aio.c 38.5 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

	struct {
116
117
118
119
120
		/*
		 * 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
121
122
		 *
		 * We batch accesses to it with a percpu version.
123
124
		 */
		atomic_t	reqs_available;
125
126
127
128
129
130
131
	} ____cacheline_aligned_in_smp;

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

132
133
	struct {
		struct mutex	ring_lock;
134
135
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;
136
137
138
139

	struct {
		unsigned	tail;
		spinlock_t	completion_lock;
140
	} ____cacheline_aligned_in_smp;
141
142

	struct page		*internal_pages[AIO_RING_PAGES];
143
	struct file		*aio_ring_file;
144
145

	unsigned		id;
146
147
};

Linus Torvalds's avatar
Linus Torvalds committed
148
/*------ sysctl variables----*/
149
150
151
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
152
153
/*----end sysctl variables---*/

154
155
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
156

157
158
159
160
161
162
163
164
165
166
167
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);
168
169
	if (IS_ERR(inode))
		return ERR_CAST(inode);
170
171
172
173
174
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

	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;
	file->private_data = ctx;
	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,
	};
	return mount_pseudo(fs_type, "aio:", NULL, &ops, 0xa10a10a1);
}

Linus Torvalds's avatar
Linus Torvalds committed
203
204
205
206
207
208
/* 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)
{
209
210
211
212
213
214
215
216
217
	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.");

218
219
	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
220

Kent Overstreet's avatar
Kent Overstreet committed
221
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
222
223
224

	return 0;
}
225
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
226

227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
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
243
244
static void aio_free_ring(struct kioctx *ctx)
{
245
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
246

247
248
249
250
251
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

252
	for (i = 0; i < ctx->nr_pages; i++) {
253
		struct page *page;
254
255
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
256
257
258
259
260
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
261
	}
Linus Torvalds's avatar
Linus Torvalds committed
262

263
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
264
		kfree(ctx->ring_pages);
265
266
		ctx->ring_pages = NULL;
	}
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
}

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;
}

284
#if IS_ENABLED(CONFIG_MIGRATION)
285
286
287
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
288
	struct kioctx *ctx;
289
	unsigned long flags;
290
	pgoff_t idx;
291
292
	int rc;

293
294
	rc = 0;

295
	/* mapping->private_lock here protects against the kioctx teardown.  */
296
297
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
	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;
317
318
319
320
	} else
		rc = -EINVAL;

	if (rc != 0)
321
		goto out_unlock;
322

323
324
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
325
	get_page(new);
326

327
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
328
	if (rc != MIGRATEPAGE_SUCCESS) {
329
		put_page(new);
330
		goto out_unlock;
331
332
	}

333
334
335
	/* 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.
336
	 */
337
338
339
340
341
	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);
342

343
344
	/* The old page is no longer accessible. */
	put_page(old);
345

346
347
348
349
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
350
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
351
}
352
#endif
Linus Torvalds's avatar
Linus Torvalds committed
353

354
355
static const struct address_space_operations aio_ctx_aops = {
	.set_page_dirty = aio_set_page_dirty,
356
#if IS_ENABLED(CONFIG_MIGRATION)
357
	.migratepage	= aio_migratepage,
358
#endif
359
360
};

Linus Torvalds's avatar
Linus Torvalds committed
361
362
363
364
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
365
	struct mm_struct *mm = current->mm;
366
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
367
	int nr_pages;
368
369
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
370
371
372
373
374
375
376

	/* 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;

377
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
378
379
380
	if (nr_pages < 0)
		return -EINVAL;

381
	file = aio_private_file(ctx, nr_pages);
382
383
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
384
		return -ENOMEM;
385
386
	}

387
388
389
390
391
392
393
394
395
396
397
398
399
400
	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;
		}
	}

401
402
403
404
405
406
407
408
409
410
411
	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);
412
413

		ctx->ring_pages[i] = page;
414
	}
415
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
416

417
418
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
419
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
420
421
	}

