aio.c 37.6 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
41
#include <linux/anon_inodes.h>
#include <linux/migrate.h>
#include <linux/ramfs.h>
Linus Torvalds's avatar
Linus Torvalds committed
42
43
44
45

#include <asm/kmap_types.h>
#include <asm/uaccess.h>

Al Viro's avatar
Al Viro committed
46
47
#include "internal.h"

48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
#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

Kent Overstreet's avatar
Kent Overstreet committed
68
69
70
71
struct kioctx_cpu {
	unsigned		reqs_available;
};

72
73
struct kioctx {
	atomic_t		users;
Kent Overstreet's avatar
Kent Overstreet committed
74
	atomic_t		dead;
75
76
77
78
79

	/* This needs improving */
	unsigned long		user_id;
	struct hlist_node	list;

Kent Overstreet's avatar
Kent Overstreet committed
80
81
82
83
84
85
86
	struct __percpu kioctx_cpu *cpu;

	/*
	 * For percpu reqs_available, number of slots we move to/from global
	 * counter at a time:
	 */
	unsigned		req_batch;
87
88
89
90
	/*
	 * This is what userspace passed to io_setup(), it's not used for
	 * anything but counting against the global max_reqs quota.
	 *
91
	 * The real limit is nr_events - 1, which will be larger (see
92
93
	 * aio_setup_ring())
	 */
94
95
	unsigned		max_reqs;

96
97
	/* Size of ringbuffer, in units of struct io_event */
	unsigned		nr_events;
98

99
100
101
102
103
104
	unsigned long		mmap_base;
	unsigned long		mmap_size;

	struct page		**ring_pages;
	long			nr_pages;

105
106
107
108
	struct rcu_head		rcu_head;
	struct work_struct	rcu_work;

	struct {
109
110
111
112
113
		/*
		 * 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
114
115
		 *
		 * We batch accesses to it with a percpu version.
116
117
		 */
		atomic_t	reqs_available;
118
119
120
121
122
123
124
	} ____cacheline_aligned_in_smp;

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

125
126
	struct {
		struct mutex	ring_lock;
127
128
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;
129
130
131
132

	struct {
		unsigned	tail;
		spinlock_t	completion_lock;
133
	} ____cacheline_aligned_in_smp;
134
135

	struct page		*internal_pages[AIO_RING_PAGES];
136
	struct file		*aio_ring_file;
137
138
};

Linus Torvalds's avatar
Linus Torvalds committed
139
/*------ sysctl variables----*/
140
141
142
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
143
144
/*----end sysctl variables---*/

145
146
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
Linus Torvalds's avatar
Linus Torvalds committed
147
148
149
150
151
152
153

/* 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)
{
154
155
	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
156

Kent Overstreet's avatar
Kent Overstreet committed
157
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds's avatar
Linus Torvalds committed
158
159
160

	return 0;
}
161
__initcall(aio_setup);
Linus Torvalds's avatar
Linus Torvalds committed
162
163
164

static void aio_free_ring(struct kioctx *ctx)
{
165
166
	int i;
	struct file *aio_ring_file = ctx->aio_ring_file;
Linus Torvalds's avatar
Linus Torvalds committed
167

168
169
170
	for (i = 0; i < ctx->nr_pages; i++) {
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
171
		put_page(ctx->ring_pages[i]);
172
	}
Linus Torvalds's avatar
Linus Torvalds committed
173

174
175
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages)
		kfree(ctx->ring_pages);
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203

	if (aio_ring_file) {
		truncate_setsize(aio_ring_file->f_inode, 0);
		pr_debug("pid(%d) i_nlink=%u d_count=%d d_unhashed=%d i_count=%d\n",
			current->pid, aio_ring_file->f_inode->i_nlink,
			aio_ring_file->f_path.dentry->d_count,
			d_unhashed(aio_ring_file->f_path.dentry),
			atomic_read(&aio_ring_file->f_inode->i_count));
		fput(aio_ring_file);
		ctx->aio_ring_file = NULL;
	}
}

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;
}

204
#if IS_ENABLED(CONFIG_MIGRATION)
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
	struct kioctx *ctx = mapping->private_data;
	unsigned long flags;
	unsigned idx = old->index;
	int rc;

	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
	put_page(old);

	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode);
	if (rc != MIGRATEPAGE_SUCCESS) {
		get_page(old);
		return rc;
	}

	get_page(new);

	spin_lock_irqsave(&ctx->completion_lock, flags);
	migrate_page_copy(new, old);
	ctx->ring_pages[idx] = new;
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

	return rc;
Linus Torvalds's avatar
Linus Torvalds committed
231
}
232
#endif
Linus Torvalds's avatar
Linus Torvalds committed
233

