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/mmu_context.h>
28 #include <linux/slab.h>
29 #include <linux/timer.h>
30 #include <linux/aio.h>
31 #include <linux/highmem.h>
32 #include <linux/workqueue.h>
33 #include <linux/security.h>
34 #include <linux/eventfd.h>
36 #include <asm/kmap_types.h>
37 #include <asm/uaccess.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));
81 __initcall(aio_setup);
83 static void aio_free_ring(struct kioctx *ctx)
85 struct aio_ring_info *info = &ctx->ring_info;
88 for (i=0; i<info->nr_pages; i++)
89 put_page(info->ring_pages[i]);
91 if (info->mmap_size) {
92 down_write(&ctx->mm->mmap_sem);
93 do_munmap(ctx->mm, info->mmap_base, info->mmap_size);
94 up_write(&ctx->mm->mmap_sem);
97 if (info->ring_pages && info->ring_pages != info->internal_pages)
98 kfree(info->ring_pages);
99 info->ring_pages = NULL;
103 static int aio_setup_ring(struct kioctx *ctx)
105 struct aio_ring *ring;
106 struct aio_ring_info *info = &ctx->ring_info;
107 unsigned nr_events = ctx->max_reqs;
111 /* Compensate for the ring buffer's head/tail overlap entry */
112 nr_events += 2; /* 1 is required, 2 for good luck */
114 size = sizeof(struct aio_ring);
115 size += sizeof(struct io_event) * nr_events;
116 nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT;
121 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event);
124 info->ring_pages = info->internal_pages;
125 if (nr_pages > AIO_RING_PAGES) {
126 info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
127 if (!info->ring_pages)
131 info->mmap_size = nr_pages * PAGE_SIZE;
132 dprintk("attempting mmap of %lu bytes\n", info->mmap_size);
133 down_write(&ctx->mm->mmap_sem);
134 info->mmap_base = do_mmap(NULL, 0, info->mmap_size,
135 PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE,
137 if (IS_ERR((void *)info->mmap_base)) {
138 up_write(&ctx->mm->mmap_sem);
144 dprintk("mmap address: 0x%08lx\n", info->mmap_base);
145 info->nr_pages = get_user_pages(current, ctx->mm,
146 info->mmap_base, nr_pages,
147 1, 0, info->ring_pages, NULL);
148 up_write(&ctx->mm->mmap_sem);
150 if (unlikely(info->nr_pages != nr_pages)) {
155 ctx->user_id = info->mmap_base;
157 info->nr = nr_events; /* trusted copy */
159 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
160 ring->nr = nr_events; /* user copy */
161 ring->id = ctx->user_id;
162 ring->head = ring->tail = 0;
163 ring->magic = AIO_RING_MAGIC;
164 ring->compat_features = AIO_RING_COMPAT_FEATURES;
165 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
166 ring->header_length = sizeof(struct aio_ring);
167 kunmap_atomic(ring, KM_USER0);
173 /* aio_ring_event: returns a pointer to the event at the given index from
174 * kmap_atomic(, km). Release the pointer with put_aio_ring_event();
176 #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
177 #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
178 #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
180 #define aio_ring_event(info, nr, km) ({ \
181 unsigned pos = (nr) + AIO_EVENTS_OFFSET; \
182 struct io_event *__event; \
183 __event = kmap_atomic( \
184 (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE], km); \
185 __event += pos % AIO_EVENTS_PER_PAGE; \
189 #define put_aio_ring_event(event, km) do { \
190 struct io_event *__event = (event); \
192 kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \
195 static void ctx_rcu_free(struct rcu_head *head)
197 struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
198 unsigned nr_events = ctx->max_reqs;
200 kmem_cache_free(kioctx_cachep, ctx);
203 spin_lock(&aio_nr_lock);
204 BUG_ON(aio_nr - nr_events > aio_nr);
206 spin_unlock(&aio_nr_lock);
211 * Called when the last user of an aio context has gone away,
212 * and the struct needs to be freed.
