aio.c 42.9 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
};

154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
/*
 * We use ki_cancel == KIOCB_CANCELLED to indicate that a kiocb has been either
 * cancelled or completed (this makes a certain amount of sense because
 * successful cancellation - io_cancel() - does deliver the completion to
 * userspace).
 *
 * And since most things don't implement kiocb cancellation and we'd really like
 * kiocb completion to be lockless when possible, we use ki_cancel to
 * synchronize cancellation and completion - we only set it to KIOCB_CANCELLED
 * with xchg() or cmpxchg(), see batch_complete_aio() and kiocb_cancel().
 */
#define KIOCB_CANCELLED		((void *) (~0ULL))

struct aio_kiocb {
	struct kiocb		common;

	struct kioctx		*ki_ctx;
	kiocb_cancel_fn		*ki_cancel;

	struct iocb __user	*ki_user_iocb;	/* user's aiocb */
	__u64			ki_user_data;	/* user's data for completion */

	struct list_head	ki_list;	/* the aio core uses this
						 * for cancellation */

	/*
	 * If the aio_resfd field of the userspace iocb is not zero,
	 * this is the underlying eventfd context to deliver events to.
	 */
	struct eventfd_ctx	*ki_eventfd;
};

Linus Torvalds's avatar
Linus Torvalds committed
186
/*------ sysctl variables----*/
187
188
189
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
190
191
/*----end sysctl variables---*/

192
193
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
194

195
196
197
198
199
200
201
202
203
204
205
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);
206
207
	if (IS_ERR(inode))
		return ERR_CAST(inode);
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236

	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,
	};
237
	return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC);
238
239
}

Linus Torvalds's avatar
Linus Torvalds committed
240
241
242
243
244
245
/* 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)
{
246
247
248
249
250
251
252
253
254
	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.");

255
	kiocb_cachep = KMEM_CACHE(aio_kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
256
	kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds's avatar
Linus Torvalds committed
257

Kent Overstreet's avatar
Kent Overstreet committed
258
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
259
260
261

	return 0;
}
262
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
263

264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
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
280
281
static void aio_free_ring(struct kioctx *ctx)
{
282
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
283

284
285
286
287
288
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

289
	for (i = 0; i < ctx->nr_pages; i++) {
290
		struct page *page;
291
292
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
293
294
295
296
297
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
298
	}
Linus Torvalds's avatar
Linus Torvalds committed
299

300
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
301
		kfree(ctx->ring_pages);
302
303
		ctx->ring_pages = NULL;
	}
304
305
306
307
}

static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma)
{
308
	vma->vm_flags |= VM_DONTEXPAND;
309
310
311
312
	vma->vm_ops = &generic_file_vm_ops;
	return 0;
}

313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
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);
}

336
337
static const struct file_operations aio_ring_fops = {
	.mmap = aio_ring_mmap,
338
	.mremap = aio_ring_remap,
339
340
};

341
#if IS_ENABLED(CONFIG_MIGRATION)
342
343
344
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
345
	struct kioctx *ctx;
346
	unsigned long flags;
347
	pgoff_t idx;
348
349
	int rc;

350
351
	rc = 0;

352
	/* mapping->private_lock here protects against the kioctx teardown.  */
353
354
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
	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;
374
375
376
377
	} else
		rc = -EINVAL;

	if (rc != 0)
378
		goto out_unlock;
379

380
381
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
382
	get_page(new);
383

384
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
385
	if (rc != MIGRATEPAGE_SUCCESS) {
386
		put_page(new);
387
		goto out_unlock;
388
389
	}

390
391
392
	/* 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.
393
	 */
394
395
396
397
398
	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);
399

400
401
	/* The old page is no longer accessible. */
	put_page(old);
402

403
404
405
406
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
407
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
408
}
409
#endif
Linus Torvalds's avatar
Linus Torvalds committed
410

411
static const struct address_space_operations aio_ctx_aops = {
412
	.set_page_dirty = __set_page_dirty_no_writeback,
413
#if IS_ENABLED(CONFIG_MIGRATION)
414
	.migratepage	= aio_migratepage,
415
#endif
416
417
};

Linus Torvalds's avatar
Linus Torvalds committed
418
419
420
421
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
422
	struct mm_struct *mm = current->mm;
423
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
424
	int nr_pages;
425
426
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
427
428
429
430
431
432
433

