct kioctx *ctx)
822 spin_lock_irq(&ctx->ctx_lock);
823 while (__aio_run_iocbs(ctx))
825 spin_unlock_irq(&ctx->ctx_lock);
830 * Work queue handler triggered to process pending
831 * retries on an ioctx. Takes on the aio issuer's
832 * mm context before running the iocbs, so that
833 * copy_xxx_user operates on the issuer's address
835 * Run on aiod's context.
837 static void aio_kick_handler(struct work_struct *work)
839 struct kioctx *ctx = container_of(work, struct kioctx, wq.work);
840 mm_segment_t oldfs = get_fs();
841 struct mm_struct *mm;
846 spin_lock_irq(&ctx->ctx_lock);
847 requeue =__aio_run_iocbs(ctx);
849 spin_unlock_irq(&ctx->ctx_lock);
853 * we're in a worker thread already, don't use queue_delayed_work,
856 queue_delayed_work(aio_wq, &ctx->wq, 0);
861 * Called by kick_iocb to queue the kiocb for retry
862 * and if required activate the aio work queue to process
865 static void try_queue_kicked_iocb(struct kiocb *iocb)
867 struct kioctx *ctx = iocb->ki_ctx;
871 /* We're supposed to be the only path putting the iocb back on the run
872 * list. If we find that the iocb is *back* on a wait queue already
873 * than retry has happened before we could queue the iocb. This also
874 * means that the retry could have completed and freed our iocb, no
876 BUG_ON((!list_empty(&iocb->ki_wait.task_list)));
878 spin_lock_irqsave(&ctx->ctx_lock, flags);
879 /* set this inside the lock so that we can't race with aio_run_iocb()
880 * testing it and putting the iocb on the run list under the lock */
881 if (!kiocbTryKick(iocb))
882 run = __queue_kicked_iocb(iocb);
883 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
890 * Called typically from a wait queue callback context
891 * (aio_wake_function) to trigger a retry of the iocb.
892 * The retry is usually executed by aio workqueue
893 * threads (See aio_kick_handler).
895 void kick_iocb(struct kiocb *iocb)
897 /* sync iocbs are easy: they can only ever be executing from a
899 if (is_sync_kiocb(iocb)) {
900 kiocbSetKicked(iocb);
901 wake_up_process(iocb->ki_obj.tsk);
905 try_queue_kicked_iocb(iocb);
907 EXPORT_SYMBOL(kick_iocb);
910 * Called when the io request on the given iocb is complete.
911 * Returns true if this is the last user of the request. The
912 * only other user of the request can be the cancellation code.
914 int aio_complete(struct kiocb *iocb, long res, long res2)
916 struct kioctx *ctx = iocb->ki_ctx;
917 struct aio_ring_info *info;
918 struct aio_ring *ring;
919 struct io_event *event;
925 * Special case handling for sync iocbs:
926 * - events go directly into the iocb for fast handling
927 * - the sync task with the iocb in its stack holds the single iocb
928 * ref, no other paths have a way to get another ref
929 * - the sync task helpfully left a reference to itself in the iocb
931 if (is_sync_kiocb(iocb)) {
932 BUG_ON(iocb->ki_users != 1);
933 iocb->ki_user_data = res;
935 wake_up_process(iocb->ki_obj.tsk);
939 info = &ctx->ring_info;
941 /* add a completion event to the ring buffer.
942 * must be done holding ctx->ctx_lock to prevent
943 * other code from messing with the tail
944 * pointer since we might be called from irq
947 spin_lock_irqsave(&ctx->ctx_lock, flags);
949 if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list))
950 list_del_init(&iocb->ki_run_list);
953 * cancelled requests don't get events, userland was given one
954 * when the event got cancelled.
