aio.c 39.5 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
2
3
4
5
6
7
8
9
10
/*
 *	An async IO implementation for Linux
 *	Written by Benjamin LaHaise <bcrl@kvack.org>
 *
 *	Implements an efficient asynchronous io interface.
 *
 *	Copyright 2000, 2001, 2002 Red Hat, Inc.  All Rights Reserved.
 *
 *	See ../COPYING for licensing terms.
 */
Kent Overstreet's avatar
Kent Overstreet committed
11
12
#define pr_fmt(fmt) "%s: " fmt, __func__

Linus Torvalds's avatar
Linus Torvalds committed
13
14
15
16
17
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/errno.h>
#include <linux/time.h>
#include <linux/aio_abi.h>
18
#include <linux/export.h>
Linus Torvalds's avatar
Linus Torvalds committed
19
#include <linux/syscalls.h>
20
#include <linux/backing-dev.h>
21
#include <linux/uio.h>
Linus Torvalds's avatar
Linus Torvalds committed
22
23
24
25
26
27

#include <linux/sched.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/mm.h>
#include <linux/mman.h>
28
#include <linux/mmu_context.h>
Kent Overstreet's avatar
Kent Overstreet committed
29
#include <linux/percpu.h>
Linus Torvalds's avatar
Linus Torvalds committed
30
31
32
33
34
35
#include <linux/slab.h>
#include <linux/timer.h>
#include <linux/aio.h>
#include <linux/highmem.h>
#include <linux/workqueue.h>
#include <linux/security.h>
36
#include <linux/eventfd.h>
Jeff Moyer's avatar
Jeff Moyer committed
37
#include <linux/blkdev.h>
38
#include <linux/compat.h>
39
40
#include <linux/migrate.h>
#include <linux/ramfs.h>
Kent Overstreet's avatar
Kent Overstreet committed
41
#include <linux/percpu-refcount.h>
42
#include <linux/mount.h>
Linus Torvalds's avatar
Linus Torvalds committed
43
44
45
46

#include <asm/kmap_types.h>
#include <asm/uaccess.h>

Al Viro's avatar
Al Viro committed
47
48
#include "internal.h"

49
50
51
52
53
54
#define AIO_RING_MAGIC			0xa10a10a1
#define AIO_RING_COMPAT_FEATURES	1
#define AIO_RING_INCOMPAT_FEATURES	0
struct aio_ring {
	unsigned	id;	/* kernel internal index number */
	unsigned	nr;	/* number of io_events */
55
56
	unsigned	head;	/* Written to by userland or under ring_lock
				 * mutex by aio_read_events_ring(). */
57
58
59
60
61
62
63
64
65
66
67
68
69
	unsigned	tail;

	unsigned	magic;
	unsigned	compat_features;
	unsigned	incompat_features;
	unsigned	header_length;	/* size of aio_ring */


	struct io_event		io_events[0];
}; /* 128 bytes + ring size */

#define AIO_RING_PAGES	8

70
71
72
73
74
75
struct kioctx_table {
	struct rcu_head	rcu;
	unsigned	nr;
	struct kioctx	*table[];
};

Kent Overstreet's avatar
Kent Overstreet committed
76
77
78
79
struct kioctx_cpu {
	unsigned		reqs_available;
};

80
struct kioctx {
Kent Overstreet's avatar
Kent Overstreet committed
81
	struct percpu_ref	users;
Kent Overstreet's avatar
Kent Overstreet committed
82
	atomic_t		dead;
83

Kent Overstreet's avatar
Kent Overstreet committed
84
85
	struct percpu_ref	reqs;

86
87
	unsigned long		user_id;

Kent Overstreet's avatar
Kent Overstreet committed
88
89
90
91
92
93
94
	struct __percpu kioctx_cpu *cpu;

	/*
	 * For percpu reqs_available, number of slots we move to/from global
	 * counter at a time:
	 */
	unsigned		req_batch;
95
96
97
98
	/*
	 * This is what userspace passed to io_setup(), it's not used for
	 * anything but counting against the global max_reqs quota.
	 *
99
	 * The real limit is nr_events - 1, which will be larger (see
100
101
	 * aio_setup_ring())
	 */
102
103
	unsigned		max_reqs;

104
105
	/* Size of ringbuffer, in units of struct io_event */
	unsigned		nr_events;
106

107
108
109
110
111
112
	unsigned long		mmap_base;
	unsigned long		mmap_size;

	struct page		**ring_pages;
	long			nr_pages;