	/* 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;

434
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
435
436
437
	if (nr_pages < 0)
		return -EINVAL;

438
	file = aio_private_file(ctx, nr_pages);
439
440
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
441
		return -ENOMEM;
442
443
	}

444
445
446
447
448
449
450
451
452
453
454
455
456
457
	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;
		}
	}

458
459
460
461
462
463
464
465
466
467
	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);
468
469

		ctx->ring_pages[i] = page;
470
	}
471
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
472

473
474
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
475
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
476
477
	}

478
479
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
480

Zach Brown's avatar
Zach Brown committed
481
	down_write(&mm->mmap_sem);
482
483
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
484
485
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
486
487
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
488
		aio_free_ring(ctx);
489
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
490
491
	}

492
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
493

494
495
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
496

497
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
498
	ring->nr = nr_events;	/* user copy */
499
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
500
501
502
503
504
	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);
505
	kunmap_atomic(ring);
506
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
507
508
509
510
511
512
513
514

	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)

515
void kiocb_set_cancel_fn(struct kiocb *iocb, kiocb_cancel_fn *cancel)
516
{
517
	struct aio_kiocb *req = container_of(iocb, struct aio_kiocb, common);
518
519
520
521
522
523
524
525
526
527
528
529
530
531
	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);

532
static int kiocb_cancel(struct aio_kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
533
{
534
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
535

536
537
538
539
540
541
542
543
	/*
	 * 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
544
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
545

546
547
548
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
549

550
	return cancel(&kiocb->common);
Kent Overstreet's avatar
Kent Overstreet committed
551
552
}

Kent Overstreet's avatar
Kent Overstreet committed
553
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
554
{
Kent Overstreet's avatar
Kent Overstreet committed
555
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
556

Kent Overstreet's avatar
Kent Overstreet committed
557
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
558

Kent Overstreet's avatar
Kent Overstreet committed
559
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
560
	free_percpu(ctx->cpu);
561
562
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
563
564
565
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
566
567
568
569
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

570
571
572
573
	/* 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
574
575
576
577
	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

Kent Overstreet's avatar
Kent Overstreet committed
578
579
580
581
582
/*
 * 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
583
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
584
{
Kent Overstreet's avatar
Kent Overstreet committed
585
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
586
	struct aio_kiocb *req;
Kent Overstreet's avatar
Kent Overstreet committed
587
588
589
590
591

	spin_lock_irq(&ctx->ctx_lock);

	while (!list_empty(&ctx->active_reqs)) {
		req = list_first_entry(&ctx->active_reqs,
592
				       struct aio_kiocb, ki_list);
Kent Overstreet's avatar
Kent Overstreet committed
593
594

		list_del_init(&req->ki_list);
595
		kiocb_cancel(req);
Kent Overstreet's avatar
Kent Overstreet committed
596
597
598
599
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
600
601
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
602
603
}

604
605
606
607
608
609
610
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);
611
	table = rcu_dereference_raw(mm->ioctx_table);
612
613
614
615
616
617
618
619
620

	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);

621
622
623
624
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
					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);
642
		old = rcu_dereference_raw(mm->ioctx_table);
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658

		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
659
660
661
662
663
664
665
666
667
668
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
669
670
671
672
673
/* 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
674
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
675
	struct kioctx *ctx;
676
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
677

Kent Overstreet's avatar
Kent Overstreet committed
678
679
680
681
682
683
684
685
686
687
688
689
	/*
	 * 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
690
691
692
693
694
695
696
	/* 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);
	}

697
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
698
699
		return ERR_PTR(-EAGAIN);

700
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
701
702
703
704
705
706
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
707
	spin_lock_init(&ctx->completion_lock);
708
	mutex_init(&ctx->ring_lock);
709
710
711
	/* 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
712
713
714
715
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

716
	if (percpu_ref_init(&ctx->users, free_ioctx_users, 0, GFP_KERNEL))
717
718
		goto err;

719
	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs, 0, GFP_KERNEL))
720
721
		goto err;

Kent Overstreet's avatar
Kent Overstreet committed
722
723
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
724
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
725