956 if (kiocbIsCancelled(iocb))
959 ring = kmap_atomic(info->ring_pages[0], KM_IRQ1);
962 event = aio_ring_event(info, tail, KM_IRQ0);
963 if (++tail >= info->nr)
966 event->obj = (u64)(unsigned long)iocb->ki_obj.user;
967 event->data = iocb->ki_user_data;
971 dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n",
972 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
975 /* after flagging the request as done, we
976 * must never even look at it again
978 smp_wmb(); /* make event visible before updating tail */
983 put_aio_ring_event(event, KM_IRQ0);
984 kunmap_atomic(ring, KM_IRQ1);
986 pr_debug("added to ring %p at [%lu]\n", iocb, tail);
989 * Check if the user asked us to deliver the result through an
990 * eventfd. The eventfd_signal() function is safe to be called
993 if (!IS_ERR(iocb->ki_eventfd))
994 eventfd_signal(iocb->ki_eventfd, 1);
997 /* everything turned out well, dispose of the aiocb. */
998 ret = __aio_put_req(ctx, iocb);
1001 * We have to order our ring_info tail store above and test
1002 * of the wait list below outside the wait lock. This is
1003 * like in wake_up_bit() where clearing a bit has to be
1004 * ordered with the unlocked test.
1008 if (waitqueue_active(&ctx->wait))
1009 wake_up(&ctx->wait);
1011 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
1016 * Pull an event off of the ioctx's event ring. Returns the number of
1017 * events fetched (0 or 1 ;-)
1018 * FIXME: make this use cmpxchg.
1019 * TODO: make the ringbuffer user mmap()able (requires FIXME).
1021 static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent)
1023 struct aio_ring_info *info = &ioctx->ring_info;
1024 struct aio_ring *ring;
1028 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
1029 dprintk("in aio_read_evt h%lu t%lu m%lu\n",
1030 (unsigned long)ring->head, (unsigned long)ring->tail,
1031 (unsigned long)ring->nr);
1033 if (ring->head == ring->tail)
1036 spin_lock(&info->ring_lock);
1038 head = ring->head % info->nr;
1039 if (head != ring->tail) {
1040 struct io_event *evp = aio_ring_event(info, head, KM_USER1);
1042 head = (head + 1) % info->nr;
1043 smp_mb(); /* finish reading the event before updatng the head */
1046 put_aio_ring_event(evp, KM_USER1);
1048 spin_unlock(&info->ring_lock);
1051 kunmap_atomic(ring, KM_USER0);
1052 dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret,
1053 (unsigned long)ring->head, (unsigned long)ring->tail);
1057 struct aio_timeout {
1058 struct timer_list timer;
1060 struct task_struct *p;
1063 static void timeout_func(unsigned long data)
1065 struct aio_timeout *to = (struct aio_timeout *)data;
1068 wake_up_process(to->p);
1071 static inline void init_timeout(struct aio_timeout *to)
1073 init_timer(&to->timer);
1074 to->timer.data = (unsigned long)to;
1075 to->timer.function = timeout_func;
1080 static inline void set_timeout(long start_jiffies, struct aio_timeout *to,
1081 const struct timespec *ts)
1083 to->timer.expires = start_jiffies + timespec_to_jiffies(ts);
1084 if (time_after(to->timer.expires, jiffies))
1085 add_timer(&to->timer);
1090 static inline void clear_timeout(struct aio_timeout *to)
1092 del_singleshot_timer_sync(&to->timer);
1095 static int read_events(struct kioctx *ctx,
1096 long min_nr, long nr,
1097 struct io_event __user *event,
1098 struct timespec __user *timeout)
1100 long start_jiffies = jiffies;
1101 struct task_struct *tsk = current;
1102 DECLARE_WAITQUEUE(wait, tsk);
1105 struct io_event ent;
1106 struct aio_timeout to;
1109 /* needed to zero any padding within an entry (there shouldn't be
1110 * any, but C is fun!