214 static void __put_ioctx(struct kioctx *ctx)
216 BUG_ON(ctx->reqs_active);
218 cancel_delayed_work(&ctx->wq);
219 cancel_work_sync(&ctx->wq.work);
223 pr_debug("__put_ioctx: freeing %p\n", ctx);
224 call_rcu(&ctx->rcu_head, ctx_rcu_free);
227 #define get_ioctx(kioctx) do { \
228 BUG_ON(atomic_read(&(kioctx)->users) <= 0); \
229 atomic_inc(&(kioctx)->users); \
231 #define put_ioctx(kioctx) do { \
232 BUG_ON(atomic_read(&(kioctx)->users) <= 0); \
233 if (unlikely(atomic_dec_and_test(&(kioctx)->users))) \
234 __put_ioctx(kioctx); \
238 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
240 static struct kioctx *ioctx_alloc(unsigned nr_events)
242 struct mm_struct *mm;
246 /* Prevent overflows */
247 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
248 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
249 pr_debug("ENOMEM: nr_events too high\n");
250 return ERR_PTR(-EINVAL);
253 if ((unsigned long)nr_events > aio_max_nr)
254 return ERR_PTR(-EAGAIN);
256 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
258 return ERR_PTR(-ENOMEM);
260 ctx->max_reqs = nr_events;
261 mm = ctx->mm = current->mm;
262 atomic_inc(&mm->mm_count);
264 atomic_set(&ctx->users, 1);
265 spin_lock_init(&ctx->ctx_lock);
266 spin_lock_init(&ctx->ring_info.ring_lock);
267 init_waitqueue_head(&ctx->wait);
269 INIT_LIST_HEAD(&ctx->active_reqs);
270 INIT_LIST_HEAD(&ctx->run_list);
271 INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler);
273 if (aio_setup_ring(ctx) < 0)
276 /* limit the number of system wide aios */
278 spin_lock_bh(&aio_nr_lock);
279 if (aio_nr + nr_events > aio_max_nr ||
280 aio_nr + nr_events < aio_nr)
283 aio_nr += ctx->max_reqs;
284 spin_unlock_bh(&aio_nr_lock);
285 if (ctx->max_reqs || did_sync)
288 /* wait for rcu callbacks to have completed before giving up */
291 ctx->max_reqs = nr_events;
294 if (ctx->max_reqs == 0)
297 /* now link into global list. */
298 spin_lock(&mm->ioctx_lock);
299 hlist_add_head_rcu(&ctx->list, &mm->ioctx_list);
300 spin_unlock(&mm->ioctx_lock);
302 dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
303 ctx, ctx->user_id, current->mm, ctx->ring_info.nr);
308 return ERR_PTR(-EAGAIN);
312 kmem_cache_free(kioctx_cachep, ctx);
313 ctx = ERR_PTR(-ENOMEM);
315 dprintk("aio: error allocating ioctx %p\n", ctx);
320 * Cancels all outstanding aio requests on an aio context. Used
321 * when the processes owning a context have all exited to encourage
322 * the rapid destruction of the kioctx.
324 static void aio_cancel_all(struct kioctx *ctx)
326 int (*cancel)(struct kiocb *, struct io_event *);
328 spin_lock_irq(&ctx->ctx_lock);
330 while (!list_empty(&ctx->active_reqs)) {
331 struct list_head *pos = ctx->active_reqs.next;
332 struct kiocb *iocb = list_kiocb(pos);
333 list_del_init(&iocb->ki_list);
334 cancel = iocb->ki_cancel;
335 kiocbSetCancelled(iocb);
338 spin_unlock_irq(&ctx->ctx_lock);
340 spin_lock_irq(&ctx->ctx_lock);
343 spin_unlock_irq(&ctx->ctx_lock);
346 static void wait_for_all_aios(struct kioctx *ctx)
348 struct task_struct *tsk = current;
349 DECLARE_WAITQUEUE(wait, tsk);
351 spin_lock_irq(&ctx->ctx_lock);
352 if (!ctx->reqs_active)
355 add_wait_queue(&ctx->wait, &wait);
356 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
357 while (ctx->reqs_active) {
358 spin_unlock_irq(&ctx->ctx_lock);
360 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
361 spin_lock_irq(&ctx->ctx_lock);
363 __set_task_state(tsk, TASK_RUNNING);
364 remove_wait_queue(&ctx->wait, &wait);
367 spin_unlock_irq(&ctx->ctx_lock);
370 /* wait_on_sync_kiocb:
371 * Waits on the given sync kiocb to complete.
373 ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
375 while (iocb->ki_users) {
376 set_current_state(TASK_UNINTERRUPTIBLE);
381 __set_current_state(TASK_RUNNING);
382 return iocb->ki_user_data;
384 EXPORT_SYMBOL(wait_on_sync_kiocb);
386 /* exit_aio: called when the last user of mm goes away. At this point,
387 * there is no way for any new requests to be submited or any of the
388 * io_* syscalls to be called on the context. However, there may be
389 * outstanding requests which hold references to the context; as they
390 * go away, they will call put_ioctx and release any pinned memory
391 * associated with the request (held via struct page * references).
393 void exit_aio(struct mm_struct *mm)
397 while (!hlist_empty(&mm->ioctx_list)) {
398 ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list);
399 hlist_del_rcu(&ctx->list);
403 wait_for_all_aios(ctx);
405 * Ensure we don't leave the ctx on the aio_wq
407 cancel_work_sync(&ctx->wq.work);
409 if (1 != atomic_read(&ctx->users))
411 "exit_aio:ioctx still alive: %d %d %d\n",
412 atomic_read(&ctx->users), ctx->dead,
419 * Allocate a slot for an aio request. Increments the users count
420 * of the kioctx so that the kioctx stays around until all requests are
421 * complete. Returns NULL if no requests are free.
423 * Returns with kiocb->users set to 2. The io submit code path holds
424 * an extra reference while submitting the i/o.
425 * This prevents races between the aio code path referencing the
426 * req (after submitting it) and aio_complete() freeing the req.