726
727
	err = aio_setup_ring(ctx);
	if (err < 0)
Kent Overstreet's avatar
Kent Overstreet committed
728
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
729

730
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
731
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
732
733
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
734

Linus Torvalds's avatar
Linus Torvalds committed
735
	/* limit the number of system wide aios */
736
	spin_lock(&aio_nr_lock);
737
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
738
	    aio_nr + nr_events < aio_nr) {
739
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
740
		err = -EAGAIN;
741
		goto err_ctx;
742
743
	}
	aio_nr += ctx->max_reqs;
744
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
745

746
747
	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
748

749
750
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
751
		goto err_cleanup;
752

753
754
755
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

Kent Overstreet's avatar
Kent Overstreet committed
756
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
757
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
758
759
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
760
761
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
762
763
err_ctx:
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
764
err:
765
	mutex_unlock(&ctx->ring_lock);
Kent Overstreet's avatar
Kent Overstreet committed
766
	free_percpu(ctx->cpu);
767
768
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Linus Torvalds's avatar
Linus Torvalds committed
769
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
770
	pr_debug("error allocating ioctx %d\n", err);
771
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
772
773
}

Kent Overstreet's avatar
Kent Overstreet committed
774
775
776
777
778
/* 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.
 */
779
static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
780
		struct completion *requests_done)
Kent Overstreet's avatar
Kent Overstreet committed
781
{
782
	struct kioctx_table *table;
783

784
785
	if (atomic_xchg(&ctx->dead, 1))
		return -EINVAL;
786
787


788
	spin_lock(&mm->ioctx_lock);
789
	table = rcu_dereference_raw(mm->ioctx_table);
790
791
792
	WARN_ON(ctx != table->table[ctx->id]);
	table->table[ctx->id] = NULL;
	spin_unlock(&mm->ioctx_lock);
793

794
795
	/* percpu_ref_kill() will do the necessary call_rcu() */
	wake_up_all(&ctx->wait);
796

797
798
799
800
801
802
803
804
	/*
	 * 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);
805

806
807
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
808

809
810
811
	ctx->requests_done = requests_done;
	percpu_ref_kill(&ctx->users);
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
812
813
}

Kent Overstreet's avatar
Kent Overstreet committed
814
815
816
817
818
819
820
/*
 * 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
821
 */
822
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
823
{
824
825
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
	int i;
826

827
828
	if (!table)
		return;
829

830
831
	for (i = 0; i < table->nr; ++i) {
		struct kioctx *ctx = table->table[i];
832
833
		struct completion requests_done =
			COMPLETION_INITIALIZER_ONSTACK(requests_done);
834

835
836
		if (!ctx)
			continue;
837
		/*
838
839
840
841
842
		 * 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.
843
		 */
844
		ctx->mmap_size = 0;
845
		kill_ioctx(mm, ctx, &requests_done);
Kent Overstreet's avatar
Kent Overstreet committed
846

847
848
		/* Wait until all IO for the context are done. */
		wait_for_completion(&requests_done);
Linus Torvalds's avatar
Linus Torvalds committed
849
	}
850
851
852

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
Linus Torvalds's avatar
Linus Torvalds committed
853
854
}

Kent Overstreet's avatar
Kent Overstreet committed
855
856
857
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
858
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
859

860
	local_irq_save(flags);
861
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
862
	kcpu->reqs_available += nr;
863

Kent Overstreet's avatar
Kent Overstreet committed
864
865
866
867
868
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

869
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
870
871
872
873
874
875
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;
876
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
877

878
	local_irq_save(flags);
879
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
	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:
898
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
899
900
901
	return ret;
}

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
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
/* 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
964
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
965
966
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
967
 */
968
static inline struct aio_kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
969
{
970
	struct aio_kiocb *req;
Kent Overstreet's avatar
Kent Overstreet committed
971

972
973
974
975
976
	if (!get_reqs_available(ctx)) {
		user_refill_reqs_available(ctx);
		if (!get_reqs_available(ctx))
			return NULL;
	}
Kent Overstreet's avatar
Kent Overstreet committed
977

978
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
979
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
980
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
981

Kent Overstreet's avatar
Kent Overstreet committed
982
983
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
984
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
985
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
986
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
987
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
988
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
989
990
}