234
235
static const struct address_space_operations aio_ctx_aops = {
	.set_page_dirty = aio_set_page_dirty,
236
#if IS_ENABLED(CONFIG_MIGRATION)
237
	.migratepage	= aio_migratepage,
238
#endif
239
240
};

Linus Torvalds's avatar
Linus Torvalds committed
241
242
243
244
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
245
	struct mm_struct *mm = current->mm;
246
	unsigned long size, populate;
Linus Torvalds's avatar
Linus Torvalds committed
247
	int nr_pages;
248
249
	int i;
	struct file *file;
Linus Torvalds's avatar
Linus Torvalds committed
250
251
252
253
254
255
256

	/* 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;

257
	nr_pages = PFN_UP(size);
Linus Torvalds's avatar
Linus Torvalds committed
258
259
260
	if (nr_pages < 0)
		return -EINVAL;

261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
	file = anon_inode_getfile_private("[aio]", &aio_ring_fops, ctx, O_RDWR);
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
		return -EAGAIN;
	}

	file->f_inode->i_mapping->a_ops = &aio_ctx_aops;
	file->f_inode->i_mapping->private_data = ctx;
	file->f_inode->i_size = PAGE_SIZE * (loff_t)nr_pages;

	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);
	}
	ctx->aio_ring_file = file;
	nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring))
			/ sizeof(struct io_event);
Linus Torvalds's avatar
Linus Torvalds committed
286

287
	ctx->ring_pages = ctx->internal_pages;
Linus Torvalds's avatar
Linus Torvalds committed
288
	if (nr_pages > AIO_RING_PAGES) {
289
290
291
		ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *),
					  GFP_KERNEL);
		if (!ctx->ring_pages)
Linus Torvalds's avatar
Linus Torvalds committed
292
293
294
			return -ENOMEM;
	}

295
296
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
297

Zach Brown's avatar
Zach Brown committed
298
	down_write(&mm->mmap_sem);
299
300
301
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
				       MAP_SHARED | MAP_POPULATE, 0, &populate);
302
	if (IS_ERR((void *)ctx->mmap_base)) {
Zach Brown's avatar
Zach Brown committed
303
		up_write(&mm->mmap_sem);
304
		ctx->mmap_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
305
306
307
		aio_free_ring(ctx);
		return -EAGAIN;
	}
308
309
310
	up_write(&mm->mmap_sem);

	mm_populate(ctx->mmap_base, populate);
Linus Torvalds's avatar
Linus Torvalds committed
311

312
313
314
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
	ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages,
				       1, 0, ctx->ring_pages, NULL);
315
316
	for (i = 0; i < ctx->nr_pages; i++)
		put_page(ctx->ring_pages[i]);
Linus Torvalds's avatar
Linus Torvalds committed
317

318
	if (unlikely(ctx->nr_pages != nr_pages)) {
Linus Torvalds's avatar
Linus Torvalds committed
319
320
321
322
		aio_free_ring(ctx);
		return -EAGAIN;
	}

323
324
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds's avatar
Linus Torvalds committed
325

326
	ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
327
328
329
330
331
332
333
	ring->nr = nr_events;	/* user copy */
	ring->id = ctx->user_id;
	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);
334
	kunmap_atomic(ring);
335
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
336
337
338
339
340
341
342
343

	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)

344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
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);

Kent Overstreet's avatar
Kent Overstreet committed
360
361
362
static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb,
			struct io_event *res)
{
363
	kiocb_cancel_fn *old, *cancel;
Kent Overstreet's avatar
Kent Overstreet committed
364
365
	int ret = -EINVAL;

366
367
368
369
370
371
372
373
374
	/*
	 * 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)
			return ret;
Kent Overstreet's avatar
Kent Overstreet committed
375

376
377
378
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
Kent Overstreet's avatar
Kent Overstreet committed
379

380
381
382
383
384
385
386
387
388
	atomic_inc(&kiocb->ki_users);
	spin_unlock_irq(&ctx->ctx_lock);

	memset(res, 0, sizeof(*res));
	res->obj = (u64)(unsigned long)kiocb->ki_obj.user;
	res->data = kiocb->ki_user_data;
	ret = cancel(kiocb, res);

	spin_lock_irq(&ctx->ctx_lock);
Kent Overstreet's avatar
Kent Overstreet committed
389
390
391
392

	return ret;
}

Kent Overstreet's avatar
Kent Overstreet committed
393
394
395
static void free_ioctx_rcu(struct rcu_head *head)
{
	struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
Kent Overstreet's avatar
Kent Overstreet committed
396
397

	free_percpu(ctx->cpu);
Kent Overstreet's avatar
Kent Overstreet committed
398
399
400
401
402
403
404
405
406
407
	kmem_cache_free(kioctx_cachep, ctx);
}