422
423
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
424

Zach Brown's avatar
Zach Brown committed
425
	down_write(&mm->mmap_sem);
426
427
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
428
429
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
430
431
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
432
		aio_free_ring(ctx);
433
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
434
435
	}

436
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
437

438
439
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
440

441
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
442
	ring->nr = nr_events;	/* user copy */
443
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
444
445
446
447
448
	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);
449
	kunmap_atomic(ring);
450
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
451
452
453
454
455
456
457
458

	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)

459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
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);

475
static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
476
{
477
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
478

479
480
481
482
483
484
485
486
	/*
	 * 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
487
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
488

489
490
491
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
492

Kent Overstreet's avatar
Kent Overstreet committed
493
	return cancel(kiocb);
Kent Overstreet's avatar
Kent Overstreet committed
494
495
}

Kent Overstreet's avatar
Kent Overstreet committed
496
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
497
{
Kent Overstreet's avatar
Kent Overstreet committed
498
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
499

Kent Overstreet's avatar
Kent Overstreet committed
500
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
501

Kent Overstreet's avatar
Kent Overstreet committed
502
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
503
	free_percpu(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
504
505
506
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
507
508
509
510
511
512
513
514
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

Kent Overstreet's avatar
Kent Overstreet committed
515
516
517
518
519
/*
 * 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
520
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
521
{
Kent Overstreet's avatar
Kent Overstreet committed
522
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
Kent Overstreet's avatar
Kent Overstreet committed
523
524
525
526
527
528
529
530
531
	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);
532
		kiocb_cancel(ctx, req);
Kent Overstreet's avatar
Kent Overstreet committed
533
534
535
536
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
537
538
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
539
540
}

541
542
543
544
545
546
547
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);
548
	rcu_read_lock();
549
	table = rcu_dereference(mm->ioctx_table);
550
551
552
553
554
555
556

	while (1) {
		if (table)
			for (i = 0; i < table->nr; i++)
				if (!table->table[i]) {
					ctx->id = i;
					table->table[i] = ctx;
557
					rcu_read_unlock();
558
559
					spin_unlock(&mm->ioctx_lock);

560
561
562
563
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
564
565
566
567
568
569
570
571
					ring = kmap_atomic(ctx->ring_pages[0]);
					ring->id = ctx->id;
					kunmap_atomic(ring);
					return 0;
				}

		new_nr = (table ? table->nr : 1) * 4;

572
		rcu_read_unlock();
573
574
575
576
577
578
579
580
581
582
		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);
583
		rcu_read_lock();
584
		old = rcu_dereference(mm->ioctx_table);
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600

		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
601
602
603
604
605
606
607
608
609
610
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
611
612
613
614
615
/* 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
616
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
617
	struct kioctx *ctx;
618
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
619

Kent Overstreet's avatar
Kent Overstreet committed
620
621
622
623
624
625
626
627
628
629
630
631
	/*
	 * 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
632
633
634
635
636
637
638
	/* 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);
	}

639
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
640
641
		return ERR_PTR(-EAGAIN);

642
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
643
644
645
646
647
648
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
649
	spin_lock_init(&ctx->completion_lock);
650
	mutex_init(&ctx->ring_lock);
651
652
653
	/* 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
654
655
656
657
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

658
659
660
661
662
663
	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
664
665
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
666
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
667

668
669
	err = aio_setup_ring(ctx);
	if (err < 0)
Kent Overstreet's avatar
Kent Overstreet committed
670
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
671

672
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
673
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
674
675
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
676

Linus Torvalds's avatar
Linus Torvalds committed
677
	/* limit the number of system wide aios */
678
	spin_lock(&aio_nr_lock);
679
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
680
	    aio_nr + nr_events < aio_nr) {
681
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
682
		err = -EAGAIN;
683
		goto err_ctx;
684
685
	}
	aio_nr += ctx->max_reqs;
686
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
687

688
689
	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
690

691
692
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
693
		goto err_cleanup;
694