1112 memset(&ent, 0, sizeof(ent));
1115 while (likely(i < nr)) {
1116 ret = aio_read_evt(ctx, &ent);
1117 if (unlikely(ret <= 0))
1120 dprintk("read event: %Lx %Lx %Lx %Lx\n",
1121 ent.data, ent.obj, ent.res, ent.res2);
1123 /* Could we split the check in two? */
1125 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1126 dprintk("aio: lost an event due to EFAULT.\n");
1131 /* Good, event copied to userland, update counts. */
1143 /* racey check, but it gets redone */
1144 if (!retry && unlikely(!list_empty(&ctx->run_list))) {
1146 aio_run_all_iocbs(ctx);
1154 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1157 set_timeout(start_jiffies, &to, &ts);
1160 while (likely(i < nr)) {
1161 add_wait_queue_exclusive(&ctx->wait, &wait);
1163 set_task_state(tsk, TASK_INTERRUPTIBLE);
1164 ret = aio_read_evt(ctx, &ent);
1169 if (unlikely(ctx->dead)) {
1173 if (to.timed_out) /* Only check after read evt */
1175 /* Try to only show up in io wait if there are ops
1177 if (ctx->reqs_active)
1181 if (signal_pending(tsk)) {
1185 /*ret = aio_read_evt(ctx, &ent);*/
1188 set_task_state(tsk, TASK_RUNNING);
1189 remove_wait_queue(&ctx->wait, &wait);
1191 if (unlikely(ret <= 0))
1195 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1196 dprintk("aio: lost an event due to EFAULT.\n");
1200 /* Good, event copied to userland, update counts. */
1211 /* Take an ioctx and remove it from the list of ioctx's. Protects
1212 * against races with itself via ->dead.
1214 static void io_destroy(struct kioctx *ioctx)
1216 struct mm_struct *mm = current->mm;
1217 struct kioctx **tmp;
1220 /* delete the entry from the list is someone else hasn't already */
1221 write_lock(&mm->ioctx_list_lock);
1222 was_dead = ioctx->dead;
1224 for (tmp = &mm->ioctx_list; *tmp && *tmp != ioctx;
1225 tmp = &(*tmp)->next)
1229 write_unlock(&mm->ioctx_list_lock);
1231 dprintk("aio_release(%p)\n", ioctx);
1232 if (likely(!was_dead))
1233 put_ioctx(ioctx); /* twice for the list */
1235 aio_cancel_all(ioctx);
1236 wait_for_all_aios(ioctx);
1239 * Wake up any waiters. The setting of ctx->dead must be seen
1240 * by other CPUs at this point. Right now, we rely on the
1241 * locking done by the above calls to ensure this consistency.
1243 wake_up(&ioctx->wait);
1244 put_ioctx(ioctx); /* once for the lookup */
1248 * Create an aio_context capable of receiving at least nr_events.
1249 * ctxp must not point to an aio_context that already exists, and
1250 * must be initialized to 0 prior to the call. On successful
1251 * creation of the aio_context, *ctxp is filled in with the resulting
1252 * handle. May fail with -EINVAL if *ctxp is not initialized,
1253 * if the specified nr_events exceeds internal limits. May fail
1254 * with -EAGAIN if the specified nr_events exceeds the user's limit
1255 * of available events. May fail with -ENOMEM if insufficient kernel
1256 * resources are available. May fail with -EFAULT if an invalid
1257 * pointer is passed for ctxp. Will fail with -ENOSYS if not
1260 asmlinkage long sys_io_setup(unsigned nr_events, aio_context_t __user *ctxp)
1262 struct kioctx *ioctx = NULL;
1266 ret = get_user(ctx, ctxp);
1271 if (unlikely(ctx || nr_events == 0)) {
1272 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
1277 ioctx = ioctx_alloc(nr_events);
1278 ret = PTR_ERR(ioctx);
1279 if (!IS_ERR(ioctx)) {
1280 ret = put_user(ioctx->user_id, ctxp);
1284 get_ioctx(ioctx); /* io_destroy() expects us to hold a ref */
1293 * Destroy the aio_context specified. May cancel any outstanding
