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>
21 #include <linux/sched.h>
23 #include <linux/file.h>
25 #include <linux/mman.h>
26 #include <linux/slab.h>
27 #include <linux/timer.h>
28 #include <linux/aio.h>
29 #include <linux/highmem.h>
30 #include <linux/workqueue.h>
31 #include <linux/security.h>
32 #include <linux/rcuref.h>
34 #include <asm/kmap_types.h>
35 #include <asm/uaccess.h>
36 #include <asm/mmu_context.h>
39 #define dprintk printk
41 #define dprintk(x...) do { ; } while (0)
44 /*------ sysctl variables----*/
45 atomic_t aio_nr
= ATOMIC_INIT(0); /* current system wide number of aio requests */
46 unsigned aio_max_nr
= 0x10000; /* system wide maximum number of aio requests */
47 /*----end sysctl variables---*/
49 static kmem_cache_t
*kiocb_cachep
;
50 static kmem_cache_t
*kioctx_cachep
;
52 static struct workqueue_struct
*aio_wq
;
54 /* Used for rare fput completion. */
55 static void aio_fput_routine(void *);
56 static DECLARE_WORK(fput_work
, aio_fput_routine
, NULL
);
58 static DEFINE_SPINLOCK(fput_lock
);
59 static LIST_HEAD(fput_head
);
61 static void aio_kick_handler(void *);
62 static void aio_queue_work(struct kioctx
*);
65 * Creates the slab caches used by the aio routines, panic on
66 * failure as this is done early during the boot sequence.
68 static int __init
aio_setup(void)
70 kiocb_cachep
= kmem_cache_create("kiocb", sizeof(struct kiocb
),
71 0, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
72 kioctx_cachep
= kmem_cache_create("kioctx", sizeof(struct kioctx
),
73 0, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
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
= kmalloc(sizeof(struct page
*) * nr_pages
, GFP_KERNEL
);
126 if (!info
->ring_pages
)
128 memset(info
->ring_pages
, 0, sizeof(struct page
*) * nr_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_ANON
|MAP_PRIVATE
,
137 if (IS_ERR((void *)info
->mmap_base
)) {
138 up_write(&ctx
->mm
->mmap_sem
);
139 printk("mmap err: %ld\n", -info
->mmap_base
);
145 dprintk("mmap address: 0x%08lx\n", info
->mmap_base
);
146 info
->nr_pages
= get_user_pages(current
, ctx
->mm
,
147 info
->mmap_base
, nr_pages
,
148 1, 0, info
->ring_pages
, NULL
);
149 up_write(&ctx
->mm
->mmap_sem
);
151 if (unlikely(info
->nr_pages
!= nr_pages
)) {
156 ctx
->user_id
= info
->mmap_base
;
158 info
->nr
= nr_events
; /* trusted copy */
160 ring
= kmap_atomic(info
->ring_pages
[0], KM_USER0
);
161 ring
->nr
= nr_events
; /* user copy */
162 ring
->id
= ctx
->user_id
;
163 ring
->head
= ring
->tail
= 0;
164 ring
->magic
= AIO_RING_MAGIC
;
165 ring
->compat_features
= AIO_RING_COMPAT_FEATURES
;
166 ring
->incompat_features
= AIO_RING_INCOMPAT_FEATURES
;
167 ring
->header_length
= sizeof(struct aio_ring
);
168 kunmap_atomic(ring
, KM_USER0
);
174 /* aio_ring_event: returns a pointer to the event at the given index from
175 * kmap_atomic(, km). Release the pointer with put_aio_ring_event();
177 #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
178 #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
179 #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
181 #define aio_ring_event(info, nr, km) ({ \
182 unsigned pos = (nr) + AIO_EVENTS_OFFSET; \
183 struct io_event *__event; \
184 __event = kmap_atomic( \
185 (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE], km); \
186 __event += pos % AIO_EVENTS_PER_PAGE; \
190 #define put_aio_ring_event(event, km) do { \
191 struct io_event *__event = (event); \
193 kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \
197 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
199 static struct kioctx
*ioctx_alloc(unsigned nr_events
)
201 struct mm_struct
*mm
;
204 /* Prevent overflows */
205 if ((nr_events
> (0x10000000U
/ sizeof(struct io_event
))) ||
206 (nr_events
> (0x10000000U
/ sizeof(struct kiocb
)))) {
207 pr_debug("ENOMEM: nr_events too high\n");
208 return ERR_PTR(-EINVAL
);
211 if (nr_events
> aio_max_nr
)
212 return ERR_PTR(-EAGAIN
);
214 ctx
= kmem_cache_alloc(kioctx_cachep
, GFP_KERNEL
);
216 return ERR_PTR(-ENOMEM
);
218 memset(ctx
, 0, sizeof(*ctx
));
219 ctx
->max_reqs
= nr_events
;
220 mm
= ctx
->mm
= current
->mm
;
221 atomic_inc(&mm
->mm_count
);
223 atomic_set(&ctx
->users
, 1);
224 spin_lock_init(&ctx
->ctx_lock
);
225 spin_lock_init(&ctx
->ring_info
.ring_lock
);
226 init_waitqueue_head(&ctx
->wait
);
228 INIT_LIST_HEAD(&ctx
->active_reqs
);
229 INIT_LIST_HEAD(&ctx
->run_list
);
230 INIT_WORK(&ctx
->wq
, aio_kick_handler
, ctx
);
232 if (aio_setup_ring(ctx
) < 0)
235 /* limit the number of system wide aios */
236 atomic_add(ctx
->max_reqs
, &aio_nr
); /* undone by __put_ioctx */
237 if (unlikely(atomic_read(&aio_nr
) > aio_max_nr
))
240 /* now link into global list. kludge. FIXME */
241 write_lock(&mm
->ioctx_list_lock
);
242 ctx
->next
= mm
->ioctx_list
;
243 mm
->ioctx_list
= ctx
;
244 write_unlock(&mm
->ioctx_list_lock
);
246 dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
247 ctx
, ctx
->user_id
, current
->mm
, ctx
->ring_info
.nr
);
251 atomic_sub(ctx
->max_reqs
, &aio_nr
);
252 ctx
->max_reqs
= 0; /* prevent __put_ioctx from sub'ing aio_nr */
254 return ERR_PTR(-EAGAIN
);
258 kmem_cache_free(kioctx_cachep
, ctx
);
259 ctx
= ERR_PTR(-ENOMEM
);
261 dprintk("aio: error allocating ioctx %p\n", ctx
);
266 * Cancels all outstanding aio requests on an aio context. Used
267 * when the processes owning a context have all exited to encourage
268 * the rapid destruction of the kioctx.
