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
static struct vfsmount *aio_mnt;

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

168
169
170
171
172
173
174
175
176
/* Backing dev info for aio fs.
 * -no dirty page accounting or writeback happens
 */
static struct backing_dev_info aio_fs_backing_dev_info = {
	.name           = "aiofs",
	.state          = 0,
	.capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_MAP_COPY,
};

177
178
179
180
181
182
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);
183
184
	if (IS_ERR(inode))
		return ERR_CAST(inode);
185
186
187

	inode->i_mapping->a_ops = &aio_ctx_aops;
	inode->i_mapping->private_data = ctx;
188
	inode->i_mapping->backing_dev_info = &aio_fs_backing_dev_info;
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
	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,
	};
215
	return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC);
216
217
}

Linus Torvalds's avatar
Linus Torvalds committed
218
219
220
221
222
223
/* 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)
{
224
225
226
227
228
229
230
231
232
	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.");

233
234
235
	if (bdi_init(&aio_fs_backing_dev_info))
		panic("Failed to init aio fs backing dev info.");

236
237
	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
238

Kent Overstreet's avatar
Kent Overstreet committed
239
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
240
241
242

	return 0;
}
243
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
244

245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
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
261
262
static void aio_free_ring(struct kioctx *ctx)
{
263
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
264

265
266
267
268
269
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

270
	for (i = 0; i < ctx->nr_pages; i++) {
271
		struct page *page;
272
273
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
274
275
276
277
278
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
279
	}
Linus Torvalds's avatar
Linus Torvalds committed
280

281
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
282
		kfree(ctx->ring_pages);
283
284
		ctx->ring_pages = NULL;
	}
285
286
287
288
289
290
291
292
293
294
295
296
}

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,
};

297
#if IS_ENABLED(CONFIG_MIGRATION)
298
299
300
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
301
	struct kioctx *ctx;
302
	unsigned long flags;
303
	pgoff_t idx;
304
305
	int rc;

306
307
	rc = 0;

308
	/* mapping->private_lock here protects against the kioctx teardown.  */
309
310
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
	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;
330
331
332
333
	} else
		rc = -EINVAL;

	if (rc != 0)
334
		goto out_unlock;
335

336
337
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
338
	get_page(new);
339

340
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
341
	if (rc != MIGRATEPAGE_SUCCESS) {
342
		put_page(new);
343
		goto out_unlock;
344
345
	}

346
347
348
	/* 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.
349
	 */
350
351
352
353
354
	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);
355

356
357
	/* The old page is no longer accessible. */
	put_page(old);
358

359
360
361
362
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
363
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
364
}
365
#endif
Linus Torvalds's avatar
Linus Torvalds committed
366

367
static const struct address_space_operations aio_ctx_aops = {
368
	.set_page_dirty = __set_page_dirty_no_writeback,
369
#if IS_ENABLED(CONFIG_MIGRATION)
370
	.migratepage	= aio_migratepage,
371
#endif
372
373
};

Linus Torvalds's avatar
Linus Torvalds committed
374
375
376
377
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
378
	struct mm_struct *mm = current->mm;
379
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
380
	int nr_pages;
381
382
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
383
384
385
386
387
388
389

	/* 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;

390
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
391
392
393
	if (nr_pages < 0)
		return -EINVAL;

394
	file = aio_private_file(ctx, nr_pages);
395
396
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
397
		return -ENOMEM;
398
399
	}

400
401
402
403
404
405
406
407
408
409
410
411
412
413
	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;
		}
	}

414
415
416
417
418
419
420
421
422
423
	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);
424
425

		ctx->ring_pages[i] = page;
426
	}
427
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
428

429
430
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
431
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
432
433
	}

434
435
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
436

Zach Brown's avatar
Zach Brown committed
437
	down_write(&mm->mmap_sem);
438
439
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
440
441
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
442
443
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
444
		aio_free_ring(ctx);
445
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
446
447
	}

448
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
449

450
451
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
452

453
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
454
	ring->nr = nr_events;	/* user copy */
455
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
456
457
458
459
460
	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);
461
	kunmap_atomic(ring);
462
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
463
464
465
466
467
468
469
470