Kent Overstreet's avatar
Kent Overstreet committed
113
	struct work_struct	free_work;
114

115
116
117
118
119
	/*
	 * signals when all in-flight requests are done
	 */
	struct completion *requests_done;

120
	struct {
121
122
123
124
125
		/*
		 * This counts the number of available slots in the ringbuffer,
		 * so we avoid overflowing it: it's decremented (if positive)
		 * when allocating a kiocb and incremented when the resulting
		 * io_event is pulled off the ringbuffer.
Kent Overstreet's avatar
Kent Overstreet committed
126
127
		 *
		 * We batch accesses to it with a percpu version.
128
129
		 */
		atomic_t	reqs_available;
130
131
132
133
134
135
136
	} ____cacheline_aligned_in_smp;

	struct {
		spinlock_t	ctx_lock;
		struct list_head active_reqs;	/* used for cancellation */
	} ____cacheline_aligned_in_smp;

137
138
	struct {
		struct mutex	ring_lock;
139
140
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;
141
142
143
144

	struct {
		unsigned	tail;
		spinlock_t	completion_lock;
145
	} ____cacheline_aligned_in_smp;
146
147

	struct page		*internal_pages[AIO_RING_PAGES];
148
	struct file		*aio_ring_file;
149
150

	unsigned		id;
151
152
};

Linus Torvalds's avatar
Linus Torvalds committed
153
/*------ sysctl variables----*/
154
155
156
static DEFINE_SPINLOCK(aio_nr_lock);
unsigned long aio_nr;		/* current system wide number of aio requests */
unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */
Linus Torvalds's avatar
Linus Torvalds committed
157
158
/*----end sysctl variables---*/

159
160
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
161

162
163
164
165
166
167
168
169
170
171
172
static struct vfsmount *aio_mnt;

static const struct file_operations aio_ring_fops;
static const struct address_space_operations aio_ctx_aops;

static struct file *aio_private_file(struct kioctx *ctx, loff_t nr_pages)
{
	struct qstr this = QSTR_INIT("[aio]", 5);
	struct file *file;
	struct path path;
	struct inode *inode = alloc_anon_inode(aio_mnt->mnt_sb);
173
174
	if (IS_ERR(inode))
		return ERR_CAST(inode);
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203

	inode->i_mapping->a_ops = &aio_ctx_aops;
	inode->i_mapping->private_data = ctx;
	inode->i_size = PAGE_SIZE * nr_pages;

	path.dentry = d_alloc_pseudo(aio_mnt->mnt_sb, &this);
	if (!path.dentry) {
		iput(inode);
		return ERR_PTR(-ENOMEM);
	}
	path.mnt = mntget(aio_mnt);

	d_instantiate(path.dentry, inode);
	file = alloc_file(&path, FMODE_READ | FMODE_WRITE, &aio_ring_fops);
	if (IS_ERR(file)) {
		path_put(&path);
		return file;
	}

	file->f_flags = O_RDWR;
	return file;
}

static struct dentry *aio_mount(struct file_system_type *fs_type,
				int flags, const char *dev_name, void *data)
{
	static const struct dentry_operations ops = {
		.d_dname	= simple_dname,
	};
204
	return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC);
205
206
}

Linus Torvalds's avatar
Linus Torvalds committed
207
208
209
210
211
212
/* aio_setup
 *	Creates the slab caches used by the aio routines, panic on
 *	failure as this is done early during the boot sequence.
 */
static int __init aio_setup(void)
{
213
214
215
216
217
218
219
220
221
	static struct file_system_type aio_fs = {
		.name		= "aio",
		.mount		= aio_mount,
		.kill_sb	= kill_anon_super,
	};
	aio_mnt = kern_mount(&aio_fs);
	if (IS_ERR(aio_mnt))
		panic("Failed to create aio fs mount.");

222
223
	kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
	kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds's avatar
Linus Torvalds committed
224

Kent Overstreet's avatar
Kent Overstreet committed
225
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
226
227
228

	return 0;
}
229
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
230

231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
static void put_aio_ring_file(struct kioctx *ctx)
{
	struct file *aio_ring_file = ctx->aio_ring_file;
	if (aio_ring_file) {
		truncate_setsize(aio_ring_file->f_inode, 0);