270 static void aio_cancel_all(struct kioctx
*ctx
)
272 int (*cancel
)(struct kiocb
*, struct io_event
*);
274 spin_lock_irq(&ctx
->ctx_lock
);
276 while (!list_empty(&ctx
->active_reqs
)) {
277 struct list_head
*pos
= ctx
->active_reqs
.next
;
278 struct kiocb
*iocb
= list_kiocb(pos
);
279 list_del_init(&iocb
->ki_list
);
280 cancel
= iocb
->ki_cancel
;
281 kiocbSetCancelled(iocb
);
284 spin_unlock_irq(&ctx
->ctx_lock
);
286 spin_lock_irq(&ctx
->ctx_lock
);
289 spin_unlock_irq(&ctx
->ctx_lock
);
292 static void wait_for_all_aios(struct kioctx
*ctx
)
294 struct task_struct
*tsk
= current
;
295 DECLARE_WAITQUEUE(wait
, tsk
);
297 if (!ctx
->reqs_active
)
300 add_wait_queue(&ctx
->wait
, &wait
);
301 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
302 while (ctx
->reqs_active
) {
304 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
306 __set_task_state(tsk
, TASK_RUNNING
);
307 remove_wait_queue(&ctx
->wait
, &wait
);
310 /* wait_on_sync_kiocb:
311 * Waits on the given sync kiocb to complete.
313 ssize_t fastcall
wait_on_sync_kiocb(struct kiocb
*iocb
)
315 while (iocb
->ki_users
) {
316 set_current_state(TASK_UNINTERRUPTIBLE
);
321 __set_current_state(TASK_RUNNING
);
322 return iocb
->ki_user_data
;
325 /* exit_aio: called when the last user of mm goes away. At this point,
326 * there is no way for any new requests to be submited or any of the
327 * io_* syscalls to be called on the context. However, there may be
328 * outstanding requests which hold references to the context; as they
329 * go away, they will call put_ioctx and release any pinned memory
330 * associated with the request (held via struct page * references).
332 void fastcall
exit_aio(struct mm_struct
*mm
)
334 struct kioctx
*ctx
= mm
->ioctx_list
;
335 mm
->ioctx_list
= NULL
;
337 struct kioctx
*next
= ctx
->next
;
341 wait_for_all_aios(ctx
);
343 * this is an overkill, but ensures we don't leave
344 * the ctx on the aio_wq
346 flush_workqueue(aio_wq
);
348 if (1 != atomic_read(&ctx
->users
))
350 "exit_aio:ioctx still alive: %d %d %d\n",
351 atomic_read(&ctx
->users
), ctx
->dead
,
359 * Called when the last user of an aio context has gone away,
360 * and the struct needs to be freed.
362 void fastcall
__put_ioctx(struct kioctx
*ctx
)
364 unsigned nr_events
= ctx
->max_reqs
;
366 if (unlikely(ctx
->reqs_active
))
369 cancel_delayed_work(&ctx
->wq
);
370 flush_workqueue(aio_wq
);
374 pr_debug("__put_ioctx: freeing %p\n", ctx
);
375 kmem_cache_free(kioctx_cachep
, ctx
);
377 atomic_sub(nr_events
, &aio_nr
);
381 * Allocate a slot for an aio request. Increments the users count
382 * of the kioctx so that the kioctx stays around until all requests are
383 * complete. Returns NULL if no requests are free.
385 * Returns with kiocb->users set to 2. The io submit code path holds
386 * an extra reference while submitting the i/o.
387 * This prevents races between the aio code path referencing the
388 * req (after submitting it) and aio_complete() freeing the req.
390 static struct kiocb
*FASTCALL(__aio_get_req(struct kioctx
*ctx
));
391 static struct kiocb fastcall
*__aio_get_req(struct kioctx
*ctx
)
393 struct kiocb
*req
= NULL
;
394 struct aio_ring
*ring
;
397 req
= kmem_cache_alloc(kiocb_cachep
, GFP_KERNEL
);
401 req
->ki_flags
= 1 << KIF_LOCKED
;
405 req
->ki_cancel
= NULL
;
406 req
->ki_retry
= NULL
;
409 INIT_LIST_HEAD(&req
->ki_run_list
);
411 /* Check if the completion queue has enough free space to
412 * accept an event from this io.
414 spin_lock_irq(&ctx
->ctx_lock
);
415 ring
= kmap_atomic(ctx
->ring_info
.ring_pages
[0], KM_USER0
);
416 if (ctx
->reqs_active
< aio_ring_avail(&ctx
->ring_info
, ring
)) {
417 list_add(&req
->ki_list
, &ctx
->active_reqs
);
422 kunmap_atomic(ring
, KM_USER0
);
423 spin_unlock_irq(&ctx
->ctx_lock
);
426 kmem_cache_free(kiocb_cachep
, req
);
433 static inline struct kiocb
*aio_get_req(struct kioctx
*ctx
)
436 /* Handle a potential starvation case -- should be exceedingly rare as
437 * requests will be stuck on fput_head only if the aio_fput_routine is
438 * delayed and the requests were the last user of the struct file.
440 req
= __aio_get_req(ctx
);
441 if (unlikely(NULL
== req
)) {
442 aio_fput_routine(NULL
);
443 req
= __aio_get_req(ctx
);
448 static inline void really_put_req(struct kioctx
*ctx
, struct kiocb
*req
)
452 kmem_cache_free(kiocb_cachep
, req
);
455 if (unlikely(!ctx
->reqs_active
&& ctx
->dead
))
459 static void aio_fput_routine(void *data
)
461 spin_lock_irq(&fput_lock
);
462 while (likely(!list_empty(&fput_head
))) {
463 struct kiocb
*req
= list_kiocb(fput_head
.next
);
464 struct kioctx
*ctx
= req
->ki_ctx
;
466 list_del(&req
->ki_list
);
467 spin_unlock_irq(&fput_lock
);
469 /* Complete the fput */
470 __fput(req
->ki_filp
);
472 /* Link the iocb into the context's free list */
473 spin_lock_irq(&ctx
->ctx_lock
);
474 really_put_req(ctx
, req
);
475 spin_unlock_irq(&ctx
->ctx_lock
);
478 spin_lock_irq(&fput_lock
);
480 spin_unlock_irq(&fput_lock
);
484 * Returns true if this put was the last user of the request.