1294 * AIOs and block on completion. Will fail with -ENOSYS if not
1295 * implemented. May fail with -EFAULT if the context pointed to
1298 asmlinkage long sys_io_destroy(aio_context_t ctx)
1300 struct kioctx *ioctx = lookup_ioctx(ctx);
1301 if (likely(NULL != ioctx)) {
1305 pr_debug("EINVAL: io_destroy: invalid context id\n");
1309 static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
1311 struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];
1315 while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
1316 ssize_t this = min((ssize_t)iov->iov_len, ret);
1317 iov->iov_base += this;
1318 iov->iov_len -= this;
1319 iocb->ki_left -= this;
1321 if (iov->iov_len == 0) {
1327 /* the caller should not have done more io than what fit in
1328 * the remaining iovecs */
1329 BUG_ON(ret > 0 && iocb->ki_left == 0);
1332 static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
1334 struct file *file = iocb->ki_filp;
1335 struct address_space *mapping = file->f_mapping;
1336 struct inode *inode = mapping->host;
1337 ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
1338 unsigned long, loff_t);
1340 unsigned short opcode;
1342 if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
1343 (iocb->ki_opcode == IOCB_CMD_PREAD)) {
1344 rw_op = file->f_op->aio_read;
1345 opcode = IOCB_CMD_PREADV;
1347 rw_op = file->f_op->aio_write;
1348 opcode = IOCB_CMD_PWRITEV;
1351 /* This matches the pread()/pwrite() logic */
1352 if (iocb->ki_pos < 0)
1356 ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
1357 iocb->ki_nr_segs - iocb->ki_cur_seg,
1360 aio_advance_iovec(iocb, ret);
1362 /* retry all partial writes. retry partial reads as long as its a
1364 } while (ret > 0 && iocb->ki_left > 0 &&
1365 (opcode == IOCB_CMD_PWRITEV ||
1366 (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
1368 /* This means we must have transferred all that we could */
1369 /* No need to retry anymore */
1370 if ((ret == 0) || (iocb->ki_left == 0))
1371 ret = iocb->ki_nbytes - iocb->ki_left;
1373 /* If we managed to write some out we return that, rather than
1374 * the eventual error. */
1375 if (opcode == IOCB_CMD_PWRITEV
1376 && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY
1377 && iocb->ki_nbytes - iocb->ki_left)
1378 ret = iocb->ki_nbytes - iocb->ki_left;
1383 static ssize_t aio_fdsync(struct kiocb *iocb)
1385 struct file *file = iocb->ki_filp;
1386 ssize_t ret = -EINVAL;
1388 if (file->f_op->aio_fsync)
1389 ret = file->f_op->aio_fsync(iocb, 1);
1393 static ssize_t aio_fsync(struct kiocb *iocb)
1395 struct file *file = iocb->ki_filp;
1396 ssize_t ret = -EINVAL;
1398 if (file->f_op->aio_fsync)
1399 ret = file->f_op->aio_fsync(iocb, 0);
1403 static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb)
1407 ret = rw_copy_check_uvector(type, (struct iovec __user *)kiocb->ki_buf,
1408 kiocb->ki_nbytes, 1,
1409 &kiocb->ki_inline_vec, &kiocb->ki_iovec);
1413 kiocb->ki_nr_segs = kiocb->ki_nbytes;
1414 kiocb->ki_cur_seg = 0;
1415 /* ki_nbytes/left now reflect bytes instead of segs */
1416 kiocb->ki_nbytes = ret;
1417 kiocb->ki_left = ret;
1424 static ssize_t aio_setup_single_vector(struct kiocb *kiocb)
1426 kiocb->ki_iovec = &kiocb->ki_inline_vec;
1427 kiocb->ki_iovec->iov_base = kiocb->ki_buf;
1428 kiocb->ki_iovec->iov_len = kiocb->ki_left;
1429 kiocb->ki_nr_segs = 1;
1430 kiocb->ki_cur_seg = 0;
1436 * Performs the initial checks and aio retry method
1437 * setup for the kiocb at the time of io submission.