		/* Prevent further access to the kioctx from migratepages */
		spin_lock(&aio_ring_file->f_inode->i_mapping->private_lock);
		aio_ring_file->f_inode->i_mapping->private_data = NULL;
		ctx->aio_ring_file = NULL;
		spin_unlock(&aio_ring_file->f_inode->i_mapping->private_lock);

		fput(aio_ring_file);
	}
}

Linus Torvalds's avatar
Linus Torvalds committed
247
248
static void aio_free_ring(struct kioctx *ctx)
{
249
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
250

251
252
253
254
255
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

256
	for (i = 0; i < ctx->nr_pages; i++) {
257
		struct page *page;
258
259
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
260
261
262
263
264
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
265
	}
Linus Torvalds's avatar
Linus Torvalds committed
266

267
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
268
		kfree(ctx->ring_pages);
269
270
		ctx->ring_pages = NULL;
	}
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
}

static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma)
{
	vma->vm_ops = &generic_file_vm_ops;
	return 0;
}

static const struct file_operations aio_ring_fops = {
	.mmap = aio_ring_mmap,
};

static int aio_set_page_dirty(struct page *page)
{
	return 0;
}

288
#if IS_ENABLED(CONFIG_MIGRATION)
289
290
291
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
292
	struct kioctx *ctx;
293
	unsigned long flags;
294
	pgoff_t idx;
295
296
	int rc;

297
298
	rc = 0;

299
	/* mapping->private_lock here protects against the kioctx teardown.  */
300
301
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
	if (!ctx) {
		rc = -EINVAL;
		goto out;
	}

	/* The ring_lock mutex.  The prevents aio_read_events() from writing
	 * to the ring's head, and prevents page migration from mucking in
	 * a partially initialized kiotx.
	 */
	if (!mutex_trylock(&ctx->ring_lock)) {
		rc = -EAGAIN;
		goto out;
	}

	idx = old->index;
	if (idx < (pgoff_t)ctx->nr_pages) {
		/* Make sure the old page hasn't already been changed */
		if (ctx->ring_pages[idx] != old)
			rc = -EAGAIN;
321
322
323
324
	} else
		rc = -EINVAL;

	if (rc != 0)
325
		goto out_unlock;
326

327
328
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
329
	get_page(new);
330

331
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
332
	if (rc != MIGRATEPAGE_SUCCESS) {
333
		put_page(new);
334
		goto out_unlock;
335
336
	}

337
338
339
	/* Take completion_lock to prevent other writes to the ring buffer
	 * while the old page is copied to the new.  This prevents new
	 * events from being lost.
340
	 */
341
342
343
344
345
	spin_lock_irqsave(&ctx->completion_lock, flags);
	migrate_page_copy(new, old);
	BUG_ON(ctx->ring_pages[idx] != old);
	ctx->ring_pages[idx] = new;
	spin_unlock_irqrestore(&ctx->completion_lock, flags);
346

347
348
	/* The old page is no longer accessible. */
	put_page(old);
349

350
351
352
353
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
354
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
355
}
356
#endif
Linus Torvalds's avatar
Linus Torvalds committed
357

358
359
static const struct address_space_operations aio_ctx_aops = {
	.set_page_dirty = aio_set_page_dirty,
360
#if IS_ENABLED(CONFIG_MIGRATION)
361
	.migratepage	= aio_migratepage,
362
#endif
363
364
};

Linus Torvalds's avatar
Linus Torvalds committed
365
366
367
368
static int aio_setup_ring(struct kioctx *ctx)
{
	struct aio_ring *ring;
	unsigned nr_events = ctx->max_reqs;
Zach Brown's avatar
Zach Brown committed
369
	struct mm_struct *mm = current->mm;
370
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
371
	int nr_pages;
372
373
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
374
375
376
377
378
379
380

	/* Compensate for the ring buffer's head/tail overlap entry */
	nr_events += 2;	/* 1 is required, 2 for good luck */

	size = sizeof(struct aio_ring);
	size += sizeof(struct io_event) * nr_events;

381
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
382
383
384
	if (nr_pages < 0)
		return -EINVAL;

385
	file = aio_private_file(ctx, nr_pages);
386
387
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
388
		return -ENOMEM;
389
390
	}

391
392
393
394
395
396
397
398
399
400
401
402
403
404
	ctx->aio_ring_file = file;
	nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring))
			/ sizeof(struct io_event);

	ctx->ring_pages = ctx->internal_pages;
	if (nr_pages > AIO_RING_PAGES) {
		ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *),
					  GFP_KERNEL);
		if (!ctx->ring_pages) {
			put_aio_ring_file(ctx);
			return -ENOMEM;
		}
	}