428 static struct kiocb *__aio_get_req(struct kioctx *ctx)
430 struct kiocb *req = NULL;
431 struct aio_ring *ring;
434 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL);
442 req->ki_cancel = NULL;
443 req->ki_retry = NULL;
446 req->ki_iovec = NULL;
447 INIT_LIST_HEAD(&req->ki_run_list);
448 req->ki_eventfd = NULL;
450 /* Check if the completion queue has enough free space to
451 * accept an event from this io.
453 spin_lock_irq(&ctx->ctx_lock);
454 ring = kmap_atomic(ctx->ring_info.ring_pages[0], KM_USER0);
455 if (ctx->reqs_active < aio_ring_avail(&ctx->ring_info, ring)) {
456 list_add(&req->ki_list, &ctx->active_reqs);
460 kunmap_atomic(ring, KM_USER0);
461 spin_unlock_irq(&ctx->ctx_lock);
464 kmem_cache_free(kiocb_cachep, req);
471 static inline struct kiocb *aio_get_req(struct kioctx *ctx)
474 /* Handle a potential starvation case -- should be exceedingly rare as
475 * requests will be stuck on fput_head only if the aio_fput_routine is
476 * delayed and the requests were the last user of the struct file.
478 req = __aio_get_req(ctx);
479 if (unlikely(NULL == req)) {
480 aio_fput_routine(NULL);
481 req = __aio_get_req(ctx);
486 static inline void really_put_req(struct kioctx *ctx, struct kiocb *req)
488 assert_spin_locked(&ctx->ctx_lock);
490 if (req->ki_eventfd != NULL)
491 eventfd_ctx_put(req->ki_eventfd);
494 if (req->ki_iovec != &req->ki_inline_vec)
495 kfree(req->ki_iovec);
496 kmem_cache_free(kiocb_cachep, req);
499 if (unlikely(!ctx->reqs_active && ctx->dead))
503 static void aio_fput_routine(struct work_struct *data)
505 spin_lock_irq(&fput_lock);
506 while (likely(!list_empty(&fput_head))) {
507 struct kiocb *req = list_kiocb(fput_head.next);
508 struct kioctx *ctx = req->ki_ctx;
510 list_del(&req->ki_list);
511 spin_unlock_irq(&fput_lock);
513 /* Complete the fput(s) */
514 if (req->ki_filp != NULL)
515 __fput(req->ki_filp);
517 /* Link the iocb into the context's free list */
518 spin_lock_irq(&ctx->ctx_lock);
519 really_put_req(ctx, req);
520 spin_unlock_irq(&ctx->ctx_lock);
523 spin_lock_irq(&fput_lock);
525 spin_unlock_irq(&fput_lock);
529 * Returns true if this put was the last user of the request.
531 static int __aio_put_req(struct kioctx *ctx, struct kiocb *req)
533 dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n",
534 req, atomic_long_read(&req->ki_filp->f_count));
536 assert_spin_locked(&ctx->ctx_lock);
539 BUG_ON(req->ki_users < 0);
540 if (likely(req->ki_users))
542 list_del(&req->ki_list); /* remove from active_reqs */
543 req->ki_cancel = NULL;
544 req->ki_retry = NULL;
547 * Try to optimize the aio and eventfd file* puts, by avoiding to
548 * schedule work in case it is not __fput() time. In normal cases,
549 * we would not be holding the last reference to the file*, so
550 * this function will be executed w/out any aio kthread wakeup.
552 if (unlikely(atomic_long_dec_and_test(&req->ki_filp->f_count))) {
554 spin_lock(&fput_lock);
555 list_add(&req->ki_list, &fput_head);
556 spin_unlock(&fput_lock);
557 queue_work(aio_wq, &fput_work);
560 really_put_req(ctx, req);
566 * Returns true if this put was the last user of the kiocb,
567 * false if the request is still in use.
569 int aio_put_req(struct kiocb *req)
571 struct kioctx *ctx = req->ki_ctx;
573 spin_lock_irq(&ctx->ctx_lock);
574 ret = __aio_put_req(ctx, req);
575 spin_unlock_irq(&ctx->ctx_lock);
578 EXPORT_SYMBOL(aio_put_req);
580 static struct kioctx *lookup_ioctx(unsigned long ctx_id)
582 struct mm_struct *mm = current->mm;
583 struct kioctx *ctx, *ret = NULL;
584 struct hlist_node *n;
588 hlist_for_each_entry_rcu(ctx, n, &mm->ioctx_list, list) {
589 if (ctx->user_id == ctx_id && !ctx->dead) {
601 * Queue up a kiocb to be retried. Assumes that the kiocb
602 * has already been marked as kicked, and places it on
603 * the retry run list for the corresponding ioctx, if it
604 * isn't already queued. Returns 1 if it actually queued
605 * the kiocb (to tell the caller to activate the work
606 * queue to process it), or 0, if it found that it was
609 static inline int __queue_kicked_iocb(struct kiocb *iocb)
611 struct kioctx *ctx = iocb->ki_ctx;
613 assert_spin_locked(&ctx->ctx_lock);
615 if (list_empty(&iocb->ki_run_list)) {
616 list_add_tail(&iocb->ki_run_list,
624 * This is the core aio execution routine. It is
625 * invoked both for initial i/o submission and
626 * subsequent retries via the aio_kick_handler.