	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)

471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
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);

487
static int kiocb_cancel(struct kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
488
{
489
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
490

491
492
493
494
495
496
497
498
	/*
	 * 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
499
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
500

501
502
503
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
504

Kent Overstreet's avatar
Kent Overstreet committed
505
	return cancel(kiocb);
Kent Overstreet's avatar
Kent Overstreet committed
506
507
}

Kent Overstreet's avatar
Kent Overstreet committed
508
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
509
{
Kent Overstreet's avatar
Kent Overstreet committed
510
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
511

Kent Overstreet's avatar
Kent Overstreet committed
512
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
513

Kent Overstreet's avatar
Kent Overstreet committed
514
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
515
	free_percpu(ctx->cpu);
516
517
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
518
519
520
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
521
522
523
524
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

525
526
527
528
	/* 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
529
530
531
532
	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

Kent Overstreet's avatar
Kent Overstreet committed
533
534
535
536
537
/*
 * 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
538
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
539
{
Kent Overstreet's avatar
Kent Overstreet committed
540
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
Kent Overstreet's avatar
Kent Overstreet committed
541
542
543
544
545
546
547
548
549
	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);
550
		kiocb_cancel(req);
Kent Overstreet's avatar
Kent Overstreet committed
551
552
553
554
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
555
556
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
557
558
}

559
560
561
562
563
564
565
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);
566
	table = rcu_dereference_raw(mm->ioctx_table);
567
568
569
570
571
572
573
574
575

	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);

576
577
578
579
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
					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);
597
		old = rcu_dereference_raw(mm->ioctx_table);
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613

		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
614
615
616
617
618
619
620
621
622
623
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
624
625
626
627
628
/* 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
629
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
630
	struct kioctx *ctx;
631
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
632

Kent Overstreet's avatar
Kent Overstreet committed
633
634
635
636
637
638
639
640
641
642
643
644
	/*
	 * 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
645
646
647
648
649
650
651
	/* 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);
	}

652
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
653
654
		return ERR_PTR(-EAGAIN);

655
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
656
657
658
659
660
661
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
662
	spin_lock_init(&ctx->completion_lock);
663
	mutex_init(&ctx->ring_lock);
664
665
666
	/* 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
667
668
669
670
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

671
	if (percpu_ref_init(&ctx->users, free_ioctx_users, 0, GFP_KERNEL))
672
673
		goto err;

674
	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs, 0, GFP_KERNEL))
675
676
		goto err;

Kent Overstreet's avatar
Kent Overstreet committed
677
678
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
679
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
680

681
682
	err = aio_setup_ring(ctx);
	if (err < 0)
Kent Overstreet's avatar
Kent Overstreet committed
683
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
684

685
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
686
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
687
688
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
689

Linus Torvalds's avatar
Linus Torvalds committed
690
	/* limit the number of system wide aios */
691
	spin_lock(&aio_nr_lock);
692
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
693
	    aio_nr + nr_events < aio_nr) {
694
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
695
		err = -EAGAIN;
696
		goto err_ctx;
697
698
	}
	aio_nr += ctx->max_reqs;
699
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
700

701
702
	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
703

704
705
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
706
		goto err_cleanup;
707

708
709
710
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

Kent Overstreet's avatar
Kent Overstreet committed
711
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
712
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
713
714
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
715
716
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
717
718
err_ctx:
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
719
err:
720
	mutex_unlock(&ctx->ring_lock);
Kent Overstreet's avatar
Kent Overstreet committed
721
	free_percpu(ctx->cpu);
722
723
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Linus Torvalds's avatar
Linus Torvalds committed
724
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
725
	pr_debug("error allocating ioctx %d\n", err);
726
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
727
728
}

Kent Overstreet's avatar
Kent Overstreet committed
729
730
731
732
733
/* 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.
 */
734
static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
735
		struct completion *requests_done)
Kent Overstreet's avatar
Kent Overstreet committed
736
{
737
	struct kioctx_table *table;
738