405
406
407
408
409
410
411
412
413
414
415
	for (i = 0; i < nr_pages; i++) {
		struct page *page;
		page = find_or_create_page(file->f_inode->i_mapping,
					   i, GFP_HIGHUSER | __GFP_ZERO);
		if (!page)
			break;
		pr_debug("pid(%d) page[%d]->count=%d\n",
			 current->pid, i, page_count(page));
		SetPageUptodate(page);
		SetPageDirty(page);
		unlock_page(page);
416
417

		ctx->ring_pages[i] = page;
418
	}
419
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
420

421
422
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
423
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
424
425
	}

426
427
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
428

Zach Brown's avatar
Zach Brown committed
429
	down_write(&mm->mmap_sem);
430
431
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
432
433
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
434
435
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
436
		aio_free_ring(ctx);
437
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
438
439
	}

440
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
441

442
443
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
444

445
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
446
	ring->nr = nr_events;	/* user copy */
447
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
448
449
450
451
452
	ring->head = ring->tail = 0;
	ring->magic = AIO_RING_MAGIC;
	ring->compat_features = AIO_RING_COMPAT_FEATURES;
	ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
	ring->header_length = sizeof(struct aio_ring);
453
	kunmap_atomic(ring);
454
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
455
456
457
458
459
460
461
462

	return 0;
}

#define AIO_EVENTS_PER_PAGE	(PAGE_SIZE / sizeof(struct io_event))
#define AIO_EVENTS_FIRST_PAGE	((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
#define AIO_EVENTS_OFFSET	(AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)

463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel)
{
	struct kioctx *ctx = req->ki_ctx;
	unsigned long flags;

	spin_lock_irqsave(&ctx->ctx_lock, flags);

	if (!req->ki_list.next)
		list_add(&req->ki_list, &ctx->active_reqs);

	req->ki_cancel = cancel;

	spin_unlock_irqrestore(&ctx->ctx_lock, flags);
}
EXPORT_SYMBOL(kiocb_set_cancel_fn);

479
static int kiocb_cancel(struct kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
480
{
481
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
482

483
484
485
486
487
488
489
490
	/*
	 * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it
	 * actually has a cancel function, hence the cmpxchg()
	 */

	cancel = ACCESS_ONCE(kiocb->ki_cancel);
	do {
		if (!cancel || cancel == KIOCB_CANCELLED)
Kent Overstreet's avatar
Kent Overstreet committed
491
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
492

493
494
495
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
496

Kent Overstreet's avatar
Kent Overstreet committed
497
	return cancel(kiocb);
Kent Overstreet's avatar
Kent Overstreet committed
498
499
}

Kent Overstreet's avatar
Kent Overstreet committed
500
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
501
{
Kent Overstreet's avatar
Kent Overstreet committed
502
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
503

Kent Overstreet's avatar
Kent Overstreet committed
504
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
505

Kent Overstreet's avatar
Kent Overstreet committed
506
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
507
	free_percpu(ctx->cpu);
508
509
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
510
511
512
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
513
514
515
516
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

517
518
519
520
	/* At this point we know that there are no any in-flight requests */
	if (ctx->requests_done)
		complete(ctx->requests_done);

Kent Overstreet's avatar
Kent Overstreet committed
521
522
523
524
	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

Kent Overstreet's avatar
Kent Overstreet committed
525
526
527
528
529
/*
 * When this function runs, the kioctx has been removed from the "hash table"
 * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted -
 * now it's safe to cancel any that need to be.
 */
Kent Overstreet's avatar
Kent Overstreet committed
530
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
531
{
Kent Overstreet's avatar
Kent Overstreet committed
532
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
Kent Overstreet's avatar
Kent Overstreet committed
533
534
535
536
537
538
539
540
541
	struct kiocb *req;

	spin_lock_irq(&ctx->ctx_lock);

	while (!list_empty(&ctx->active_reqs)) {
		req = list_first_entry(&ctx->active_reqs,
				       struct kiocb, ki_list);

		list_del_init(&req->ki_list);
542
		kiocb_cancel(req);
Kent Overstreet's avatar
Kent Overstreet committed
543
544
545
546
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
547
548
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
549
550
}