695
696
697
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

Kent Overstreet's avatar
Kent Overstreet committed
698
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
699
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
700
701
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
702
703
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
704
705
err_ctx:
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
706
err:
707
	mutex_unlock(&ctx->ring_lock);
Kent Overstreet's avatar
Kent Overstreet committed
708
	free_percpu(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
709
	free_percpu(ctx->reqs.pcpu_count);
Kent Overstreet's avatar
Kent Overstreet committed
710
	free_percpu(ctx->users.pcpu_count);
Linus Torvalds's avatar
Linus Torvalds committed
711
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
712
	pr_debug("error allocating ioctx %d\n", err);
713
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
714
715
}

Kent Overstreet's avatar
Kent Overstreet committed
716
717
718
719
720
/* 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.
 */
721
static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx)
Kent Overstreet's avatar
Kent Overstreet committed
722
723
{
	if (!atomic_xchg(&ctx->dead, 1)) {
724
725
726
		struct kioctx_table *table;

		spin_lock(&mm->ioctx_lock);
727
		rcu_read_lock();
728
		table = rcu_dereference(mm->ioctx_table);
729
730
731

		WARN_ON(ctx != table->table[ctx->id]);
		table->table[ctx->id] = NULL;
732
		rcu_read_unlock();
733
734
		spin_unlock(&mm->ioctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
735
736
		/* percpu_ref_kill() will do the necessary call_rcu() */
		wake_up_all(&ctx->wait);
737

Kent Overstreet's avatar
Kent Overstreet committed
738
		/*
739
740
741
742
743
		 * 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).
Kent Overstreet's avatar
Kent Overstreet committed
744
		 */
Kent Overstreet's avatar
Kent Overstreet committed
745
		aio_nr_sub(ctx->max_reqs);
746
747
748
749

		if (ctx->mmap_size)
			vm_munmap(ctx->mmap_base, ctx->mmap_size);

Kent Overstreet's avatar
Kent Overstreet committed
750
		percpu_ref_kill(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
751
	}
Linus Torvalds's avatar
Linus Torvalds committed
752
753
754
755
756
}

/* wait_on_sync_kiocb:
 *	Waits on the given sync kiocb to complete.
 */
Kent Overstreet's avatar
Kent Overstreet committed
757
ssize_t wait_on_sync_kiocb(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
758
{
Kent Overstreet's avatar
Kent Overstreet committed
759
	while (!req->ki_ctx) {
Linus Torvalds's avatar
Linus Torvalds committed
760
		set_current_state(TASK_UNINTERRUPTIBLE);
Kent Overstreet's avatar
Kent Overstreet committed
761
		if (req->ki_ctx)
Linus Torvalds's avatar
Linus Torvalds committed
762
			break;
763
		io_schedule();
Linus Torvalds's avatar
Linus Torvalds committed
764
765
	}
	__set_current_state(TASK_RUNNING);
Kent Overstreet's avatar
Kent Overstreet committed
766
	return req->ki_user_data;
Linus Torvalds's avatar
Linus Torvalds committed
767
}
768
EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds's avatar
Linus Torvalds committed
769

Kent Overstreet's avatar
Kent Overstreet committed
770
771
772
773
774
775
776
/*
 * 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
777
 */
778
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
779
{
780
	struct kioctx_table *table;
781
	struct kioctx *ctx;
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
	unsigned i = 0;

	while (1) {
		rcu_read_lock();
		table = rcu_dereference(mm->ioctx_table);

		do {
			if (!table || i >= table->nr) {
				rcu_read_unlock();
				rcu_assign_pointer(mm->ioctx_table, NULL);
				if (table)
					kfree(table);
				return;
			}

			ctx = table->table[i++];
		} while (!ctx);

		rcu_read_unlock();
801

802
803
804
805
806
807
808
809
		/*
		 * We don't need to bother with munmap() here -
		 * exit_mmap(mm) is coming and it'll unmap everything.
		 * Since aio_free_ring() uses non-zero ->mmap_size
		 * as indicator that it needs to unmap the area,
		 * just set it to 0; aio_free_ring() is the only
		 * place that uses ->mmap_size, so it's safe.
		 */
810
		ctx->mmap_size = 0;
Kent Overstreet's avatar
Kent Overstreet committed
811

