aio.c 37.8 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
55
56
57
58
59
60
61
62
63
64
65
66
67
68
#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 */
	unsigned	head;
	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

69
70
71
72
73
74
struct kioctx_table {
	struct rcu_head	rcu;
	unsigned	nr;
	struct kioctx	*table[];
};

Kent Overstreet's avatar
Kent Overstreet committed
75
76
77
78
struct kioctx_cpu {
	unsigned		reqs_available;
};

79
struct kioctx {
Kent Overstreet's avatar
Kent Overstreet committed
80
	struct percpu_ref	users;
Kent Overstreet's avatar
Kent Overstreet committed
81
	atomic_t		dead;
82

Kent Overstreet's avatar
Kent Overstreet committed
83
84
	struct percpu_ref	reqs;

85
86
	unsigned long		user_id;

Kent Overstreet's avatar
Kent Overstreet committed
87
88
89
90
91
92
93
	struct __percpu kioctx_cpu *cpu;

	/*
	 * For percpu reqs_available, number of slots we move to/from global
	 * counter at a time:
	 */
	unsigned		req_batch;
94
95
96
97
	/*
	 * This is what userspace passed to io_setup(), it's not used for
	 * anything but counting against the global max_reqs quota.
	 *
98
	 * The real limit is nr_events - 1, which will be larger (see
99
100
	 * aio_setup_ring())
	 */
101
102
	unsigned		max_reqs;

103
104
	/* Size of ringbuffer, in units of struct io_event */
	unsigned		nr_events;
105

106
107
108
109
110
111
	unsigned long		mmap_base;
	unsigned long		mmap_size;

	struct page		**ring_pages;
	long			nr_pages;

Kent Overstreet's avatar
Kent Overstreet committed
112
	struct work_struct	free_work;
113
114

	struct {
115
116
117
118
119
		/*
		 * 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
120
121
		 *
		 * We batch accesses to it with a percpu version.
122
123
		 */
		atomic_t	reqs_available;
124
125
126
127
128
129
130
	} ____cacheline_aligned_in_smp;

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

131
132
	struct {
		struct mutex	ring_lock;
133
134
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;
135
136
137
138

	struct {
		unsigned	tail;
		spinlock_t	completion_lock;
139
	} ____cacheline_aligned_in_smp;
140
141

	struct page		*internal_pages[AIO_RING_PAGES];
142
	struct file		*aio_ring_file;
143
144

	unsigned		id;
145
146
};

Linus Torvalds's avatar
Linus Torvalds committed
147
/*------ sysctl variables----*/
148
149
150
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
151
152
/*----end sysctl variables---*/

153
154
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
155

156
157
158
159
160
161
162
163
164
165
166
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);
167
168
	if (IS_ERR(inode))
		return ERR_CAST(inode);
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201

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

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

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

	file->f_flags = O_RDWR;
	file->private_data = ctx;
	return file;
}

static struct dentry *aio_mount(struct file_system_type *fs_type,
				int flags, const char *dev_name, void *data)
{
	static const struct dentry_operations ops = {
		.d_dname	= simple_dname,
	};
	return mount_pseudo(fs_type, "aio:", NULL, &ops, 0xa10a10a1);
}

Linus Torvalds's avatar
Linus Torvalds committed
202
203
204
205
206
207
/* 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)
{
208
209
210
211
212
213
214
215
216
	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.");

217
218
	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
219

Kent Overstreet's avatar
Kent Overstreet committed
220
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
221
222
223

	return 0;
}
224
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
225

226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
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
242
243
static void aio_free_ring(struct kioctx *ctx)
{
244
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
245

246
	for (i = 0; i < ctx->nr_pages; i++) {
247
		struct page *page;
248
249
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
250
251
252
253
254
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
255
	}
Linus Torvalds's avatar
Linus Torvalds committed
256

257
258
	put_aio_ring_file(ctx);

259
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
260
		kfree(ctx->ring_pages);
261
262
		ctx->ring_pages = NULL;
	}
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
}

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

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

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

280
#if IS_ENABLED(CONFIG_MIGRATION)
281
282
283
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
284
	struct kioctx *ctx;
285
286
287
	unsigned long flags;
	int rc;

288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
	rc = 0;

	/* Make sure the old page hasn't already been changed */
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
	if (ctx) {
		pgoff_t idx;
		spin_lock_irqsave(&ctx->completion_lock, flags);
		idx = old->index;
		if (idx < (pgoff_t)ctx->nr_pages) {
			if (ctx->ring_pages[idx] != old)
				rc = -EAGAIN;
		} else
			rc = -EINVAL;
		spin_unlock_irqrestore(&ctx->completion_lock, flags);
	} else
		rc = -EINVAL;
	spin_unlock(&mapping->private_lock);

	if (rc != 0)
		return rc;

310
311
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
312
	get_page(new);
313

314
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
315
	if (rc != MIGRATEPAGE_SUCCESS) {
316
		put_page(new);
317
318
319
		return rc;
	}

320
321
322
323
324
325
326
327
328
329
330
	/* We can potentially race against kioctx teardown here.  Use the
	 * address_space's private data lock to protect the mapping's
	 * private_data.
	 */
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
	if (ctx) {
		pgoff_t idx;
		spin_lock_irqsave(&ctx->completion_lock, flags);
		migrate_page_copy(new, old);
		idx = old->index;
331
332
333
334
335
336
337
338
		if (idx < (pgoff_t)ctx->nr_pages) {
			/* And only do the move if things haven't changed */
			if (ctx->ring_pages[idx] == old)
				ctx->ring_pages[idx] = new;
			else
				rc = -EAGAIN;
		} else
			rc = -EINVAL;
339
340
341
342
		spin_unlock_irqrestore(&ctx->completion_lock, flags);
	} else
		rc = -EBUSY;
	spin_unlock(&mapping->private_lock);
343

344
345
346
347
348
	if (rc == MIGRATEPAGE_SUCCESS)
		put_page(old);
	else
		put_page(new);

349
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
350
}
351
#endif
Linus Torvalds's avatar
Linus Torvalds committed
352

353
354
static const struct address_space_operations aio_ctx_aops = {
	.set_page_dirty = aio_set_page_dirty,
355
#if IS_ENABLED(CONFIG_MIGRATION)
356
	.migratepage	= aio_migratepage,
357
#endif
358
359
};

Linus Torvalds's avatar
Linus Torvalds committed
360
361
362
363
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
364
	struct mm_struct *mm = current->mm;
365
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
366
	int nr_pages;
367
368
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
369
370
371
372
373
374
375

	/* 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;

376
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
377
378
379
	if (nr_pages < 0)
		return -EINVAL;

380
	file = aio_private_file(ctx, nr_pages);
381
382
383
384
385
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
		return -EAGAIN;
	}

386
387
388
389
390
391
392
393
394
395
396
397
398
399
	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;
		}
	}

400
401
402
403
404
405
406
407
408
409
410
	for (i = 0; i < nr_pages; i++) {
		struct page *page;
		page = find_or_create_page(file->f_inode->i_mapping,
					   i, GFP_HIGHUSER | __GFP_ZERO);
		if (!page)
			break;
		pr_debug("pid(%d) page[%d]->count=%d\n",
			 current->pid, i, page_count(page));
		SetPageUptodate(page);
		SetPageDirty(page);
		unlock_page(page);
411
412

		ctx->ring_pages[i] = page;
413
	}
414
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
415

416
417
418
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
		return -EAGAIN;
Linus Torvalds's avatar
Linus Torvalds committed
419
420
	}

421
422
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
423

Zach Brown's avatar
Zach Brown committed
424
	down_write(&mm->mmap_sem);
425
426
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
427
428
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
429
430
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
431
432
433
434
		aio_free_ring(ctx);
		return -EAGAIN;
	}

435
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
436

437
438
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
439

440
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
441
	ring->nr = nr_events;	/* user copy */
442
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
443
444
445
446
447
	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);
448
	kunmap_atomic(ring);
449
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
450
451
452
453
454
455
456
457

	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)

458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
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);

474
static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
475
{
476
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
477

478
479
480
481
482
483
484
485
	/*
	 * 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
486
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
487

488
489
490
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
491

Kent Overstreet's avatar
Kent Overstreet committed
492
	return cancel(kiocb);
Kent Overstreet's avatar
Kent Overstreet committed
493
494
}

Kent Overstreet's avatar
Kent Overstreet committed
495
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
496
{
Kent Overstreet's avatar
Kent Overstreet committed
497
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
498

Kent Overstreet's avatar
Kent Overstreet committed
499
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
500

Kent Overstreet's avatar
Kent Overstreet committed
501
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
502
	free_percpu(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
503
504
505
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
506
507
508
509
510
511
512
513
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

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

Kent Overstreet's avatar
Kent Overstreet committed
514
515
516
517
518
/*
 * 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
519
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
520
{
Kent Overstreet's avatar
Kent Overstreet committed
521
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
Kent Overstreet's avatar
Kent Overstreet committed
522
523
524
525
526
527
528
529
530
	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);
531
		kiocb_cancel(ctx, req);
Kent Overstreet's avatar
Kent Overstreet committed
532
533
534
535
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
536
537
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
538
539
}

540
541
542
543
544
545
546
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);
547
	rcu_read_lock();
548
	table = rcu_dereference(mm->ioctx_table);
549
550
551
552
553
554
555

	while (1) {
		if (table)
			for (i = 0; i < table->nr; i++)
				if (!table->table[i]) {
					ctx->id = i;
					table->table[i] = ctx;
556
					rcu_read_unlock();
557
558
559
560
561
562
563
564
565
566
					spin_unlock(&mm->ioctx_lock);

					ring = kmap_atomic(ctx->ring_pages[0]);
					ring->id = ctx->id;
					kunmap_atomic(ring);
					return 0;
				}

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

567
		rcu_read_unlock();
568
569
570
571
572
573
574
575
576
577
		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);
578
		rcu_read_lock();
579
		old = rcu_dereference(mm->ioctx_table);
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595

		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
596
597
598
599
600
601
602
603
604
605
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
606
607
608
609
610
/* 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
611
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
612
	struct kioctx *ctx;
613
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
614

Kent Overstreet's avatar
Kent Overstreet committed
615
616
617
618
619
620
621
622
623
624
625
626
	/*
	 * 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
627
628
629
630
631
632
633
	/* 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);
	}

634
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
635
636
		return ERR_PTR(-EAGAIN);

637
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
638
639
640
641
642
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

Kent Overstreet's avatar
Kent Overstreet committed
643
644
645
646
647
	if (percpu_ref_init(&ctx->users, free_ioctx_users))
		goto err;

	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs))
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
648

Linus Torvalds's avatar
Linus Torvalds committed
649
	spin_lock_init(&ctx->ctx_lock);
650
	spin_lock_init(&ctx->completion_lock);
651
	mutex_init(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
652
653
654
655
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

Kent Overstreet's avatar
Kent Overstreet committed
656
657
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
658
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
659

Kent Overstreet's avatar
Kent Overstreet committed
660
	if (aio_setup_ring(ctx) < 0)
Kent Overstreet's avatar
Kent Overstreet committed
661
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
662

663
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
664
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
665
666
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
667

Linus Torvalds's avatar
Linus Torvalds committed
668
	/* limit the number of system wide aios */
669
	spin_lock(&aio_nr_lock);
670
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
671
	    aio_nr + nr_events < aio_nr) {
672
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
673
		err = -EAGAIN;
674
		goto err_ctx;
675
676
	}
	aio_nr += ctx->max_reqs;
677
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
678

679
680
	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
681

682
683
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
684
		goto err_cleanup;
685

Kent Overstreet's avatar
Kent Overstreet committed
686
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
687
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
688
689
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
690
691
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
692
693
err_ctx:
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
694
err:
Kent Overstreet's avatar
Kent Overstreet committed
695
	free_percpu(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
696
	free_percpu(ctx->reqs.pcpu_count);
Kent Overstreet's avatar
Kent Overstreet committed
697
	free_percpu(ctx->users.pcpu_count);
Linus Torvalds's avatar
Linus Torvalds committed
698
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
699
	pr_debug("error allocating ioctx %d\n", err);
700
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
701
702
}

Kent Overstreet's avatar
Kent Overstreet committed
703
704
705
706
707
/* 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.
 */
708
static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx)
Kent Overstreet's avatar
Kent Overstreet committed
709
710
{
	if (!atomic_xchg(&ctx->dead, 1)) {
711
712
713
		struct kioctx_table *table;

		spin_lock(&mm->ioctx_lock);
714
		rcu_read_lock();
715
		table = rcu_dereference(mm->ioctx_table);
716
717
718

		WARN_ON(ctx != table->table[ctx->id]);
		table->table[ctx->id] = NULL;
719
		rcu_read_unlock();
720
721
		spin_unlock(&mm->ioctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
722
723
		/* percpu_ref_kill() will do the necessary call_rcu() */
		wake_up_all(&ctx->wait);
724

Kent Overstreet's avatar
Kent Overstreet committed
725
		/*
726
727
728
729
730
		 * It'd be more correct to do this in free_ioctx(), after all
		 * the outstanding kiocbs have finished - but by then io_destroy
		 * has already returned, so io_setup() could potentially return
		 * -EAGAIN with no ioctxs actually in use (as far as userspace
		 *  could tell).
Kent Overstreet's avatar
Kent Overstreet committed
731
		 */
Kent Overstreet's avatar
Kent Overstreet committed
732
		aio_nr_sub(ctx->max_reqs);
733
734
735
736

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

Kent Overstreet's avatar
Kent Overstreet committed
737
		percpu_ref_kill(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
738
	}
Linus Torvalds's avatar
Linus Torvalds committed
739
740
741
742
743
}

/* wait_on_sync_kiocb:
 *	Waits on the given sync kiocb to complete.
 */
Kent Overstreet's avatar
Kent Overstreet committed
744
ssize_t wait_on_sync_kiocb(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
745
{
Kent Overstreet's avatar
Kent Overstreet committed
746
	while (!req->ki_ctx) {
Linus Torvalds's avatar
Linus Torvalds committed
747
		set_current_state(TASK_UNINTERRUPTIBLE);
Kent Overstreet's avatar
Kent Overstreet committed
748
		if (req->ki_ctx)
Linus Torvalds's avatar
Linus Torvalds committed
749
			break;
750
		io_schedule();
Linus Torvalds's avatar
Linus Torvalds committed
751
752
	}
	__set_current_state(TASK_RUNNING);
Kent Overstreet's avatar
Kent Overstreet committed
753
	return req->ki_user_data;
Linus Torvalds's avatar
Linus Torvalds committed
754
}
755
EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds's avatar
Linus Torvalds committed
756

Kent Overstreet's avatar
Kent Overstreet committed
757
758
759
760
761
762
763
/*
 * 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
764
 */
765
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
766
{
767
	struct kioctx_table *table;
768
	struct kioctx *ctx;
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
	unsigned i = 0;

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

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

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

		rcu_read_unlock();
788

789
790
791
792
793
794
795
796
		/*
		 * We don't need to bother with munmap() here -
		 * exit_mmap(mm) is coming and it'll unmap everything.
		 * Since aio_free_ring() uses non-zero ->mmap_size
		 * as indicator that it needs to unmap the area,
		 * just set it to 0; aio_free_ring() is the only
		 * place that uses ->mmap_size, so it's safe.
		 */
797
		ctx->mmap_size = 0;
Kent Overstreet's avatar
Kent Overstreet committed
798

799
		kill_ioctx(mm, ctx);
Linus Torvalds's avatar
Linus Torvalds committed
800
801
802
	}
}

Kent Overstreet's avatar
Kent Overstreet committed
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;

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

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

	preempt_enable();
}

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

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

	if (!kcpu->reqs_available) {
		int old, avail = atomic_read(&ctx->reqs_available);

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

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

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

	ret = true;
	kcpu->reqs_available--;
out:
	preempt_enable();
	return ret;
}

Linus Torvalds's avatar
Linus Torvalds committed
849
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
850
851
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
852
 */
Kent Overstreet's avatar
Kent Overstreet committed
853
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
854
{
Kent Overstreet's avatar
Kent Overstreet committed
855
856
	struct kiocb *req;

Kent Overstreet's avatar
Kent Overstreet committed
857
	if (!get_reqs_available(ctx))
Kent Overstreet's avatar
Kent Overstreet committed
858
859
		return NULL;

860
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
861
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
862
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
863

Kent Overstreet's avatar
Kent Overstreet committed
864
865
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
866
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
867
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
868
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
869
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
870
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
871
872
}

873
static void kiocb_free(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
874
{
875
876
	if (req->ki_filp)
		fput(req->ki_filp);
877
878
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
879
880
881
	kmem_cache_free(kiocb_cachep, req);
}

882
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
883
{
884
	struct aio_ring __user *ring  = (void __user *)ctx_id;
885
	struct mm_struct *mm = current->mm;
886
	struct kioctx *ctx, *ret = NULL;
887
888
889
890
891
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
892

893
	rcu_read_lock();
894
	table = rcu_dereference(mm->ioctx_table);
895

896
897
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
898

899
	ctx = table->table[id];
900
	if (ctx && ctx->user_id == ctx_id) {
901
902
903
904
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
905
	rcu_read_unlock();
906
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
907
908
909
910
911
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
912
void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
913
914
915
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
916
	struct io_event	*ev_page, *event;
Linus Torvalds's avatar
Linus Torvalds committed
917
	unsigned long	flags;
Kent Overstreet's avatar
Kent Overstreet committed
918
	unsigned tail, pos;
Linus Torvalds's avatar
Linus Torvalds committed
919

920
921
922
923
924
925
	/*
	 * 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
926
927
928
	 */
	if (is_sync_kiocb(iocb)) {
		iocb->ki_user_data = res;
Kent Overstreet's avatar
Kent Overstreet committed
929
930
		smp_wmb();
		iocb->ki_ctx = ERR_PTR(-EXDEV);
Linus Torvalds's avatar
Linus Torvalds committed
931
		wake_up_process(iocb->ki_obj.tsk);
932
		return;
Linus Torvalds's avatar
Linus Torvalds committed
933
934
	}

935
936
937
938
939
940
941
	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);
	}
942

943
944
	/*
	 * Add a completion event to the ring buffer. Must be done holding
945
	 * ctx->completion_lock to prevent other code from messing with the tail
946
947
948
949
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

950
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
951
952
	pos = tail + AIO_EVENTS_OFFSET;

953
	if (++tail >= ctx->nr_events)
954
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
955

956
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
957
958
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

Linus Torvalds's avatar
Linus Torvalds committed
959
960
961
962
963
	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
964
	kunmap_atomic(ev_page);
965
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
966
967

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreet's avatar
Kent Overstreet committed
968
969
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
970
971
972
973
974
975

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

976
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
977

978
	ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreet's avatar
Kent Overstreet committed
979
	ring->tail = tail;
980
	kunmap_atomic(ring);
981
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
982

983
984
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
985
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
986
987
988
989
990
991

	/*
	 * 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.
	 */
992
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
993
994
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
995
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
996
	kiocb_free(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
997

998
999
1000
1001
1002
1003
1004
1005
	/*
	 * 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
1006
1007
1008
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
1009
	percpu_ref_put(&ctx->reqs);
Linus Torvalds's avatar
Linus Torvalds committed
1010
}
1011
EXPORT_SYMBOL(aio_complete);
Linus Torvalds's avatar
Linus Torvalds committed
1012

1013
1014
1015
/* aio_read_events
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
Linus Torvalds's avatar
Linus Torvalds committed
1016
 */
1017
1018
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
1019
1020
{
	struct aio_ring *ring;
1021
	unsigned head, tail, pos;
1022
1023
1024
	long ret = 0;
	int copy_ret;

1025
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1026

1027
	ring = kmap_atomic(ctx->ring_pages[0]);
1028
	head = ring->head;
1029
	tail = ring->tail;
1030
1031
	kunmap_atomic(ring);

1032
	pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
1033

1034
	if (head == tail)
Linus Torvalds's avatar
Linus Torvalds committed
1035
1036
		goto out;

1037
1038
1039
1040
1041
	while (ret < nr) {
		long avail;
		struct io_event *ev;
		struct page *page;

1042
1043
		avail = (head <= tail ?  tail : ctx->nr_events) - head;
		if (head == tail)
1044
1045
1046
1047
1048
1049
1050
			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;
1051
		page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
		pos %= AIO_EVENTS_PER_PAGE;

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

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

		ret += avail;
		head += avail;
1066
		head %= ctx->nr_events;
Linus Torvalds's avatar
Linus Torvalds committed
1067
1068
	}

1069
	ring = kmap_atomic(ctx->ring_pages[0]);
1070
	ring->head = head;
1071
	kunmap_atomic(ring);
1072
	flush_dcache_page(ctx->ring_pages[0]);
1073

1074
	pr_debug("%li  h%u t%u\n", ret, head, tail);
1075

Kent Overstreet's avatar
Kent Overstreet committed
1076
	put_reqs_available(ctx, ret);
1077
out:
1078
	mutex_unlock(&ctx->ring_lock);
1079

Linus Torvalds's avatar
Linus Torvalds committed
1080
1081
1082
	return ret;
}

1083
1084
static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr,
			    struct io_event __user *event, long *i)
Linus Torvalds's avatar
Linus Torvalds committed
1085
{
1086
	long ret = aio_read_events_ring(ctx, event + *i, nr - *i);
Linus Torvalds's avatar
Linus Torvalds committed
1087

1088
1089
	if (ret > 0)
		*i += ret;
Linus Torvalds's avatar
Linus Torvalds committed
1090

1091
1092
	if (unlikely(atomic_read(&ctx->dead)))
		ret = -EINVAL;
Linus Torvalds's avatar
Linus Torvalds committed
1093

1094
1095
	if (!*i)
		*i = ret;
Linus Torvalds's avatar
Linus Torvalds committed
1096

1097
	return ret < 0 || *i >= min_nr;
Linus Torvalds's avatar
Linus Torvalds committed
1098
1099
}

1100
static long read_events(struct kioctx *ctx, long min_nr, long nr,
Linus Torvalds's avatar
Linus Torvalds committed
1101
1102
1103
			struct io_event __user *event,
			struct timespec __user *timeout)
{
1104
1105
	ktime_t until = { .tv64 = KTIME_MAX };
	long ret = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1106
1107
1108

	if (timeout) {
		struct timespec	ts;
1109

Linus Torvalds's avatar
Linus Torvalds committed
1110
		if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1111
			return -EFAULT;
Linus Torvalds's avatar
Linus Torvalds committed
1112

1113
		until = timespec_to_ktime(ts);
Linus Torvalds's avatar
Linus Torvalds committed
1114
1115
	}

1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
	/*
	 * Note that aio_read_events() is being called as the conditional - i.e.
	 * we're calling it after prepare_to_wait() has set task state to
	 * TASK_INTERRUPTIBLE.
	 *
	 * But aio_read_events() can block, and if it blocks it's going to flip
	 * the task state back to TASK_RUNNING.
	 *
	 * This should be ok, provided it doesn't flip the state back to
	 * TASK_RUNNING and return 0 too much - that causes us to spin. That
	 * will only happen if the mutex_lock() call blocks, and we then find
	 * the ringbuffer empty. So in practice we should be ok, but it's
	 * something to be aware of when touching this code.
	 */
	wait_event_interruptible_hrtimeout(ctx->wait,
			aio_read_events(ctx, min_nr, nr, event, &ret), until);
Linus Torvalds's avatar
Linus Torvalds committed
1132

1133
1134
	if (!ret && signal_pending(current))
		ret = -EINTR;
Linus Torvalds's avatar
Linus Torvalds committed
1135

1136
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
}

/* sys_io_setup:
 *	Create an aio_context capable of receiving at least nr_events.
 *	ctxp must not point to an aio_context that already exists, and
 *	must be initialized to 0 prior to the call.  On successful
 *	creation of the aio_context, *ctxp is filled in with the resulting 
 *	handle.  May fail with -EINVAL if *ctxp is not initialized,
 *	if the specified nr_events exceeds internal limits.  May fail 
 *	with -EAGAIN if the specified nr_events exceeds the user's limit 
 *	of available events.  May fail with -ENOMEM if insufficient kernel
 *	resources are available.  May fail with -EFAULT if an invalid
 *	pointer is passed for ctxp.  Will fail with -ENOSYS if not
 *	implemented.
 */
1152
SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
Linus Torvalds's avatar
Linus Torvalds committed
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
{
	struct kioctx *ioctx = NULL;
	unsigned long ctx;
	long ret;

	ret = get_user(ctx, ctxp);
	if (unlikely(ret))
		goto out;

	ret = -EINVAL;
1163
1164
1165
	if (unlikely(ctx || nr_events == 0)) {
		pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
		         ctx, nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
1166
1167
1168
1169
1170
1171
1172
		goto out;
	}

	ioctx = ioctx_alloc(nr_events);
	ret = PTR_ERR(ioctx);
	if (!IS_ERR(ioctx)) {
		ret = put_user(ioctx->user_id, ctxp);
1173
		if (ret)
1174
			kill_ioctx(current->mm, ioctx);
Kent Overstreet's avatar
Kent Overstreet committed
1175
		percpu_ref_put(&ioctx->users);
Linus Torvalds's avatar
Linus Torvalds committed
1176
1177
1178
1179
1180
1181
1182
1183
1184
	}

out:
	return ret;
}

/* sys_io_destroy:
 *	Destroy the aio_context specified.  May cancel any outstanding 
 *	AIOs and block on completion.  Will fail with -ENOSYS if not
1185
 *	implemented.  May fail with -EINVAL if the context pointed to
Linus Torvalds's avatar
Linus Torvalds committed
1186
1187
 *	is invalid.
 */
1188
SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)