551
552
553
554
555
556
557
static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm)
{
	unsigned i, new_nr;
	struct kioctx_table *table, *old;
	struct aio_ring *ring;

	spin_lock(&mm->ioctx_lock);
558
	table = rcu_dereference_raw(mm->ioctx_table);
559
560
561
562
563
564
565
566
567

	while (1) {
		if (table)
			for (i = 0; i < table->nr; i++)
				if (!table->table[i]) {
					ctx->id = i;
					table->table[i] = ctx;
					spin_unlock(&mm->ioctx_lock);

568
569
570
571
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
					ring = kmap_atomic(ctx->ring_pages[0]);
					ring->id = ctx->id;
					kunmap_atomic(ring);
					return 0;
				}

		new_nr = (table ? table->nr : 1) * 4;
		spin_unlock(&mm->ioctx_lock);

		table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) *
				new_nr, GFP_KERNEL);
		if (!table)
			return -ENOMEM;

		table->nr = new_nr;

		spin_lock(&mm->ioctx_lock);
589
		old = rcu_dereference_raw(mm->ioctx_table);
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605

		if (!old) {
			rcu_assign_pointer(mm->ioctx_table, table);
		} else if (table->nr > old->nr) {
			memcpy(table->table, old->table,
			       old->nr * sizeof(struct kioctx *));

			rcu_assign_pointer(mm->ioctx_table, table);
			kfree_rcu(old, rcu);
		} else {
			kfree(table);
			table = old;
		}
	}
}

Kent Overstreet's avatar
Kent Overstreet committed
606
607
608
609
610
611
612
613
614
615
static void aio_nr_sub(unsigned nr)
{
	spin_lock(&aio_nr_lock);
	if (WARN_ON(aio_nr - nr > aio_nr))
		aio_nr = 0;
	else
		aio_nr -= nr;
	spin_unlock(&aio_nr_lock);
}

Linus Torvalds's avatar
Linus Torvalds committed
616
617
618
619
620
/* ioctx_alloc
 *	Allocates and initializes an ioctx.  Returns an ERR_PTR if it failed.
 */
static struct kioctx *ioctx_alloc(unsigned nr_events)
{
Zach Brown's avatar
Zach Brown committed
621
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
622
	struct kioctx *ctx;
623
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
624

Kent Overstreet's avatar
Kent Overstreet committed
625
626
627
628
629
630
631
632
633
634
635
636
	/*
	 * We keep track of the number of available ringbuffer slots, to prevent
	 * overflow (reqs_available), and we also use percpu counters for this.
	 *
	 * So since up to half the slots might be on other cpu's percpu counters
	 * and unavailable, double nr_events so userspace sees what they
	 * expected: additionally, we move req_batch slots to/from percpu
	 * counters at a time, so make sure that isn't 0:
	 */
	nr_events = max(nr_events, num_possible_cpus() * 4);
	nr_events *= 2;

Linus Torvalds's avatar
Linus Torvalds committed
637
638
639
640
641
642
643
	/* Prevent overflows */
	if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
	    (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
		pr_debug("ENOMEM: nr_events too high\n");
		return ERR_PTR(-EINVAL);
	}

644
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
645
646
		return ERR_PTR(-EAGAIN);

647
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
648
649
650
651
652
653
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
654
	spin_lock_init(&ctx->completion_lock);
655
	mutex_init(&ctx->ring_lock);
656
657
658
	/* Protect against page migration throughout kiotx setup by keeping
	 * the ring_lock mutex held until setup is complete. */
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
659
660
661
662
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

663
664
665
666
667
668
	if (percpu_ref_init(&ctx->users, free_ioctx_users))
		goto err;

	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs))
		goto err;

Kent Overstreet's avatar
Kent Overstreet committed
669
670
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
671
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
672

673
674
	err = aio_setup_ring(ctx);
	if (err < 0)
Kent Overstreet's avatar
Kent Overstreet committed
675
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
676

677
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
678
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
679
680
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
681

Linus Torvalds's avatar
Linus Torvalds committed
682
	/* limit the number of system wide aios */
683
	spin_lock(&aio_nr_lock);
684
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
685
	    aio_nr + nr_events < aio_nr) {
686
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
687
		err = -EAGAIN;
688
		goto err_ctx;
689
690
	}
	aio_nr += ctx->max_reqs;
691
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
692

693
694
	percpu_ref_get(&ctx->users);	/* io_setup() will drop this ref */
	percpu_ref_get(&ctx->reqs);	/* free_ioctx_users() will drop this */
Kent Overstreet's avatar
Kent Overstreet committed
695