1439 static ssize_t aio_setup_iocb(struct kiocb *kiocb)
1441 struct file *file = kiocb->ki_filp;
1444 switch (kiocb->ki_opcode) {
1445 case IOCB_CMD_PREAD:
1447 if (unlikely(!(file->f_mode & FMODE_READ)))
1450 if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
1453 ret = security_file_permission(file, MAY_READ);
1456 ret = aio_setup_single_vector(kiocb);
1460 if (file->f_op->aio_read)
1461 kiocb->ki_retry = aio_rw_vect_retry;
1463 case IOCB_CMD_PWRITE:
1465 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1468 if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
1471 ret = security_file_permission(file, MAY_WRITE);
1474 ret = aio_setup_single_vector(kiocb);
1478 if (file->f_op->aio_write)
1479 kiocb->ki_retry = aio_rw_vect_retry;
1481 case IOCB_CMD_PREADV:
1483 if (unlikely(!(file->f_mode & FMODE_READ)))
1485 ret = security_file_permission(file, MAY_READ);
1488 ret = aio_setup_vectored_rw(READ, kiocb);
1492 if (file->f_op->aio_read)
1493 kiocb->ki_retry = aio_rw_vect_retry;
1495 case IOCB_CMD_PWRITEV:
1497 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1499 ret = security_file_permission(file, MAY_WRITE);
1502 ret = aio_setup_vectored_rw(WRITE, kiocb);
1506 if (file->f_op->aio_write)
1507 kiocb->ki_retry = aio_rw_vect_retry;
1509 case IOCB_CMD_FDSYNC:
1511 if (file->f_op->aio_fsync)
1512 kiocb->ki_retry = aio_fdsync;
1514 case IOCB_CMD_FSYNC:
1516 if (file->f_op->aio_fsync)
1517 kiocb->ki_retry = aio_fsync;
1520 dprintk("EINVAL: io_submit: no operation provided\n");
1524 if (!kiocb->ki_retry)
1531 * aio_wake_function:
1532 * wait queue callback function for aio notification,
1533 * Simply triggers a retry of the operation via kick_iocb.
1535 * This callback is specified in the wait queue entry in
1539 * This routine is executed with the wait queue lock held.
1540 * Since kick_iocb acquires iocb->ctx->ctx_lock, it nests
1541 * the ioctx lock inside the wait queue lock. This is safe
1542 * because this callback isn't used for wait queues which
1543 * are nested inside ioctx lock (i.e. ctx->wait)
1545 static int aio_wake_function(wait_queue_t *wait, unsigned mode,
1546 int sync, void *key)
1548 struct kiocb *iocb = container_of(wait, struct kiocb, ki_wait);
1550 list_del_init(&wait->task_list);
1555 int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
1562 /* enforce forwards compatibility on users */
1563 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
1564 pr_debug("EINVAL: io_submit: reserve field set\n");
1568 /* prevent overflows */
1570 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1571 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1572 ((ssize_t)iocb->aio_nbytes < 0)
1574 pr_debug("EINVAL: io_submit: overflow check\n");
1578 file = fget(iocb->aio_fildes);
1579 if (unlikely(!file))
1582 req = aio_get_req(ctx); /* returns with 2 references to req */
1583 if (unlikely(!req)) {
1587 req->ki_filp = file;
1588 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1590 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
1591 * instance of the file* now. The file descriptor must be
1592 * an eventfd() fd, and will be signaled for each completed
1593 * event using the eventfd_signal() function.
1595 req->ki_eventfd = eventfd_fget((int) iocb->aio_resfd);
1596 if (unlikely(IS_ERR(req->ki_eventfd))) {
1597 ret = PTR_ERR(req->ki_eventfd);
1602 ret = put_user(req->ki_key, &user_iocb->aio_key);
1603 if (unlikely(ret)) {
1604 dprintk("EFAULT: aio_key\n");
1608 req->ki_obj.user = user_iocb;
1609 req->ki_user_data = iocb->aio_data;
1610 req->ki_pos = iocb->aio_offset;
1612 req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf;
1613 req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
1614 req->ki_opcode = iocb->aio_lio_opcode;
1615 init_waitqueue_func_entry(&req->ki_wait, aio_wake_function);
1616 INIT_LIST_HEAD(&req->ki_wait.task_list);
1618 ret = aio_setup_iocb(req);
1623 spin_lock_irq(&ctx->ctx_lock);
1625 if (!list_empty(&ctx->run_list)) {
1626 /* drain the run list */
1627 while (__aio_run_iocbs(ctx))
1630 spin_unlock_irq(&ctx->ctx_lock);
1631 aio_put_req(req); /* drop extra ref to req */
1635 aio_put_req(req); /* drop extra ref to req */
1636 aio_put_req(req); /* drop i/o ref to req */
1641 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1642 * the number of iocbs queued. May return -EINVAL if the aio_context
1643 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1644 * *iocbpp[0] is not properly initialized, if the operation specified
1645 * is invalid for the file descriptor in the iocb. May fail with
1646 * -EFAULT if any of the data structures point to invalid data. May
1647 * fail with -EBADF if the file descriptor specified in the first
1648 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1649 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1650 * fail with -ENOSYS if not implemented.