627 * Expects to be invoked with iocb->ki_ctx->lock
628 * already held. The lock is released and reacquired
629 * as needed during processing.
631 * Calls the iocb retry method (already setup for the
632 * iocb on initial submission) for operation specific
633 * handling, but takes care of most of common retry
634 * execution details for a given iocb. The retry method
635 * needs to be non-blocking as far as possible, to avoid
636 * holding up other iocbs waiting to be serviced by the
637 * retry kernel thread.
639 * The trickier parts in this code have to do with
640 * ensuring that only one retry instance is in progress
641 * for a given iocb at any time. Providing that guarantee
642 * simplifies the coding of individual aio operations as
643 * it avoids various potential races.
645 static ssize_t aio_run_iocb(struct kiocb *iocb)
647 struct kioctx *ctx = iocb->ki_ctx;
648 ssize_t (*retry)(struct kiocb *);
651 if (!(retry = iocb->ki_retry)) {
652 printk("aio_run_iocb: iocb->ki_retry = NULL\n");
657 * We don't want the next retry iteration for this
658 * operation to start until this one has returned and
659 * updated the iocb state. However, wait_queue functions
660 * can trigger a kick_iocb from interrupt context in the
661 * meantime, indicating that data is available for the next
662 * iteration. We want to remember that and enable the
663 * next retry iteration _after_ we are through with
666 * So, in order to be able to register a "kick", but
667 * prevent it from being queued now, we clear the kick
668 * flag, but make the kick code *think* that the iocb is
669 * still on the run list until we are actually done.
670 * When we are done with this iteration, we check if
671 * the iocb was kicked in the meantime and if so, queue
675 kiocbClearKicked(iocb);
678 * This is so that aio_complete knows it doesn't need to
679 * pull the iocb off the run list (We can't just call
680 * INIT_LIST_HEAD because we don't want a kick_iocb to
681 * queue this on the run list yet)
683 iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL;
684 spin_unlock_irq(&ctx->ctx_lock);
686 /* Quit retrying if the i/o has been cancelled */
687 if (kiocbIsCancelled(iocb)) {
689 aio_complete(iocb, ret, 0);
690 /* must not access the iocb after this */
695 * Now we are all set to call the retry method in async
698 BUG_ON(current->io_wait != NULL);
699 current->io_wait = &iocb->ki_wq;
701 current->io_wait = NULL;
703 if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) {
704 BUG_ON(!list_empty(&iocb->ki_wq.wait.task_list));
705 aio_complete(iocb, ret, 0);
708 spin_lock_irq(&ctx->ctx_lock);
710 if (-EIOCBRETRY == ret) {
712 * OK, now that we are done with this iteration
713 * and know that there is more left to go,
714 * this is where we let go so that a subsequent
715 * "kick" can start the next iteration
717 iocb->ki_retry = retry;
719 /* will make __queue_kicked_iocb succeed from here on */
720 INIT_LIST_HEAD(&iocb->ki_run_list);
721 /* we must queue the next iteration ourselves, if it
722 * has already been kicked */
723 if (kiocbIsKicked(iocb)) {
724 __queue_kicked_iocb(iocb);
727 * __queue_kicked_iocb will always return 1 here, because
728 * iocb->ki_run_list is empty at this point so it should
729 * be safe to unconditionally queue the context into the
740 * Process all pending retries queued on the ioctx
742 * Assumes it is operating within the aio issuer's mm
745 static int __aio_run_iocbs(struct kioctx *ctx)
748 struct list_head run_list;
750 assert_spin_locked(&ctx->ctx_lock);
752 list_replace_init(&ctx->run_list, &run_list);
753 while (!list_empty(&run_list)) {
754 iocb = list_entry(run_list.next, struct kiocb,
756 list_del(&iocb->ki_run_list);
758 * Hold an extra reference while retrying i/o.
760 iocb->ki_users++; /* grab extra reference */
762 __aio_put_req(ctx, iocb);
764 if (!list_empty(&ctx->run_list))
769 static void aio_queue_work(struct kioctx * ctx)
771 unsigned long timeout;
773 * if someone is waiting, get the work started right
774 * away, otherwise, use a longer delay
777 if (waitqueue_active(&ctx->wait))
781 queue_delayed_work(aio_wq, &ctx->wq, timeout);
787 * Process all pending retries queued on the ioctx
789 * Assumes it is operating within the aio issuer's mm
792 static inline void aio_run_iocbs(struct kioctx *ctx)
796 spin_lock_irq(&ctx->ctx_lock);
798 requeue = __aio_run_iocbs(ctx);
799 spin_unlock_irq(&ctx->ctx_lock);
805 * just like aio_run_iocbs, but keeps running them until
806 * the list stays empty
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, don't use queue_delayed_work,
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 /* We're supposed to be the only path putting the iocb back on the run
860 * list. If we find that the iocb is *back* on a wait queue already
861 * than retry has happened before we could queue the iocb. This also
862 * means that the retry could have completed and freed our iocb, no
864 BUG_ON((!list_empty(&iocb->ki_wq.wait.task_list)));
866 spin_lock_irqsave(&ctx->ctx_lock, flags);
867 /* set this inside the lock so that we can't race with aio_run_iocb()
868 * testing it and putting the iocb on the run list under the lock */
869 if (!kiocbTryKick(iocb))
870 run = __queue_kicked_iocb(iocb);
871 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
878 * Called typically from a wait queue callback context
879 * (aio_wake_function) to trigger a retry of the iocb.