696
697
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
698
		goto err_cleanup;
699

700
701
702
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

Kent Overstreet's avatar
Kent Overstreet committed
703
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
704
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
705
706
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
707
708
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
709
710
err_ctx:
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
711
err:
712
	mutex_unlock(&ctx->ring_lock);
Kent Overstreet's avatar
Kent Overstreet committed
713
	free_percpu(ctx->cpu);
714
715
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Linus Torvalds's avatar
Linus Torvalds committed
716
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
717
	pr_debug("error allocating ioctx %d\n", err);
718
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
719
720
}

Kent Overstreet's avatar
Kent Overstreet committed
721
722
723
724
725
/* kill_ioctx
 *	Cancels all outstanding aio requests on an aio context.  Used
 *	when the processes owning a context have all exited to encourage
 *	the rapid destruction of the kioctx.
 */
726
static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
727
		struct completion *requests_done)
Kent Overstreet's avatar
Kent Overstreet committed
728
{
729
	struct kioctx_table *table;
730

731
732
	if (atomic_xchg(&ctx->dead, 1))
		return -EINVAL;
733
734


735
	spin_lock(&mm->ioctx_lock);
736
	table = rcu_dereference_raw(mm->ioctx_table);
737
738
739
	WARN_ON(ctx != table->table[ctx->id]);
	table->table[ctx->id] = NULL;
	spin_unlock(&mm->ioctx_lock);
740

741
742
	/* percpu_ref_kill() will do the necessary call_rcu() */
	wake_up_all(&ctx->wait);
743

744
745
746
747
748
749
750
751
	/*
	 * It'd be more correct to do this in free_ioctx(), after all
	 * the outstanding kiocbs have finished - but by then io_destroy
	 * has already returned, so io_setup() could potentially return
	 * -EAGAIN with no ioctxs actually in use (as far as userspace
	 *  could tell).
	 */
	aio_nr_sub(ctx->max_reqs);
752

753
754
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
755

756
757
758
	ctx->requests_done = requests_done;
	percpu_ref_kill(&ctx->users);
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
759
760
761
762
763
}

/* wait_on_sync_kiocb:
 *	Waits on the given sync kiocb to complete.
 */
Kent Overstreet's avatar
Kent Overstreet committed
764
ssize_t wait_on_sync_kiocb(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
765
{
Kent Overstreet's avatar
Kent Overstreet committed
766
	while (!req->ki_ctx) {
Linus Torvalds's avatar
Linus Torvalds committed
767
		set_current_state(TASK_UNINTERRUPTIBLE);
Kent Overstreet's avatar
Kent Overstreet committed
768
		if (req->ki_ctx)
Linus Torvalds's avatar
Linus Torvalds committed
769
			break;
770
		io_schedule();
Linus Torvalds's avatar
Linus Torvalds committed
771
772
	}
	__set_current_state(TASK_RUNNING);
Kent Overstreet's avatar
Kent Overstreet committed
773
	return req->ki_user_data;
Linus Torvalds's avatar
Linus Torvalds committed
774
}
775
EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds's avatar
Linus Torvalds committed
776

Kent Overstreet's avatar
Kent Overstreet committed
777
778
779
780
781
782
783
/*
 * exit_aio: called when the last user of mm goes away.  At this point, there is
 * no way for any new requests to be submited or any of the io_* syscalls to be
 * called on the context.
 *
 * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on
 * them.
Linus Torvalds's avatar
Linus Torvalds committed
784
 */
785
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
786
{
787
788
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
	int i;
789

790
791
	if (!table)
		return;
792

793
794
	for (i = 0; i < table->nr; ++i) {
		struct kioctx *ctx = table->table[i];
795

796
797
		if (!ctx)
			continue;
798
		/*
799
800
801
802
803
		 * We don't need to bother with munmap() here - exit_mmap(mm)
		 * is coming and it'll unmap everything. And we simply can't,
		 * this is not necessarily our ->mm.
		 * Since kill_ioctx() uses non-zero ->mmap_size as indicator
		 * that it needs to unmap the area, just set it to 0.
804
		 */
805
		ctx->mmap_size = 0;
806
		kill_ioctx(mm, ctx, NULL);
Linus Torvalds's avatar
Linus Torvalds committed
807
	}
808
809
810

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
Linus Torvalds's avatar
Linus Torvalds committed
811
812
}