/*
 * 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.
 */
static void free_ioctx(struct kioctx *ctx)
{
408
	struct aio_ring *ring;
Kent Overstreet's avatar
Kent Overstreet committed
409
410
	struct io_event res;
	struct kiocb *req;
Kent Overstreet's avatar
Kent Overstreet committed
411
	unsigned cpu, head, avail;
Kent Overstreet's avatar
Kent Overstreet committed
412
413
414
415
416
417
418
419
420
421
422
423
424

	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);
		kiocb_cancel(ctx, req, &res);
	}

	spin_unlock_irq(&ctx->ctx_lock);

Kent Overstreet's avatar
Kent Overstreet committed
425
426
427
428
429
430
431
	for_each_possible_cpu(cpu) {
		struct kioctx_cpu *kcpu = per_cpu_ptr(ctx->cpu, cpu);

		atomic_add(kcpu->reqs_available, &ctx->reqs_available);
		kcpu->reqs_available = 0;
	}

432
	ring = kmap_atomic(ctx->ring_pages[0]);
433
434
435
	head = ring->head;
	kunmap_atomic(ring);

436
	while (atomic_read(&ctx->reqs_available) < ctx->nr_events - 1) {
437
		wait_event(ctx->wait,
438
439
440
			   (head != ctx->tail) ||
			   (atomic_read(&ctx->reqs_available) >=
			    ctx->nr_events - 1));
441

442
		avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head;
443

444
		atomic_add(avail, &ctx->reqs_available);
445
		head += avail;
446
		head %= ctx->nr_events;
447
448
	}

449
	WARN_ON(atomic_read(&ctx->reqs_available) > ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470

	aio_free_ring(ctx);

	pr_debug("freeing %p\n", ctx);

	/*
	 * Here the call_rcu() is between the wait_event() for reqs_active to
	 * hit 0, and freeing the ioctx.
	 *
	 * aio_complete() decrements reqs_active, but it has to touch the ioctx
	 * after to issue a wakeup so we use rcu.
	 */
	call_rcu(&ctx->rcu_head, free_ioctx_rcu);
}

static void put_ioctx(struct kioctx *ctx)
{
	if (unlikely(atomic_dec_and_test(&ctx->users)))
		free_ioctx(ctx);
}

Linus Torvalds's avatar
Linus Torvalds committed
471
472
473
474
475
/* 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
476
	struct mm_struct *mm = current->mm;
Linus Torvalds's avatar
Linus Torvalds committed
477
	struct kioctx *ctx;
478
	int err = -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
479

Kent Overstreet's avatar
Kent Overstreet committed
480
481
482
483
484
485
486
487
488
489
490
491
	/*
	 * 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
492
493
494
495
496
497
498
	/* 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);
	}

499
	if (!nr_events || (unsigned long)nr_events > aio_max_nr)
Linus Torvalds's avatar
Linus Torvalds committed
500
501
		return ERR_PTR(-EAGAIN);

502
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
503
504
505
506
507
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

Al Viro's avatar
Al Viro committed
508
	atomic_set(&ctx->users, 2);
Kent Overstreet's avatar
Kent Overstreet committed
509
	atomic_set(&ctx->dead, 0);
Linus Torvalds's avatar
Linus Torvalds committed
510
	spin_lock_init(&ctx->ctx_lock);
511
	spin_lock_init(&ctx->completion_lock);
512
	mutex_init(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
513
514
515
516
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

Kent Overstreet's avatar
Kent Overstreet committed
517
518
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
Linus Torvalds's avatar
Linus Torvalds committed
519
520
		goto out_freectx;

Kent Overstreet's avatar
Kent Overstreet committed
521
522
523
	if (aio_setup_ring(ctx) < 0)
		goto out_freepcpu;

524
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreet's avatar
Kent Overstreet committed
525
526
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
	BUG_ON(!ctx->req_batch);
527