880 * The retry is usually executed by aio workqueue
881 * threads (See aio_kick_handler).
883 void kick_iocb(struct kiocb *iocb)
885 /* sync iocbs are easy: they can only ever be executing from a
887 if (is_sync_kiocb(iocb)) {
888 kiocbSetKicked(iocb);
889 wake_up_process(iocb->ki_obj.tsk);
893 try_queue_kicked_iocb(iocb);
895 EXPORT_SYMBOL(kick_iocb);
898 * Called when the io request on the given iocb is complete.
899 * Returns true if this is the last user of the request. The
900 * only other user of the request can be the cancellation code.
902 int aio_complete(struct kiocb *iocb, long res, long res2)
904 struct kioctx *ctx = iocb->ki_ctx;
905 struct aio_ring_info *info;
906 struct aio_ring *ring;
907 struct io_event *event;
913 * Special case handling for sync iocbs:
914 * - events go directly into the iocb for fast handling
915 * - the sync task with the iocb in its stack holds the single iocb
916 * ref, no other paths have a way to get another ref
917 * - the sync task helpfully left a reference to itself in the iocb
919 if (is_sync_kiocb(iocb)) {
920 BUG_ON(iocb->ki_users != 1);
921 iocb->ki_user_data = res;
923 wake_up_process(iocb->ki_obj.tsk);
927 info = &ctx->ring_info;
929 /* add a completion event to the ring buffer.
930 * must be done holding ctx->ctx_lock to prevent
931 * other code from messing with the tail
932 * pointer since we might be called from irq
935 spin_lock_irqsave(&ctx->ctx_lock, flags);
937 if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list))
938 list_del_init(&iocb->ki_run_list);
941 * cancelled requests don't get events, userland was given one
942 * when the event got cancelled.
944 if (kiocbIsCancelled(iocb))
947 ring = kmap_atomic(info->ring_pages[0], KM_IRQ1);
950 event = aio_ring_event(info, tail, KM_IRQ0);
951 if (++tail >= info->nr)
954 event->obj = (u64)(unsigned long)iocb->ki_obj.user;
955 event->data = iocb->ki_user_data;
959 dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n",
960 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
963 /* after flagging the request as done, we
964 * must never even look at it again
966 smp_wmb(); /* make event visible before updating tail */
971 put_aio_ring_event(event, KM_IRQ0);
972 kunmap_atomic(ring, KM_IRQ1);
974 pr_debug("added to ring %p at [%lu]\n", iocb, tail);
977 * Check if the user asked us to deliver the result through an
978 * eventfd. The eventfd_signal() function is safe to be called
981 if (iocb->ki_eventfd != NULL)
982 eventfd_signal(iocb->ki_eventfd, 1);
985 /* everything turned out well, dispose of the aiocb. */
986 ret = __aio_put_req(ctx, iocb);
989 * We have to order our ring_info tail store above and test
990 * of the wait list below outside the wait lock. This is
991 * like in wake_up_bit() where clearing a bit has to be
992 * ordered with the unlocked test.
996 if (waitqueue_active(&ctx->wait))
999 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
1002 EXPORT_SYMBOL(aio_complete);
1005 * Pull an event off of the ioctx's event ring. Returns the number of
1006 * events fetched (0 or 1 ;-)
1007 * FIXME: make this use cmpxchg.
1008 * TODO: make the ringbuffer user mmap()able (requires FIXME).
1010 static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent)
1012 struct aio_ring_info *info = &ioctx->ring_info;
1013 struct aio_ring *ring;
1017 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
1018 dprintk("in aio_read_evt h%lu t%lu m%lu\n",
1019 (unsigned long)ring->head, (unsigned long)ring->tail,
1020 (unsigned long)ring->nr);
1022 if (ring->head == ring->tail)
1025 spin_lock(&info->ring_lock);
1027 head = ring->head % info->nr;
1028 if (head != ring->tail) {
1029 struct io_event *evp = aio_ring_event(info, head, KM_USER1);
1031 head = (head + 1) % info->nr;
1032 smp_mb(); /* finish reading the event before updatng the head */
1035 put_aio_ring_event(evp, KM_USER1);
1037 spin_unlock(&info->ring_lock);
1040 kunmap_atomic(ring, KM_USER0);
1041 dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret,
1042 (unsigned long)ring->head, (unsigned long)ring->tail);
1046 struct aio_timeout {
1047 struct timer_list timer;
1049 struct task_struct *p;
1052 static void timeout_func(unsigned long data)
1054 struct aio_timeout *to = (struct aio_timeout *)data;
1057 wake_up_process(to->p);
1060 static inline void init_timeout(struct aio_timeout *to)
1062 setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to);
1067 static inline void set_timeout(long start_jiffies, struct aio_timeout *to,
1068 const struct timespec *ts)
1070 to->timer.expires = start_jiffies + timespec_to_jiffies(ts);
1071 if (time_after(to->timer.expires, jiffies))
1072 add_timer(&to->timer);
1077 static inline void clear_timeout(struct aio_timeout *to)
1079 del_singleshot_timer_sync(&to->timer);
1082 static int read_events(struct kioctx *ctx,
1083 long min_nr, long nr,
1084 struct io_event __user *event,
1085 struct timespec __user *timeout)
1087 long start_jiffies = jiffies;
1088 struct task_struct *tsk = current;
1089 DECLARE_WAITQUEUE(wait, tsk);
1092 struct io_event ent;
1093 struct aio_timeout to;
1096 /* needed to zero any padding within an entry (there shouldn't be
1097 * any, but C is fun!