Kent Overstreet's avatar
Kent Overstreet committed
813
814
815
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
816
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
817

818
	local_irq_save(flags);
819
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
820
	kcpu->reqs_available += nr;
821

Kent Overstreet's avatar
Kent Overstreet committed
822
823
824
825
826
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

827
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
828
829
830
831
832
833
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;
834
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
835

836
	local_irq_save(flags);
837
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
	if (!kcpu->reqs_available) {
		int old, avail = atomic_read(&ctx->reqs_available);

		do {
			if (avail < ctx->req_batch)
				goto out;

			old = avail;
			avail = atomic_cmpxchg(&ctx->reqs_available,
					       avail, avail - ctx->req_batch);
		} while (avail != old);

		kcpu->reqs_available += ctx->req_batch;
	}

	ret = true;
	kcpu->reqs_available--;
out:
856
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
857
858
859
	return ret;
}

Linus Torvalds's avatar
Linus Torvalds committed
860
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
861
862
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
863
 */
Kent Overstreet's avatar
Kent Overstreet committed
864
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
865
{
Kent Overstreet's avatar
Kent Overstreet committed
866
867
	struct kiocb *req;

Kent Overstreet's avatar
Kent Overstreet committed
868
	if (!get_reqs_available(ctx))
Kent Overstreet's avatar
Kent Overstreet committed
869
870
		return NULL;

871
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
872
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
873
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
874

Kent Overstreet's avatar
Kent Overstreet committed
875
876
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
877
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
878
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
879
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
880
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
881
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
882
883
}

884
static void kiocb_free(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
885
{
886
887
	if (req->ki_filp)
		fput(req->ki_filp);
888
889
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
890
891
892
	kmem_cache_free(kiocb_cachep, req);
}

893
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
894
{
895
	struct aio_ring __user *ring  = (void __user *)ctx_id;
896
	struct mm_struct *mm = current->mm;
897
	struct kioctx *ctx, *ret = NULL;
898
899
900
901
902
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
903

904
	rcu_read_lock();
905
	table = rcu_dereference(mm->ioctx_table);
906

907
908
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
909

910
	ctx = table->table[id];
911
	if (ctx && ctx->user_id == ctx_id) {
912
913
914
915
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
916
	rcu_read_unlock();
917
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
918
919
920
921
922
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
923
void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
924
925
926
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
927
	struct io_event	*ev_page, *event;
Linus Torvalds's avatar
Linus Torvalds committed
928
	unsigned long	flags;
Kent Overstreet's avatar
Kent Overstreet committed
929
	unsigned tail, pos;
Linus Torvalds's avatar
Linus Torvalds committed
930

931
932
933
934
935
936
	/*
	 * Special case handling for sync iocbs:
	 *  - events go directly into the iocb for fast handling
	 *  - the sync task with the iocb in its stack holds the single iocb
	 *    ref, no other paths have a way to get another ref
	 *  - the sync task helpfully left a reference to itself in the iocb
Linus Torvalds's avatar
Linus Torvalds committed
937
938
939
	 */
	if (is_sync_kiocb(iocb)) {
		iocb->ki_user_data = res;
Kent Overstreet's avatar
Kent Overstreet committed
940
941
		smp_wmb();
		iocb->ki_ctx = ERR_PTR(-EXDEV);
Linus Torvalds's avatar
Linus Torvalds committed
942
		wake_up_process(iocb->ki_obj.tsk);
943
		return;
Linus Torvalds's avatar
Linus Torvalds committed
944
945
	}

946
947
948
949
950
951
952
	if (iocb->ki_list.next) {
		unsigned long flags;

		spin_lock_irqsave(&ctx->ctx_lock, flags);
		list_del(&iocb->ki_list);
		spin_unlock_irqrestore(&ctx->ctx_lock, flags);
	}
953

954
955
	/*
	 * Add a completion event to the ring buffer. Must be done holding
956
	 * ctx->completion_lock to prevent other code from messing with the tail
957
958
959
960
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

961
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
962
963
	pos = tail + AIO_EVENTS_OFFSET;

964
	if (++tail >= ctx->nr_events)
965
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
966

967
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
968
969
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

Linus Torvalds's avatar
Linus Torvalds committed
970
971
972
973
974
	event->obj = (u64)(unsigned long)iocb->ki_obj.user;
	event->data = iocb->ki_user_data;
	event->res = res;
	event->res2 = res2;