739
740
	if (atomic_xchg(&ctx->dead, 1))
		return -EINVAL;
741
742


743
	spin_lock(&mm->ioctx_lock);
744
	table = rcu_dereference_raw(mm->ioctx_table);
745
746
747
	WARN_ON(ctx != table->table[ctx->id]);
	table->table[ctx->id] = NULL;
	spin_unlock(&mm->ioctx_lock);
748

749
750
	/* percpu_ref_kill() will do the necessary call_rcu() */
	wake_up_all(&ctx->wait);
751

752
753
754
755
756
757
758
759
	/*
	 * 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);
760

761
762
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
763

764
765
766
	ctx->requests_done = requests_done;
	percpu_ref_kill(&ctx->users);
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
767
768
769
770
771
}

/* wait_on_sync_kiocb:
 *	Waits on the given sync kiocb to complete.
 */
Kent Overstreet's avatar
Kent Overstreet committed
772
ssize_t wait_on_sync_kiocb(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
773
{
Kent Overstreet's avatar
Kent Overstreet committed
774
	while (!req->ki_ctx) {
Linus Torvalds's avatar
Linus Torvalds committed
775
		set_current_state(TASK_UNINTERRUPTIBLE);
Kent Overstreet's avatar
Kent Overstreet committed
776
		if (req->ki_ctx)
Linus Torvalds's avatar
Linus Torvalds committed
777
			break;
778
		io_schedule();
Linus Torvalds's avatar
Linus Torvalds committed
779
780
	}
	__set_current_state(TASK_RUNNING);
Kent Overstreet's avatar
Kent Overstreet committed
781
	return req->ki_user_data;
Linus Torvalds's avatar
Linus Torvalds committed
782
}
783
EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds's avatar
Linus Torvalds committed
784

Kent Overstreet's avatar
Kent Overstreet committed
785
786
787
788
789
790
791
/*
 * 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
792
 */
793
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
794
{
795
796
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
	int i;
797

798
799
	if (!table)
		return;
800

801
802
	for (i = 0; i < table->nr; ++i) {
		struct kioctx *ctx = table->table[i];
803
804
		struct completion requests_done =
			COMPLETION_INITIALIZER_ONSTACK(requests_done);
805

806
807
		if (!ctx)
			continue;
808
		/*
809
810
811
812
813
		 * 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.
814
		 */
815
		ctx->mmap_size = 0;
816
		kill_ioctx(mm, ctx, &requests_done);
Kent Overstreet's avatar
Kent Overstreet committed
817

818
819
		/* Wait until all IO for the context are done. */
		wait_for_completion(&requests_done);
Linus Torvalds's avatar
Linus Torvalds committed
820
	}
821
822
823

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
Linus Torvalds's avatar
Linus Torvalds committed
824
825
}

Kent Overstreet's avatar
Kent Overstreet committed
826
827
828
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
829
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
830

831
	local_irq_save(flags);
832
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
833
	kcpu->reqs_available += nr;
834

Kent Overstreet's avatar
Kent Overstreet committed
835
836
837
838
839
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

840
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
841
842
843
844
845
846
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;
847
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
848

849
	local_irq_save(flags);
850
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
	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:
869
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
870
871
872
	return ret;
}

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
931
932
933
934
/* 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
935
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
936
937
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
938
 */
Kent Overstreet's avatar
Kent Overstreet committed
939
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
940
{
Kent Overstreet's avatar
Kent Overstreet committed
941
942
	struct kiocb *req;

943
944
945
946
947
	if (!get_reqs_available(ctx)) {
		user_refill_reqs_available(ctx);
		if (!get_reqs_available(ctx))
			return NULL;
	}
Kent Overstreet's avatar
Kent Overstreet committed
948

949
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
950
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
951
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
952

Kent Overstreet's avatar
Kent Overstreet committed
953
954
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
955
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
956
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
957
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
958
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
959
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
960
961
}

962
static void kiocb_free(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
963
{
964
965
	if (req->ki_filp)
		fput(req->ki_filp);
966
967
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
968
969
970
	kmem_cache_free(kiocb_cachep, req);
}

