aio.c 42.2 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

	struct {
		unsigned	tail;
144
		unsigned	completed_events;
145
		spinlock_t	completion_lock;
146
	} ____cacheline_aligned_in_smp;
147
148

	struct page		*internal_pages[AIO_RING_PAGES];
149
	struct file		*aio_ring_file;
150
151

	unsigned		id;
152
153
};

Linus Torvalds's avatar
Linus Torvalds committed
154
/*------ sysctl variables----*/
155
156
157
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
158
159
/*----end sysctl variables---*/

160
161
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
162

163
164
165
166
167
168
169
170
171
172
173
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);
174
175
	if (IS_ERR(inode))
		return ERR_CAST(inode);
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
204

	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,
	};
205
	return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC);
206
207
}

Linus Torvalds's avatar
Linus Torvalds committed
208
209
210
211
212
213
/* 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)
{
214
215
216
217
218
219
220
221
222
	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.");

223
224
	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
225

Kent Overstreet's avatar
Kent Overstreet committed
226
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
227
228
229

	return 0;
}
230
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
231

232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
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
248
249
static void aio_free_ring(struct kioctx *ctx)
{
250
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
251

252
253
254
255
256
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

257
	for (i = 0; i < ctx->nr_pages; i++) {
258
		struct page *page;
259
260
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
261
262
263
264
265
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
266
	}
Linus Torvalds's avatar
Linus Torvalds committed
267

268
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
269
		kfree(ctx->ring_pages);
270
271
		ctx->ring_pages = NULL;
	}
272
273
274
275
}

static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma)
{
276
	vma->vm_flags |= VM_DONTEXPAND;
277
278
279
280
	vma->vm_ops = &generic_file_vm_ops;
	return 0;
}

281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
static void aio_ring_remap(struct file *file, struct vm_area_struct *vma)
{
	struct mm_struct *mm = vma->vm_mm;
	struct kioctx_table *table;
	int i;

	spin_lock(&mm->ioctx_lock);
	rcu_read_lock();
	table = rcu_dereference(mm->ioctx_table);
	for (i = 0; i < table->nr; i++) {
		struct kioctx *ctx;

		ctx = table->table[i];
		if (ctx && ctx->aio_ring_file == file) {
			ctx->user_id = ctx->mmap_base = vma->vm_start;
			break;
		}
	}

	rcu_read_unlock();
	spin_unlock(&mm->ioctx_lock);
}

304
305
static const struct file_operations aio_ring_fops = {
	.mmap = aio_ring_mmap,
306
	.mremap = aio_ring_remap,
307
308
};

309
#if IS_ENABLED(CONFIG_MIGRATION)
310
311
312
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
313
	struct kioctx *ctx;
314
	unsigned long flags;
315
	pgoff_t idx;
316
317
	int rc;

318
319
	rc = 0;

320
	/* mapping->private_lock here protects against the kioctx teardown.  */
321
322
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
	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;
342
343
344
345
	} else
		rc = -EINVAL;

	if (rc != 0)
346
		goto out_unlock;
347

348
349
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
350
	get_page(new);
351

352
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
353
	if (rc != MIGRATEPAGE_SUCCESS) {
354
		put_page(new);
355
		goto out_unlock;
356
357
	}

358
359
360
	/* 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.
361
	 */
362
363
364
365
366
	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);
367

368
369
	/* The old page is no longer accessible. */
	put_page(old);
370

371
372
373
374
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
375
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
376
}
377
#endif
Linus Torvalds's avatar
Linus Torvalds committed
378

379
static const struct address_space_operations aio_ctx_aops = {
380
	.set_page_dirty = __set_page_dirty_no_writeback,
381
#if IS_ENABLED(CONFIG_MIGRATION)
382
	.migratepage	= aio_migratepage,
383
#endif
384
385
};

Linus Torvalds's avatar
Linus Torvalds committed
386
387
388
389
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
390
	struct mm_struct *mm = current->mm;
391
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
392
	int nr_pages;
393
394
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
395
396
397
398
399
400
401

	/* 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;

402
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
403
404
405
	if (nr_pages < 0)
		return -EINVAL;

406
	file = aio_private_file(ctx, nr_pages);
407
408
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
409
		return -ENOMEM;
410
411
	}

412
413
414
415
416
417
418
419
420
421
422
423
424
425
	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;
		}
	}

426
427
428
429
430
431
432
433
434
435
	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);
		unlock_page(page);
436
437

		ctx->ring_pages[i] = page;
438
	}
439
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
440

441
442
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
443
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
444
445
	}