Kent Overstreet's avatar
Kent Overstreet committed
975
	kunmap_atomic(ev_page);
976
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
977
978

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreet's avatar
Kent Overstreet committed
979
980
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
981
982
983
984
985
986

	/* after flagging the request as done, we
	 * must never even look at it again
	 */
	smp_wmb();	/* make event visible before updating tail */

987
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
988

989
	ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreet's avatar
Kent Overstreet committed
990
	ring->tail = tail;
991
	kunmap_atomic(ring);
992
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
993

994
995
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
996
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
997
998
999
1000
1001
1002

	/*
	 * Check if the user asked us to deliver the result through an
	 * eventfd. The eventfd_signal() function is safe to be called
	 * from IRQ context.
	 */
1003
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
1004
1005
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
1006
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
1007
	kiocb_free(iocb);
1008
	put_reqs_available(ctx, 1);
Linus Torvalds's avatar
Linus Torvalds committed
1009

1010
1011
1012
1013
1014
1015
1016
1017
	/*
	 * We have to order our ring_info tail store above and test
	 * of the wait list below outside the wait lock.  This is
	 * like in wake_up_bit() where clearing a bit has to be
	 * ordered with the unlocked test.
	 */
	smp_mb();

Linus Torvalds's avatar
Linus Torvalds committed
1018
1019
1020
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
1021
	percpu_ref_put(&ctx->reqs);
Linus Torvalds's avatar
Linus Torvalds committed
1022
}
1023
EXPORT_SYMBOL(aio_complete);
Linus Torvalds's avatar
Linus Torvalds committed
1024

Gu Zheng's avatar
Gu Zheng committed
1025
/* aio_read_events_ring
1026
1027
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
Linus Torvalds's avatar
Linus Torvalds committed
1028
 */
1029
1030
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
1031
1032
{
	struct aio_ring *ring;
1033
	unsigned head, tail, pos;
1034
1035
1036
	long ret = 0;
	int copy_ret;

1037
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1038

1039
	/* Access to ->ring_pages here is protected by ctx->ring_lock. */
1040
	ring = kmap_atomic(ctx->ring_pages[0]);
1041
	head = ring->head;
1042
	tail = ring->tail;
1043
1044
	kunmap_atomic(ring);

1045
	pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
1046

1047
	if (head == tail)
Linus Torvalds's avatar
Linus Torvalds committed
1048
1049
		goto out;

1050
1051
1052
	head %= ctx->nr_events;
	tail %= ctx->nr_events;

1053
1054
1055
1056
1057
	while (ret < nr) {
		long avail;
		struct io_event *ev;
		struct page *page;

1058
1059
		avail = (head <= tail ?  tail : ctx->nr_events) - head;
		if (head == tail)
1060
1061
1062
1063
1064
1065
1066
			break;

		avail = min(avail, nr - ret);
		avail = min_t(long, avail, AIO_EVENTS_PER_PAGE -
			    ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE));

		pos = head + AIO_EVENTS_OFFSET;
1067
		page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
		pos %= AIO_EVENTS_PER_PAGE;

		ev = kmap(page);
		copy_ret = copy_to_user(event + ret, ev + pos,
					sizeof(*ev) * avail);
		kunmap(page);

		if (unlikely(copy_ret)) {
			ret = -EFAULT;
			goto out;
		}

		ret += avail;
		head += avail;
1082
		head %= ctx->nr_events;
Linus Torvalds's avatar
Linus Torvalds committed
1083
1084
	}

1085
	ring = kmap_atomic(ctx->ring_pages[0]);
1086
	ring->head = head;
1087
	kunmap_atomic(ring);
1088
	flush_dcache_page(ctx->ring_pages[0]);
1089

1090
	pr_debug("%li  h%u t%u\n", ret, head, tail);
1091
out:
1092
	mutex_unlock(&ctx->ring_lock);
1093

Linus Torvalds's avatar
Linus Torvalds committed
1094
1095
1096
	return ret;
}

1097
1098
static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr,
			    struct io_event __user *event, long *i)
Linus Torvalds's avatar
Linus Torvalds committed
1099
{
1100
	long ret = aio_read_events_ring(ctx, event + *i, nr - *i);
Linus Torvalds's avatar
Linus Torvalds committed
1101

1102
1103
	if (ret > 0)
		*i += ret;
Linus Torvalds's avatar
Linus Torvalds committed
1104

1105
1106
	if (unlikely(