971
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
972
{
973
	struct aio_ring __user *ring  = (void __user *)ctx_id;
974
	struct mm_struct *mm = current->mm;
975
	struct kioctx *ctx, *ret = NULL;
976
977
978
979
980
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
981

982
	rcu_read_lock();
983
	table = rcu_dereference(mm->ioctx_table);
984

985
986
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
987

988
	ctx = table->table[id];
989
	if (ctx && ctx->user_id == ctx_id) {
990
991
992
993
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
994
	rcu_read_unlock();
995
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
996
997
998
999
1000
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
1001
void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
1002
1003
1004
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
1005
	struct io_event	*ev_page, *event;
1006
	unsigned tail, pos, head;
Linus Torvalds's avatar
Linus Torvalds committed
1007
1008
	unsigned long	flags;

1009
1010
1011
1012
1013
1014
	/*
	 * 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
1015
1016
1017
	 */
	if (is_sync_kiocb(iocb)) {
		iocb->ki_user_data = res;
Kent Overstreet's avatar
Kent Overstreet committed
1018
1019
		smp_wmb();
		iocb->ki_ctx = ERR_PTR(-EXDEV);
Linus Torvalds's avatar
Linus Torvalds committed
1020
		wake_up_process(iocb->ki_obj.tsk);
1021
		return;
Linus Torvalds's avatar
Linus Torvalds committed
1022
1023
	}

1024
1025
1026
1027
1028
1029
1030
	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);
	}
1031

1032
1033
	/*
	 * Add a completion event to the ring buffer. Must be done holding
1034
	 * ctx->completion_lock to prevent other code from messing with the tail
1035
1036
1037
1038
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

1039
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
1040
1041
	pos = tail + AIO_EVENTS_OFFSET;

1042
	if (++tail >= ctx->nr_events)
1043
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1044

1045
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1046
1047
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

Linus Torvalds's avatar
Linus Torvalds committed
1048
1049
1050
1051
1052
	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
1053
	kunmap_atomic(ev_page);
1054
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1055
1056

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreet's avatar
Kent Overstreet committed
1057
1058
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
1059
1060
1061
1062
1063
1064

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

1065
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
1066

1067
	ring = kmap_atomic(ctx->ring_pages[0]);
1068
	head = ring->head;
Kent Overstreet's avatar
Kent Overstreet committed
1069
	ring->tail = tail;
1070
	kunmap_atomic(ring);
1071
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
1072

1073
1074
1075
	ctx->completed_events++;
	if (ctx->completed_events > 1)
		refill_reqs_available(ctx, head, tail);
1076
1077
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
1078
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
1079
1080
1081
1082
1083
1084

	/*
	 * 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.
	 */
1085
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
1086
1087
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
1088
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
1089
	kiocb_free(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
1090

1091
1092
1093
1094
1095
1096
1097
1098
	/*
	 * 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
1099
1100
1101
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
1102
	percpu_ref_put(&ctx->reqs);
Linus Torvalds's avatar
Linus Torvalds committed
1103
}
1104
EXPORT_SYMBOL(aio_complete);
Linus Torvalds's avatar
Linus Torvalds committed
1105

Gu Zheng's avatar
Gu Zheng committed
1106
/* aio_read_events_ring
1107
1108
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
Linus Torvalds's avatar
Linus Torvalds committed
1109
 */
1110
1111
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
1112
1113
{
	struct aio_ring *ring;
1114
	unsigned head, tail, pos;
1115
1116
1117
	long ret = 0;
	int copy_ret;

1118
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1119

1120
	/* Access to ->ring_pages here is protected by ctx->ring_lock. */
1121
	ring = kmap_atomic(ctx->ring_pages[0]);
1122
	head = ring->head;
1123
	tail = ring->tail;
1124
1125
	kunmap_atomic(ring);

1126
1127
1128
1129
1130
1131
	/*
	 * 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();

1132
	pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
1133

1134
	if (head == tail)
Linus Torvalds's avatar
Linus Torvalds committed
1135
1136
		goto out;

1137
1138
1139
	head %= ctx->nr_events;
	tail %= ctx->nr_events;