486 static int __aio_put_req(struct kioctx
*ctx
, struct kiocb
*req
)
488 dprintk(KERN_DEBUG
"aio_put(%p): f_count=%d\n",
489 req
, atomic_read(&req
->ki_filp
->f_count
));
492 if (unlikely(req
->ki_users
< 0))
494 if (likely(req
->ki_users
))
496 list_del(&req
->ki_list
); /* remove from active_reqs */
497 req
->ki_cancel
= NULL
;
498 req
->ki_retry
= NULL
;
500 /* Must be done under the lock to serialise against cancellation.
501 * Call this aio_fput as it duplicates fput via the fput_work.
503 if (unlikely(rcuref_dec_and_test(&req
->ki_filp
->f_count
))) {
505 spin_lock(&fput_lock
);
506 list_add(&req
->ki_list
, &fput_head
);
507 spin_unlock(&fput_lock
);
508 queue_work(aio_wq
, &fput_work
);
510 really_put_req(ctx
, req
);
515 * Returns true if this put was the last user of the kiocb,
516 * false if the request is still in use.
518 int fastcall
aio_put_req(struct kiocb
*req
)
520 struct kioctx
*ctx
= req
->ki_ctx
;
522 spin_lock_irq(&ctx
->ctx_lock
);
523 ret
= __aio_put_req(ctx
, req
);
524 spin_unlock_irq(&ctx
->ctx_lock
);
530 /* Lookup an ioctx id. ioctx_list is lockless for reads.
531 * FIXME: this is O(n) and is only suitable for development.
533 struct kioctx
*lookup_ioctx(unsigned long ctx_id
)
535 struct kioctx
*ioctx
;
536 struct mm_struct
*mm
;
539 read_lock(&mm
->ioctx_list_lock
);
540 for (ioctx
= mm
->ioctx_list
; ioctx
; ioctx
= ioctx
->next
)
541 if (likely(ioctx
->user_id
== ctx_id
&& !ioctx
->dead
)) {
545 read_unlock(&mm
->ioctx_list_lock
);
550 static int lock_kiocb_action(void *param
)
556 static inline void lock_kiocb(struct kiocb
*iocb
)
558 wait_on_bit_lock(&iocb
->ki_flags
, KIF_LOCKED
, lock_kiocb_action
,
559 TASK_UNINTERRUPTIBLE
);
562 static inline void unlock_kiocb(struct kiocb
*iocb
)
564 kiocbClearLocked(iocb
);
565 smp_mb__after_clear_bit();
566 wake_up_bit(&iocb
->ki_flags
, KIF_LOCKED
);
571 * Makes the calling kernel thread take on the specified
573 * Called by the retry thread execute retries within the
574 * iocb issuer's mm context, so that copy_from/to_user
575 * operations work seamlessly for aio.
576 * (Note: this routine is intended to be called only
577 * from a kernel thread context)
579 static void use_mm(struct mm_struct
*mm
)
581 struct mm_struct
*active_mm
;
582 struct task_struct
*tsk
= current
;
585 tsk
->flags
|= PF_BORROWED_MM
;
586 active_mm
= tsk
->active_mm
;
587 atomic_inc(&mm
->mm_count
);
591 * Note that on UML this *requires* PF_BORROWED_MM to be set, otherwise
592 * it won't work. Update it accordingly if you change it here
594 activate_mm(active_mm
, mm
);
602 * Reverses the effect of use_mm, i.e. releases the
603 * specified mm context which was earlier taken on
604 * by the calling kernel thread
605 * (Note: this routine is intended to be called only
606 * from a kernel thread context)
608 * Comments: Called with ctx->ctx_lock held. This nests
609 * task_lock instead ctx_lock.
611 static void unuse_mm(struct mm_struct
*mm
)
613 struct task_struct
*tsk
= current
;
616 tsk
->flags
&= ~PF_BORROWED_MM
;
618 /* active_mm is still 'mm' */
619 enter_lazy_tlb(mm
, tsk
);
624 * Queue up a kiocb to be retried. Assumes that the kiocb
625 * has already been marked as kicked, and places it on
626 * the retry run list for the corresponding ioctx, if it
627 * isn't already queued. Returns 1 if it actually queued
628 * the kiocb (to tell the caller to activate the work
629 * queue to process it), or 0, if it found that it was
632 * Should be called with the spin lock iocb->ki_ctx->ctx_lock
635 static inline int __queue_kicked_iocb(struct kiocb
*iocb
)
637 struct kioctx
*ctx
= iocb
->ki_ctx
;
639 if (list_empty(&iocb
->ki_run_list
)) {
640 list_add_tail(&iocb
->ki_run_list
,
648 * This is the core aio execution routine. It is
649 * invoked both for initial i/o submission and
650 * subsequent retries via the aio_kick_handler.
651 * Expects to be invoked with iocb->ki_ctx->lock
652 * already held. The lock is released and reaquired
653 * as needed during processing.
655 * Calls the iocb retry method (already setup for the
656 * iocb on initial submission) for operation specific
657 * handling, but takes care of most of common retry
658 * execution details for a given iocb. The retry method
659 * needs to be non-blocking as far as possible, to avoid
660 * holding up other iocbs waiting to be serviced by the
661 * retry kernel thread.
663 * The trickier parts in this code have to do with
664 * ensuring that only one retry instance is in progress
665 * for a given iocb at any time. Providing that guarantee
666 * simplifies the coding of individual aio operations as
667 * it avoids various potential races.