Linus Torvalds's avatar
Linus Torvalds committed
528
	/* limit the number of system wide aios */
529
	spin_lock(&aio_nr_lock);
530
531
	if (aio_nr + nr_events > aio_max_nr ||
	    aio_nr + nr_events < aio_nr) {
532
		spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
533
		goto out_cleanup;
534
535
	}
	aio_nr += ctx->max_reqs;
536
	spin_unlock(&aio_nr_lock);
Linus Torvalds's avatar
Linus Torvalds committed
537

Jeff Moyer's avatar
Jeff Moyer committed
538
	/* now link into global list. */
539
540
541
	spin_lock(&mm->ioctx_lock);
	hlist_add_head_rcu(&ctx->list, &mm->ioctx_list);
	spin_unlock(&mm->ioctx_lock);
Linus Torvalds's avatar
Linus Torvalds committed
542

Kent Overstreet's avatar
Kent Overstreet committed
543
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
544
		 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
545
546
547
	return ctx;

out_cleanup:
548
549
	err = -EAGAIN;
	aio_free_ring(ctx);
Kent Overstreet's avatar
Kent Overstreet committed
550
551
out_freepcpu:
	free_percpu(ctx->cpu);
Linus Torvalds's avatar
Linus Torvalds committed
552
out_freectx:
553
554
	if (ctx->aio_ring_file)
		fput(ctx->aio_ring_file);
Linus Torvalds's avatar
Linus Torvalds committed
555
	kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreet's avatar
Kent Overstreet committed
556
	pr_debug("error allocating ioctx %d\n", err);
557
	return ERR_PTR(err);
Linus Torvalds's avatar
Linus Torvalds committed
558
559
}

Kent Overstreet's avatar
Kent Overstreet committed
560
static void kill_ioctx_work(struct work_struct *work)
Linus Torvalds's avatar
Linus Torvalds committed
561
{
Kent Overstreet's avatar
Kent Overstreet committed
562
	struct kioctx *ctx = container_of(work, struct kioctx, rcu_work);
563

Kent Overstreet's avatar
Kent Overstreet committed
564
565
566
	wake_up_all(&ctx->wait);
	put_ioctx(ctx);
}
Kent Overstreet's avatar
Kent Overstreet committed
567

Kent Overstreet's avatar
Kent Overstreet committed
568
569
570
static void kill_ioctx_rcu(struct rcu_head *head)
{
	struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
Linus Torvalds's avatar
Linus Torvalds committed
571

Kent Overstreet's avatar
Kent Overstreet committed
572
573
574
	INIT_WORK(&ctx->rcu_work, kill_ioctx_work);
	schedule_work(&ctx->rcu_work);
}
Linus Torvalds's avatar
Linus Torvalds committed
575

Kent Overstreet's avatar
Kent Overstreet committed
576
577
578
579
580
581
582
583
584
/* 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.
 */
static void kill_ioctx(struct kioctx *ctx)
{
	if (!atomic_xchg(&ctx->dead, 1)) {
		hlist_del_rcu(&ctx->list);
585

Kent Overstreet's avatar
Kent Overstreet committed
586
		/*
587
588
589
590
591
		 * 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
592
		 */
593
594
595
596
597
598
599
600
601
602
		spin_lock(&aio_nr_lock);
		BUG_ON(aio_nr - ctx->max_reqs > aio_nr);
		aio_nr -= ctx->max_reqs;
		spin_unlock(&aio_nr_lock);

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

		/* Between hlist_del_rcu() and dropping the initial ref */
		call_rcu(&ctx->rcu_head, kill_ioctx_rcu);
Kent Overstreet's avatar
Kent Overstreet committed
603
	}
Linus Torvalds's avatar
Linus Torvalds committed
604
605
606
607
608
}

/* wait_on_sync_kiocb:
 *	Waits on the given sync kiocb to complete.
 */
609
ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
Linus Torvalds's avatar
Linus Torvalds committed
610
{
611
	while (atomic_read(&iocb->ki_users)) {
Linus Torvalds's avatar
Linus Torvalds committed
612
		set_current_state(TASK_UNINTERRUPTIBLE);
613
		if (!atomic_read(&iocb->ki_users))
Linus Torvalds's avatar
Linus Torvalds committed
614
			break;
615
		io_schedule();
Linus Torvalds's avatar
Linus Torvalds committed
616
617
618
619
	}
	__set_current_state(TASK_RUNNING);
	return iocb->ki_user_data;
}
620
EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds's avatar
Linus Torvalds committed
621

Kent Overstreet's avatar
Kent Overstreet committed
622
623
624
625
626
627
628
/*
 * 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
629
 */
630
void exit_aio(struct mm_struct *mm)
Linus Torvalds's avatar
Linus Torvalds committed
631
{
632
	struct kioctx *ctx;
Kent Overstreet's avatar
Kent Overstreet committed
633
	struct hlist_node *n;
634

Kent Overstreet's avatar
Kent Overstreet committed
635
	hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) {
Linus Torvalds's avatar
Linus Torvalds committed
636
637
638
		if (1 != atomic_read(&ctx->users))
			printk(KERN_DEBUG
				"exit_aio:ioctx still alive: %d %d %d\n",
Kent Overstreet's avatar
Kent Overstreet committed
639
640
				atomic_read(&ctx->users),
				atomic_read(&ctx->dead),
641
				atomic_read(&ctx->reqs_available));
642
643
644
645
646
647
648
649
		/*
		 * 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.
		 */
650
		ctx->mmap_size = 0;
Kent Overstreet's avatar
Kent Overstreet committed
651

652
		kill_ioctx(ctx);
Linus Torvalds's avatar
Linus Torvalds committed
653
654
655
	}
}

Kent Overstreet's avatar
Kent Overstreet committed
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
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
702
/* aio_get_req
703
 *	Allocate a slot for an aio request.  Increments the ki_users count
Linus Torvalds's avatar
Linus Torvalds committed
704
705
706
 * of the kioctx so that the kioctx stays around until all requests are
 * complete.  Returns NULL if no requests are free.
 *
707
 * Returns with kiocb->ki_users set to 2.  The io submit code path holds
Linus Torvalds's avatar
Linus Torvalds committed
708
709
710
711
 * an extra reference while submitting the i/o.
 * This prevents races between the aio code path referencing the
 * req (after submitting it) and aio_complete() freeing the req.
 */
Kent Overstreet's avatar
Kent Overstreet committed
712
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
713
{
Kent Overstreet's avatar
Kent Overstreet committed
714
715
	struct kiocb *req;

Kent Overstreet's avatar
Kent Overstreet committed
716
	if (!get_reqs_available(ctx))
Kent Overstreet's avatar
Kent Overstreet committed
717
718
		return NULL;

719
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds's avatar
Linus Torvalds committed
720
	if (unlikely(!req))
Kent Overstreet's avatar
Kent Overstreet committed
721
		goto out_put;
Linus Torvalds's avatar
Linus Torvalds committed
722

723
	atomic_set(&req->ki_users, 2);
Linus Torvalds's avatar
Linus Torvalds committed
724
	req->ki_ctx = ctx;
Jeff Moyer's avatar
Jeff Moyer committed
725
	return req;
Kent Overstreet's avatar
Kent Overstreet committed
726
out_put:
Kent Overstreet's avatar
Kent Overstreet committed
727
	put_reqs_available(ctx, 1);
Kent Overstreet's avatar
Kent Overstreet committed
728
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
729
730
}

731
static void kiocb_free(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
732
{
733
734
	if (req->ki_filp)
		fput(req->ki_filp);
735
736
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds's avatar
Linus Torvalds committed
737
738
	if (req->ki_dtor)
		req->ki_dtor(req);
739
740
	if (req->ki_iovec != &req->ki_inline_vec)
		kfree(req->ki_iovec);
Linus Torvalds's avatar
Linus Torvalds committed
741
742
743
	kmem_cache_free(kiocb_cachep, req);
}

744
void aio_put_req(struct kiocb *req)
Linus Torvalds's avatar
Linus Torvalds committed
745
{
746
747
	if (atomic_dec_and_test(&req->ki_users))
		kiocb_free(req);
Linus Torvalds's avatar
Linus Torvalds committed
748
}
749
EXPORT_SYMBOL(aio_put_req);
Linus Torvalds's avatar
Linus Torvalds committed
750

751
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds's avatar
Linus Torvalds committed
752
{
753
	struct mm_struct *mm = current->mm;
754
	struct kioctx *ctx, *ret = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
755

756
757
	rcu_read_lock();

758
	hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) {
Kent Overstreet's avatar
Kent Overstreet committed
759
760
		if (ctx->user_id == ctx_id) {
			atomic_inc(&ctx->users);
761
			ret = ctx;
Linus Torvalds's avatar
Linus Torvalds committed
762
763
			break;
		}
764
	}
Linus Torvalds's avatar
Linus Torvalds committed
765

766
	rcu_read_unlock();
767
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
768
769
770
771
772
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
773
void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds's avatar
Linus Torvalds committed
774
775
776
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
Kent Overstreet's avatar
Kent Overstreet committed
777
	struct io_event	*ev_page, *event;
Linus Torvalds's avatar
Linus Torvalds committed
778
	unsigned long	flags;
Kent Overstreet's avatar
Kent Overstreet committed
779
	unsigned tail, pos;
Linus Torvalds's avatar
Linus Torvalds committed
780

781
782
783
784
785
786
	/*
	 * 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
787
788
	 */
	if (is_sync_kiocb(iocb)) {
789
		BUG_ON(atomic_read(&iocb->ki_users) != 1);
Linus Torvalds's avatar
Linus Torvalds committed
790
		iocb->ki_user_data = res;
791
		atomic_set(&iocb->ki_users, 0);
Linus Torvalds's avatar
Linus Torvalds committed
792
		wake_up_process(iocb->ki_obj.tsk);
793
		return;
Linus Torvalds's avatar
Linus Torvalds committed
794
795
	}

Kent Overstreet's avatar
Kent Overstreet committed
796
797
	/*
	 * Take rcu_read_lock() in case the kioctx is being destroyed, as we
798
	 * need to issue a wakeup after incrementing reqs_available.
Linus Torvalds's avatar
Linus Torvalds committed
799
	 */
Kent Overstreet's avatar
Kent Overstreet committed
800
	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
801

802
803
804
805
806
807
808
	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);
	}
809

Linus Torvalds's avatar
Linus Torvalds committed
810
811
812
813
	/*
	 * cancelled requests don't get events, userland was given one
	 * when the event got cancelled.
	 */
814
	if (unlikely(xchg(&iocb->ki_cancel,
815
			  KIOCB_CANCELLED) == KIOCB_CANCELLED)) {
Kent Overstreet's avatar
Kent Overstreet committed
816
817
818
819
		/*
		 * Can't use the percpu reqs_available here - could race with
		 * free_ioctx()
		 */
820
		atomic_inc(&ctx->reqs_available);
821
		/* Still need the wake_up in case free_ioctx is waiting */
Linus Torvalds's avatar
Linus Torvalds committed
822
		goto put_rq;
823
	}
Linus Torvalds's avatar
Linus Torvalds committed
824

825
826
	/*
	 * Add a completion event to the ring buffer. Must be done holding
827
	 * ctx->completion_lock to prevent other code from messing with the tail
828
829
830
831
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

832
	tail = ctx->tail;
Kent Overstreet's avatar
Kent Overstreet committed
833
834
	pos = tail + AIO_EVENTS_OFFSET;

835
	if (++tail >= ctx->nr_events)
836
		tail = 0;
Linus Torvalds's avatar
Linus Torvalds committed
837

838
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
839
840
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

Linus Torvalds's avatar
Linus Torvalds committed
841
842
843
844
845
	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
846
	kunmap_atomic(ev_page);
847
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet's avatar
Kent Overstreet committed
848
849

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreet's avatar
Kent Overstreet committed
850
851
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
Linus Torvalds's avatar
Linus Torvalds committed
852
853
854
855
856
857

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

858
	ctx->tail = tail;
Linus Torvalds's avatar
Linus Torvalds committed
859

860
	ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreet's avatar
Kent Overstreet committed
861
	ring->tail = tail;
862
	kunmap_atomic(ring);
863
	flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds's avatar
Linus Torvalds committed
864

865
866
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

Kent Overstreet's avatar
Kent Overstreet committed
867
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi's avatar
Davide Libenzi committed
868
869
870
871
872
873

	/*
	 * 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.
	 */
874
	if (iocb->ki_eventfd != NULL)
Davide Libenzi's avatar
Davide Libenzi committed
875
876
		eventfd_signal(iocb->ki_eventfd, 1);

Linus Torvalds's avatar
Linus Torvalds committed
877
878
put_rq:
	/* everything turned out well, dispose of the aiocb. */
879
	aio_put_req(iocb);
Linus Torvalds's avatar
Linus Torvalds committed
880

881
882
883
884
885
886
887
888
	/*
	 * 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
889
890
891
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

Kent Overstreet's avatar
Kent Overstreet committed
892
	rcu_read_unlock();
Linus Torvalds's avatar
Linus Torvalds committed
893
}
894
EXPORT_SYMBOL(aio_complete);
Linus Torvalds's avatar
Linus Torvalds committed
895

896
897
898
/* 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
899
 */
900
901
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
Linus Torvalds's avatar
Linus Torvalds committed
902
903
{
	struct aio_ring *ring;
904
905
906
907
	unsigned head, pos;
	long ret = 0;
	int copy_ret;

908
	mutex_lock(&ctx->ring_lock);
Linus Torvalds's avatar
Linus Torvalds committed
909

910
	ring = kmap_atomic(ctx->ring_pages[0]);
911
912
913
	head = ring->head;
	kunmap_atomic(ring);

914
	pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events);
Linus Torvalds's avatar
Linus Torvalds committed
915

916
	if (head == ctx->tail)
Linus Torvalds's avatar
Linus Torvalds committed
917
918
		goto out;

919
920
921
922
923
	while (ret < nr) {
		long avail;
		struct io_event *ev;
		struct page *page;

924
925
		avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head;
		if (head == ctx->tail)
926
927
928
929
930
931
932
			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;
933
		page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
934
935
936
937
938
939
940
941
942
943
944
945
946
947
		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;
948
		head %= ctx->nr_events;
Linus Torvalds's avatar
Linus Torvalds committed
949
950
	}

951
	ring = kmap_atomic(ctx->ring_pages[0]);
952
	ring->head = head;
953
	kunmap_atomic(ring);
954
	flush_dcache_page(ctx->ring_pages[0]);
955

956
	pr_debug("%li  h%u t%u\n", ret, head, ctx->tail);
957

Kent Overstreet's avatar
Kent Overstreet committed
958
	put_reqs_available(ctx, ret);
959
out:
960
	mutex_unlock(&ctx->ring_lock);
961

Linus Torvalds's avatar
Linus Torvalds committed
962
963
964
	return ret;
}

965
966
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
967
{
968
	long ret = aio_read_events_ring(ctx, event + *i, nr - *i);
Linus Torvalds's avatar
Linus Torvalds committed
969

970
971
	if (ret > 0)
		*i += ret;
Linus Torvalds's avatar
Linus Torvalds committed
972

973
974
	if (unlikely(atomic_read(&ctx->dead)))
		ret = -EINVAL;
Linus Torvalds's avatar
Linus Torvalds committed
975

976
977
	if (!*i)
		*i = ret;
Linus Torvalds's avatar
Linus Torvalds committed
978

979
	return ret < 0 || *i >= min_nr;
Linus Torvalds's avatar
Linus Torvalds committed
980
981
}

982
static long read_events(struct kioctx *ctx, long min_nr, long nr,
Linus Torvalds's avatar
Linus Torvalds committed
983
984
985
			struct io_event __user *event,
			struct timespec __user *timeout)
{
986
987
	ktime_t until = { .tv64 = KTIME_MAX };
	long ret = 0;
Linus Torvalds's avatar
Linus Torvalds committed
988
989
990

	if (timeout) {
		struct timespec	ts;
991

Linus Torvalds's avatar
Linus Torvalds committed
992
		if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
993
			return -EFAULT;
Linus Torvalds's avatar
Linus Torvalds committed
994

995
		until = timespec_to_ktime(ts);
Linus Torvalds's avatar
Linus Torvalds committed
996
997
	}

998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
	/*
	 * 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
1014

1015
1016
	if (!ret && signal_pending(current))
		ret = -EINTR;
Linus Torvalds's avatar
Linus Torvalds committed
1017

1018
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
}

/* 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.
 */
1034
SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
Linus Torvalds's avatar
Linus Torvalds committed
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
{
	struct kioctx *ioctx = NULL;
	unsigned long ctx;
	long ret;

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

	ret = -EINVAL;
1045
1046
1047
	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
1048
1049
1050
1051
1052
1053
1054
		goto out;
	}

	ioctx = ioctx_alloc(nr_events);
	ret = PTR_ERR(ioctx);
	if (!IS_ERR(ioctx)) {
		ret = put_user(ioctx->user_id, ctxp);
1055
		if (ret)
Kent Overstreet's avatar
Kent Overstreet committed
1056
			kill_ioctx(ioctx);
1057
		put_ioctx(ioctx);
Linus Torvalds's avatar
Linus Torvalds committed
1058
1059
1060
1061
1062
1063
1064
1065
1066
	}

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
1067
 *	implemented.  May fail with -EINVAL if the context pointed to
Linus Torvalds's avatar
Linus Torvalds committed
1068
1069
 *	is invalid.
 */
1070
SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
Linus Torvalds's avatar
Linus Torvalds committed
1071
1072
1073
{
	struct kioctx *ioctx = lookup_ioctx(ctx);
	if (likely(NULL != ioctx)) {
Kent Overstreet's avatar
Kent Overstreet committed
1074
		kill_ioctx(ioctx);
1075
		put_ioctx(ioctx);
Linus Torvalds's avatar
Linus Torvalds committed
1076
1077
1078
1079
1080
1081
		return 0;
	}
	pr_debug("EINVAL: io_destroy: invalid context id\n");
	return -EINVAL;
}