1099 memset(&ent, 0, sizeof(ent));
1102 while (likely(i < nr)) {
1103 ret = aio_read_evt(ctx, &ent);
1104 if (unlikely(ret <= 0))
1107 dprintk("read event: %Lx %Lx %Lx %Lx\n",
1108 ent.data, ent.obj, ent.res, ent.res2);
1110 /* Could we split the check in two? */
1112 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1113 dprintk("aio: lost an event due to EFAULT.\n");
1118 /* Good, event copied to userland, update counts. */
1130 /* racey check, but it gets redone */
1131 if (!retry && unlikely(!list_empty(&ctx->run_list))) {
1133 aio_run_all_iocbs(ctx);
1141 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1144 set_timeout(start_jiffies, &to, &ts);
1147 while (likely(i < nr)) {
1148 add_wait_queue_exclusive(&ctx->wait, &wait);
1150 set_task_state(tsk, TASK_INTERRUPTIBLE);
1151 ret = aio_read_evt(ctx, &ent);
1156 if (unlikely(ctx->dead)) {
1160 if (to.timed_out) /* Only check after read evt */
1162 /* Try to only show up in io wait if there are ops
1164 if (ctx->reqs_active)
1168 if (signal_pending(tsk)) {
1172 /*ret = aio_read_evt(ctx, &ent);*/
1175 set_task_state(tsk, TASK_RUNNING);
1176 remove_wait_queue(&ctx->wait, &wait);
1178 if (unlikely(ret <= 0))
1182 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1183 dprintk("aio: lost an event due to EFAULT.\n");
1187 /* Good, event copied to userland, update counts. */
1195 destroy_timer_on_stack(&to.timer);
1199 /* Take an ioctx and remove it from the list of ioctx's. Protects
1200 * against races with itself via ->dead.
1202 static void io_destroy(struct kioctx *ioctx)
1204 struct mm_struct *mm = current->mm;
1207 /* delete the entry from the list is someone else hasn't already */
1208 spin_lock(&mm->ioctx_lock);
1209 was_dead = ioctx->dead;
1211 hlist_del_rcu(&ioctx->list);
1212 spin_unlock(&mm->ioctx_lock);
1214 dprintk("aio_release(%p)\n", ioctx);
1215 if (likely(!was_dead))
1216 put_ioctx(ioctx); /* twice for the list */
1218 aio_cancel_all(ioctx);
1219 wait_for_all_aios(ioctx);
1222 * Wake up any waiters. The setting of ctx->dead must be seen
1223 * by other CPUs at this point. Right now, we rely on the
1224 * locking done by the above calls to ensure this consistency.
1226 wake_up(&ioctx->wait);
1227 put_ioctx(ioctx); /* once for the lookup */
1231 * Create an aio_context capable of receiving at least nr_events.