669 static ssize_t
aio_run_iocb(struct kiocb
*iocb
)
671 struct kioctx
*ctx
= iocb
->ki_ctx
;
672 ssize_t (*retry
)(struct kiocb
*);
675 if (iocb
->ki_retried
++ > 1024*1024) {
676 printk("Maximal retry count. Bytes done %Zd\n",
677 iocb
->ki_nbytes
- iocb
->ki_left
);
681 if (!(iocb
->ki_retried
& 0xff)) {
682 pr_debug("%ld retry: %d of %d\n", iocb
->ki_retried
,
683 iocb
->ki_nbytes
- iocb
->ki_left
, iocb
->ki_nbytes
);
686 if (!(retry
= iocb
->ki_retry
)) {
687 printk("aio_run_iocb: iocb->ki_retry = NULL\n");
692 * We don't want the next retry iteration for this
693 * operation to start until this one has returned and
694 * updated the iocb state. However, wait_queue functions
695 * can trigger a kick_iocb from interrupt context in the
696 * meantime, indicating that data is available for the next
697 * iteration. We want to remember that and enable the
698 * next retry iteration _after_ we are through with
701 * So, in order to be able to register a "kick", but
702 * prevent it from being queued now, we clear the kick
703 * flag, but make the kick code *think* that the iocb is
704 * still on the run list until we are actually done.
705 * When we are done with this iteration, we check if
706 * the iocb was kicked in the meantime and if so, queue
710 kiocbClearKicked(iocb
);
713 * This is so that aio_complete knows it doesn't need to
714 * pull the iocb off the run list (We can't just call
715 * INIT_LIST_HEAD because we don't want a kick_iocb to
716 * queue this on the run list yet)
718 iocb
->ki_run_list
.next
= iocb
->ki_run_list
.prev
= NULL
;
719 spin_unlock_irq(&ctx
->ctx_lock
);
721 /* Quit retrying if the i/o has been cancelled */
722 if (kiocbIsCancelled(iocb
)) {
724 aio_complete(iocb
, ret
, 0);
725 /* must not access the iocb after this */
730 * Now we are all set to call the retry method in async
731 * context. By setting this thread's io_wait context
732 * to point to the wait queue entry inside the currently
733 * running iocb for the duration of the retry, we ensure
734 * that async notification wakeups are queued by the
735 * operation instead of blocking waits, and when notified,
736 * cause the iocb to be kicked for continuation (through
737 * the aio_wake_function callback).
739 BUG_ON(current
->io_wait
!= NULL
);
740 current
->io_wait
= &iocb
->ki_wait
;
742 current
->io_wait
= NULL
;
744 if (-EIOCBRETRY
!= ret
) {
745 if (-EIOCBQUEUED
!= ret
) {
746 BUG_ON(!list_empty(&iocb
->ki_wait
.task_list
));
747 aio_complete(iocb
, ret
, 0);
748 /* must not access the iocb after this */
752 * Issue an additional retry to avoid waiting forever if
753 * no waits were queued (e.g. in case of a short read).
755 if (list_empty(&iocb
->ki_wait
.task_list
))
756 kiocbSetKicked(iocb
);
759 spin_lock_irq(&ctx
->ctx_lock
);
761 if (-EIOCBRETRY
== ret
) {
763 * OK, now that we are done with this iteration
764 * and know that there is more left to go,
765 * this is where we let go so that a subsequent
766 * "kick" can start the next iteration
769 /* will make __queue_kicked_iocb succeed from here on */
770 INIT_LIST_HEAD(&iocb
->ki_run_list
);
771 /* we must queue the next iteration ourselves, if it
772 * has already been kicked */
773 if (kiocbIsKicked(iocb
)) {
774 __queue_kicked_iocb(iocb
);
777 * __queue_kicked_iocb will always return 1 here, because
778 * iocb->ki_run_list is empty at this point so it should
779 * be safe to unconditionally queue the context into the
790 * Process all pending retries queued on the ioctx
792 * Assumes it is operating within the aio issuer's mm
793 * context. Expects to be called with ctx->ctx_lock held
795 static int __aio_run_iocbs(struct kioctx
*ctx
)
800 list_splice_init(&ctx
->run_list
, &run_list
);
801 while (!list_empty(&run_list
)) {
802 iocb
= list_entry(run_list
.next
, struct kiocb
,
804 list_del(&iocb
->ki_run_list
);
806 * Hold an extra reference while retrying i/o.
808 iocb
->ki_users
++; /* grab extra reference */
812 if (__aio_put_req(ctx
, iocb
)) /* drop extra ref */
815 if (!list_empty(&ctx
->run_list
))
820 static void aio_queue_work(struct kioctx
* ctx
)
822 unsigned long timeout
;
824 * if someone is waiting, get the work started right
825 * away, otherwise, use a longer delay
828 if (waitqueue_active(&ctx
->wait
))
832 queue_delayed_work(aio_wq
, &ctx
->wq
, timeout
);
838 * Process all pending retries queued on the ioctx
840 * Assumes it is operating within the aio issuer's mm
843 static inline void aio_run_iocbs(struct kioctx
*ctx
)
847 spin_lock_irq(&ctx
->ctx_lock
);
849 requeue
= __aio_run_iocbs(ctx
);
850 spin_unlock_irq(&ctx
->ctx_lock
);
856 * just like aio_run_iocbs, but keeps running them until
857 * the list stays empty
859 static inline void aio_run_all_iocbs(struct kioctx
*ctx
)
861 spin_lock_irq(&ctx
->ctx_lock
);
862 while (__aio_run_iocbs(ctx
))
864 spin_unlock_irq(&ctx
->ctx_lock
);
869 * Work queue handler triggered to process pending
870 * retries on an ioctx. Takes on the aio issuer's
871 * mm context before running the iocbs, so that
872 * copy_xxx_user operates on the issuer's address
874 * Run on aiod's context.
876 static void aio_kick_handler(void *data
)
878 struct kioctx
*ctx
= data
;
879 mm_segment_t oldfs
= get_fs();
884 spin_lock_irq(&ctx
->ctx_lock
);
885 requeue
=__aio_run_iocbs(ctx
);
887 spin_unlock_irq(&ctx
->ctx_lock
);
890 * we're in a worker thread already, don't use queue_delayed_work,
893 queue_work(aio_wq
, &ctx
->wq
);
898 * Called by kick_iocb to queue the kiocb for retry
899 * and if required activate the aio work queue to process
902 static void try_queue_kicked_iocb(struct kiocb
*iocb
)
904 struct kioctx
*ctx
= iocb
->ki_ctx
;
908 /* We're supposed to be the only path putting the iocb back on the run
909 * list. If we find that the iocb is *back* on a wait queue already
910 * than retry has happened before we could queue the iocb. This also
911 * means that the retry could have completed and freed our iocb, no
913 BUG_ON((!list_empty(&iocb
->ki_wait
.task_list
)));
915 spin_lock_irqsave(&ctx
->ctx_lock
, flags
);
916 /* set this inside the lock so that we can't race with aio_run_iocb()
917 * testing it and putting the iocb on the run list under the lock */
918 if (!kiocbTryKick(iocb
))
919 run
= __queue_kicked_iocb(iocb
);
920 spin_unlock_irqrestore(&ctx
->ctx_lock
, flags
);
927 * Called typically from a wait queue callback context
928 * (aio_wake_function) to trigger a retry of the iocb.
929 * The retry is usually executed by aio workqueue
930 * threads (See aio_kick_handler).
932 void fastcall
kick_iocb(struct kiocb
*iocb
)
934 /* sync iocbs are easy: they can only ever be executing from a
936 if (is_sync_kiocb(iocb
)) {
937 kiocbSetKicked(iocb
);
938 wake_up_process(iocb
->ki_obj
.tsk
);
942 try_queue_kicked_iocb(iocb
);
944 EXPORT_SYMBOL(kick_iocb
);
947 * Called when the io request on the given iocb is complete.
948 * Returns true if this is the last user of the request. The
949 * only other user of the request can be the cancellation code.
951 int fastcall
aio_complete(struct kiocb
*iocb
, long res
, long res2
)
953 struct kioctx
*ctx
= iocb
->ki_ctx
;
954 struct aio_ring_info
*info
;
955 struct aio_ring
*ring
;
956 struct io_event
*event
;
961 /* Special case handling for sync iocbs: events go directly
962 * into the iocb for fast handling. Note that this will not
963 * work if we allow sync kiocbs to be cancelled. in which
964 * case the usage count checks will have to move under ctx_lock
967 if (is_sync_kiocb(iocb
)) {
970 iocb
->ki_user_data
= res
;
971 if (iocb
->ki_users
== 1) {
975 spin_lock_irq(&ctx
->ctx_lock
);
977 ret
= (0 == iocb
->ki_users
);
978 spin_unlock_irq(&ctx
->ctx_lock
);
980 /* sync iocbs put the task here for us */
981 wake_up_process(iocb
->ki_obj
.tsk
);
985 info
= &ctx
->ring_info
;
987 /* add a completion event to the ring buffer.
988 * must be done holding ctx->ctx_lock to prevent
989 * other code from messing with the tail
990 * pointer since we might be called from irq
993 spin_lock_irqsave(&ctx
->ctx_lock
, flags
);
995 if (iocb
->ki_run_list
.prev
&& !list_empty(&iocb
->ki_run_list
))
996 list_del_init(&iocb
->ki_run_list
);
999 * cancelled requests don't get events, userland was given one
1000 * when the event got cancelled.
1002 if (kiocbIsCancelled(iocb
))
1005 ring
= kmap_atomic(info
->ring_pages
[0], KM_IRQ1
);
1008 event
= aio_ring_event(info
, tail
, KM_IRQ0
);
1009 if (++tail
>= info
->nr
)
1012 event
->obj
= (u64
)(unsigned long)iocb
->ki_obj
.user
;
1013 event
->data
= iocb
->ki_user_data
;
1017 dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n",
1018 ctx
, tail
, iocb
, iocb
->ki_obj
.user
, iocb
->ki_user_data
,
1021 /* after flagging the request as done, we
1022 * must never even look at it again
1024 smp_wmb(); /* make event visible before updating tail */
1029 put_aio_ring_event(event
, KM_IRQ0
);
1030 kunmap_atomic(ring
, KM_IRQ1
);
1032 pr_debug("added to ring %p at [%lu]\n", iocb
, tail
);
1034 pr_debug("%ld retries: %d of %d\n", iocb
->ki_retried
,
1035 iocb
->ki_nbytes
- iocb
->ki_left
, iocb
->ki_nbytes
);
1037 /* everything turned out well, dispose of the aiocb. */
1038 ret
= __aio_put_req(ctx
, iocb
);
1040 spin_unlock_irqrestore(&ctx
->ctx_lock
, flags
);
1042 if (waitqueue_active(&ctx
->wait
))
1043 wake_up(&ctx
->wait
);
1052 * Pull an event off of the ioctx's event ring. Returns the number of
1053 * events fetched (0 or 1 ;-)
1054 * FIXME: make this use cmpxchg.
1055 * TODO: make the ringbuffer user mmap()able (requires FIXME).
1057 static int aio_read_evt(struct kioctx
*ioctx
, struct io_event
*ent
)
1059 struct aio_ring_info
*info
= &ioctx
->ring_info
;
1060 struct aio_ring
*ring
;
1064 ring
= kmap_atomic(info
->ring_pages
[0], KM_USER0
);
1065 dprintk("in aio_read_evt h%lu t%lu m%lu\n",
1066 (unsigned long)ring
->head
, (unsigned long)ring
->tail
,
1067 (unsigned long)ring
->nr
);
1069 if (ring
->head
== ring
->tail
)
1072 spin_lock(&info
->ring_lock
);
1074 head
= ring
->head
% info
->nr
;
1075 if (head
!= ring
->tail
) {
1076 struct io_event
*evp
= aio_ring_event(info
, head
, KM_USER1
);
1078 head
= (head
+ 1) % info
->nr
;
1079 smp_mb(); /* finish reading the event before updatng the head */
1082 put_aio_ring_event(evp
, KM_USER1
);
1084 spin_unlock(&info
->ring_lock
);
1087 kunmap_atomic(ring
, KM_USER0
);
1088 dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret
,
1089 (unsigned long)ring
->head
, (unsigned long)ring
->tail
);
1093 struct aio_timeout
{
1094 struct timer_list timer
;
1096 struct task_struct
*p
;
1099 static void timeout_func(unsigned long data
)
1101 struct aio_timeout
*to
= (struct aio_timeout
*)data
;
1104 wake_up_process(to
->p
);
1107 static inline void init_timeout(struct aio_timeout
*to
)
1109 init_timer(&to
->timer
);
1110 to
->timer
.data
= (unsigned long)to
;
1111 to
->timer
.function
= timeout_func
;
1116 static inline void set_timeout(long start_jiffies
, struct aio_timeout
*to
,
1117 const struct timespec
*ts
)
1119 to
->timer
.expires
= start_jiffies
+ timespec_to_jiffies(ts
);
1120 if (time_after(to
->timer
.expires
, jiffies
))
1121 add_timer(&to
->timer
);
1126 static inline void clear_timeout(struct aio_timeout
*to
)
1128 del_singleshot_timer_sync(&to
->timer
);
1131 static int read_events(struct kioctx
*ctx
,
1132 long min_nr
, long nr
,
1133 struct io_event __user
*event
,
1134 struct timespec __user
*timeout
)
1136 long start_jiffies
= jiffies
;
1137 struct task_struct
*tsk
= current
;
1138 DECLARE_WAITQUEUE(wait
, tsk
);
1141 struct io_event ent
;
1142 struct aio_timeout to
;
1145 /* needed to zero any padding within an entry (there shouldn't be
1146 * any, but C is fun!