446
447
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
448

Zach Brown's avatar
Zach Brown committed
449
	down_write(&mm->mmap_sem);
450
451
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
452
453
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
454
455
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
456
		aio_free_ring(ctx);
457
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
458
459
	}

460
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
461

462
463
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
464

465
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
466
	ring->nr = nr_events;	/* user copy */
467
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
468
469
470
471
472
	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);
473
	kunmap_atomic(ring);
474
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
475
476
477
478
479
480
481
482

	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)

483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
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);

499
static int kiocb_cancel(struct kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
500
{
501
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
502

503
504
505
506
507
508
509
510
	/*
	 * 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
511
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
512

513
514
515
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
516

Kent Overstreet's avatar
Kent Overstreet committed
517
	return cancel(kiocb);
Kent Overstreet's avatar
Kent Overstreet committed
518
519
}

Kent Overstreet's avatar
Kent Overstreet committed
520
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
521
{
Kent Overstreet's avatar
Kent Overstreet committed
522
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
523

Kent Overstreet's avatar
Kent Overstreet committed
524
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
525

Kent Overstreet's avatar
Kent Overstreet committed
526
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
527
	free_percpu(ctx->cpu);
528
529
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
530
531
532
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
533
534
535
536
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

537
538
539
540
	/* 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
541
542
543
544
	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

Kent Overstreet's avatar
Kent Overstreet committed
545
546
547
548
549
/*
 * 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
550
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
551
{
Kent Overstreet's avatar
Kent Overstreet committed
552
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
Kent Overstreet's avatar
Kent Overstreet committed
553
554
555
556
557
558
559
560
561
	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);
562
		kiocb_cancel(req);
Kent Overstreet's avatar
Kent Overstreet committed
563
564
565
566
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
567
568
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
569
570
}

571
572
573
574
575
576
577
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);
578
	table = rcu_dereference_raw(mm->ioctx_table);
579
580
581
582
583
584
585
586
587

	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);

588
589
590
591
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
					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);
609
		old = rcu_dereference_raw(mm->ioctx_table);
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625

		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
626
627
628
629
630
631
632
633
634
635
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
636
637
638
639
640
/* 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
641
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
642
	struct kioctx *ctx;
643
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
644

Kent Overstreet's avatar
Kent Overstreet committed
645
646
647
648
649
650
651
652
653
654
655
656
	/*
	 * 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
657
658
659
660
661
662
663
	/* 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);
	}

664
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
665
666
		return ERR_PTR(-EAGAIN);

667
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
668
669
670
671
672
673
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
674
	spin_lock_init(&ctx->completion_lock);
675
	mutex_init(&ctx->ring_lock);
676
677
678
	/* 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
679
680
681
682
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

683
	if (percpu_ref_init(&ctx->users, free_ioctx_users, 0, GFP_KERNEL))
684
685
		goto err;

686
	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs, 0, GFP_KERNEL))
687
688
		goto err;

Kent Overstreet's avatar
Kent Overstreet committed
689
690
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
691
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
692

693
694
	err = aio_setup_ring(ctx);
	if (err < 0)
Kent Overstreet's avatar
Kent Overstreet committed
695
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
696

697
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
698
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
699
700
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
701

Linus Torvalds's avatar
Linus Torvalds committed
702
	/* limit the number of system wide aios */
703
	spin_lock(&aio_nr_lock);
704
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
705
	    aio_nr + nr_events < aio_nr) {
706
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
707
		err = -EAGAIN;
708
		goto err_ctx;
709
710
	}
	aio_nr += ctx->max_reqs;
711
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
712

713
714
	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
715

716
717
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
718
		goto err_cleanup;
719

720
721
722
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

Kent Overstreet's avatar
Kent Overstreet committed
723
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
724
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
725
726
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
727
728
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
729
730
err_ctx:
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
731
err:
732
	mutex_unlock(&ctx->ring_lock);
Kent Overstreet's avatar
Kent Overstreet committed
733
	free_percpu(ctx->cpu);
734
735
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Linus Torvalds's avatar
Linus Torvalds committed
736
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
737
	pr_debug("error allocating ioctx %d\n", err);
738
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
739
740
}

Kent Overstreet's avatar
Kent Overstreet committed
741
742
743
744
745
/* 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.
 */
746
static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
747
		struct completion *requests_done)
Kent Overstreet's avatar
Kent Overstreet committed
748
{
749
	struct kioctx_table *table;
750

751
752
	if (atomic_xchg(&ctx->dead, 1))
		return -EINVAL;
753
754


755
	spin_lock(&mm->ioctx_lock);
756
	table = rcu_dereference_raw(mm->ioctx_table);
757
758
759
	WARN_ON(ctx != table->table[ctx->id]);
	table->table[ctx->id] = NULL;
	spin_unlock(&mm->ioctx_lock);
760