1652 asmlinkage long sys_io_submit(aio_context_t ctx_id, long nr,
1653 struct iocb __user * __user *iocbpp)
1659 if (unlikely(nr < 0))
1662 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1665 ctx = lookup_ioctx(ctx_id);
1666 if (unlikely(!ctx)) {
1667 pr_debug("EINVAL: io_submit: invalid context id\n");
1672 * AKPM: should this return a partial result if some of the IOs were
1673 * successfully submitted?
1675 for (i=0; i<nr; i++) {
1676 struct iocb __user *user_iocb;
1679 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1684 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1689 ret = io_submit_one(ctx, user_iocb, &tmp);
1699 * Finds a given iocb for cancellation.
1701 static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
1704 struct list_head *pos;
1706 assert_spin_locked(&ctx->ctx_lock);
1708 /* TODO: use a hash or array, this sucks. */
1709 list_for_each(pos, &ctx->active_reqs) {
1710 struct kiocb *kiocb = list_kiocb(pos);
1711 if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key)
1718 * Attempts to cancel an iocb previously passed to io_submit. If
1719 * the operation is successfully cancelled, the resulting event is
1720 * copied into the memory pointed to by result without being placed
1721 * into the completion queue and 0 is returned. May fail with
1722 * -EFAULT if any of the data structures pointed to are invalid.
1723 * May fail with -EINVAL if aio_context specified by ctx_id is
1724 * invalid. May fail with -EAGAIN if the iocb specified was not
1725 * cancelled. Will fail with -ENOSYS if not implemented.
1727 asmlinkage long sys_io_cancel(aio_context_t ctx_id, struct iocb __user *iocb,
1728 struct io_event __user *result)
1730 int (*cancel)(struct kiocb *iocb, struct io_event *res);
1732 struct kiocb *kiocb;
1736 ret = get_user(key, &iocb->aio_key);
1740 ctx = lookup_ioctx(ctx_id);
1744 spin_lock_irq(&ctx->ctx_lock);
1746 kiocb = lookup_kiocb(ctx, iocb, key);
1747 if (kiocb && kiocb->ki_cancel) {
1748 cancel = kiocb->ki_cancel;
1750 kiocbSetCancelled(kiocb);
1753 spin_unlock_irq(&ctx->ctx_lock);
1755 if (NULL != cancel) {
1756 struct io_event tmp;
1757 pr_debug("calling cancel\n");
1758 memset(&tmp, 0, sizeof(tmp));
1759 tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user;
1760 tmp.data = kiocb->ki_user_data;
1761 ret = cancel(kiocb, &tmp);
1763 /* Cancellation succeeded -- copy the result
1764 * into the user's buffer.
1766 if (copy_to_user(result, &tmp, sizeof(tmp)))
1778 * Attempts to read at least min_nr events and up to nr events from
1779 * the completion queue for the aio_context specified by ctx_id. May
1780 * fail with -EINVAL if ctx_id is invalid, if min_nr is out of range,
1781 * if nr is out of range, if when is out of range. May fail with
1782 * -EFAULT if any of the memory specified to is invalid. May return
1783 * 0 or < min_nr if no events are available and the timeout specified
1784 * by when has elapsed, where when == NULL specifies an infinite
1785 * timeout. Note that the timeout pointed to by when is relative and
1786 * will be updated if not NULL and the operation blocks. Will fail
1787 * with -ENOSYS if not implemented.
1789 asmlinkage long sys_io_getevents(aio_context_t ctx_id,
1792 struct io_event __user *events,
1793 struct timespec __user *timeout)
1795 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1798 if (likely(ioctx)) {
1799 if (likely(min_nr <= nr && min_nr >= 0 && nr >= 0))
1800 ret = read_events(ioctx, min_nr, nr, events, timeout);
1804 asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout);
1808 __initcall(aio_setup);
1810 EXPORT_SYMBOL(aio_complete);
1811 EXPORT_SYMBOL(aio_put_req);
1812 EXPORT_SYMBOL(wait_on_sync_kiocb);