1
2
3
4
5
6
7
8
9
10
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/backing-dev.h>
19#include <linux/uio.h>
20
21#define DEBUG 0
22
23#include <linux/sched.h>
24#include <linux/fs.h>
25#include <linux/file.h>
26#include <linux/mm.h>
27#include <linux/mman.h>
28#include <linux/mmu_context.h>
29#include <linux/slab.h>
30#include <linux/timer.h>
31#include <linux/aio.h>
32#include <linux/highmem.h>
33#include <linux/workqueue.h>
34#include <linux/security.h>
35#include <linux/eventfd.h>
36#include <linux/blkdev.h>
37#include <linux/mempool.h>
38#include <linux/hash.h>
39#include <linux/compat.h>
40
41#include <asm/kmap_types.h>
42#include <asm/uaccess.h>
43
44#if DEBUG > 1
45#define dprintk printk
46#else
47#define dprintk(x...) do { ; } while (0)
48#endif
49
50
51static DEFINE_SPINLOCK(aio_nr_lock);
52unsigned long aio_nr;
53unsigned long aio_max_nr = 0x10000;
54
55
56static struct kmem_cache *kiocb_cachep;
57static struct kmem_cache *kioctx_cachep;
58
59static struct workqueue_struct *aio_wq;
60
61
62static void aio_fput_routine(struct work_struct *);
63static DECLARE_WORK(fput_work, aio_fput_routine);
64
65static DEFINE_SPINLOCK(fput_lock);
66static LIST_HEAD(fput_head);
67
68#define AIO_BATCH_HASH_BITS 3
69#define AIO_BATCH_HASH_SIZE (1 << AIO_BATCH_HASH_BITS)
70struct aio_batch_entry {
71 struct hlist_node list;
72 struct address_space *mapping;
73};
74mempool_t *abe_pool;
75
76static void aio_kick_handler(struct work_struct *);
77static void aio_queue_work(struct kioctx *);
78
79
80
81
82
83static int __init aio_setup(void)
84{
85 kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
86 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
87
88 aio_wq = create_workqueue("aio");
89 abe_pool = mempool_create_kmalloc_pool(1, sizeof(struct aio_batch_entry));
90 BUG_ON(!aio_wq || !abe_pool);
91
92 pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page));
93
94 return 0;
95}
96__initcall(aio_setup);
97
98static void aio_free_ring(struct kioctx *ctx)
99{
100 struct aio_ring_info *info = &ctx->ring_info;
101 long i;
102
103 for (i=0; i<info->nr_pages; i++)
104 put_page(info->ring_pages[i]);
105
106 if (info->mmap_size) {
107 down_write(&ctx->mm->mmap_sem);
108 do_munmap(ctx->mm, info->mmap_base, info->mmap_size);
109 up_write(&ctx->mm->mmap_sem);
110 }
111
112 if (info->ring_pages && info->ring_pages != info->internal_pages)
113 kfree(info->ring_pages);
114 info->ring_pages = NULL;
115 info->nr = 0;
116}
117
118static int aio_setup_ring(struct kioctx *ctx)
119{
120 struct aio_ring *ring;
121 struct aio_ring_info *info = &ctx->ring_info;
122 unsigned nr_events = ctx->max_reqs;
123 unsigned long size;
124 int nr_pages;
125
126
127 nr_events += 2;
128
129 size = sizeof(struct aio_ring);
130 size += sizeof(struct io_event) * nr_events;
131 nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT;
132
133 if (nr_pages < 0)
134 return -EINVAL;
135
136 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event);
137
138 info->nr = 0;
139 info->ring_pages = info->internal_pages;
140 if (nr_pages > AIO_RING_PAGES) {
141 info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
142 if (!info->ring_pages)
143 return -ENOMEM;
144 }
145
146 info->mmap_size = nr_pages * PAGE_SIZE;
147 dprintk("attempting mmap of %lu bytes\n", info->mmap_size);
148 down_write(&ctx->mm->mmap_sem);
149 info->mmap_base = do_mmap(NULL, 0, info->mmap_size,
150 PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE,
151 0);
152 if (IS_ERR((void *)info->mmap_base)) {
153 up_write(&ctx->mm->mmap_sem);
154 info->mmap_size = 0;
155 aio_free_ring(ctx);
156 return -EAGAIN;
157 }
158
159 dprintk("mmap address: 0x%08lx\n", info->mmap_base);
160 info->nr_pages = get_user_pages(current, ctx->mm,
161 info->mmap_base, nr_pages,
162 1, 0, info->ring_pages, NULL);
163 up_write(&ctx->mm->mmap_sem);
164
165 if (unlikely(info->nr_pages != nr_pages)) {
166 aio_free_ring(ctx);
167 return -EAGAIN;
168 }
169
170 ctx->user_id = info->mmap_base;
171
172 info->nr = nr_events;
173
174 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
175 ring->nr = nr_events;
176 ring->id = ctx->user_id;
177 ring->head = ring->tail = 0;
178 ring->magic = AIO_RING_MAGIC;
179 ring->compat_features = AIO_RING_COMPAT_FEATURES;
180 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
181 ring->header_length = sizeof(struct aio_ring);
182 kunmap_atomic(ring, KM_USER0);
183
184 return 0;
185}
186
187
188
189
190
191#define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
192#define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
193#define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
194
195#define aio_ring_event(info, nr, km) ({ \
196 unsigned pos = (nr) + AIO_EVENTS_OFFSET; \
197 struct io_event *__event; \
198 __event = kmap_atomic( \