761
762
	/* percpu_ref_kill() will do the necessary call_rcu() */
	wake_up_all(&ctx->wait);
763

764
765
766
767
768
769
770
771
	/*
	 * 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);
772

773
774
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
775

776
777
778
	ctx->requests_done = requests_done;
	percpu_ref_kill(&ctx->users);
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
779
780
}

Kent Overstreet's avatar
Kent Overstreet committed
781
782
783
784
785
786
787
/*
 * 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
788
 */
789
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
790
{
791
792
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
	int i;
793

794
795
	if (!table)
		return;
796

797
798
	for (i = 0; i < table->nr; ++i) {
		struct kioctx *ctx = table->table[i];
799
800
		struct completion requests_done =
			COMPLETION_INITIALIZER_ONSTACK(requests_done);
801

802
803
		if (!ctx)
			continue;
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. 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.
810
		 */
811
		ctx->mmap_size = 0;
812
		kill_ioctx(mm, ctx, &requests_done);
Kent Overstreet's avatar
Kent Overstreet committed
813

814
815
		/* Wait until all IO for the context are done. */
		wait_for_completion(&requests_done);
Linus Torvalds's avatar
Linus Torvalds committed
816
	}
817
818
819

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
Linus Torvalds's avatar
Linus Torvalds committed
820
821
}

Kent Overstreet's avatar
Kent Overstreet committed
822
823
824
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
825
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
826

827
	local_irq_save(flags);
828
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
829
	kcpu->reqs_available += nr;
830

Kent Overstreet's avatar
Kent Overstreet committed
831
832
833
834
835
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

836
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
837
838
839
840
841
842
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;
843
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
844

845
	local_irq_save(flags);
846
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
	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:
865
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
866
867
868
	return ret;
}

869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
/* refill_reqs_available
 *	Updates the reqs_available reference counts used for tracking the
 *	number of free slots in the completion ring.  This can be called
 *	from aio_complete() (to optimistically update reqs_available) or
 *	from aio_get_req() (the we're out of events case).  It must be
 *	called holding ctx->completion_lock.
 */
static void refill_reqs_available(struct kioctx *ctx, unsigned head,
                                  unsigned tail)
{
	unsigned events_in_ring, completed;

	/* Clamp head since userland can write to it. */
	head %= ctx->nr_events;
	if (head <= tail)
		events_in_ring = tail - head;
	else
		events_in_ring = ctx->nr_events - (head - tail);

	completed = ctx->completed_events;
	if (events_in_ring < completed)
		completed -= events_in_ring;
	else
		completed = 0;

	if (!completed)
		return;

	ctx->completed_events -= completed;
	put_reqs_available(ctx, completed);
}

/* user_refill_reqs_available
 *	Called to refill reqs_available when aio_get_req() encounters an
 *	out of space in the completion ring.
 */
static void user_refill_reqs_available(struct kioctx *ctx)
{
	spin_lock_irq(&ctx->completion_lock);
	if (ctx->completed_events) {
		struct aio_ring *ring;
		unsigned head;

		/* Access of ring->head may race with aio_read_events_ring()
		 * here, but that's okay since whether we read the old version
		 * or the new version, and either will be valid.  The important
		 * part is that head cannot pass tail since we prevent
		 * aio_complete() from updating tail by holding
		 * ctx->completion_lock.  Even if head is invalid, the check
		 * against ctx->completed_events below will make sure we do the
		 * safe/right thing.
		 */
		ring = kmap_atomic(ctx->ring_pages[0]);
		head = ring->head;
		kunmap_atomic(ring);

		refill_reqs_available(ctx, head, ctx->tail);
	}

	spin_unlock_irq(&ctx->completion_lock);
}

Linus Torvalds's avatar
Linus Torvalds committed
931
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
932
933
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
934
 */
Kent Overstreet's avatar
Kent Overstreet committed
935
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
936
{
Kent Overstreet's avatar
Kent Overstreet committed
937
938
	struct kiocb *req;

939
940
941
942
943
	if (!get_reqs_available(ctx)) {
		user_refill_reqs_available(ctx);
		if (!get_reqs_available(ctx))
			return NULL;
	}
Kent Overstreet's avatar
Kent Overstreet committed
944

945
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
946
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
947
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
948

Kent Overstreet's avatar
Kent Overstreet committed
949
950
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
951
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
952
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
953
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
954
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
955
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
956
957
}

958
static void kiocb_free(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
959
{
960
961
	if (req->ki_filp)
		fput(req->ki_filp);
962
963
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
964
965
966
	kmem_cache_free(kiocb_cachep, req);
}

967
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
968
{
969
	struct aio_ring __user *ring  = (void __user *)ctx_id;
970
	struct mm_struct *mm = current->mm;
971
	struct kioctx *ctx, *ret = NULL;
972
973
974
975
976
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
977

978
	rcu_read_lock();
979
	table = rcu_dereference(mm->ioctx_table);
980

981
982
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
983

984
	ctx = table->table[id];
985
	if (ctx && ctx->user_id == ctx_id) {
986
987
988
989
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
990
	rcu_read_unlock();
991
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
992
993
994
995
996
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
997
void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
998
999
1000
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
1001
	struct io_event	*ev_page, *event;
1002
	unsigned tail, pos, head;
Linus Torvalds's avatar
Linus Torvalds committed
1003
1004
	unsigned long	flags;

1005
1006
1007
1008
1009
1010
	/*
	 * 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
1011
	 */
1012
	BUG_ON(is_sync_kiocb(iocb));
Linus Torvalds's avatar
Linus Torvalds committed
1013

1014
1015
1016
1017
1018
1019
1020
	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);
	}
1021

1022
1023
	/*
	 * Add a completion event to the ring buffer. Must be done holding
1024
	 * ctx->completion_lock to prevent other code from messing with the tail
1025
1026
1027
1028
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

1029
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
1030
1031
	pos = tail + AIO_EVENTS_OFFSET;

1032
	if (++tail >= ctx->nr_events)
1033
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1034

1035
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1036
1037
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

Linus Torvalds's avatar
Linus Torvalds committed
1038
1039
1040
1041
1042
	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
1043
	kunmap_atomic(ev_page);
1044
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1045
1046

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreet's avatar
Kent Overstreet committed
1047
1048
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
1049
1050
1051
1052
1053
1054

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

1055
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
1056

1057
	ring = kmap_atomic(ctx->ring_pages[0]);
1058
	head = ring->head;
Kent Overstreet's avatar
Kent Overstreet committed
1059
	ring->tail = tail;
1060
	kunmap_atomic(ring);
1061
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
1062

1063
1064
1065
	ctx->completed_events++;
	if (ctx->completed_events > 1)
		refill_reqs_available(ctx, head, tail);
1066
1067
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
1068
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
1069
1070
1071
1072
1073
1074

	/*
	 * 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.
	 */
1075
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
1076
1077
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
1078
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
1079
	kiocb_free(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
1080

1081
1082
1083
1084
1085
1086
1087
1088
	/*
	 * 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
1089
1090
1091
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
1092
	percpu_ref_put(&ctx->reqs);
Linus Torvalds's avatar
Linus Torvalds committed
1093
}
1094
EXPORT_SYMBOL(aio_complete);
Linus Torvalds's avatar
Linus Torvalds committed
1095

Gu Zheng's avatar
Gu Zheng committed
1096
/* aio_read_events_ring
1097
1098
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
Linus Torvalds's avatar
Linus Torvalds committed
1099
 */
1100
1101
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
1102
1103
{
	struct aio_ring *ring;
1104
	unsigned head, tail, pos;
1105
1106
1107
	long ret = 0;
	int copy_ret;

1108
1109
1110
1111
1112
1113
1114
	/*
	 * The mutex can block and wake us up and that will cause
	 * wait_event_interruptible_hrtimeout() to schedule without sleeping
	 * and repeat. This should be rare enough that it doesn't cause
	 * peformance issues. See the comment in read_events() for more detail.
	 */
	sched_annotate_sleep();
1115
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1116

1117
	/* Access to ->ring_pages here is protected by ctx->ring_lock. */
1118
	ring = kmap_atomic(ctx->ring_pages[0]);
1119
	head = ring->head;
1120
	tail = ring->tail;
1121
1122
	kunmap_atomic(ring);

1123
1124
1125
1126
1127
1128
	/*
	 * Ensure that once we've read the current tail pointer, that
	 * we also see the events that were stored up to the tail.
	 */
	smp_rmb();

1129
	pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
1130

1131
	if (head == tail)
Linus Torvalds's avatar
Linus Torvalds committed
1132
1133
		goto out;

1134
1135
1136
	head %= ctx->nr_events;
	tail %= ctx->nr_events;

1137
1138
1139
1140
1141
	while (ret < nr) {
		long avail;
		struct io_event *ev;
		struct page *page;

1142
1143
		avail = (head <= tail ?  tail : ctx->nr_events) - head;
		if (head == tail)
1144
1145
1146
1147
1148
1149
1150
			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;
1151
		page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
		pos %= AIO_EVENTS_PER_PAGE;

		ev = kmap(page);
		copy_ret = copy_to_user(event + ret, ev + pos,