1148 memset(&ent
, 0, sizeof(ent
));
1151 while (likely(i
< nr
)) {
1152 ret
= aio_read_evt(ctx
, &ent
);
1153 if (unlikely(ret
<= 0))
1156 dprintk("read event: %Lx %Lx %Lx %Lx\n",
1157 ent
.data
, ent
.obj
, ent
.res
, ent
.res2
);
1159 /* Could we split the check in two? */
1161 if (unlikely(copy_to_user(event
, &ent
, sizeof(ent
)))) {
1162 dprintk("aio: lost an event due to EFAULT.\n");
1167 /* Good, event copied to userland, update counts. */
1179 /* racey check, but it gets redone */
1180 if (!retry
&& unlikely(!list_empty(&ctx
->run_list
))) {
1182 aio_run_all_iocbs(ctx
);
1190 if (unlikely(copy_from_user(&ts
, timeout
, sizeof(ts
))))
1193 set_timeout(start_jiffies
, &to
, &ts
);
1196 while (likely(i
< nr
)) {
1197 add_wait_queue_exclusive(&ctx
->wait
, &wait
);
1199 set_task_state(tsk
, TASK_INTERRUPTIBLE
);
1200 ret
= aio_read_evt(ctx
, &ent
);
1206 if (to
.timed_out
) /* Only check after read evt */
1209 if (signal_pending(tsk
)) {
1213 /*ret = aio_read_evt(ctx, &ent);*/
1216 set_task_state(tsk
, TASK_RUNNING
);
1217 remove_wait_queue(&ctx
->wait
, &wait
);
1219 if (unlikely(ret
<= 0))
1223 if (unlikely(copy_to_user(event
, &ent
, sizeof(ent
)))) {
1224 dprintk("aio: lost an event due to EFAULT.\n");
1228 /* Good, event copied to userland, update counts. */
1239 /* Take an ioctx and remove it from the list of ioctx's. Protects
1240 * against races with itself via ->dead.
1242 static void io_destroy(struct kioctx
*ioctx
)
1244 struct mm_struct
*mm
= current
->mm
;
1245 struct kioctx
**tmp
;
1248 /* delete the entry from the list is someone else hasn't already */
1249 write_lock(&mm
->ioctx_list_lock
);
1250 was_dead
= ioctx
->dead
;
1252 for (tmp
= &mm
->ioctx_list
; *tmp
&& *tmp
!= ioctx
;
1253 tmp
= &(*tmp
)->next
)
1257 write_unlock(&mm
->ioctx_list_lock
);
1259 dprintk("aio_release(%p)\n", ioctx
);
1260 if (likely(!was_dead
))
1261 put_ioctx(ioctx
); /* twice for the list */
1263 aio_cancel_all(ioctx
);
1264 wait_for_all_aios(ioctx
);
1265 put_ioctx(ioctx
); /* once for the lookup */
1269 * Create an aio_context capable of receiving at least nr_events.
1270 * ctxp must not point to an aio_context that already exists, and
1271 * must be initialized to 0 prior to the call. On successful
1272 * creation of the aio_context, *ctxp is filled in with the resulting
1273 * handle. May fail with -EINVAL if *ctxp is not initialized,
1274 * if the specified nr_events exceeds internal limits. May fail
1275 * with -EAGAIN if the specified nr_events exceeds the user's limit
1276 * of available events. May fail with -ENOMEM if insufficient kernel
1277 * resources are available. May fail with -EFAULT if an invalid
1278 * pointer is passed for ctxp. Will fail with -ENOSYS if not
1281 asmlinkage
long sys_io_setup(unsigned nr_events
, aio_context_t __user
*ctxp
)
1283 struct kioctx
*ioctx
= NULL
;
1287 ret
= get_user(ctx
, ctxp
);
1292 if (unlikely(ctx
|| (int)nr_events
<= 0)) {
1293 pr_debug("EINVAL: io_setup: ctx or nr_events > max\n");
1297 ioctx
= ioctx_alloc(nr_events
);
1298 ret
= PTR_ERR(ioctx
);
1299 if (!IS_ERR(ioctx
)) {
1300 ret
= put_user(ioctx
->user_id
, ctxp
);
1304 get_ioctx(ioctx
); /* io_destroy() expects us to hold a ref */
1313 * Destroy the aio_context specified. May cancel any outstanding
1314 * AIOs and block on completion. Will fail with -ENOSYS if not
1315 * implemented. May fail with -EFAULT if the context pointed to
1318 asmlinkage
long sys_io_destroy(aio_context_t ctx
)
1320 struct kioctx
*ioctx
= lookup_ioctx(ctx
);
1321 if (likely(NULL
!= ioctx
)) {
1325 pr_debug("EINVAL: io_destroy: invalid context id\n");
1330 * Default retry method for aio_read (also used for first time submit)
1331 * Responsible for updating iocb state as retries progress
1333 static ssize_t
aio_pread(struct kiocb
*iocb
)
1335 struct file
*file
= iocb
->ki_filp
;
1336 struct address_space
*mapping
= file
->f_mapping
;
1337 struct inode
*inode
= mapping
->host
;
1340 ret
= file
->f_op
->aio_read(iocb
, iocb
->ki_buf
,
1341 iocb
->ki_left
, iocb
->ki_pos
);
1344 * Can't just depend on iocb->ki_left to determine
1345 * whether we are done. This may have been a short read.
1348 iocb
->ki_buf
+= ret
;
1349 iocb
->ki_left
-= ret
;
1351 * For pipes and sockets we return once we have
1352 * some data; for regular files we retry till we
1353 * complete the entire read or find that we can't
1354 * read any more data (e.g short reads).
1356 if (!S_ISFIFO(inode
->i_mode
) && !S_ISSOCK(inode
->i_mode
))
1360 /* This means we must have transferred all that we could */
1361 /* No need to retry anymore */
1362 if ((ret
== 0) || (iocb
->ki_left
== 0))
1363 ret
= iocb
->ki_nbytes
- iocb
->ki_left
;
1369 * Default retry method for aio_write (also used for first time submit)
1370 * Responsible for updating iocb state as retries progress
1372 static ssize_t
aio_pwrite(struct kiocb
*iocb
)
1374 struct file
*file
= iocb
->ki_filp
;
1377 ret
= file
->f_op
->aio_write(iocb
, iocb
->ki_buf
,
1378 iocb
->ki_left
, iocb
->ki_pos
);
1381 iocb
->ki_buf
+= ret
;
1382 iocb
->ki_left
-= ret
;
1387 /* This means we must have transferred all that we could */
1388 /* No need to retry anymore */
1389 if ((ret
== 0) || (iocb
->ki_left
== 0))
1390 ret
= iocb
->ki_nbytes
- iocb
->ki_left
;
1395 static ssize_t
aio_fdsync(struct kiocb
*iocb
)
1397 struct file
*file
= iocb
->ki_filp
;
1398 ssize_t ret
= -EINVAL
;
1400 if (file
->f_op
->aio_fsync
)
1401 ret
= file
->f_op
->aio_fsync(iocb
, 1);
1405 static ssize_t
aio_fsync(struct kiocb
*iocb
)
1407 struct file
*file
= iocb
->ki_filp
;
1408 ssize_t ret
= -EINVAL
;
1410 if (file
->f_op
->aio_fsync
)
1411 ret
= file
->f_op
->aio_fsync(iocb
, 0);
1417 * Performs the initial checks and aio retry method
1418 * setup for the kiocb at the time of io submission.
1420 static ssize_t
aio_setup_iocb(struct kiocb
*kiocb
)
1422 struct file
*file
= kiocb
->ki_filp
;
1425 switch (kiocb
->ki_opcode
) {
1426 case IOCB_CMD_PREAD
:
1428 if (unlikely(!(file
->f_mode
& FMODE_READ
)))
1431 if (unlikely(!access_ok(VERIFY_WRITE
, kiocb
->ki_buf
,
1435 if (file
->f_op
->aio_read
)
1436 kiocb
->ki_retry
= aio_pread
;
1438 case IOCB_CMD_PWRITE
:
1440 if (unlikely(!(file
->f_mode
& FMODE_WRITE
)))
1443 if (unlikely(!access_ok(VERIFY_READ
, kiocb
->ki_buf
,
1447 if (file
->f_op
->aio_write
)
1448 kiocb
->ki_retry
= aio_pwrite
;
1450 case IOCB_CMD_FDSYNC
:
1452 if (file
->f_op
->aio_fsync
)
1453 kiocb
->ki_retry
= aio_fdsync
;
1455 case IOCB_CMD_FSYNC
:
1457 if (file
->f_op
->aio_fsync
)
1458 kiocb
->ki_retry
= aio_fsync
;
1461 dprintk("EINVAL: io_submit: no operation provided\n");
1465 if (!kiocb
->ki_retry
)
1472 * aio_wake_function:
1473 * wait queue callback function for aio notification,
1474 * Simply triggers a retry of the operation via kick_iocb.
1476 * This callback is specified in the wait queue entry in
1477 * a kiocb (current->io_wait points to this wait queue
1478 * entry when an aio operation executes; it is used
1479 * instead of a synchronous wait when an i/o blocking
1480 * condition is encountered during aio).
1483 * This routine is executed with the wait queue lock held.
1484 * Since kick_iocb acquires iocb->ctx->ctx_lock, it nests
1485 * the ioctx lock inside the wait queue lock. This is safe
1486 * because this callback isn't used for wait queues which
1487 * are nested inside ioctx lock (i.e. ctx->wait)
1489 static int aio_wake_function(wait_queue_t
*wait
, unsigned mode
,
1490 int sync
, void *key
)
1492 struct kiocb
*iocb
= container_of(wait
, struct kiocb
, ki_wait
);
1494 list_del_init(&wait
->task_list
);
1499 int fastcall
io_submit_one(struct kioctx
*ctx
, struct iocb __user
*user_iocb
,
1506 /* enforce forwards compatibility on users */
1507 if (unlikely(iocb
->aio_reserved1
|| iocb
->aio_reserved2
||
1508 iocb
->aio_reserved3
)) {
1509 pr_debug("EINVAL: io_submit: reserve field set\n");
1513 /* prevent overflows */
1515 (iocb
->aio_buf
!= (unsigned long)iocb
->aio_buf
) ||
1516 (iocb
->aio_nbytes
!= (size_t)iocb
->aio_nbytes
) ||
1517 ((ssize_t
)iocb
->aio_nbytes
< 0)
1519 pr_debug("EINVAL: io_submit: overflow check\n");
1523 file
= fget(iocb
->aio_fildes
);
1524 if (unlikely(!file
))
1527 req
= aio_get_req(ctx
); /* returns with 2 references to req */
1528 if (unlikely(!req
)) {
1533 req
->ki_filp
= file
;
1534 ret
= put_user(req
->ki_key
, &user_iocb
->aio_key
);
1535 if (unlikely(ret
)) {
1536 dprintk("EFAULT: aio_key\n");
1540 req
->ki_obj
.user
= user_iocb
;
1541 req
->ki_user_data
= iocb
->aio_data
;
1542 req
->ki_pos
= iocb
->aio_offset
;
1544 req
->ki_buf
= (char __user
*)(unsigned long)iocb
->aio_buf
;
1545 req
->ki_left
= req
->ki_nbytes
= iocb
->aio_nbytes
;
1546 req
->ki_opcode
= iocb
->aio_lio_opcode
;
1547 init_waitqueue_func_entry(&req
->ki_wait
, aio_wake_function
);
1548 INIT_LIST_HEAD(&req
->ki_wait
.task_list
);
1549 req
->ki_retried
= 0;
1551 ret
= aio_setup_iocb(req
);
1556 spin_lock_irq(&ctx
->ctx_lock
);
1559 if (!list_empty(&ctx
->run_list
)) {
1560 /* drain the run list */
1561 while (__aio_run_iocbs(ctx
))
1564 spin_unlock_irq(&ctx
->ctx_lock
);
1565 aio_put_req(req
); /* drop extra ref to req */
1569 aio_put_req(req
); /* drop extra ref to req */
1570 aio_put_req(req
); /* drop i/o ref to req */
1575 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1576 * the number of iocbs queued. May return -EINVAL if the aio_context
1577 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1578 * *iocbpp[0] is not properly initialized, if the operation specified
1579 * is invalid for the file descriptor in the iocb. May fail with
1580 * -EFAULT if any of the data structures point to invalid data. May
1581 * fail with -EBADF if the file descriptor specified in the first
1582 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1583 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1584 * fail with -ENOSYS if not implemented.
1586 asmlinkage
long sys_io_submit(aio_context_t ctx_id
, long nr
,
1587 struct iocb __user
* __user
*iocbpp
)
1593 if (unlikely(nr
< 0))
1596 if (unlikely(!access_ok(VERIFY_READ
, iocbpp
, (nr
*sizeof(*iocbpp
)))))
1599 ctx
= lookup_ioctx(ctx_id
);
1600 if (unlikely(!ctx
)) {
1601 pr_debug("EINVAL: io_submit: invalid context id\n");
1606 * AKPM: should this return a partial result if some of the IOs were
1607 * successfully submitted?
1609 for (i
=0; i
<nr
; i
++) {
1610 struct iocb __user
*user_iocb
;
1613 if (unlikely(__get_user(user_iocb
, iocbpp
+ i
))) {
1618 if (unlikely(copy_from_user(&tmp
, user_iocb
, sizeof(tmp
)))) {
1623 ret
= io_submit_one(ctx
, user_iocb
, &tmp
);
1633 * Finds a given iocb for cancellation.
1634 * MUST be called with ctx->ctx_lock held.
1636 static struct kiocb
*lookup_kiocb(struct kioctx
*ctx
, struct iocb __user
*iocb
,
1639 struct list_head
*pos
;
1640 /* TODO: use a hash or array, this sucks. */
1641 list_for_each(pos
, &ctx
->active_reqs
) {
1642 struct kiocb
*kiocb
= list_kiocb(pos
);
1643 if (kiocb
->ki_obj
.user
== iocb
&& kiocb
->ki_key
== key
)
1650 * Attempts to cancel an iocb previously passed to io_submit. If
1651 * the operation is successfully cancelled, the resulting event is
1652 * copied into the memory pointed to by result without being placed
1653 * into the completion queue and 0 is returned. May fail with
1654 * -EFAULT if any of the data structures pointed to are invalid.
1655 * May fail with -EINVAL if aio_context specified by ctx_id is
1656 * invalid. May fail with -EAGAIN if the iocb specified was not
1657 * cancelled. Will fail with -ENOSYS if not implemented.
1659 asmlinkage
long sys_io_cancel(aio_context_t ctx_id
, struct iocb __user
*iocb
,
1660 struct io_event __user
*result
)
1662 int (*cancel
)(struct kiocb
*iocb
, struct io_event
*res
);
1664 struct kiocb
*kiocb
;
1668 ret
= get_user(key
, &iocb
->aio_key
);
1672 ctx
= lookup_ioctx(ctx_id
);
1676 spin_lock_irq(&ctx
->ctx_lock
);
1678 kiocb
= lookup_kiocb(ctx
, iocb
, key
);
1679 if (kiocb
&& kiocb
->ki_cancel
) {
1680 cancel
= kiocb
->ki_cancel
;
1682 kiocbSetCancelled(kiocb
);
1685 spin_unlock_irq(&ctx
->ctx_lock
);
1687 if (NULL
!= cancel
) {
1688 struct io_event tmp
;
1689 pr_debug("calling cancel\n");
1691 memset(&tmp
, 0, sizeof(tmp
));
1692 tmp
.obj
= (u64
)(unsigned long)kiocb
->ki_obj
.user
;
1693 tmp
.data
= kiocb
->ki_user_data
;
1694 ret
= cancel(kiocb
, &tmp
);
1696 /* Cancellation succeeded -- copy the result
1697 * into the user's buffer.
1699 if (copy_to_user(result
, &tmp
, sizeof(tmp
)))
1702 unlock_kiocb(kiocb
);
1712 * Attempts to read at least min_nr events and up to nr events from
1713 * the completion queue for the aio_context specified by ctx_id. May
1714 * fail with -EINVAL if ctx_id is invalid, if min_nr is out of range,
1715 * if nr is out of range, if when is out of range. May fail with
1716 * -EFAULT if any of the memory specified to is invalid. May return
1717 * 0 or < min_nr if no events are available and the timeout specified
1718 * by when has elapsed, where when == NULL specifies an infinite
1719 * timeout. Note that the timeout pointed to by when is relative and
1720 * will be updated if not NULL and the operation blocks. Will fail
1721 * with -ENOSYS if not implemented.
1723 asmlinkage
long sys_io_getevents(aio_context_t ctx_id
,
1726 struct io_event __user
*events
,
1727 struct timespec __user
*timeout
)
1729 struct kioctx
*ioctx
= lookup_ioctx(ctx_id
);
1732 if (likely(ioctx
)) {
1733 if (likely(min_nr
<= nr
&& min_nr
>= 0 && nr
>= 0))
1734 ret
= read_events(ioctx
, min_nr
, nr
, events
, timeout
);
1741 __initcall(aio_setup
);
1743 EXPORT_SYMBOL(aio_complete
);
1744 EXPORT_SYMBOL(aio_put_req
);
1745 EXPORT_SYMBOL(wait_on_sync_kiocb
);