1232 * ctxp must not point to an aio_context that already exists, and
1233 * must be initialized to 0 prior to the call. On successful
1234 * creation of the aio_context, *ctxp is filled in with the resulting
1235 * handle. May fail with -EINVAL if *ctxp is not initialized,
1236 * if the specified nr_events exceeds internal limits. May fail
1237 * with -EAGAIN if the specified nr_events exceeds the user's limit
1238 * of available events. May fail with -ENOMEM if insufficient kernel
1239 * resources are available. May fail with -EFAULT if an invalid
1240 * pointer is passed for ctxp. Will fail with -ENOSYS if not
1243 SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
1245 struct kioctx *ioctx = NULL;
1249 ret = get_user(ctx, ctxp);
1254 if (unlikely(ctx || nr_events == 0)) {
1255 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
1260 ioctx = ioctx_alloc(nr_events);
1261 ret = PTR_ERR(ioctx);
1262 if (!IS_ERR(ioctx)) {
1263 ret = put_user(ioctx->user_id, ctxp);
1267 get_ioctx(ioctx); /* io_destroy() expects us to hold a ref */
1276 * Destroy the aio_context specified. May cancel any outstanding
1277 * AIOs and block on completion. Will fail with -ENOSYS if not
1278 * implemented. May fail with -EFAULT if the context pointed to
1281 SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
1283 struct kioctx *ioctx = lookup_ioctx(ctx);
1284 if (likely(NULL != ioctx)) {
1288 pr_debug("EINVAL: io_destroy: invalid context id\n");
1292 static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
1294 struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];
1298 while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
1299 ssize_t this = min((ssize_t)iov->iov_len, ret);
1300 iov->iov_base += this;
1301 iov->iov_len -= this;
1302 iocb->ki_left -= this;
1304 if (iov->iov_len == 0) {
1310 /* the caller should not have done more io than what fit in
1311 * the remaining iovecs */
1312 BUG_ON(ret > 0 && iocb->ki_left == 0);
1315 static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
1317 struct file *file = iocb->ki_filp;
1318 struct address_space *mapping = file->f_mapping;
1319 struct inode *inode = mapping->host;
1320 ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
1321 unsigned long, loff_t);
1323 unsigned short opcode;
1325 if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
1326 (iocb->ki_opcode == IOCB_CMD_PREAD)) {
1327 rw_op = file->f_op->aio_read;
1328 opcode = IOCB_CMD_PREADV;
1330 rw_op = file->f_op->aio_write;
1331 opcode = IOCB_CMD_PWRITEV;
1334 /* This matches the pread()/pwrite() logic */
1335 if (iocb->ki_pos < 0)
1339 ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
1340 iocb->ki_nr_segs - iocb->ki_cur_seg,
1343 aio_advance_iovec(iocb, ret);
1345 /* retry all partial writes. retry partial reads as long as its a
1347 } while (ret > 0 && iocb->ki_left > 0 &&
1348 (opcode == IOCB_CMD_PWRITEV ||
1349 (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
1351 /* This means we must have transferred all that we could */
1352 /* No need to retry anymore */
1353 if ((ret == 0) || (iocb->ki_left == 0))
1354 ret = iocb->ki_nbytes - iocb->ki_left;
1356 /* If we managed to write some out we return that, rather than
1357 * the eventual error. */
1358 if (opcode == IOCB_CMD_PWRITEV
1359 && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY
1360 && iocb->ki_nbytes - iocb->ki_left)
1361 ret = iocb->ki_nbytes - iocb->ki_left;
1366 static ssize_t aio_fdsync(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, 1);
1376 static ssize_t aio_fsync(struct kiocb *iocb)
1378 struct file *file = iocb->ki_filp;
1379 ssize_t ret = -EINVAL;
1381 if (file->f_op->aio_fsync)
1382 ret = file->f_op->aio_fsync(iocb, 0);
1386 static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb)
1390 ret = rw_copy_check_uvector(type, (struct iovec __user *)kiocb->ki_buf,
1391 kiocb->ki_nbytes, 1,
1392 &kiocb->ki_inline_vec, &kiocb->ki_iovec);
1396 kiocb->ki_nr_segs = kiocb->ki_nbytes;
1397 kiocb->ki_cur_seg = 0;
1398 /* ki_nbytes/left now reflect bytes instead of segs */
1399 kiocb->ki_nbytes = ret;
1400 kiocb->ki_left = ret;
1407 static ssize_t aio_setup_single_vector(struct kiocb *kiocb)
1409 kiocb->ki_iovec = &kiocb->ki_inline_vec;
1410 kiocb->ki_iovec->iov_base = kiocb->ki_buf;
1411 kiocb->ki_iovec->iov_len = kiocb->ki_left;
1412 kiocb->ki_nr_segs = 1;
1413 kiocb->ki_cur_seg = 0;
1419 * Performs the initial checks and aio retry method
1420 * setup for the kiocb at the time of io submission.
1422 static ssize_t aio_setup_iocb(struct kiocb *kiocb)
1424 struct file *file = kiocb->ki_filp;
1427 switch (kiocb->ki_opcode) {
1428 case IOCB_CMD_PREAD:
1430 if (unlikely(!(file->f_mode & FMODE_READ)))
1433 if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
1436 ret = security_file_permission(file, MAY_READ);
1439 ret = aio_setup_single_vector(kiocb);
1443 if (file->f_op->aio_read)
1444 kiocb->ki_retry = aio_rw_vect_retry;
1446 case IOCB_CMD_PWRITE:
1448 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1451 if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
1454 ret = security_file_permission(file, MAY_WRITE);
1457 ret = aio_setup_single_vector(kiocb);
1461 if (file->f_op->aio_write)
1462 kiocb->ki_retry = aio_rw_vect_retry;
1464 case IOCB_CMD_PREADV:
1466 if (unlikely(!(file->f_mode & FMODE_READ)))
1468 ret = security_file_permission(file, MAY_READ);
1471 ret = aio_setup_vectored_rw(READ, kiocb);
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 = security_file_permission(file, MAY_WRITE);
1485 ret = aio_setup_vectored_rw(WRITE, kiocb);
1489 if (file->f_op->aio_write)
1490 kiocb->ki_retry = aio_rw_vect_retry;
1492 case IOCB_CMD_FDSYNC:
1494 if (file->f_op->aio_fsync)
1495 kiocb->ki_retry = aio_fdsync;
1497 case IOCB_CMD_FSYNC:
1499 if (file->f_op->aio_fsync)
1500 kiocb->ki_retry = aio_fsync;
1503 dprintk("EINVAL: io_submit: no operation provided\n");
1507 if (!kiocb->ki_retry)
1514 * aio_wake_function:
1515 * wait queue callback function for aio notification,
1516 * Simply triggers a retry of the operation via kick_iocb.