991
static void kiocb_free(struct aio_kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
992
{
993
994
	if (req->common.ki_filp)
		fput(req->common.ki_filp);
995
996
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
997
998
999
	kmem_cache_free(kiocb_cachep, req);
}

1000
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
1001
{
1002
	struct aio_ring __user *ring  = (void __user *)ctx_id;
1003
	struct mm_struct *mm = current->mm;
1004
	struct kioctx *ctx, *ret = NULL;
1005
1006
1007
1008
1009
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1010

1011
	rcu_read_lock();
1012
	table = rcu_dereference(mm->ioctx_table);
1013

1014
1015
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
1016

1017
	ctx = table->table[id];
1018
	if (ctx && ctx->user_id == ctx_id) {
1019
1020
1021
1022
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
1023
	rcu_read_unlock();
1024
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
1025
1026
1027
1028
1029
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
1030
static void aio_complete(struct kiocb *kiocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
1031
{
1032
	struct aio_kiocb *iocb = container_of(kiocb, struct aio_kiocb, common);
Linus Torvalds's avatar
Linus Torvalds committed
1033
1034
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
1035
	struct io_event	*ev_page, *event;
1036
	unsigned tail, pos, head;
Linus Torvalds's avatar
Linus Torvalds committed
1037
1038
	unsigned long	flags;

1039
1040
1041
1042
1043
1044
	/*
	 * 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
1045
	 */
1046
	BUG_ON(is_sync_kiocb(kiocb));
Linus Torvalds's avatar
Linus Torvalds committed
1047

1048
1049
1050
1051
1052
1053
1054
	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);
	}
1055

1056
1057
	/*
	 * Add a completion event to the ring buffer. Must be done holding
1058
	 * ctx->completion_lock to prevent other code from messing with the tail
1059
1060
1061
1062
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

1063
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
1064
1065
	pos = tail + AIO_EVENTS_OFFSET;

1066
	if (++tail >= ctx->nr_events)
1067
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1068

1069
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1070
1071
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

1072
	event->obj = (u64)(unsigned long)iocb->ki_user_iocb;
Linus Torvalds's avatar
Linus Torvalds committed
1073
1074
1075
1076
	event->data = iocb->ki_user_data;
	event->res = res;
	event->res2 = res2;

Kent Overstreet's avatar
Kent Overstreet committed
1077
	kunmap_atomic(ev_page);
1078
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1079
1080

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
1081
		 ctx, tail, iocb, iocb->ki_user_iocb, iocb->ki_user_data,
Kent Overstreet's avatar
Kent Overstreet committed
1082
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
1083
1084
1085
1086
1087
1088

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

1089
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
1090

1091
	ring = kmap_atomic(ctx->ring_pages[0]);
1092
	head = ring->head;
Kent Overstreet's avatar
Kent Overstreet committed
1093
	ring->tail = tail;
1094
	kunmap_atomic(ring);
1095
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
1096

1097
1098
1099
	ctx->completed_events++;
	if (ctx->completed_events > 1)
		refill_reqs_available(ctx, head, tail);
1100
1101
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
1102
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
1103
1104
1105
1106
1107
1108

	/*
	 * 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.
	 */
1109
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
1110
1111
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
1112
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
1113
	kiocb_free(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
1114

1115
1116
1117
1118
1119
1120
1121
1122
	/*
	 * 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
1123
1124
1125
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
1126
	percpu_ref_put(&ctx->reqs);
Linus Torvalds's avatar
Linus Torvalds committed
1127
1128
}

Gu Zheng's avatar
Gu Zheng committed
1129
/* aio_read_events_ring
1130
1131
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
Linus Torvalds's avatar
Linus Torvalds committed
1132
 */
1133
1134
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
1135
1136
{
	struct aio_ring *ring;
1137
	unsigned head, tail, pos;
1138
1139
1140
	long ret = 0;
	int copy_ret;

1141
1142
1143
1144
1145
1146
1147
	/*
	 * 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();
1148
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1149

1150
	/* Access to ->ring_pages here is protected by ctx->ring_lock. */
1151
	ring = kmap_atomic(ctx->ring_pages[0]);
1152
	head = ring->head;
1153
	tail = ring->tail;
1154
1155
	kunmap_atomic(ring);

1156
1157
1158
1159
1160
1161
	/*
	 * 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();