199 (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE], km); \
200 __event += pos % AIO_EVENTS_PER_PAGE; \
201 __event; \
202})
203
204#define put_aio_ring_event(event, km) do { \
205 struct io_event *__event = (event); \
206 (void)__event; \
207 kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \
208} while(0)
209
210static void ctx_rcu_free(struct rcu_head *head)
211{
212 struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
213 unsigned nr_events = ctx->max_reqs;
214
215 kmem_cache_free(kioctx_cachep, ctx);
216
217 if (nr_events) {
218 spin_lock(&aio_nr_lock);
219 BUG_ON(aio_nr - nr_events > aio_nr);
220 aio_nr -= nr_events;
221 spin_unlock(&aio_nr_lock);
222 }
223}
224
225
226
227
228
229static void __put_ioctx(struct kioctx *ctx)
230{
231 BUG_ON(ctx->reqs_active);
232
233 cancel_delayed_work(&ctx->wq);
234 cancel_work_sync(&ctx->wq.work);
235 aio_free_ring(ctx);
236 mmdrop(ctx->mm);
237 ctx->mm = NULL;
238 pr_debug("__put_ioctx: freeing %p\n", ctx);
239 call_rcu(&ctx->rcu_head, ctx_rcu_free);
240}
241
242static inline void get_ioctx(struct kioctx *kioctx)
243{
244 BUG_ON(atomic_read(&kioctx->users) <= 0);
245 atomic_inc(&kioctx->users);
246}
247
248static inline int try_get_ioctx(struct kioctx *kioctx)
249{
250 return atomic_inc_not_zero(&kioctx->users);
251}
252
253static inline void put_ioctx(struct kioctx *kioctx)
254{
255 BUG_ON(atomic_read(&kioctx->users) <= 0);
256 if (unlikely(atomic_dec_and_test(&kioctx->users)))
257 __put_ioctx(kioctx);
258}
259
260
261
262
263static struct kioctx *ioctx_alloc(unsigned nr_events)
264{
265 struct mm_struct *mm;
266 struct kioctx *ctx;
267 int did_sync = 0;
268
269
270 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
271 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
272 pr_debug("ENOMEM: nr_events too high\n");
273 return ERR_PTR(-EINVAL);
274 }
275
276 if ((unsigned long)nr_events > aio_max_nr)
277 return ERR_PTR(-EAGAIN);
278
279 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
280 if (!ctx)
281 return ERR_PTR(-ENOMEM);
282
283 ctx->max_reqs = nr_events;
284 mm = ctx->mm = current->mm;
285 atomic_inc(&mm->mm_count);
286
287 atomic_set(&ctx->users, 1);
288 spin_lock_init(&ctx->ctx_lock);
289 spin_lock_init(&ctx->ring_info.ring_lock);
290 init_waitqueue_head(&ctx->wait);
291
292 INIT_LIST_HEAD(&ctx->active_reqs);
293 INIT_LIST_HEAD(&ctx->run_list);
294 INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler);
295
296 if (aio_setup_ring(ctx) < 0)
297 goto out_freectx;
298
299
300 do {
301 spin_lock_bh(&aio_nr_lock);
302 if (aio_nr + nr_events > aio_max_nr ||
303 aio_nr + nr_events < aio_nr)
304 ctx->max_reqs = 0;
305 else
306 aio_nr += ctx->max_reqs;
307 spin_unlock_bh(&aio_nr_lock);
308 if (ctx->max_reqs || did_sync)
309 break;
310
311
312 synchronize_rcu();
313 did_sync = 1;
314 ctx->max_reqs = nr_events;
315 } while (1);
316
317 if (ctx->max_reqs == 0)
318 goto out_cleanup;
319
320
321 spin_lock(&mm->ioctx_lock);
322 hlist_add_head_rcu(&ctx->list, &mm->ioctx_list);
323 spin_unlock(&mm->ioctx_lock);
324
325 dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
326 ctx, ctx->user_id, current->mm, ctx->ring_info.nr);
327 return ctx;
328
329out_cleanup:
330 __put_ioctx(ctx);
331 return ERR_PTR(-EAGAIN);
332
333out_freectx:
334 mmdrop(mm);
335 kmem_cache_free(kioctx_cachep, ctx);
336 ctx = ERR_PTR(-ENOMEM);
337
338 dprintk("aio: error allocating ioctx %p\n", ctx);
339 return ctx;
340}
341
342
343
344
345
346
347static void aio_cancel_all(struct kioctx *ctx)
348{
349 int (*cancel)(struct kiocb *, struct io_event *);
350 struct io_event res;
351 spin_lock_irq(&ctx->ctx_lock);
352 ctx->dead = 1;
353 while (!list_empty(&ctx->active_reqs)) {
354 struct list_head *pos = ctx->active_reqs.next;
355 struct kiocb *iocb = list_kiocb(pos);
356 list_del_init(&iocb->ki_list);
357 cancel = iocb->ki_cancel;
358 kiocbSetCancelled(iocb);
359 if (cancel) {
360 iocb->ki_users++;
361 spin_unlock_irq(&ctx->ctx_lock);
362 cancel(iocb, &res);
363 spin_lock_irq(&ctx->ctx_lock);
364 }
365 }
366 spin_unlock_irq(&ctx->ctx_lock);
367}
368
369static void wait_for_all_aios(struct kioctx *ctx)
370{
371 struct task_struct *tsk = current;
372 DECLARE_WAITQUEUE(wait, tsk);
373
374 spin_lock_irq(&ctx->ctx_lock);
375 if (!ctx->reqs_active)
376 goto out;
377
378 add_wait_queue(&ctx->wait, &wait);
379 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
380 while (ctx->reqs_active) {
381 spin_unlock_irq(&ctx->ctx_lock);
382 io_schedule();
383 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
384 spin_lock_irq(&ctx->ctx_lock);
385 }
386 __set_task_state(tsk, TASK_RUNNING);
387 remove_wait_queue(&ctx->wait, &wait);
388
389out:
390 spin_unlock_irq(&ctx->ctx_lock);
391}
392
393
394
395
396ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
397{
398 while (iocb->ki_users) {
399 set_current_state(TASK_UNINTERRUPTIBLE);
400 if (!iocb->ki_users)
401 break;
402 io_schedule();
403 }
404 __set_current_state(TASK_RUNNING);
405 return iocb->ki_user_data;
406}
407EXPORT_SYMBOL(wait_on_sync_kiocb);
408
409
410
411
412
413
414
415
416void exit_aio(struct mm_struct *mm)
417{
418 struct kioctx *ctx;
419
420 while (!hlist_empty(&mm->ioctx_list)) {
421 ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list);
422 hlist_del_rcu(&ctx->list);
423
424 aio_cancel_all(ctx);
425
426 wait_for_all_aios(ctx);
427
428
429
430 cancel_work_sync(&ctx->wq.work);
431
432 if (1 != atomic_read(&ctx->users))
433 printk(KERN_DEBUG
434 "exit_aio:ioctx still alive: %d %d %d\n",
435 atomic_read(&ctx->users), ctx->dead,
436 ctx->reqs_active);
437 put_ioctx(ctx);
438 }
439}
440
441
442
443
444
445
446
447
448
449
450
451static struct kiocb *__aio_get_req(struct kioctx *ctx)
452{
453 struct kiocb *req = NULL;
454 struct aio_ring *ring;
455 int okay = 0;
456
457 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL);
458 if (unlikely(!req))
459 return NULL;
460
461 req->ki_flags = 0;
462 req->ki_users = 2;
463 req->ki_key = 0;
464 req->ki_ctx = ctx;
465 req->ki_cancel = NULL;
466 req->ki_retry = NULL;
467 req->ki_dtor = NULL;
468 req->private = NULL;
469 req->ki_iovec = NULL;
470 INIT_LIST_HEAD(&req->ki_run_list);
471 req->ki_eventfd = NULL;
472
473
474
475
476 spin_lock_irq(&ctx->ctx_lock);
477 ring = kmap_atomic(ctx->ring_info.ring_pages[0], KM_USER0);
478 if (ctx->reqs_active < aio_ring_avail(&ctx->ring_info, ring)) {
479 list_add(&req->ki_list, &ctx->active_reqs);
480 ctx->reqs_active++;
481 okay = 1;
482 }
483 kunmap_atomic(ring, KM_USER0);
484 spin_unlock_irq(&ctx->ctx_lock);
485
486 if (!okay) {
487 kmem_cache_free(kiocb_cachep, req);
488 req = NULL;
489 }
490
491 return req;
492}
493
494static inline struct kiocb *aio_get_req(struct kioctx *ctx)
495{
496 struct kiocb *req;
497
498
499
500
501 req = __aio_get_req(ctx);
502 if (unlikely(NULL == req)) {
503 aio_fput_routine(NULL);
504 req = __aio_get_req(ctx);
505 }
506 return req;
507}
508
509static inline void really_put_req(struct kioctx *ctx, struct kiocb *req)
510{
511 assert_spin_locked(&ctx->ctx_lock);
512
513 if (req->ki_eventfd != NULL)
514 eventfd_ctx_put(req->ki_eventfd);
515 if (req->ki_dtor)
516 req->ki_dtor(req);
517 if (req->ki_iovec != &req->ki_inline_vec)
518 kfree(req->ki_iovec);
519 kmem_cache_free(kiocb_cachep, req);
520 ctx->reqs_active--;
521
522 if (unlikely(!ctx->reqs_active && ctx->dead))
523 wake_up(&ctx->wait);
524}
525
526static void aio_fput_routine(struct work_struct *data)
527{
528 spin_lock_irq(&fput_lock);
529 while (likely(!list_empty(&fput_head))) {
530 struct kiocb *req = list_kiocb(fput_head.next);
531 struct kioctx *ctx = req->ki_ctx;
532
533 list_del(&req->ki_list);
534 spin_unlock_irq(&fput_lock);
535
536
537 if (req->ki_filp != NULL)
538 fput(req->ki_filp);
539
540
541 spin_lock_irq(&ctx->ctx_lock);
542 really_put_req(ctx, req);
543 spin_unlock_irq(&ctx->ctx_lock);
544
545 put_ioctx(ctx);
546 spin_lock_irq(&fput_lock);
547 }
548 spin_unlock_irq(&fput_lock);
549}
550
551
552
553
554static int __aio_put_req(struct kioctx *ctx, struct kiocb *req)
555{
556 dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n",
557 req, atomic_long_read(&req->ki_filp->f_count));
558
559 assert_spin_locked(&ctx->ctx_lock);
560
561 req->ki_users--;
562 BUG_ON(req->ki_users < 0);
563 if (likely(req->ki_users))
564 return 0;
565 list_del(&req->ki_list);
566 req->ki_cancel = NULL;
567 req->ki_retry = NULL;
568
569
570
571
572
573
574
575 if (unlikely(!fput_atomic(req->ki_filp))) {
576 get_ioctx(ctx);
577 spin_lock(&fput_lock);
578 list_add(&req->ki_list, &fput_head);
579 spin_unlock(&fput_lock);
580 queue_work(aio_wq, &fput_work);
581 } else {
582 req->ki_filp = NULL;
583 really_put_req(ctx, req);
584 }
585 return 1;
586}
587
588
589
590
591
592int aio_put_req(struct kiocb *req)
593{
594 struct kioctx *ctx = req->ki_ctx;
595 int ret;
596 spin_lock_irq(&ctx->ctx_lock);
597 ret = __aio_put_req(ctx, req);
598 spin_unlock_irq(&ctx->ctx_lock);
599 return ret;
600}
601EXPORT_SYMBOL(aio_put_req);
602
603static struct kioctx *lookup_ioctx(unsigned long ctx_id)
604{
605 struct mm_struct *mm = current->mm;
606 struct kioctx *ctx, *ret = NULL;
607 struct hlist_node *n;
608
609 rcu_read_lock();
610
611 hlist_for_each_entry_rcu(ctx, n, &mm->ioctx_list, list) {
612
613
614
615
616
617
618 if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){
619 ret = ctx;
620 break;
621 }
622 }
623
624 rcu_read_unlock();
625 return ret;
626}
627
628
629
630
631
632
633
634
635
636
637static inline int __queue_kicked_iocb(struct kiocb *iocb)
638{
639 struct kioctx *ctx = iocb->ki_ctx;
640
641 assert_spin_locked(&ctx->ctx_lock);
642
643 if (list_empty(&iocb->ki_run_list)) {
644 list_add_tail(&iocb->ki_run_list,
645 &ctx->run_list);
646 return 1;
647 }
648 return 0;
649}
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673static ssize_t aio_run_iocb(struct kiocb *iocb)
674{
675 struct kioctx *ctx = iocb->ki_ctx;
676 ssize_t (*retry)(struct kiocb *);
677 ssize_t ret;
678
679 if (!(retry = iocb->ki_retry)) {
680 printk("aio_run_iocb: iocb->ki_retry = NULL\n");
681 return 0;
682 }
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703 kiocbClearKicked(iocb);
704
705
706
707
708
709
710
711 iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL;
712 spin_unlock_irq(&ctx->ctx_lock);
713
714
715 if (kiocbIsCancelled(iocb)) {
716 ret = -EINTR;
717 aio_complete(iocb, ret, 0);
718
719 goto out;
720 }
721
722
723
724
725
726 ret = retry(iocb);
727
728 if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) {
729
730
731
732
733 if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
734 ret == -ERESTARTNOHAND || ret == -ERESTART_RESTARTBLOCK))
735 ret = -EINTR;
736 aio_complete(iocb, ret, 0);
737 }
738out:
739 spin_lock_irq(&ctx->ctx_lock);
740
741 if (-EIOCBRETRY == ret) {
742
743
744
745
746
747
748
749
750 INIT_LIST_HEAD(&iocb->ki_run_list);
751
752
753 if (kiocbIsKicked(iocb)) {
754 __queue_kicked_iocb(iocb);
755
756
757
758
759
760
761
762 aio_queue_work(ctx);
763 }
764 }
765 return ret;
766}
767
768
769
770
771
772
773
774
775static int __aio_run_iocbs(struct kioctx *ctx)
776{
777 struct kiocb *iocb;
778 struct list_head run_list;
779
780 assert_spin_locked(&ctx->ctx_lock);
781
782 list_replace_init(&ctx->run_list, &run_list);
783 while (!list_empty(&run_list)) {
784 iocb = list_entry(run_list.next, struct kiocb,
785 ki_run_list);
786 list_del(&iocb->ki_run_list);
787
788
789
790 iocb->ki_users++;
791 aio_run_iocb(iocb);
792 __aio_put_req(ctx, iocb);
793 }
794 if (!list_empty(&ctx->run_list))
795 return 1;
796 return 0;
797}
798
799static void aio_queue_work(struct kioctx * ctx)
800{
801 unsigned long timeout;
802
803
804
805
806 smp_mb();
807 if (waitqueue_active(&ctx->wait))
808 timeout = 1;
809 else
810 timeout = HZ/10;
811 queue_delayed_work(aio_wq, &ctx->wq, timeout);
812}
813
814
815
816
817
818
819
820
821static inline void aio_run_all_iocbs(struct kioctx *ctx)
822{
823 spin_lock_irq(&ctx->ctx_lock);
824 while (__aio_run_iocbs(ctx))
825 ;
826 spin_unlock_irq(&ctx->ctx_lock);
827}
828
829
830
831
832
833
834
835
836
837
838static void aio_kick_handler(struct work_struct *work)
839{
840 struct kioctx *ctx = container_of(work, struct kioctx, wq.work);
841 mm_segment_t oldfs = get_fs();
842 struct mm_struct *mm;
843 int requeue;
844
845 set_fs(USER_DS);
846 use_mm(ctx->mm);
847 spin_lock_irq(&ctx->ctx_lock);
848 requeue =__aio_run_iocbs(ctx);
849 mm = ctx->mm;
850 spin_unlock_irq(&ctx->ctx_lock);
851 unuse_mm(mm);
852 set_fs(oldfs);
853
854
855
856 if (requeue)
857 queue_delayed_work(aio_wq, &ctx->wq, 0);
858}
859
860
861
862
863
864
865
866static void try_queue_kicked_iocb(struct kiocb *iocb)
867{
868 struct kioctx *ctx = iocb->ki_ctx;
869 unsigned long flags;
870 int run = 0;
871
872 spin_lock_irqsave(&ctx->ctx_lock, flags);
873
874
875 if (!kiocbTryKick(iocb))
876 run = __queue_kicked_iocb(iocb);
877 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
878 if (run)
879 aio_queue_work(ctx);
880}
881
882
883
884
885
886
887
888
889void kick_iocb(struct kiocb *iocb)
890{
891
892
893 if (is_sync_kiocb(iocb)) {
894 kiocbSetKicked(iocb);
895 wake_up_process(iocb->ki_obj.tsk);
896 return;
897 }
898
899 try_queue_kicked_iocb(iocb);
900}
901EXPORT_SYMBOL(kick_iocb);
902
903
904
905
906
907
908int aio_complete(struct kiocb *iocb, long res, long res2)
909{
910 struct kioctx *ctx = iocb->ki_ctx;
911 struct aio_ring_info *info;
912 struct aio_ring *ring;
913 struct io_event *event;
914 unsigned long flags;
915 unsigned long tail;
916 int ret;
917
918
919
920
921
922
923
924
925 if (is_sync_kiocb(iocb)) {
926 BUG_ON(iocb->ki_users != 1);
927 iocb->ki_user_data = res;
928 iocb->ki_users = 0;
929 wake_up_process(iocb->ki_obj.tsk);
930 return 1;
931 }
932
933 info = &ctx->ring_info;
934
935
936
937
938
939
940
941 spin_lock_irqsave(&ctx->ctx_lock, flags);
942
943 if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list))
944 list_del_init(&iocb->ki_run_list);
945
946
947
948
949
950 if (kiocbIsCancelled(iocb))
951 goto put_rq;
952
953 ring = kmap_atomic(info->ring_pages[0], KM_IRQ1);
954
955 tail = info->tail;
956 event = aio_ring_event(info, tail, KM_IRQ0);
957 if (++tail >= info->nr)
958 tail = 0;
959
960 event->obj = (u64)(unsigned long)iocb->ki_obj.user;
961 event->data = iocb->ki_user_data;
962 event->res = res;
963 event->res2 = res2;
964
965 dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n",
966 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
967 res, res2);
968
969
970
971
972 smp_wmb();
973
974 info->tail = tail;
975 ring->tail = tail;
976
977 put_aio_ring_event(event, KM_IRQ0);
978 kunmap_atomic(ring, KM_IRQ1);
979
980 pr_debug("added to ring %p at [%lu]\n", iocb, tail);
981
982
983
984
985
986
987 if (iocb->ki_eventfd != NULL)
988 eventfd_signal(iocb->ki_eventfd, 1);
989
990put_rq:
991
992 ret = __aio_put_req(ctx, iocb);
993
994
995
996
997
998
999
1000 smp_mb();
1001
1002 if (waitqueue_active(&ctx->wait))
1003 wake_up(&ctx->wait);
1004
1005 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
1006 return ret;
1007}
1008EXPORT_SYMBOL(aio_complete);
1009
1010
1011
1012
1013
1014
1015
1016static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent)
1017{
1018 struct aio_ring_info *info = &ioctx->ring_info;
1019 struct aio_ring *ring;
1020 unsigned long head;
1021 int ret = 0;
1022
1023 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
1024 dprintk("in aio_read_evt h%lu t%lu m%lu\n",
1025 (unsigned long)ring->head, (unsigned long)ring->tail,
1026 (unsigned long)ring->nr);
1027
1028 if (ring->head == ring->tail)
1029 goto out;
1030
1031 spin_lock(&info->ring_lock);
1032
1033 head = ring->head % info->nr;
1034 if (head != ring->tail) {
1035 struct io_event *evp = aio_ring_event(info, head, KM_USER1);
1036 *ent = *evp;
1037 head = (head + 1) % info->nr;
1038 smp_mb();
1039 ring->head = head;
1040 ret = 1;
1041 put_aio_ring_event(evp, KM_USER1);
1042 }
1043 spin_unlock(&info->ring_lock);
1044
1045out:
1046 kunmap_atomic(ring, KM_USER0);
1047 dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret,
1048 (unsigned long)ring->head, (unsigned long)ring->tail);
1049 return ret;
1050}
1051
1052struct aio_timeout {
1053 struct timer_list timer;
1054 int timed_out;
1055 struct task_struct *p;
1056};
1057
1058static void timeout_func(unsigned long data)
1059{
1060 struct aio_timeout *to = (struct aio_timeout *)data;
1061
1062 to->timed_out = 1;
1063 wake_up_process(to->p);
1064}
1065
1066static inline void init_timeout(struct aio_timeout *to)
1067{
1068 setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to);
1069 to->timed_out = 0;
1070 to->p = current;
1071}
1072
1073static inline void set_timeout(long start_jiffies, struct aio_timeout *to,
1074 const struct timespec *ts)
1075{
1076 to->timer.expires = start_jiffies + timespec_to_jiffies(ts);
1077 if (time_after(to->timer.expires, jiffies))
1078 add_timer(&to->timer);
1079 else
1080 to->timed_out = 1;
1081}
1082
1083static inline void clear_timeout(struct aio_timeout *to)
1084{
1085 del_singleshot_timer_sync(&to->timer);
1086}
1087
1088static int read_events(struct kioctx *ctx,
1089 long min_nr, long nr,
1090 struct io_event __user *event,
1091 struct timespec __user *timeout)
1092{
1093 long start_jiffies = jiffies;
1094 struct task_struct *tsk = current;
1095 DECLARE_WAITQUEUE(wait, tsk);
1096 int ret;
1097 int i = 0;
1098 struct io_event ent;
1099 struct aio_timeout to;
1100 int retry = 0;
1101
1102
1103
1104
1105 memset(&ent, 0, sizeof(ent));
1106retry:
1107 ret = 0;
1108 while (likely(i < nr)) {
1109 ret = aio_read_evt(ctx, &ent);
1110 if (unlikely(ret <= 0))
1111 break;
1112
1113 dprintk("read event: %Lx %Lx %Lx %Lx\n",
1114 ent.data, ent.obj, ent.res, ent.res2);
1115
1116
1117 ret = -EFAULT;
1118 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1119 dprintk("aio: lost an event due to EFAULT.\n");
1120 break;
1121 }
1122 ret = 0;
1123
1124
1125 event ++;
1126 i ++;
1127 }
1128
1129 if (min_nr <= i)
1130 return i;
1131 if (ret)
1132 return ret;
1133
1134
1135
1136
1137 if (!retry && unlikely(!list_empty(&ctx->run_list))) {
1138 retry = 1;
1139 aio_run_all_iocbs(ctx);
1140 goto retry;
1141 }
1142
1143 init_timeout(&to);
1144 if (timeout) {
1145 struct timespec ts;
1146 ret = -EFAULT;
1147 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1148 goto out;
1149
1150 set_timeout(start_jiffies, &to, &ts);
1151 }
1152
1153 while (likely(i < nr)) {
1154 add_wait_queue_exclusive(&ctx->wait, &wait);
1155 do {
1156 set_task_state(tsk, TASK_INTERRUPTIBLE);
1157 ret = aio_read_evt(ctx, &ent);
1158 if (ret)
1159 break;
1160 if (min_nr <= i)
1161 break;
1162 if (unlikely(ctx->dead)) {
1163 ret = -EINVAL;
1164 break;
1165 }
1166 if (to.timed_out)
1167 break;
1168
1169
1170 if (ctx->reqs_active)
1171 io_schedule();
1172 else
1173 schedule();
1174 if (signal_pending(tsk)) {
1175 ret = -EINTR;
1176 break;
1177 }
1178
1179 } while (1) ;
1180
1181 set_task_state(tsk, TASK_RUNNING);
1182 remove_wait_queue(&ctx->wait, &wait);
1183
1184 if (unlikely(ret <= 0))
1185 break;
1186
1187 ret = -EFAULT;
1188 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1189 dprintk("aio: lost an event due to EFAULT.\n");
1190 break;
1191 }
1192
1193
1194 event ++;
1195 i ++;
1196 }
1197
1198 if (timeout)
1199 clear_timeout(&to);
1200out:
1201 destroy_timer_on_stack(&to.timer);
1202 return i ? i : ret;
1203}
1204
1205
1206
1207
1208static void io_destroy(struct kioctx *ioctx)
1209{
1210 struct mm_struct *mm = current->mm;
1211 int was_dead;
1212
1213
1214 spin_lock(&mm->ioctx_lock);
1215 was_dead = ioctx->dead;
1216 ioctx->dead = 1;
1217 hlist_del_rcu(&ioctx->list);
1218 spin_unlock(&mm->ioctx_lock);
1219
1220 dprintk("aio_release(%p)\n", ioctx);
1221 if (likely(!was_dead))
1222 put_ioctx(ioctx);
1223
1224 aio_cancel_all(ioctx);
1225 wait_for_all_aios(ioctx);
1226
1227
1228
1229
1230
1231
1232 wake_up(&ioctx->wait);
1233 put_ioctx(ioctx);
1234}
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
1250{
1251 struct kioctx *ioctx = NULL;
1252 unsigned long ctx;
1253 long ret;
1254
1255 ret = get_user(ctx, ctxp);
1256 if (unlikely(ret))
1257 goto out;
1258
1259 ret = -EINVAL;
1260 if (unlikely(ctx || nr_events == 0)) {
1261 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
1262 ctx, nr_events);
1263 goto out;
1264 }
1265
1266 ioctx = ioctx_alloc(nr_events);
1267 ret = PTR_ERR(ioctx);
1268 if (!IS_ERR(ioctx)) {
1269 ret = put_user(ioctx->user_id, ctxp);
1270 if (!ret)
1271 return 0;
1272
1273 get_ioctx(ioctx);
1274 io_destroy(ioctx);
1275 }
1276
1277out:
1278 return ret;
1279}
1280
1281
1282
1283
1284
1285
1286
1287SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
1288{
1289 struct kioctx *ioctx = lookup_ioctx(ctx);
1290 if (likely(NULL != ioctx)) {
1291 io_destroy(ioctx);
1292 return 0;
1293 }
1294 pr_debug("EINVAL: io_destroy: invalid context id\n");
1295 return -EINVAL;
1296}
1297
1298static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
1299{
1300 struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];
1301
1302 BUG_ON(ret <= 0);
1303
1304 while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
1305 ssize_t this = min((ssize_t)iov->iov_len, ret);
1306 iov->iov_base += this;
1307 iov->iov_len -= this;
1308 iocb->ki_left -= this;
1309 ret -= this;
1310 if (iov->iov_len == 0) {
1311 iocb->ki_cur_seg++;
1312 iov++;
1313 }
1314 }
1315
1316
1317
1318 BUG_ON(ret > 0 && iocb->ki_left == 0);
1319}
1320
1321static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
1322{
1323 struct file *file = iocb->ki_filp;
1324 struct address_space *mapping = file->f_mapping;
1325 struct inode *inode = mapping->host;
1326 ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
1327 unsigned long, loff_t);
1328 ssize_t ret = 0;
1329 unsigned short opcode;
1330
1331 if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
1332 (iocb->ki_opcode == IOCB_CMD_PREAD)) {
1333 rw_op = file->f_op->aio_read;
1334 opcode = IOCB_CMD_PREADV;
1335 } else {
1336 rw_op = file->f_op->aio_write;
1337 opcode = IOCB_CMD_PWRITEV;
1338 }
1339
1340
1341 if (iocb->ki_pos < 0)
1342 return -EINVAL;
1343
1344 do {
1345 ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
1346 iocb->ki_nr_segs - iocb->ki_cur_seg,
1347 iocb->ki_pos);
1348 if (ret > 0)
1349 aio_advance_iovec(iocb, ret);
1350
1351
1352
1353 } while (ret > 0 && iocb->ki_left > 0 &&
1354 (opcode == IOCB_CMD_PWRITEV ||
1355 (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
1356
1357
1358
1359 if ((ret == 0) || (iocb->ki_left == 0))
1360 ret = iocb->ki_nbytes - iocb->ki_left;
1361
1362
1363
1364 if (opcode == IOCB_CMD_PWRITEV
1365 && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY
1366 && iocb->ki_nbytes - iocb->ki_left)
1367 ret = iocb->ki_nbytes - iocb->ki_left;
1368
1369 return ret;
1370}
1371
1372static ssize_t aio_fdsync(struct kiocb *iocb)
1373{
1374 struct file *file = iocb->ki_filp;
1375 ssize_t ret = -EINVAL;
1376
1377 if (file->f_op->aio_fsync)
1378 ret = file->f_op->aio_fsync(iocb, 1);
1379 return ret;
1380}
1381
1382static ssize_t aio_fsync(struct kiocb *iocb)
1383{
1384 struct file *file = iocb->ki_filp;
1385 ssize_t ret = -EINVAL;
1386
1387 if (file->f_op->aio_fsync)
1388 ret = file->f_op->aio_fsync(iocb, 0);
1389 return ret;
1390}
1391
1392static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat)
1393{
1394 ssize_t ret;
1395
1396#ifdef CONFIG_COMPAT
1397 if (compat)
1398 ret = compat_rw_copy_check_uvector(type,
1399 (struct compat_iovec __user *)kiocb->ki_buf,
1400 kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
1401 &kiocb->ki_iovec);
1402 else
1403#endif
1404 ret = rw_copy_check_uvector(type,
1405 (struct iovec __user *)kiocb->ki_buf,
1406 kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
1407 &kiocb->ki_iovec);
1408 if (ret < 0)
1409 goto out;
1410
1411 kiocb->ki_nr_segs = kiocb->ki_nbytes;
1412 kiocb->ki_cur_seg = 0;
1413
1414 kiocb->ki_nbytes = ret;
1415 kiocb->ki_left = ret;
1416
1417 ret = 0;
1418out:
1419 return ret;
1420}
1421
1422static ssize_t aio_setup_single_vector(struct kiocb *kiocb)
1423{
1424 kiocb->ki_iovec = &kiocb->ki_inline_vec;
1425 kiocb->ki_iovec->iov_base = kiocb->ki_buf;
1426 kiocb->ki_iovec->iov_len = kiocb->ki_left;
1427 kiocb->ki_nr_segs = 1;
1428 kiocb->ki_cur_seg = 0;
1429 return 0;
1430}
1431
1432
1433
1434
1435
1436
1437static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat)
1438{
1439 struct file *file = kiocb->ki_filp;
1440 ssize_t ret = 0;
1441
1442 switch (kiocb->ki_opcode) {
1443 case IOCB_CMD_PREAD:
1444 ret = -EBADF;
1445 if (unlikely(!(file->f_mode & FMODE_READ)))
1446 break;
1447 ret = -EFAULT;
1448 if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
1449 kiocb->ki_left)))
1450 break;
1451 ret = security_file_permission(file, MAY_READ);
1452 if (unlikely(ret))
1453 break;
1454 ret = aio_setup_single_vector(kiocb);
1455 if (ret)
1456 break;
1457 ret = -EINVAL;
1458 if (file->f_op->aio_read)
1459 kiocb->ki_retry = aio_rw_vect_retry;
1460 break;
1461 case IOCB_CMD_PWRITE:
1462 ret = -EBADF;
1463 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1464 break;
1465 ret = -EFAULT;
1466 if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
1467 kiocb->ki_left)))
1468 break;
1469 ret = security_file_permission(file, MAY_WRITE);
1470 if (unlikely(ret))
1471 break;
1472 ret = aio_setup_single_vector(kiocb);
1473 if (ret)
1474 break;
1475 ret = -EINVAL;
1476 if (file->f_op->aio_write)
1477 kiocb->ki_retry = aio_rw_vect_retry;
1478 break;
1479 case IOCB_CMD_PREADV:
1480 ret = -EBADF;
1481 if (unlikely(!(file->f_mode & FMODE_READ)))
1482 break;
1483 ret = security_file_permission(file, MAY_READ);
1484 if (unlikely(ret))
1485 break;
1486 ret = aio_setup_vectored_rw(READ, kiocb, compat);
1487 if (ret)
1488 break;
1489 ret = -EINVAL;
1490 if (file->f_op->aio_read)
1491 kiocb->ki_retry = aio_rw_vect_retry;
1492 break;
1493 case IOCB_CMD_PWRITEV:
1494 ret = -EBADF;
1495 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1496 break;
1497 ret = security_file_permission(file, MAY_WRITE);
1498 if (unlikely(ret))
1499 break;
1500 ret = aio_setup_vectored_rw(WRITE, kiocb, compat);
1501 if (ret)
1502 break;
1503 ret = -EINVAL;
1504 if (file->f_op->aio_write)
1505 kiocb->ki_retry = aio_rw_vect_retry;
1506 break;
1507 case IOCB_CMD_FDSYNC:
1508 ret = -EINVAL;
1509 if (file->f_op->aio_fsync)
1510 kiocb->ki_retry = aio_fdsync;
1511 break;
1512 case IOCB_CMD_FSYNC:
1513 ret = -EINVAL;
1514 if (file->f_op->aio_fsync)
1515 kiocb->ki_retry = aio_fsync;
1516 break;
1517 default:
1518 dprintk("EINVAL: io_submit: no operation provided\n");
1519 ret = -EINVAL;
1520 }
1521
1522 if (!kiocb->ki_retry)
1523 return ret;
1524
1525 return 0;
1526}
1527
1528static void aio_batch_add(struct address_space *mapping,
1529 struct hlist_head *batch_hash)
1530{
1531 struct aio_batch_entry *abe;
1532 struct hlist_node *pos;
1533 unsigned bucket;
1534
1535 bucket = hash_ptr(mapping, AIO_BATCH_HASH_BITS);
1536 hlist_for_each_entry(abe, pos, &batch_hash[bucket], list) {
1537 if (abe->mapping == mapping)
1538 return;
1539 }
1540
1541 abe = mempool_alloc(abe_pool, GFP_KERNEL);
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554 ihold(mapping->host);
1555 abe->mapping = mapping;
1556 hlist_add_head(&abe->list, &batch_hash[bucket]);
1557 return;
1558}
1559
1560static void aio_batch_free(struct hlist_head *batch_hash)
1561{
1562 struct aio_batch_entry *abe;
1563 struct hlist_node *pos, *n;
1564 int i;
1565
1566 for (i = 0; i < AIO_BATCH_HASH_SIZE; i++) {
1567 hlist_for_each_entry_safe(abe, pos, n, &batch_hash[i], list) {
1568 blk_run_address_space(abe->mapping);
1569 iput(abe->mapping->host);
1570 hlist_del(&abe->list);
1571 mempool_free(abe, abe_pool);
1572 }
1573 }
1574}
1575
1576static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
1577 struct iocb *iocb, struct hlist_head *batch_hash,
1578 bool compat)
1579{
1580 struct kiocb *req;
1581 struct file *file;
1582 ssize_t ret;
1583
1584
1585 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
1586 pr_debug("EINVAL: io_submit: reserve field set\n");
1587 return -EINVAL;
1588 }
1589
1590
1591 if (unlikely(
1592 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1593 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1594 ((ssize_t)iocb->aio_nbytes < 0)
1595 )) {
1596 pr_debug("EINVAL: io_submit: overflow check\n");
1597 return -EINVAL;
1598 }
1599
1600 file = fget(iocb->aio_fildes);
1601 if (unlikely(!file))
1602 return -EBADF;
1603
1604 req = aio_get_req(ctx);
1605 if (unlikely(!req)) {
1606 fput(file);
1607 return -EAGAIN;
1608 }
1609 req->ki_filp = file;
1610 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1611
1612
1613
1614
1615
1616
1617 req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd);
1618 if (IS_ERR(req->ki_eventfd)) {
1619 ret = PTR_ERR(req->ki_eventfd);
1620 req->ki_eventfd = NULL;
1621 goto out_put_req;
1622 }
1623 }
1624
1625 ret = put_user(req->ki_key, &user_iocb->aio_key);
1626 if (unlikely(ret)) {
1627 dprintk("EFAULT: aio_key\n");
1628 goto out_put_req;
1629 }
1630
1631 req->ki_obj.user = user_iocb;
1632 req->ki_user_data = iocb->aio_data;
1633 req->ki_pos = iocb->aio_offset;
1634
1635 req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf;
1636 req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
1637 req->ki_opcode = iocb->aio_lio_opcode;
1638
1639 ret = aio_setup_iocb(req, compat);
1640
1641 if (ret)
1642 goto out_put_req;
1643
1644 spin_lock_irq(&ctx->ctx_lock);
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657 if (ctx->dead) {
1658 spin_unlock_irq(&ctx->ctx_lock);
1659 ret = -EINVAL;
1660 goto out_put_req;
1661 }
1662 aio_run_iocb(req);
1663 if (!list_empty(&ctx->run_list)) {
1664
1665 while (__aio_run_iocbs(ctx))
1666 ;
1667 }
1668 spin_unlock_irq(&ctx->ctx_lock);
1669 if (req->ki_opcode == IOCB_CMD_PREAD ||
1670 req->ki_opcode == IOCB_CMD_PREADV ||
1671 req->ki_opcode == IOCB_CMD_PWRITE ||
1672 req->ki_opcode == IOCB_CMD_PWRITEV)
1673 aio_batch_add(file->f_mapping, batch_hash);
1674
1675 aio_put_req(req);
1676 return 0;
1677
1678out_put_req:
1679 aio_put_req(req);
1680 aio_put_req(req);
1681 return ret;
1682}
1683
1684long do_io_submit(aio_context_t ctx_id, long nr,
1685 struct iocb __user *__user *iocbpp, bool compat)
1686{
1687 struct kioctx *ctx;
1688 long ret = 0;
1689 int i;
1690 struct hlist_head batch_hash[AIO_BATCH_HASH_SIZE] = { { 0, }, };
1691
1692 if (unlikely(nr < 0))
1693 return -EINVAL;
1694
1695 if (unlikely(nr > LONG_MAX/sizeof(*iocbpp)))
1696 nr = LONG_MAX/sizeof(*iocbpp);
1697
1698 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1699 return -EFAULT;
1700
1701 ctx = lookup_ioctx(ctx_id);
1702 if (unlikely(!ctx)) {
1703 pr_debug("EINVAL: io_submit: invalid context id\n");
1704 return -EINVAL;
1705 }
1706
1707
1708
1709
1710
1711 for (i=0; i<nr; i++) {
1712 struct iocb __user *user_iocb;
1713 struct iocb tmp;
1714
1715 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1716 ret = -EFAULT;
1717 break;
1718 }
1719
1720 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1721 ret = -EFAULT;
1722 break;
1723 }
1724
1725 ret = io_submit_one(ctx, user_iocb, &tmp, batch_hash, compat);
1726 if (ret)
1727 break;
1728 }
1729 aio_batch_free(batch_hash);
1730
1731 put_ioctx(ctx);
1732 return i ? i : ret;
1733}
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr,
1748 struct iocb __user * __user *, iocbpp)
1749{
1750 return do_io_submit(ctx_id, nr, iocbpp, 0);
1751}
1752
1753
1754
1755
1756static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
1757 u32 key)
1758{
1759 struct list_head *pos;
1760
1761 assert_spin_locked(&ctx->ctx_lock);
1762
1763
1764 list_for_each(pos, &ctx->active_reqs) {
1765 struct kiocb *kiocb = list_kiocb(pos);
1766 if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key)
1767 return kiocb;
1768 }
1769 return NULL;
1770}
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb,
1783 struct io_event __user *, result)
1784{
1785 int (*cancel)(struct kiocb *iocb, struct io_event *res);
1786 struct kioctx *ctx;
1787 struct kiocb *kiocb;
1788 u32 key;
1789 int ret;
1790
1791 ret = get_user(key, &iocb->aio_key);
1792 if (unlikely(ret))
1793 return -EFAULT;
1794
1795 ctx = lookup_ioctx(ctx_id);
1796 if (unlikely(!ctx))
1797 return -EINVAL;
1798
1799 spin_lock_irq(&ctx->ctx_lock);
1800 ret = -EAGAIN;
1801 kiocb = lookup_kiocb(ctx, iocb, key);
1802 if (kiocb && kiocb->ki_cancel) {
1803 cancel = kiocb->ki_cancel;
1804 kiocb->ki_users ++;
1805 kiocbSetCancelled(kiocb);
1806 } else
1807 cancel = NULL;
1808 spin_unlock_irq(&ctx->ctx_lock);
1809
1810 if (NULL != cancel) {
1811 struct io_event tmp;
1812 pr_debug("calling cancel\n");
1813 memset(&tmp, 0, sizeof(tmp));
1814 tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user;
1815 tmp.data = kiocb->ki_user_data;
1816 ret = cancel(kiocb, &tmp);
1817 if (!ret) {
1818
1819
1820
1821 if (copy_to_user(result, &tmp, sizeof(tmp)))
1822 ret = -EFAULT;
1823 }
1824 } else
1825 ret = -EINVAL;
1826
1827 put_ioctx(ctx);
1828
1829 return ret;
1830}
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id,
1846 long, min_nr,
1847 long, nr,
1848 struct io_event __user *, events,
1849 struct timespec __user *, timeout)
1850{
1851 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1852 long ret = -EINVAL;
1853
1854 if (likely(ioctx)) {
1855 if (likely(min_nr <= nr && min_nr >= 0))
1856 ret = read_events(ioctx, min_nr, nr, events, timeout);
1857 put_ioctx(ioctx);
1858 }
1859
1860 asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout);
1861 return ret;
1862}
1863