1518 * This callback is specified in the wait queue entry in
1522 * This routine is executed with the wait queue lock held.
1523 * Since kick_iocb acquires iocb->ctx->ctx_lock, it nests
1524 * the ioctx lock inside the wait queue lock. This is safe
1525 * because this callback isn't used for wait queues which
1526 * are nested inside ioctx lock (i.e. ctx->wait)
1528 static int aio_wake_function(wait_queue_t *wait, unsigned mode,
1529 int sync, void *arg)
1531 struct wait_bit_key *key = arg;
1532 struct wait_bit_queue *wb
1533 = container_of(wait, struct wait_bit_queue, wait);
1534 struct kiocb *iocb = container_of(wb, struct kiocb, ki_wq);
1536 if (!wait_bit_cleared(wb, key))
1539 list_del_init(&wait->task_list);
1544 static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
1551 /* enforce forwards compatibility on users */
1552 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
1553 pr_debug("EINVAL: io_submit: reserve field set\n");
1557 /* prevent overflows */
1559 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1560 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1561 ((ssize_t)iocb->aio_nbytes < 0)
1563 pr_debug("EINVAL: io_submit: overflow check\n");
1567 file = fget(iocb->aio_fildes);
1568 if (unlikely(!file))
1571 req = aio_get_req(ctx); /* returns with 2 references to req */
1572 if (unlikely(!req)) {
1576 req->ki_filp = file;
1577 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1579 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
1580 * instance of the file* now. The file descriptor must be
1581 * an eventfd() fd, and will be signaled for each completed
1582 * event using the eventfd_signal() function.
1584 req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd);
1585 if (IS_ERR(req->ki_eventfd)) {
1586 ret = PTR_ERR(req->ki_eventfd);
1587 req->ki_eventfd = NULL;
1592 ret = put_user(req->ki_key, &user_iocb->aio_key);
1593 if (unlikely(ret)) {
1594 dprintk("EFAULT: aio_key\n");
1598 req->ki_obj.user = user_iocb;
1599 req->ki_user_data = iocb->aio_data;
1600 req->ki_pos = iocb->aio_offset;
1602 req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf;
1603 req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
1604 req->ki_opcode = iocb->aio_lio_opcode;
1605 init_waitqueue_func_entry(&req->ki_wq.wait, aio_wake_function);
1606 INIT_LIST_HEAD(&req->ki_wq.wait.task_list);
1608 ret = aio_setup_iocb(req);
1613 spin_lock_irq(&ctx->ctx_lock);
1615 if (!list_empty(&ctx->run_list)) {
1616 /* drain the run list */
1617 while (__aio_run_iocbs(ctx))
1620 spin_unlock_irq(&ctx->ctx_lock);
1621 aio_put_req(req); /* drop extra ref to req */
1625 aio_put_req(req); /* drop extra ref to req */
1626 aio_put_req(req); /* drop i/o ref to req */
1631 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1632 * the number of iocbs queued. May return -EINVAL if the aio_context
1633 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1634 * *iocbpp[0] is not properly initialized, if the operation specified
1635 * is invalid for the file descriptor in the iocb. May fail with
1636 * -EFAULT if any of the data structures point to invalid data. May
1637 * fail with -EBADF if the file descriptor specified in the first
1638 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1639 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1640 * fail with -ENOSYS if not implemented.
1642 SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr,
1643 struct iocb __user * __user *, iocbpp)
1649 if (unlikely(nr < 0))
1652 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1655 ctx = lookup_ioctx(ctx_id);
1656 if (unlikely(!ctx)) {
1657 pr_debug("EINVAL: io_submit: invalid context id\n");
1662 * AKPM: should this return a partial result if some of the IOs were
1663 * successfully submitted?
1665 for (i=0; i<nr; i++) {
1666 struct iocb __user *user_iocb;
1669 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1674 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1679 ret = io_submit_one(ctx, user_iocb, &tmp);
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. May
1770 * fail with -EINVAL if ctx_id is invalid, if min_nr is out of range,
1771 * if nr is out of range, if when is out of range. May fail with
1772 * -EFAULT if any of the memory specified to is invalid. May return
1773 * 0 or < min_nr if no events are available and the timeout specified
1774 * by when has elapsed, where when == NULL specifies an infinite
1775 * timeout. Note that the timeout pointed to by when is relative and
1776 * will be updated if not NULL and the operation blocks. Will fail
1777 * with -ENOSYS if not implemented.
1779 SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id,
1782 struct io_event __user *, events,
1783 struct timespec __user *, timeout)
1785 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1788 if (likely(ioctx)) {
1789 if (likely(min_nr <= nr && min_nr >= 0 && nr >= 0))
1790 ret = read_events(ioctx, min_nr, nr, events, timeout);
1794 asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout);