aio.c 43 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
81
82
83
84
struct ctx_rq_wait {
	struct completion comp;
	atomic_t count;
};

85
struct kioctx {
Kent Overstreet's avatar
Kent Overstreet committed
86
	struct percpu_ref	users;
Kent Overstreet's avatar
Kent Overstreet committed
87
	atomic_t		dead;
88

Kent Overstreet's avatar
Kent Overstreet committed
89
90
	struct percpu_ref	reqs;

91
92
	unsigned long		user_id;

Kent Overstreet's avatar
Kent Overstreet committed
93
94
95
96
97
98
99
	struct __percpu kioctx_cpu *cpu;

	/*
	 * For percpu reqs_available, number of slots we move to/from global
	 * counter at a time:
	 */
	unsigned		req_batch;
100
101
102
103
	/*
	 * This is what userspace passed to io_setup(), it's not used for
	 * anything but counting against the global max_reqs quota.
	 *
104
	 * The real limit is nr_events - 1, which will be larger (see
105
106
	 * aio_setup_ring())
	 */
107
108
	unsigned		max_reqs;

109
110
	/* Size of ringbuffer, in units of struct io_event */
	unsigned		nr_events;
111

112
113
114
115
116
117
	unsigned long		mmap_base;
	unsigned long		mmap_size;

	struct page		**ring_pages;
	long			nr_pages;

Kent Overstreet's avatar
Kent Overstreet committed
118
	struct work_struct	free_work;
119

120
121
122
	/*
	 * signals when all in-flight requests are done
	 */
123
	struct ctx_rq_wait	*rq_wait;
124

125
	struct {
126
127
128
129
130
		/*
		 * 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
131
132
		 *
		 * We batch accesses to it with a percpu version.
133
134
		 */
		atomic_t	reqs_available;
135
136
137
138
139
140
141
	} ____cacheline_aligned_in_smp;

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

142
143
	struct {
		struct mutex	ring_lock;
144
145
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;
146
147
148

	struct {
		unsigned	tail;
149
		unsigned	completed_events;
150
		spinlock_t	completion_lock;
151
	} ____cacheline_aligned_in_smp;
152
153

	struct page		*internal_pages[AIO_RING_PAGES];
154
	struct file		*aio_ring_file;
155
156

	unsigned		id;
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
186
187
188
189
190
/*
 * 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
191
/*------ sysctl variables----*/
192
193
194
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
195
196
/*----end sysctl variables---*/

197
198
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
199

200
201
202
203
204
205
206
207
208
209
210
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);
211
212
	if (IS_ERR(inode))
		return ERR_CAST(inode);
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241

	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,
	};
242
	return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC);
243
244
}

Linus Torvalds's avatar
Linus Torvalds committed
245
246
247
248
249
250
/* 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)
{
251
252
253
254
255
256
257
258
259
	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.");

260
	kiocb_cachep = KMEM_CACHE(aio_kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
261
	kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds's avatar
Linus Torvalds committed
262

Kent Overstreet's avatar
Kent Overstreet committed
263
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
264
265
266

	return 0;
}
267
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
268

269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
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
285
286
static void aio_free_ring(struct kioctx *ctx)
{
287
	int i;
Linus Torvalds's avatar
Linus Torvalds committed
288

289
290
291
292
293
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

294
	for (i = 0; i < ctx->nr_pages; i++) {
295
		struct page *page;
296
297
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
298
299
300
301
302
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
303
	}
Linus Torvalds's avatar
Linus Torvalds committed
304

305
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
306
		kfree(ctx->ring_pages);
307
308
		ctx->ring_pages = NULL;
	}
309
310
}

311
static int aio_ring_mremap(struct vm_area_struct *vma)
312
{
313
	struct file *file = vma->vm_file;
314
315
	struct mm_struct *mm = vma->vm_mm;
	struct kioctx_table *table;
Al Viro's avatar
Al Viro committed
316
	int i, res = -EINVAL;
317
318
319
320
321
322
323
324
325

	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) {
Al Viro's avatar
Al Viro committed
326
327
328
329
			if (!atomic_read(&ctx->dead)) {
				ctx->user_id = ctx->mmap_base = vma->vm_start;
				res = 0;
			}
330
331
332
333
334
335
			break;
		}
	}

	rcu_read_unlock();
	spin_unlock(&mm->ioctx_lock);
Al Viro's avatar
Al Viro committed
336
	return res;
337
338
}

339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
static const struct vm_operations_struct aio_ring_vm_ops = {
	.mremap		= aio_ring_mremap,
#if IS_ENABLED(CONFIG_MMU)
	.fault		= filemap_fault,
	.map_pages	= filemap_map_pages,
	.page_mkwrite	= filemap_page_mkwrite,
#endif
};

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

355
356
357
358
static const struct file_operations aio_ring_fops = {
	.mmap = aio_ring_mmap,
};

359
#if IS_ENABLED(CONFIG_MIGRATION)
360
361
362
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
363
	struct kioctx *ctx;
364
	unsigned long flags;
365
	pgoff_t idx;
366
367
	int rc;

368
369
	rc = 0;

370
	/* mapping->private_lock here protects against the kioctx teardown.  */
371
372
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
	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;
392
393
394
395
	} else
		rc = -EINVAL;

	if (rc != 0)
396
		goto out_unlock;
397

398
399
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
400
	get_page(new);
401

402
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
403
	if (rc != MIGRATEPAGE_SUCCESS) {
404
		put_page(new);
405
		goto out_unlock;
406
407
	}

408
409
410
	/* 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.
411
	 */
412
413
414
415
416
	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);
417

418
419
	/* The old page is no longer accessible. */
	put_page(old);
420

421
422
423
424
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
425
	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
426
}
427
#endif
Linus Torvalds's avatar
Linus Torvalds committed
428

429
static const struct address_space_operations aio_ctx_aops = {
430
	.set_page_dirty = __set_page_dirty_no_writeback,
431
#if IS_ENABLED(CONFIG_MIGRATION)
432
	.migratepage	= aio_migratepage,
433
#endif
434
435
};

Linus Torvalds's avatar
Linus Torvalds committed
436
437
438
439
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
440
	struct mm_struct *mm = current->mm;
441
	unsigned long size, unused;
Linus Torvalds's avatar
Linus Torvalds committed
442
	int nr_pages;
443
444
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
445
446
447
448
449
450
451

	/* 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;

452
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
453
454
455
	if (nr_pages < 0)
		return -EINVAL;

456
	file = aio_private_file(ctx, nr_pages);
457
458
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
459
		return -ENOMEM;
460
461
	}

462
463
464
465
466
467
468
469
470
471
472
473
474
475
	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;
		}
	}

476
477
478
479
480
481
482
483
484
485
	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);
486
487

		ctx->ring_pages[i] = page;
488
	}
489
	ctx->nr_pages = i;
Linus Torvalds's avatar
Linus Torvalds committed
490

491
492
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
493
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
494
495
	}

496
497
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
498

Zach Brown's avatar
Zach Brown committed
499
	down_write(&mm->mmap_sem);
500
501
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
502
503
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
504
505
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
506
		aio_free_ring(ctx);
507
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
508
509
	}

510
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
511

512
513
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
514

515
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
516
	ring->nr = nr_events;	/* user copy */
517
	ring->id = ~0U;
Linus Torvalds's avatar
Linus Torvalds committed
518
519
520
521
522
	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);
523
	kunmap_atomic(ring);
524
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
525
526
527
528
529
530
531
532

	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)

533
void kiocb_set_cancel_fn(struct kiocb *iocb, kiocb_cancel_fn *cancel)
534
{
535
	struct aio_kiocb *req = container_of(iocb, struct aio_kiocb, common);
536
537
538
539
540
541
542
543
544
545
546
547
548
549
	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);

550
static int kiocb_cancel(struct aio_kiocb *kiocb)
Kent Overstreet's avatar
Kent Overstreet committed
551
{
552
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
553

554
555
556
557
558
559
560
561
	/*
	 * 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
562
			return -EINVAL;
Kent Overstreet's avatar
Kent Overstreet committed
563

564
565
566
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
567

568
	return cancel(&kiocb->common);
Kent Overstreet's avatar
Kent Overstreet committed
569
570
}

Kent Overstreet's avatar
Kent Overstreet committed
571
static void free_ioctx(struct work_struct *work)
Kent Overstreet's avatar
Kent Overstreet committed
572
{
Kent Overstreet's avatar
Kent Overstreet committed
573
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreet's avatar
Kent Overstreet committed
574

Kent Overstreet's avatar
Kent Overstreet committed
575
	pr_debug("freeing %p\n", ctx);
Kent Overstreet's avatar
Kent Overstreet committed
576

Kent Overstreet's avatar
Kent Overstreet committed
577
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
578
	free_percpu(ctx->cpu);
579
580
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Kent Overstreet's avatar
Kent Overstreet committed
581
582
583
	kmem_cache_free(kioctx_cachep, ctx);
}

Kent Overstreet's avatar
Kent Overstreet committed
584
585
586
587
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

588
	/* At this point we know that there are no any in-flight requests */
589
590
	if (ctx->rq_wait && atomic_dec_and_test(&ctx->rq_wait->count))
		complete(&ctx->rq_wait->comp);
591

Kent Overstreet's avatar
Kent Overstreet committed
592
593
594
595
	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

Kent Overstreet's avatar
Kent Overstreet committed
596
597
598
599
600
/*
 * 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
601
static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet's avatar
Kent Overstreet committed
602
{
Kent Overstreet's avatar
Kent Overstreet committed
603
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
604
	struct aio_kiocb *req;
Kent Overstreet's avatar
Kent Overstreet committed
605
606
607
608
609

	spin_lock_irq(&ctx->ctx_lock);

	while (!list_empty(&ctx->active_reqs)) {
		req = list_first_entry(&ctx->active_reqs,
610
				       struct aio_kiocb, ki_list);
Kent Overstreet's avatar
Kent Overstreet committed
611
612

		list_del_init(&req->ki_list);
613
		kiocb_cancel(req);
Kent Overstreet's avatar
Kent Overstreet committed
614
615
616
617
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
618
619
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
Kent Overstreet's avatar
Kent Overstreet committed
620
621
}

622
623
624
625
626
627
628
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);
629
	table = rcu_dereference_raw(mm->ioctx_table);
630
631
632
633
634
635
636
637
638

	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);

639
640
641
642
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
					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);
660
		old = rcu_dereference_raw(mm->ioctx_table);
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676

		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
677
678
679
680
681
682
683
684
685
686
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
687
688
689
690
691
/* 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
692
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
693
	struct kioctx *ctx;
694
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
695

Kent Overstreet's avatar
Kent Overstreet committed
696
697
698
699
700
701
702
703
704
705
706
707
	/*
	 * 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
708
	/* Prevent overflows */
709
	if (nr_events > (0x10000000U / sizeof(struct io_event))) {
Linus Torvalds's avatar
Linus Torvalds committed
710
711
712
713
		pr_debug("ENOMEM: nr_events too high\n");
		return ERR_PTR(-EINVAL);
	}

714
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds's avatar
Linus Torvalds committed
715
716
		return ERR_PTR(-EAGAIN);

717
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
718
719
720
721
722
723
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
724
	spin_lock_init(&ctx->completion_lock);
725
	mutex_init(&ctx->ring_lock);
726
727
728
	/* 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
729
730
731
732
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

733
	if (percpu_ref_init(&ctx->users, free_ioctx_users, 0, GFP_KERNEL))
734
735
		goto err;

736
	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs, 0, GFP_KERNEL))
737
738
		goto err;

Kent Overstreet's avatar
Kent Overstreet committed
739
740
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Kent Overstreet's avatar
Kent Overstreet committed
741
		goto err;
Linus Torvalds's avatar
Linus Torvalds committed
742

743
744
	err = aio_setup_ring(ctx);
	if (err < 0)
Kent Overstreet's avatar
Kent Overstreet committed
745
		goto err;
Kent Overstreet's avatar
Kent Overstreet committed
746

747
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
748
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
749
750
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
751

Linus Torvalds's avatar
Linus Torvalds committed
752
	/* limit the number of system wide aios */
753
	spin_lock(&aio_nr_lock);
754
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
755
	    aio_nr + nr_events < aio_nr) {
756
		spin_unlock(&aio_nr_lock);
Kent Overstreet's avatar
Kent Overstreet committed
757
		err = -EAGAIN;
758
		goto err_ctx;
759
760
	}
	aio_nr += ctx->max_reqs;
761
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
762

763
764
	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
765

766
767
	err = ioctx_add_table(ctx, mm);
	if (err)
Kent Overstreet's avatar
Kent Overstreet committed
768
		goto err_cleanup;
769

770
771
772
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

Kent Overstreet's avatar
Kent Overstreet committed
773
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
774
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
775
776
	return ctx;

Kent Overstreet's avatar
Kent Overstreet committed
777
778
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
779
err_ctx:
780
781
782
	atomic_set(&ctx->dead, 1);
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
783
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
784
err:
785
	mutex_unlock(&ctx->ring_lock);
Kent Overstreet's avatar
Kent Overstreet committed
786
	free_percpu(ctx->cpu);
787
788
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
Linus Torvalds's avatar
Linus Torvalds committed
789
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
790
	pr_debug("error allocating ioctx %d\n", err);
791
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
792
793
}

Kent Overstreet's avatar
Kent Overstreet committed
794
795
796
797
798
/* 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.
 */
799
static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
800
		      struct ctx_rq_wait *wait)
Kent Overstreet's avatar
Kent Overstreet committed
801
{
802
	struct kioctx_table *table;
803

Al Viro's avatar
Al Viro committed
804
805
806
	spin_lock(&mm->ioctx_lock);
	if (atomic_xchg(&ctx->dead, 1)) {
		spin_unlock(&mm->ioctx_lock);
807
		return -EINVAL;
Al Viro's avatar
Al Viro committed
808
	}
809

810
	table = rcu_dereference_raw(mm->ioctx_table);
811
812
813
	WARN_ON(ctx != table->table[ctx->id]);
	table->table[ctx->id] = NULL;
	spin_unlock(&mm->ioctx_lock);
814

815
816
	/* percpu_ref_kill() will do the necessary call_rcu() */
	wake_up_all(&ctx->wait);
817

818
819
820
821
822
823
824
825
	/*
	 * 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);
826

827
828
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
829

830
	ctx->rq_wait = wait;
831
832
	percpu_ref_kill(&ctx->users);
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
833
834
}

Kent Overstreet's avatar
Kent Overstreet committed
835
836
837
838
839
840
841
/*
 * 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
842
 */
843
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
844
{
845
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
846
847
	struct ctx_rq_wait wait;
	int i, skipped;
848

849
850
	if (!table)
		return;
851

852
853
854
855
	atomic_set(&wait.count, table->nr);
	init_completion(&wait.comp);

	skipped = 0;
856
857
	for (i = 0; i < table->nr; ++i) {
		struct kioctx *ctx = table->table[i];
858

859
860
		if (!ctx) {
			skipped++;
861
			continue;
862
863
		}

864
		/*
865
866
867
868
869
		 * 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.
870
		 */
871
		ctx->mmap_size = 0;
872
873
		kill_ioctx(mm, ctx, &wait);
	}
Kent Overstreet's avatar
Kent Overstreet committed
874

875
	if (!atomic_sub_and_test(skipped, &wait.count)) {
876
		/* Wait until all IO for the context are done. */
877
		wait_for_completion(&wait.comp);
Linus Torvalds's avatar
Linus Torvalds committed
878
	}
879
880
881

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
Linus Torvalds's avatar
Linus Torvalds committed
882
883
}

Kent Overstreet's avatar
Kent Overstreet committed
884
885
886
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
887
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
888

889
	local_irq_save(flags);
890
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
891
	kcpu->reqs_available += nr;
892

Kent Overstreet's avatar
Kent Overstreet committed
893
894
895
896
897
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

898
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
899
900
901
902
903
904
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;
905
	unsigned long flags;
Kent Overstreet's avatar
Kent Overstreet committed
906

907
	local_irq_save(flags);
908
	kcpu = this_cpu_ptr(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
	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:
927
	local_irq_restore(flags);
Kent Overstreet's avatar
Kent Overstreet committed
928
929
930
	return ret;
}

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
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
/* 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
993
/* aio_get_req
Kent Overstreet's avatar
Kent Overstreet committed
994
995
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
Linus Torvalds's avatar
Linus Torvalds committed
996
 */
997
static inline struct aio_kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
998
{
999
	struct aio_kiocb *req;
Kent Overstreet's avatar
Kent Overstreet committed
1000

1001
1002
1003
1004
1005
	if (!get_reqs_available(ctx)) {
		user_refill_reqs_available(ctx);
		if (!get_reqs_available(ctx))
			return NULL;
	}
Kent Overstreet's avatar
Kent Overstreet committed
1006

1007
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
1008
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
1009
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
1010

Kent Overstreet's avatar
Kent Overstreet committed
1011
1012
	percpu_ref_get(&ctx->reqs);

Linus Torvalds's avatar
Linus Torvalds committed
1013
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
1014
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
1015
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
1016
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
1017
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1018
1019
}

1020
static void kiocb_free(struct aio_kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
1021
{
1022
1023
	if (req->common.ki_filp)
		fput(req->common.ki_filp);
1024
1025
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
1026
1027
1028
	kmem_cache_free(kiocb_cachep, req);
}

1029
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
1030
{
1031
	struct aio_ring __user *ring  = (void __user *)ctx_id;
1032
	struct mm_struct *mm = current->mm;
1033
	struct kioctx *ctx, *ret = NULL;
1034
1035
1036
1037
1038
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1039

1040
	rcu_read_lock();
1041
	table = rcu_dereference(mm->ioctx_table);
1042

1043
1044
	if (!table || id >= table->nr)
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
1045

1046
	ctx = table->table[id];
1047
	if (ctx && ctx->user_id == ctx_id) {
1048
1049
1050
1051
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
1052
	rcu_read_unlock();
1053
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
1054
1055
1056
1057
1058
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
1059
static void aio_complete(struct kiocb *kiocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
1060
{
1061
	struct aio_kiocb *iocb = container_of(kiocb, struct aio_kiocb, common);
Linus Torvalds's avatar
Linus Torvalds committed
1062
1063
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
1064
	struct io_event	*ev_page, *event;
1065
	unsigned tail, pos, head;
Linus Torvalds's avatar
Linus Torvalds committed
1066
1067
	unsigned long	flags;

1068
1069
1070
1071
1072
1073
	/*
	 * 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
1074
	 */
1075
	BUG_ON(is_sync_kiocb(kiocb));
Linus Torvalds's avatar
Linus Torvalds committed
1076

1077
1078
1079
1080
1081
1082
1083
	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);
	}
1084

1085
1086
	/*
	 * Add a completion event to the ring buffer. Must be done holding
1087
	 * ctx->completion_lock to prevent other code from messing with the tail
1088
1089
1090
1091
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

1092
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
1093
1094
	pos = tail + AIO_EVENTS_OFFSET;

1095
	if (++tail >= ctx->nr_events)
1096
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1097

1098
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1099
1100
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

1101
	event->obj = (u64)(unsigned long)iocb->ki_user_iocb;
Linus Torvalds's avatar
Linus Torvalds committed
1102
1103
1104
1105
	event->data = iocb->ki_user_data;
	event->res = res;
	event->res2 = res2;

Kent Overstreet's avatar
Kent Overstreet committed
1106
	kunmap_atomic(ev_page);
1107
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
1108
1109

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
1110
		 ctx, tail, iocb, iocb->ki_user_iocb, iocb->ki_user_data,
Kent Overstreet's avatar
Kent Overstreet committed
1111
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
1112
1113
1114
1115
1116
1117

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

1118
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
1119

1120
	ring = kmap_atomic(ctx->ring_pages[0]);
1121
	head = ring->head;
Kent Overstreet's avatar
Kent Overstreet committed
1122
	ring->tail = tail;
1123
	kunmap_atomic(ring);
1124
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
1125

1126
1127
1128
	ctx->completed_events++;
	if (ctx->completed_events > 1)
		refill_reqs_available(ctx, head, tail);
1129
1130
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
1131
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
1132
1133
1134
1135
1136
1137

	/*
	 * 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.
	 */
1138
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
1139
1140
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
1141
	/* everything turned out well, dispose of the aiocb. */
Kent Overstreet's avatar
Kent Overstreet committed
1142
	kiocb_free(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
1143