812
		kill_ioctx(mm, ctx);
Linus Torvalds's avatar
Linus Torvalds committed
813
814
815
	}
}

Kent Overstreet's avatar
Kent Overstreet committed
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;

	preempt_disable();
	kcpu = this_cpu_ptr(ctx->cpu);

	kcpu->reqs_available += nr;
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

	preempt_enable();
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;

	preempt_disable();
	kcpu = this_cpu_ptr(ctx->cpu);

	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:
	preempt_enable();
	return ret;
}

Linus Torvalds's avatar
Linus Torvalds committed
862
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
863
864
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
865
 */
Kent Overstreet's avatar
Kent Overstreet committed
866
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
867
{
Kent Overstreet's avatar
Kent Overstreet committed
868
869
	struct kiocb *req;

Kent Overstreet's avatar
Kent Overstreet committed
870
	if (!get_reqs_available(ctx))
Kent Overstreet's avatar
Kent Overstreet committed
871
872
		return NULL;

873
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
874
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
875
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
876

Kent Overstreet's avatar
Kent Overstreet committed
877
878
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
879
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
880
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
881
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
882
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
883
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
884
885
}

886
static void kiocb_free(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
887
{
888
889
	if (req->ki_filp)
		fput(req->ki_filp);
890
891
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
892
893
894
	kmem_cache_free(kiocb_cachep, req);
}

895
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
896
{
897
	struct aio_ring __user *ring  = (void __user *)ctx_id;
898
	struct mm_struct *mm = current->mm;
899
	struct kioctx *ctx, *ret = NULL;
900
901
902
903
904
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
905

906
	rcu_read_lock();
907
	table = rcu_dereference(mm->ioctx_table);
908

909
910
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
911

912
	ctx = table->table[id];
913
	if (ctx && ctx->user_id == ctx_id) {
914
915
916
917
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
918
	rcu_read_unlock();
919
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
920
921
922
923
924
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
925
void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
926
927
928
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
929
	struct io_event	*ev_page, *event;
Linus Torvalds's avatar
Linus Torvalds committed
930
	unsigned long	flags;
Kent Overstreet's avatar
Kent Overstreet committed
931
	unsigned tail, pos;
Linus Torvalds's avatar
Linus Torvalds committed
932

933
934
935
936
937
938
	/*
	 * 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
939
940
941
	 */
	if (is_sync_kiocb(iocb)) {
		iocb->ki_user_data = res;
Kent Overstreet's avatar
Kent Overstreet committed
942
943
		smp_wmb();
		iocb->ki_ctx = ERR_PTR(-EXDEV);
Linus Torvalds's avatar
Linus Torvalds committed
944
		wake_up_process(iocb->ki_obj.tsk);
945
		return;
Linus Torvalds's avatar
Linus Torvalds committed
946
947
	}

948
949
950
951
952
953
954
	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);
	}
955

956
957
	/*
	 * Add a completion event to the ring buffer. Must be done holding
958
	 * ctx->completion_lock to prevent other code from messing with the tail
959
960
961
962
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

963
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
964
965
	pos = tail + AIO_EVENTS_OFFSET;

966
	if (++tail >= ctx->nr_events)
967
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
968

969
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
970
971
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

Linus Torvalds's avatar
Linus Torvalds committed
972
973
974
975
976
	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
977
	kunmap_atomic(ev_page);
978
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
979
980

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreet's avatar
Kent Overstreet committed
981
982
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
983
984
985
986
987
988

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

989
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
990

991
	ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreet's avatar
Kent Overstreet committed
992
	ring->tail = tail;
993
	kunmap_atomic(ring);
994
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
995

996
997
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
998
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
999
1000
1001
1002
1003
1004

	/*
	 * 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.
	 */
1005
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
1006
1007
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
1008
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
1009
	kiocb_free(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
1010