1082
static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
Linus Torvalds's avatar
Linus Torvalds committed
1083
{
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
	struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];

	BUG_ON(ret <= 0);

	while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
		ssize_t this = min((ssize_t)iov->iov_len, ret);
		iov->iov_base += this;
		iov->iov_len -= this;
		iocb->ki_left -= this;
		ret -= this;
		if (iov->iov_len == 0) {
			iocb->ki_cur_seg++;
			iov++;
1097
		}
1098
	}
Linus Torvalds's avatar
Linus Torvalds committed
1099

1100
1101
1102
	/* the caller should not have done more io than what fit in
	 * the remaining iovecs */
	BUG_ON(ret > 0 && iocb->ki_left == 0);
Linus Torvalds's avatar
Linus Torvalds committed
1103
1104
}

Kent Overstreet's avatar
Kent Overstreet committed
1105
1106
1107
1108
typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *,
			    unsigned long, loff_t);

static ssize_t aio_rw_vect_retry(struct kiocb *iocb, int rw, aio_rw_op *rw_op)
Linus Torvalds's avatar
Linus Torvalds committed
1109
1110
{
	struct file *file = iocb->ki_filp;
1111
1112
	struct address_space *mapping = file->f_mapping;
	struct inode *inode = mapping->host;
Linus Torvalds's avatar
Linus Torvalds committed
1113
1114
	ssize_t ret = 0;

1115
1116
1117
1118
	/* This matches the pread()/pwrite() logic */
	if (iocb->ki_pos < 0)
		return -EINVAL;

Kent Overstreet's avatar
Kent Overstreet committed
1119
	if (rw == WRITE)
1120
		file_start_write(file);
1121
	do {
1122
1123
1124
1125
1126
1127
1128
1129
1130
		ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
			    iocb->ki_nr_segs - iocb->ki_cur_seg,
			    iocb->ki_pos);
		if (ret > 0)
			aio_advance_iovec(iocb, ret);

	/* retry all partial writes.  retry partial reads as long as its a
	 * regular file. */
	} while (ret > 0 && iocb->ki_left > 0 &&
Kent Overstreet's avatar
Kent Overstreet committed
1131
		 (rw == WRITE ||
1132
		  (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
Kent Overstreet's avatar
Kent Overstreet committed
1133
	if (rw == WRITE)
1134
		file_end_write(file);
Linus Torvalds's avatar
Linus Torvalds committed
1135

1136
1137
	/* This means we must have transferred all that we could */
	/* No need to retry anymore */
Linus Torvalds's avatar
Linus Torvalds committed
1138
1139
1140
	if ((ret == 0) || (iocb->ki_left == 0))
		ret = iocb->ki_nbytes - iocb->ki_left;

1141
1142
	/* If we managed to write some out we return that, rather than
	 * the eventual error. */
Kent Overstreet's avatar
Kent Overstreet committed
1143
	if (rw == WRITE
Zach Brown's avatar
Zach Brown committed
1144
	    && ret < 0 && ret != -EIOCBQUEUED
1145
1146
1147
	    && iocb->ki_nbytes - iocb->ki_left)
		ret = iocb->ki_nbytes - iocb->ki_left;

Linus Torvalds's avatar
Linus Torvalds committed
1148
1149
1150
	return ret;
}

Kent Overstreet's avatar
Kent Overstreet committed
1151
static ssize_t aio_setup_vectored_rw(int rw, struct kiocb *kiocb, bool compat)
1152
1153
1154
{
	ssize_t ret;

Kent Overstreet's avatar
Kent Overstreet committed
1155
1156
	kiocb->ki_nr_segs = kiocb->ki_nbytes;

1157
1158
#ifdef CONFIG_COMPAT
	if (compat)
Kent Overstreet's avatar
Kent Overstreet committed
1159
		ret = compat_rw_copy_check_uvector(rw,
1160
				(struct compat_iovec __user *)kiocb->ki_buf,
Kent Overstreet's avatar
Kent Overstreet committed
1161
				kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec,
1162
				&kiocb->ki_iovec);
1163
1164
	else
#endif
Kent Overstreet's avatar
Kent Overstreet committed
1165
		ret = rw_copy_check_uvector(rw,
1166
				(struct iovec __user *)kiocb->ki_buf,
Kent Overstreet's avatar
Kent Overstreet committed
1167
				kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec,
1168