1011
1012
1013
1014
1015
1016
1017
1018
	/*
	 * 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
1019
1020
1021
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
1022
	percpu_ref_put(&ctx->reqs);
Linus Torvalds's avatar
Linus Torvalds committed
1023
}
1024
EXPORT_SYMBOL(aio_complete);
Linus Torvalds's avatar
Linus Torvalds committed
1025

1026
1027
1028
/* aio_read_events
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
Linus Torvalds's avatar
Linus Torvalds committed
1029
 */
1030
1031
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
1032
1033
{
	struct aio_ring *ring;
1034
	unsigned head, tail, pos;
1035
1036
1037
	long ret = 0;
	int copy_ret;

1038
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1039

1040
	/* Access to ->ring_pages here is protected by ctx->ring_lock. */
1041
	ring = kmap_atomic(ctx->ring_pages[0]);
1042
	head = ring->head;
1043
	tail = ring->tail;
1044
1045
	kunmap_atomic(ring);

1046
	pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
1047

1048
	if (head == tail)
Linus Torvalds's avatar
Linus Torvalds committed
1049
1050
		goto out;

1051
1052
1053
1054
1055
	while (ret < nr) {
		long avail;
		struct io_event *ev;
		struct page *page;

1056
1057
		avail = (head <= tail ?  tail : ctx->nr_events) - head;
		if (head == tail)
1058
1059
1060
1061
1062
1063
1064
			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;
1065
		page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
		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;
1080
		head %= ctx->nr_events;
Linus Torvalds's avatar
Linus Torvalds committed
1081
1082
	}

1083
	ring = kmap_atomic(ctx->ring_pages[0]);
1084
	ring->head = head;
1085
	kunmap_atomic(ring);
1086
	flush_dcache_page(ctx->ring_pages[0]);
1087

1088
	pr_debug("%li  h%u t%u\n", ret, head, tail);
1089

Kent Overstreet's avatar
Kent Overstreet committed
1090
	put_reqs_available(ctx, ret);
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(atomic_read(&ctx->dead)))
		ret = -EINVAL;
Linus Torvalds's avatar
Linus Torvalds committed
1107

1108
1109
	if (!*i)
		*i = ret;
Linus Torvalds's avatar
Linus Torvalds committed
1110

1111
	return ret < 0 || *i >= min_nr;
Linus Torvalds's avatar
Linus Torvalds committed
1112
1113
}

1114
static long read_events(struct kioctx *ctx, long min_nr, long nr,
Linus Torvalds's avatar
Linus Torvalds committed
1115
1116
1117
			struct io_event __user *event,
			struct timespec __user *timeout)
{
1118
1119
	ktime_t until = { .tv64 = KTIME_MAX };
	long ret = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1120
1121
1122

	if (timeout) {
		struct timespec	ts;
1123

Linus Torvalds's avatar
Linus Torvalds committed
1124
		if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1125
			return -EFAULT;
Linus Torvalds's avatar
Linus Torvalds committed
1126

1127
		until = timespec_to_ktime(ts);
Linus Torvalds's avatar
Linus Torvalds committed
1128
1129
	}

1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
	/*
	 * Note that aio_read_events() is being called as the conditional - i.e.
	 * we're calling it after prepare_to_wait() has set task state to
	 * TASK_INTERRUPTIBLE.
	 *
	 * But aio_read_events() can block, and if it blocks it's going to flip
	 * the task state back to TASK_RUNNING.
	 *
	 * This should be ok, provided it doesn't flip the state back to
	 * TASK_RUNNING and return 0 too much - that causes us to spin. That
	 * will only happen if the mutex_lock() call blocks, and we then find
	 * the ringbuffer empty. So in practice we should be ok, but it's
	 * something to be aware of when touching this code.
	 */
	wait_event_interruptible_hrtimeout(ctx->wait,
			aio_read_events(ctx, min_nr, nr, event, &ret), until);
Linus Torvalds's avatar
Linus Torvalds committed
1146